drivers/md/dm-kcopyd.c

   1 /*
   2  * Copyright (C) 2002 Sistina Software (UK) Limited.
   3  * Copyright (C) 2006 Red Hat GmbH
   4  *
   5  * This file is released under the GPL.
   6  *
   7  * Kcopyd provides a simple interface for copying an area of one
   8  * block-device to one or more other block-devices, with an asynchronous
   9  * completion notification.
  10  */
  11
  12 #include <linux/types.h>
  13 #include <asm/atomic.h>
  14 #include <linux/blkdev.h>
  15 #include <linux/fs.h>
  16 #include <linux/init.h>
  17 #include <linux/list.h>
  18 #include <linux/mempool.h>
  19 #include <linux/module.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/slab.h>
  22 #include <linux/vmalloc.h>
  23 #include <linux/workqueue.h>
  24 #include <linux/mutex.h>
  25 #include <linux/dm-kcopyd.h>
  26
  27 #include "dm.h"
  28
  29 /*-----------------------------------------------------------------
  30  * Each kcopyd client has its own little pool of preallocated
  31  * pages for kcopyd io.
  32  *---------------------------------------------------------------*/
  33 struct dm_kcopyd_client {
  34         spinlock_t lock;
  35         struct page_list *pages;
  36         unsigned int nr_pages;
  37         unsigned int nr_free_pages;
  38
  39         struct dm_io_client *io_client;
  40
  41         wait_queue_head_t destroyq;
  42         atomic_t nr_jobs;
  43
  44         mempool_t *job_pool;
  45
  46         struct workqueue_struct *kcopyd_wq;
  47         struct work_struct kcopyd_work;
  48
  49 /*
  50  * We maintain three lists of jobs:
  51  *
  52  * i)   jobs waiting for pages
  53  * ii)  jobs that have pages, and are waiting for the io to be issued.
  54  * iii) jobs that have completed.
  55  *
  56  * All three of these are protected by job_lock.
  57  */
  58         spinlock_t job_lock;
  59         struct list_head complete_jobs;
  60         struct list_head io_jobs;
  61         struct list_head pages_jobs;
  62 };
  63
  64 static void wake(struct dm_kcopyd_client *kc)
  65 {
  66         queue_work(kc->kcopyd_wq, &kc->kcopyd_work);
  67 }
  68
  69 static struct page_list *alloc_pl(void)
  70 {
  71         struct page_list *pl;
  72
  73         pl = kmalloc(sizeof(*pl), GFP_KERNEL);
  74         if (!pl)
  75                 return NULL;
  76
  77         pl->page = alloc_page(GFP_KERNEL);
  78         if (!pl->page) {
  79                 kfree(pl);
  80                 return NULL;
  81         }
  82
  83         return pl;
  84 }
  85
  86 static void free_pl(struct page_list *pl)
  87 {
  88         __free_page(pl->page);
  89         kfree(pl);
  90 }
  91
  92 static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
  93                             unsigned int nr, struct page_list **pages)
  94 {
  95         struct page_list *pl;
  96
  97         spin_lock(&kc->lock);
  98         if (kc->nr_free_pages < nr) {
  99                 spin_unlock(&kc->lock);
 100                 return -ENOMEM;
 101         }
 102
 103         kc->nr_free_pages -= nr;
 104         for (*pages = pl = kc->pages; --nr; pl = pl->next)
 105                 ;
 106
 107         kc->pages = pl->next;
 108         pl->next = NULL;
 109
 110         spin_unlock(&kc->lock);
 111
 112         return 0;
 113 }
 114
 115 static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
 116 {
 117         struct page_list *cursor;
 118
 119         spin_lock(&kc->lock);
 120         for (cursor = pl; cursor->next; cursor = cursor->next)
 121                 kc->nr_free_pages++;
 122
 123         kc->nr_free_pages++;
 124         cursor->next = kc->pages;
 125         kc->pages = pl;
 126         spin_unlock(&kc->lock);
 127 }
 128
 129 /*
 130  * These three functions resize the page pool.
 131  */
 132 static void drop_pages(struct page_list *pl)
 133 {
 134         struct page_list *next;
 135
 136         while (pl) {
 137                 next = pl->next;
 138                 free_pl(pl);
 139                 pl = next;
 140         }
 141 }
 142
 143 static int client_alloc_pages(struct dm_kcopyd_client *kc, unsigned int nr)
 144 {
 145         unsigned int i;
 146         struct page_list *pl = NULL, *next;
 147
 148         for (i = 0; i < nr; i++) {
 149                 next = alloc_pl();
 150                 if (!next) {
 151                         if (pl)
 152                                 drop_pages(pl);
 153                         return -ENOMEM;
 154                 }
 155                 next->next = pl;
 156                 pl = next;
 157         }
 158
 159         kcopyd_put_pages(kc, pl);
 160         kc->nr_pages += nr;
 161         return 0;
 162 }
 163
 164 static void client_free_pages(struct dm_kcopyd_client *kc)
 165 {
 166         BUG_ON(kc->nr_free_pages != kc->nr_pages);
 167         drop_pages(kc->pages);
 168         kc->pages = NULL;
 169         kc->nr_free_pages = kc->nr_pages = 0;
 170 }
 171
 172 /*-----------------------------------------------------------------
 173  * kcopyd_jobs need to be allocated by the *clients* of kcopyd,
 174  * for this reason we use a mempool to prevent the client from
 175  * ever having to do io (which could cause a deadlock).
 176  *---------------------------------------------------------------*/
 177 struct kcopyd_job {
 178         struct dm_kcopyd_client *kc;
 179         struct list_head list;
 180         unsigned long flags;
 181
 182         /*
 183          * Error state of the job.
 184          */
 185         int read_err;
 186         unsigned long write_err;
 187
 188         /*
 189          * Either READ or WRITE
 190          */
 191         int rw;
 192         struct dm_io_region source;
 193
 194         /*
 195          * The destinations for the transfer.
 196          */
 197         unsigned int num_dests;
 198         struct dm_io_region dests[DM_KCOPYD_MAX_REGIONS];
 199
 200         sector_t offset;
 201         unsigned int nr_pages;
 202         struct page_list *pages;
 203
 204         /*
 205          * Set this to ensure you are notified when the job has
 206          * completed.  'context' is for callback to use.
 207          */
 208         dm_kcopyd_notify_fn fn;
 209         void *context;
 210
 211         /*
 212          * These fields are only used if the job has been split
 213          * into more manageable parts.
 214          */
 215         struct mutex lock;
 216         atomic_t sub_jobs;
 217         sector_t progress;
 218 };
 219
 220 /* FIXME: this should scale with the number of pages */
 221 #define MIN_JOBS 512
 222
 223 static struct kmem_cache *_job_cache;
 224
 225 int __init dm_kcopyd_init(void)
 226 {
 227         _job_cache = KMEM_CACHE(kcopyd_job, 0);
 228         if (!_job_cache)
 229                 return -ENOMEM;
 230
 231         return 0;
 232 }
 233
 234 void dm_kcopyd_exit(void)
 235 {
 236         kmem_cache_destroy(_job_cache);
 237         _job_cache = NULL;
 238 }
 239
 240 /*
 241  * Functions to push and pop a job onto the head of a given job
 242  * list.
 243  */
 244 static struct kcopyd_job *pop(struct list_head *jobs,
 245                               struct dm_kcopyd_client *kc)
 246 {
 247         struct kcopyd_job *job = NULL;
 248         unsigned long flags;
 249
 250         spin_lock_irqsave(&kc->job_lock, flags);
 251
 252         if (!list_empty(jobs)) {
 253                 job = list_entry(jobs->next, struct kcopyd_job, list);
 254                 list_del(&job->list);
 255         }
 256         spin_unlock_irqrestore(&kc->job_lock, flags);
 257
 258         return job;
 259 }
 260
 261 static void push(struct list_head *jobs, struct kcopyd_job *job)
 262 {
 263         unsigned long flags;
 264         struct dm_kcopyd_client *kc = job->kc;
 265
 266         spin_lock_irqsave(&kc->job_lock, flags);
 267         list_add_tail(&job->list, jobs);
 268         spin_unlock_irqrestore(&kc->job_lock, flags);
 269 }
 270
 271
 272 static void push_head(struct list_head *jobs, struct kcopyd_job *job)
 273 {
 274         unsigned long flags;
 275         struct dm_kcopyd_client *kc = job->kc;
 276
 277         spin_lock_irqsave(&kc->job_lock, flags);
 278         list_add(&job->list, jobs);
 279         spin_unlock_irqrestore(&kc->job_lock, flags);
 280 }
 281
 282 /*
 283  * These three functions process 1 item from the corresponding
 284  * job list.
 285  *
 286  * They return:
 287  * < 0: error
 288  *   0: success
 289  * > 0: can't process yet.
 290  */
 291 static int run_complete_job(struct kcopyd_job *job)
 292 {
 293         void *context = job->context;
 294         int read_err = job->read_err;
 295         unsigned long write_err = job->write_err;
 296         dm_kcopyd_notify_fn fn = job->fn;
 297         struct dm_kcopyd_client *kc = job->kc;
 298
 299         kcopyd_put_pages(kc, job->pages);
 300         mempool_free(job, kc->job_pool);
 301         fn(read_err, write_err, context);
 302
 303         if (atomic_dec_and_test(&kc->nr_jobs))
 304                 wake_up(&kc->destroyq);
 305
 306         return 0;
 307 }
 308
 309 static void complete_io(unsigned long error, void *context)
 310 {
 311         struct kcopyd_job *job = (struct kcopyd_job *) context;
 312         struct dm_kcopyd_client *kc = job->kc;
 313
 314         if (error) {
 315                 if (job->rw == WRITE)
 316                         job->write_err |= error;
 317                 else
 318                         job->read_err = 1;
 319
 320                 if (!test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
 321                         push(&kc->complete_jobs, job);
 322                         wake(kc);
 323                         return;
 324                 }
 325         }
 326
 327         if (job->rw == WRITE)
 328                 push(&kc->complete_jobs, job);
 329
 330         else {
 331                 job->rw = WRITE;
 332                 push(&kc->io_jobs, job);
 333         }
 334
 335         wake(kc);
 336 }
 337
 338 /*
 339  * Request io on as many buffer heads as we can currently get for
 340  * a particular job.
 341  */
 342 static int run_io_job(struct kcopyd_job *job)
 343 {
 344         int r;
 345         struct dm_io_request io_req = {
 346                 .bi_rw = job->rw | (1 << BIO_RW_SYNC),
 347                 .mem.type = DM_IO_PAGE_LIST,
 348                 .mem.ptr.pl = job->pages,
 349                 .mem.offset = job->offset,
 350                 .notify.fn = complete_io,
 351                 .notify.context = job,
 352                 .client = job->kc->io_client,
 353         };
 354
 355         if (job->rw == READ)
 356                 r = dm_io(&io_req, 1, &job->source, NULL);
 357         else
 358                 r = dm_io(&io_req, job->num_dests, job->dests, NULL);
 359
 360         return r;
 361 }
 362
 363 static int run_pages_job(struct kcopyd_job *job)
 364 {
 365         int r;
 366
 367         job->nr_pages = dm_div_up(job->dests[0].count + job->offset,
 368                                   PAGE_SIZE >> 9);
 369         r = kcopyd_get_pages(job->kc, job->nr_pages, &job->pages);
 370         if (!r) {
 371                 /* this job is ready for io */
 372                 push(&job->kc->io_jobs, job);
 373                 return 0;
 374         }
 375
 376         if (r == -ENOMEM)
 377                 /* can't complete now */
 378                 return 1;
 379
 380         return r;
 381 }
 382
 383 /*
 384  * Run through a list for as long as possible.  Returns the count
 385  * of successful jobs.
 386  */
 387 static int process_jobs(struct list_head *jobs, struct dm_kcopyd_client *kc,
 388                         int (*fn) (struct kcopyd_job *))
 389 {
 390         struct kcopyd_job *job;
 391         int r, count = 0;
 392
 393         while ((job = pop(jobs, kc))) {
 394
 395                 r = fn(job);
 396
 397                 if (r < 0) {
 398                         /* error this rogue job */
 399                         if (job->rw == WRITE)
 400                                 job->write_err = (unsigned long) -1L;
 401                         else
 402                                 job->read_err = 1;
 403                         push(&kc->complete_jobs, job);
 404                         break;
 405                 }
 406
 407                 if (r > 0) {
 408                         /*
 409                          * We couldn't service this job ATM, so
 410                          * push this job back onto the list.
 411                          */
 412                         push_head(jobs, job);
 413                         break;
 414                 }
 415
 416                 count++;
 417         }
 418
 419         return count;
 420 }
 421
 422 /*
 423  * kcopyd does this every time it's woken up.
 424  */
 425 static void do_work(struct work_struct *work)
 426 {
 427         struct dm_kcopyd_client *kc = container_of(work,
 428                                         struct dm_kcopyd_client, kcopyd_work);
 429
 430         /*
 431          * The order that these are called is *very* important.
 432          * complete jobs can free some pages for pages jobs.
 433          * Pages jobs when successful will jump onto the io jobs
 434          * list.  io jobs call wake when they complete and it all
 435          * starts again.
 436          */
 437         process_jobs(&kc->complete_jobs, kc, run_complete_job);
 438         process_jobs(&kc->pages_jobs, kc, run_pages_job);
 439         process_jobs(&kc->io_jobs, kc, run_io_job);
 440 }
 441
 442 /*
 443  * If we are copying a small region we just dispatch a single job
 444  * to do the copy, otherwise the io has to be split up into many
 445  * jobs.
 446  */
 447 static void dispatch_job(struct kcopyd_job *job)
 448 {
 449         struct dm_kcopyd_client *kc = job->kc;
 450         atomic_inc(&kc->nr_jobs);
 451         push(&kc->pages_jobs, job);
 452         wake(kc);
 453 }
 454
 455 #define SUB_JOB_SIZE 128
 456 static void segment_complete(int read_err, unsigned long write_err,
 457                              void *context)
 458 {
 459         /* FIXME: tidy this function */
 460         sector_t progress = 0;
 461         sector_t count = 0;
 462         struct kcopyd_job *job = (struct kcopyd_job *) context;
 463
 464         mutex_lock(&job->lock);
 465
 466         /* update the error */
 467         if (read_err)
 468                 job->read_err = 1;
 469
 470         if (write_err)
 471                 job->write_err |= write_err;
 472
 473         /*
 474          * Only dispatch more work if there hasn't been an error.
 475          */
 476         if ((!job->read_err && !job->write_err) ||
 477             test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
 478                 /* get the next chunk of work */
 479                 progress = job->progress;
 480                 count = job->source.count - progress;
 481                 if (count) {
 482                         if (count > SUB_JOB_SIZE)
 483                                 count = SUB_JOB_SIZE;
 484
 485                         job->progress += count;
 486                 }
 487         }
 488         mutex_unlock(&job->lock);
 489
 490         if (count) {
 491                 int i;
 492                 struct kcopyd_job *sub_job = mempool_alloc(job->kc->job_pool,
 493                                                            GFP_NOIO);
 494
 495                 *sub_job = *job;
 496                 sub_job->source.sector += progress;
 497                 sub_job->source.count = count;
 498
 499                 for (i = 0; i < job->num_dests; i++) {
 500                         sub_job->dests[i].sector += progress;
 501                         sub_job->dests[i].count = count;
 502                 }
 503
 504                 sub_job->fn = segment_complete;
 505                 sub_job->context = job;
 506                 dispatch_job(sub_job);
 507
 508         } else if (atomic_dec_and_test(&job->sub_jobs)) {
 509
 510                 /*
 511                  * To avoid a race we must keep the job around
 512                  * until after the notify function has completed.
 513                  * Otherwise the client may try and stop the job
 514                  * after we've completed.
 515                  */
 516                 job->fn(read_err, write_err, job->context);
 517                 mempool_free(job, job->kc->job_pool);
 518         }
 519 }
 520
 521 /*
 522  * Create some little jobs that will do the move between
 523  * them.
 524  */
 525 #define SPLIT_COUNT 8
 526 static void split_job(struct kcopyd_job *job)
 527 {
 528         int i;
 529
 530         atomic_set(&job->sub_jobs, SPLIT_COUNT);
 531         for (i = 0; i < SPLIT_COUNT; i++)
 532                 segment_complete(0, 0u, job);
 533 }
 534
 535 int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
 536                    unsigned int num_dests, struct dm_io_region *dests,
 537                    unsigned int flags, dm_kcopyd_notify_fn fn, void *context)
 538 {
 539         struct kcopyd_job *job;
 540
 541         /*
 542          * Allocate a new job.
 543          */
 544         job = mempool_alloc(kc->job_pool, GFP_NOIO);
 545
 546         /*
 547          * set up for the read.
 548          */
 549         job->kc = kc;
 550         job->flags = flags;
 551         job->read_err = 0;
 552         job->write_err = 0;
 553         job->rw = READ;
 554
 555         job->source = *from;
 556
 557         job->num_dests = num_dests;
 558         memcpy(&job->dests, dests, sizeof(*dests) * num_dests);
 559
 560         job->offset = 0;
 561         job->nr_pages = 0;
 562         job->pages = NULL;
 563
 564         job->fn = fn;
 565         job->context = context;
 566
 567         if (job->source.count < SUB_JOB_SIZE)
 568                 dispatch_job(job);
 569
 570         else {
 571                 mutex_init(&job->lock);
 572                 job->progress = 0;
 573                 split_job(job);
 574         }
 575
 576         return 0;
 577 }
 578 EXPORT_SYMBOL(dm_kcopyd_copy);
 579
 580 /*
 581  * Cancels a kcopyd job, eg. someone might be deactivating a
 582  * mirror.
 583  */
 584 #if 0
 585 int kcopyd_cancel(struct kcopyd_job *job, int block)
 586 {
 587         /* FIXME: finish */
 588         return -1;
 589 }
 590 #endif  /*  0  */
 591
 592 /*-----------------------------------------------------------------
 593  * Client setup
 594  *---------------------------------------------------------------*/
 595 int dm_kcopyd_client_create(unsigned int nr_pages,
 596                             struct dm_kcopyd_client **result)
 597 {
 598         int r = -ENOMEM;
 599         struct dm_kcopyd_client *kc;
 600
 601         kc = kmalloc(sizeof(*kc), GFP_KERNEL);
 602         if (!kc)
 603                 return -ENOMEM;
 604
 605         spin_lock_init(&kc->lock);
 606         spin_lock_init(&kc->job_lock);
 607         INIT_LIST_HEAD(&kc->complete_jobs);
 608         INIT_LIST_HEAD(&kc->io_jobs);
 609         INIT_LIST_HEAD(&kc->pages_jobs);
 610
 611         kc->job_pool = mempool_create_slab_pool(MIN_JOBS, _job_cache);
 612         if (!kc->job_pool)
 613                 goto bad_slab;
 614
 615         INIT_WORK(&kc->kcopyd_work, do_work);
 616         kc->kcopyd_wq = create_singlethread_workqueue("kcopyd");
 617         if (!kc->kcopyd_wq)
 618                 goto bad_workqueue;
 619
 620         kc->pages = NULL;
 621         kc->nr_pages = kc->nr_free_pages = 0;
 622         r = client_alloc_pages(kc, nr_pages);
 623         if (r)
 624                 goto bad_client_pages;
 625
 626         kc->io_client = dm_io_client_create(nr_pages);
 627         if (IS_ERR(kc->io_client)) {
 628                 r = PTR_ERR(kc->io_client);
 629                 goto bad_io_client;
 630         }
 631
 632         init_waitqueue_head(&kc->destroyq);
 633         atomic_set(&kc->nr_jobs, 0);
 634
 635         *result = kc;
 636         return 0;
 637
 638 bad_io_client:
 639         client_free_pages(kc);
 640 bad_client_pages:
 641         destroy_workqueue(kc->kcopyd_wq);
 642 bad_workqueue:
 643         mempool_destroy(kc->job_pool);
 644 bad_slab:
 645         kfree(kc);
 646
 647         return r;
 648 }
 649 EXPORT_SYMBOL(dm_kcopyd_client_create);
 650
 651 void dm_kcopyd_client_destroy(struct dm_kcopyd_client *kc)
 652 {
 653         /* Wait for completion of all jobs submitted by this client. */
 654         wait_event(kc->destroyq, !atomic_read(&kc->nr_jobs));
 655
 656         BUG_ON(!list_empty(&kc->complete_jobs));
 657         BUG_ON(!list_empty(&kc->io_jobs));
 658         BUG_ON(!list_empty(&kc->pages_jobs));
 659         destroy_workqueue(kc->kcopyd_wq);
 660         dm_io_client_destroy(kc->io_client);
 661         client_free_pages(kc);
 662         mempool_destroy(kc->job_pool);
 663         kfree(kc);
 664 }
 665 EXPORT_SYMBOL(dm_kcopyd_client_destroy);