drivers/md/persistent-data/dm-block-manager.c

   1 /*
   2  * Copyright (C) 2011 Red Hat, Inc.
   3  *
   4  * This file is released under the GPL.
   5  */
   6 #include "dm-block-manager.h"
   7 #include "dm-persistent-data-internal.h"
   8
   9 #include <linux/dm-bufio.h>
  10 #include <linux/crc32c.h>
  11 #include <linux/module.h>
  12 #include <linux/slab.h>
  13 #include <linux/rwsem.h>
  14 #include <linux/device-mapper.h>
  15 #include <linux/stacktrace.h>
  16 #include <linux/sched/task.h>
  17
  18 #define DM_MSG_PREFIX "block manager"
  19
  20 /*----------------------------------------------------------------*/
  21
  22 #ifdef CONFIG_DM_DEBUG_BLOCK_MANAGER_LOCKING
  23
  24 /*
  25  * This is a read/write semaphore with a couple of differences.
  26  *
  27  * i) There is a restriction on the number of concurrent read locks that
  28  * may be held at once.  This is just an implementation detail.
  29  *
  30  * ii) Recursive locking attempts are detected and return EINVAL.  A stack
  31  * trace is also emitted for the previous lock acquisition.
  32  *
  33  * iii) Priority is given to write locks.
  34  */
  35 #define MAX_HOLDERS 4
  36 #define MAX_STACK 10
  37
  38 struct stack_store {
  39         unsigned int    nr_entries;
  40         unsigned long   entries[MAX_STACK];
  41 };
  42
  43 struct block_lock {
  44         spinlock_t lock;
  45         __s32 count;
  46         struct list_head waiters;
  47         struct task_struct *holders[MAX_HOLDERS];
  48
  49 #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING
  50         struct stack_store traces[MAX_HOLDERS];
  51 #endif
  52 };
  53
  54 struct waiter {
  55         struct list_head list;
  56         struct task_struct *task;
  57         int wants_write;
  58 };
  59
  60 static unsigned __find_holder(struct block_lock *lock,
  61                               struct task_struct *task)
  62 {
  63         unsigned i;
  64
  65         for (i = 0; i < MAX_HOLDERS; i++)
  66                 if (lock->holders[i] == task)
  67                         break;
  68
  69         BUG_ON(i == MAX_HOLDERS);
  70         return i;
  71 }
  72
  73 /* call this *after* you increment lock->count */
  74 static void __add_holder(struct block_lock *lock, struct task_struct *task)
  75 {
  76         unsigned h = __find_holder(lock, NULL);
  77 #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING
  78         struct stack_store *t;
  79 #endif
  80
  81         get_task_struct(task);
  82         lock->holders[h] = task;
  83
  84 #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING
  85         t = lock->traces + h;
  86         t->nr_entries = stack_trace_save(t->entries, MAX_STACK, 2);
  87 #endif
  88 }
  89
  90 /* call this *before* you decrement lock->count */
  91 static void __del_holder(struct block_lock *lock, struct task_struct *task)
  92 {
  93         unsigned h = __find_holder(lock, task);
  94         lock->holders[h] = NULL;
  95         put_task_struct(task);
  96 }
  97
  98 static int __check_holder(struct block_lock *lock)
  99 {
 100         unsigned i;
 101
 102         for (i = 0; i < MAX_HOLDERS; i++) {
 103                 if (lock->holders[i] == current) {
 104                         DMERR("recursive lock detected in metadata");
 105 #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING
 106                         DMERR("previously held here:");
 107                         stack_trace_print(lock->traces[i].entries,
 108                                           lock->traces[i].nr_entries, 4);
 109
 110                         DMERR("subsequent acquisition attempted here:");
 111                         dump_stack();
 112 #endif
 113                         return -EINVAL;
 114                 }
 115         }
 116
 117         return 0;
 118 }
 119
 120 static void __wait(struct waiter *w)
 121 {
 122         for (;;) {
 123                 set_current_state(TASK_UNINTERRUPTIBLE);
 124
 125                 if (!w->task)
 126                         break;
 127
 128                 schedule();
 129         }
 130
 131         set_current_state(TASK_RUNNING);
 132 }
 133
 134 static void __wake_waiter(struct waiter *w)
 135 {
 136         struct task_struct *task;
 137
 138         list_del(&w->list);
 139         task = w->task;
 140         smp_mb();
 141         w->task = NULL;
 142         wake_up_process(task);
 143 }
 144
 145 /*
 146  * We either wake a few readers or a single writer.
 147  */
 148 static void __wake_many(struct block_lock *lock)
 149 {
 150         struct waiter *w, *tmp;
 151
 152         BUG_ON(lock->count < 0);
 153         list_for_each_entry_safe(w, tmp, &lock->waiters, list) {
 154                 if (lock->count >= MAX_HOLDERS)
 155                         return;
 156
 157                 if (w->wants_write) {
 158                         if (lock->count > 0)
 159                                 return; /* still read locked */
 160
 161                         lock->count = -1;
 162                         __add_holder(lock, w->task);
 163                         __wake_waiter(w);
 164                         return;
 165                 }
 166
 167                 lock->count++;
 168                 __add_holder(lock, w->task);
 169                 __wake_waiter(w);
 170         }
 171 }
 172
 173 static void bl_init(struct block_lock *lock)
 174 {
 175         int i;
 176
 177         spin_lock_init(&lock->lock);
 178         lock->count = 0;
 179         INIT_LIST_HEAD(&lock->waiters);
 180         for (i = 0; i < MAX_HOLDERS; i++)
 181                 lock->holders[i] = NULL;
 182 }
 183
 184 static int __available_for_read(struct block_lock *lock)
 185 {
 186         return lock->count >= 0 &&
 187                 lock->count < MAX_HOLDERS &&
 188                 list_empty(&lock->waiters);
 189 }
 190
 191 static int bl_down_read(struct block_lock *lock)
 192 {
 193         int r;
 194         struct waiter w;
 195
 196         spin_lock(&lock->lock);
 197         r = __check_holder(lock);
 198         if (r) {
 199                 spin_unlock(&lock->lock);
 200                 return r;
 201         }
 202
 203         if (__available_for_read(lock)) {
 204                 lock->count++;
 205                 __add_holder(lock, current);
 206                 spin_unlock(&lock->lock);
 207                 return 0;
 208         }
 209
 210         get_task_struct(current);
 211
 212         w.task = current;
 213         w.wants_write = 0;
 214         list_add_tail(&w.list, &lock->waiters);
 215         spin_unlock(&lock->lock);
 216
 217         __wait(&w);
 218         put_task_struct(current);
 219         return 0;
 220 }
 221
 222 static int bl_down_read_nonblock(struct block_lock *lock)
 223 {
 224         int r;
 225
 226         spin_lock(&lock->lock);
 227         r = __check_holder(lock);
 228         if (r)
 229                 goto out;
 230
 231         if (__available_for_read(lock)) {
 232                 lock->count++;
 233                 __add_holder(lock, current);
 234                 r = 0;
 235         } else
 236                 r = -EWOULDBLOCK;
 237
 238 out:
 239         spin_unlock(&lock->lock);
 240         return r;
 241 }
 242
 243 static void bl_up_read(struct block_lock *lock)
 244 {
 245         spin_lock(&lock->lock);
 246         BUG_ON(lock->count <= 0);
 247         __del_holder(lock, current);
 248         --lock->count;
 249         if (!list_empty(&lock->waiters))
 250                 __wake_many(lock);
 251         spin_unlock(&lock->lock);
 252 }
 253
 254 static int bl_down_write(struct block_lock *lock)
 255 {
 256         int r;
 257         struct waiter w;
 258
 259         spin_lock(&lock->lock);
 260         r = __check_holder(lock);
 261         if (r) {
 262                 spin_unlock(&lock->lock);
 263                 return r;
 264         }
 265
 266         if (lock->count == 0 && list_empty(&lock->waiters)) {
 267                 lock->count = -1;
 268                 __add_holder(lock, current);
 269                 spin_unlock(&lock->lock);
 270                 return 0;
 271         }
 272
 273         get_task_struct(current);
 274         w.task = current;
 275         w.wants_write = 1;
 276
 277         /*
 278          * Writers given priority. We know there's only one mutator in the
 279          * system, so ignoring the ordering reversal.
 280          */
 281         list_add(&w.list, &lock->waiters);
 282         spin_unlock(&lock->lock);
 283
 284         __wait(&w);
 285         put_task_struct(current);
 286
 287         return 0;
 288 }
 289
 290 static void bl_up_write(struct block_lock *lock)
 291 {
 292         spin_lock(&lock->lock);
 293         __del_holder(lock, current);
 294         lock->count = 0;
 295         if (!list_empty(&lock->waiters))
 296                 __wake_many(lock);
 297         spin_unlock(&lock->lock);
 298 }
 299
 300 static void report_recursive_bug(dm_block_t b, int r)
 301 {
 302         if (r == -EINVAL)
 303                 DMERR("recursive acquisition of block %llu requested.",
 304                       (unsigned long long) b);
 305 }
 306
 307 #else  /* !CONFIG_DM_DEBUG_BLOCK_MANAGER_LOCKING */
 308
 309 #define bl_init(x) do { } while (0)
 310 #define bl_down_read(x) 0
 311 #define bl_down_read_nonblock(x) 0
 312 #define bl_up_read(x) do { } while (0)
 313 #define bl_down_write(x) 0
 314 #define bl_up_write(x) do { } while (0)
 315 #define report_recursive_bug(x, y) do { } while (0)
 316
 317 #endif /* CONFIG_DM_DEBUG_BLOCK_MANAGER_LOCKING */
 318
 319 /*----------------------------------------------------------------*/
 320
 321 /*
 322  * Block manager is currently implemented using dm-bufio.  struct
 323  * dm_block_manager and struct dm_block map directly onto a couple of
 324  * structs in the bufio interface.  I want to retain the freedom to move
 325  * away from bufio in the future.  So these structs are just cast within
 326  * this .c file, rather than making it through to the public interface.
 327  */
 328 static struct dm_buffer *to_buffer(struct dm_block *b)
 329 {
 330         return (struct dm_buffer *) b;
 331 }
 332
 333 dm_block_t dm_block_location(struct dm_block *b)
 334 {
 335         return dm_bufio_get_block_number(to_buffer(b));
 336 }
 337 EXPORT_SYMBOL_GPL(dm_block_location);
 338
 339 void *dm_block_data(struct dm_block *b)
 340 {
 341         return dm_bufio_get_block_data(to_buffer(b));
 342 }
 343 EXPORT_SYMBOL_GPL(dm_block_data);
 344
 345 struct buffer_aux {
 346         struct dm_block_validator *validator;
 347         int write_locked;
 348
 349 #ifdef CONFIG_DM_DEBUG_BLOCK_MANAGER_LOCKING
 350         struct block_lock lock;
 351 #endif
 352 };
 353
 354 static void dm_block_manager_alloc_callback(struct dm_buffer *buf)
 355 {
 356         struct buffer_aux *aux = dm_bufio_get_aux_data(buf);
 357         aux->validator = NULL;
 358         bl_init(&aux->lock);
 359 }
 360
 361 static void dm_block_manager_write_callback(struct dm_buffer *buf)
 362 {
 363         struct buffer_aux *aux = dm_bufio_get_aux_data(buf);
 364         if (aux->validator) {
 365                 aux->validator->prepare_for_write(aux->validator, (struct dm_block *) buf,
 366                          dm_bufio_get_block_size(dm_bufio_get_client(buf)));
 367         }
 368 }
 369
 370 /*----------------------------------------------------------------
 371  * Public interface
 372  *--------------------------------------------------------------*/
 373 struct dm_block_manager {
 374         struct dm_bufio_client *bufio;
 375         bool read_only:1;
 376 };
 377
 378 struct dm_block_manager *dm_block_manager_create(struct block_device *bdev,
 379                                                  unsigned block_size,
 380                                                  unsigned max_held_per_thread)
 381 {
 382         int r;
 383         struct dm_block_manager *bm;
 384
 385         bm = kmalloc(sizeof(*bm), GFP_KERNEL);
 386         if (!bm) {
 387                 r = -ENOMEM;
 388                 goto bad;
 389         }
 390
 391         bm->bufio = dm_bufio_client_create(bdev, block_size, max_held_per_thread,
 392                                            sizeof(struct buffer_aux),
 393                                            dm_block_manager_alloc_callback,
 394                                            dm_block_manager_write_callback);
 395         if (IS_ERR(bm->bufio)) {
 396                 r = PTR_ERR(bm->bufio);
 397                 kfree(bm);
 398                 goto bad;
 399         }
 400
 401         bm->read_only = false;
 402
 403         return bm;
 404
 405 bad:
 406         return ERR_PTR(r);
 407 }
 408 EXPORT_SYMBOL_GPL(dm_block_manager_create);
 409
 410 void dm_block_manager_destroy(struct dm_block_manager *bm)
 411 {
 412         dm_bufio_client_destroy(bm->bufio);
 413         kfree(bm);
 414 }
 415 EXPORT_SYMBOL_GPL(dm_block_manager_destroy);
 416
 417 unsigned dm_bm_block_size(struct dm_block_manager *bm)
 418 {
 419         return dm_bufio_get_block_size(bm->bufio);
 420 }
 421 EXPORT_SYMBOL_GPL(dm_bm_block_size);
 422
 423 dm_block_t dm_bm_nr_blocks(struct dm_block_manager *bm)
 424 {
 425         return dm_bufio_get_device_size(bm->bufio);
 426 }
 427
 428 static int dm_bm_validate_buffer(struct dm_block_manager *bm,
 429                                  struct dm_buffer *buf,
 430                                  struct buffer_aux *aux,
 431                                  struct dm_block_validator *v)
 432 {
 433         if (unlikely(!aux->validator)) {
 434                 int r;
 435                 if (!v)
 436                         return 0;
 437                 r = v->check(v, (struct dm_block *) buf, dm_bufio_get_block_size(bm->bufio));
 438                 if (unlikely(r)) {
 439                         DMERR_LIMIT("%s validator check failed for block %llu", v->name,
 440                                     (unsigned long long) dm_bufio_get_block_number(buf));
 441                         return r;
 442                 }
 443                 aux->validator = v;
 444         } else {
 445                 if (unlikely(aux->validator != v)) {
 446                         DMERR_LIMIT("validator mismatch (old=%s vs new=%s) for block %llu",
 447                                     aux->validator->name, v ? v->name : "NULL",
 448                                     (unsigned long long) dm_bufio_get_block_number(buf));
 449                         return -EINVAL;
 450                 }
 451         }
 452
 453         return 0;
 454 }
 455 int dm_bm_read_lock(struct dm_block_manager *bm, dm_block_t b,
 456                     struct dm_block_validator *v,
 457                     struct dm_block **result)
 458 {
 459         struct buffer_aux *aux;
 460         void *p;
 461         int r;
 462
 463         p = dm_bufio_read(bm->bufio, b, (struct dm_buffer **) result);
 464         if (IS_ERR(p))
 465                 return PTR_ERR(p);
 466
 467         aux = dm_bufio_get_aux_data(to_buffer(*result));
 468         r = bl_down_read(&aux->lock);
 469         if (unlikely(r)) {
 470                 dm_bufio_release(to_buffer(*result));
 471                 report_recursive_bug(b, r);
 472                 return r;
 473         }
 474
 475         aux->write_locked = 0;
 476
 477         r = dm_bm_validate_buffer(bm, to_buffer(*result), aux, v);
 478         if (unlikely(r)) {
 479                 bl_up_read(&aux->lock);
 480                 dm_bufio_release(to_buffer(*result));
 481                 return r;
 482         }
 483
 484         return 0;
 485 }
 486 EXPORT_SYMBOL_GPL(dm_bm_read_lock);
 487
 488 int dm_bm_write_lock(struct dm_block_manager *bm,
 489                      dm_block_t b, struct dm_block_validator *v,
 490                      struct dm_block **result)
 491 {
 492         struct buffer_aux *aux;
 493         void *p;
 494         int r;
 495
 496         if (bm->read_only)
 497                 return -EPERM;
 498
 499         p = dm_bufio_read(bm->bufio, b, (struct dm_buffer **) result);
 500         if (IS_ERR(p))
 501                 return PTR_ERR(p);
 502
 503         aux = dm_bufio_get_aux_data(to_buffer(*result));
 504         r = bl_down_write(&aux->lock);
 505         if (r) {
 506                 dm_bufio_release(to_buffer(*result));
 507                 report_recursive_bug(b, r);
 508                 return r;
 509         }
 510
 511         aux->write_locked = 1;
 512
 513         r = dm_bm_validate_buffer(bm, to_buffer(*result), aux, v);
 514         if (unlikely(r)) {
 515                 bl_up_write(&aux->lock);
 516                 dm_bufio_release(to_buffer(*result));
 517                 return r;
 518         }
 519
 520         return 0;
 521 }
 522 EXPORT_SYMBOL_GPL(dm_bm_write_lock);
 523
 524 int dm_bm_read_try_lock(struct dm_block_manager *bm,
 525                         dm_block_t b, struct dm_block_validator *v,
 526                         struct dm_block **result)
 527 {
 528         struct buffer_aux *aux;
 529         void *p;
 530         int r;
 531
 532         p = dm_bufio_get(bm->bufio, b, (struct dm_buffer **) result);
 533         if (IS_ERR(p))
 534                 return PTR_ERR(p);
 535         if (unlikely(!p))
 536                 return -EWOULDBLOCK;
 537
 538         aux = dm_bufio_get_aux_data(to_buffer(*result));
 539         r = bl_down_read_nonblock(&aux->lock);
 540         if (r < 0) {
 541                 dm_bufio_release(to_buffer(*result));
 542                 report_recursive_bug(b, r);
 543                 return r;
 544         }
 545         aux->write_locked = 0;
 546
 547         r = dm_bm_validate_buffer(bm, to_buffer(*result), aux, v);
 548         if (unlikely(r)) {
 549                 bl_up_read(&aux->lock);
 550                 dm_bufio_release(to_buffer(*result));
 551                 return r;
 552         }
 553
 554         return 0;
 555 }
 556
 557 int dm_bm_write_lock_zero(struct dm_block_manager *bm,
 558                           dm_block_t b, struct dm_block_validator *v,
 559                           struct dm_block **result)
 560 {
 561         int r;
 562         struct buffer_aux *aux;
 563         void *p;
 564
 565         if (bm->read_only)
 566                 return -EPERM;
 567
 568         p = dm_bufio_new(bm->bufio, b, (struct dm_buffer **) result);
 569         if (IS_ERR(p))
 570                 return PTR_ERR(p);
 571
 572         memset(p, 0, dm_bm_block_size(bm));
 573
 574         aux = dm_bufio_get_aux_data(to_buffer(*result));
 575         r = bl_down_write(&aux->lock);
 576         if (r) {
 577                 dm_bufio_release(to_buffer(*result));
 578                 return r;
 579         }
 580
 581         aux->write_locked = 1;
 582         aux->validator = v;
 583
 584         return 0;
 585 }
 586 EXPORT_SYMBOL_GPL(dm_bm_write_lock_zero);
 587
 588 void dm_bm_unlock(struct dm_block *b)
 589 {
 590         struct buffer_aux *aux;
 591         aux = dm_bufio_get_aux_data(to_buffer(b));
 592
 593         if (aux->write_locked) {
 594                 dm_bufio_mark_buffer_dirty(to_buffer(b));
 595                 bl_up_write(&aux->lock);
 596         } else
 597                 bl_up_read(&aux->lock);
 598
 599         dm_bufio_release(to_buffer(b));
 600 }
 601 EXPORT_SYMBOL_GPL(dm_bm_unlock);
 602
 603 int dm_bm_flush(struct dm_block_manager *bm)
 604 {
 605         if (bm->read_only)
 606                 return -EPERM;
 607
 608         return dm_bufio_write_dirty_buffers(bm->bufio);
 609 }
 610 EXPORT_SYMBOL_GPL(dm_bm_flush);
 611
 612 void dm_bm_prefetch(struct dm_block_manager *bm, dm_block_t b)
 613 {
 614         dm_bufio_prefetch(bm->bufio, b, 1);
 615 }
 616
 617 bool dm_bm_is_read_only(struct dm_block_manager *bm)
 618 {
 619         return bm->read_only;
 620 }
 621 EXPORT_SYMBOL_GPL(dm_bm_is_read_only);
 622
 623 void dm_bm_set_read_only(struct dm_block_manager *bm)
 624 {
 625         bm->read_only = true;
 626 }
 627 EXPORT_SYMBOL_GPL(dm_bm_set_read_only);
 628
 629 void dm_bm_set_read_write(struct dm_block_manager *bm)
 630 {
 631         bm->read_only = false;
 632 }
 633 EXPORT_SYMBOL_GPL(dm_bm_set_read_write);
 634
 635 u32 dm_bm_checksum(const void *data, size_t len, u32 init_xor)
 636 {
 637         return crc32c(~(u32) 0, data, len) ^ init_xor;
 638 }
 639 EXPORT_SYMBOL_GPL(dm_bm_checksum);
 640
 641 /*----------------------------------------------------------------*/
 642
 643 MODULE_LICENSE("GPL");
 644 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
 645 MODULE_DESCRIPTION("Immutable metadata library for dm");
 646
 647 /*----------------------------------------------------------------*/