fs/fscache/page.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* Cache page management and data I/O routines
   3  *
   4  * Copyright (C) 2004-2008 Red Hat, Inc. All Rights Reserved.
   5  * Written by David Howells ([email protected])
   6  */
   7
   8 #define FSCACHE_DEBUG_LEVEL PAGE
   9 #include <linux/module.h>
  10 #include <linux/fscache-cache.h>
  11 #include <linux/buffer_head.h>
  12 #include <linux/pagevec.h>
  13 #include <linux/slab.h>
  14 #include "internal.h"
  15
  16 /*
  17  * check to see if a page is being written to the cache
  18  */
  19 bool __fscache_check_page_write(struct fscache_cookie *cookie, struct page *page)
  20 {
  21         void *val;
  22
  23         rcu_read_lock();
  24         val = radix_tree_lookup(&cookie->stores, page->index);
  25         rcu_read_unlock();
  26         trace_fscache_check_page(cookie, page, val, 0);
  27
  28         return val != NULL;
  29 }
  30 EXPORT_SYMBOL(__fscache_check_page_write);
  31
  32 /*
  33  * wait for a page to finish being written to the cache
  34  */
  35 void __fscache_wait_on_page_write(struct fscache_cookie *cookie, struct page *page)
  36 {
  37         wait_queue_head_t *wq = bit_waitqueue(&cookie->flags, 0);
  38
  39         trace_fscache_page(cookie, page, fscache_page_write_wait);
  40
  41         wait_event(*wq, !__fscache_check_page_write(cookie, page));
  42 }
  43 EXPORT_SYMBOL(__fscache_wait_on_page_write);
  44
  45 /*
  46  * wait for a page to finish being written to the cache. Put a timeout here
  47  * since we might be called recursively via parent fs.
  48  */
  49 static
  50 bool release_page_wait_timeout(struct fscache_cookie *cookie, struct page *page)
  51 {
  52         wait_queue_head_t *wq = bit_waitqueue(&cookie->flags, 0);
  53
  54         return wait_event_timeout(*wq, !__fscache_check_page_write(cookie, page),
  55                                   HZ);
  56 }
  57
  58 /*
  59  * decide whether a page can be released, possibly by cancelling a store to it
  60  * - we're allowed to sleep if __GFP_DIRECT_RECLAIM is flagged
  61  */
  62 bool __fscache_maybe_release_page(struct fscache_cookie *cookie,
  63                                   struct page *page,
  64                                   gfp_t gfp)
  65 {
  66         struct page *xpage;
  67         void *val;
  68
  69         _enter("%p,%p,%x", cookie, page, gfp);
  70
  71         trace_fscache_page(cookie, page, fscache_page_maybe_release);
  72
  73 try_again:
  74         rcu_read_lock();
  75         val = radix_tree_lookup(&cookie->stores, page->index);
  76         if (!val) {
  77                 rcu_read_unlock();
  78                 fscache_stat(&fscache_n_store_vmscan_not_storing);
  79                 __fscache_uncache_page(cookie, page);
  80                 return true;
  81         }
  82
  83         /* see if the page is actually undergoing storage - if so we can't get
  84          * rid of it till the cache has finished with it */
  85         if (radix_tree_tag_get(&cookie->stores, page->index,
  86                                FSCACHE_COOKIE_STORING_TAG)) {
  87                 rcu_read_unlock();
  88                 goto page_busy;
  89         }
  90
  91         /* the page is pending storage, so we attempt to cancel the store and
  92          * discard the store request so that the page can be reclaimed */
  93         spin_lock(&cookie->stores_lock);
  94         rcu_read_unlock();
  95
  96         if (radix_tree_tag_get(&cookie->stores, page->index,
  97                                FSCACHE_COOKIE_STORING_TAG)) {
  98                 /* the page started to undergo storage whilst we were looking,
  99                  * so now we can only wait or return */
 100                 spin_unlock(&cookie->stores_lock);
 101                 goto page_busy;
 102         }
 103
 104         xpage = radix_tree_delete(&cookie->stores, page->index);
 105         trace_fscache_page(cookie, page, fscache_page_radix_delete);
 106         spin_unlock(&cookie->stores_lock);
 107
 108         if (xpage) {
 109                 fscache_stat(&fscache_n_store_vmscan_cancelled);
 110                 fscache_stat(&fscache_n_store_radix_deletes);
 111                 ASSERTCMP(xpage, ==, page);
 112         } else {
 113                 fscache_stat(&fscache_n_store_vmscan_gone);
 114         }
 115
 116         wake_up_bit(&cookie->flags, 0);
 117         trace_fscache_wake_cookie(cookie);
 118         if (xpage)
 119                 put_page(xpage);
 120         __fscache_uncache_page(cookie, page);
 121         return true;
 122
 123 page_busy:
 124         /* We will wait here if we're allowed to, but that could deadlock the
 125          * allocator as the work threads writing to the cache may all end up
 126          * sleeping on memory allocation, so we may need to impose a timeout
 127          * too. */
 128         if (!(gfp & __GFP_DIRECT_RECLAIM) || !(gfp & __GFP_FS)) {
 129                 fscache_stat(&fscache_n_store_vmscan_busy);
 130                 return false;
 131         }
 132
 133         fscache_stat(&fscache_n_store_vmscan_wait);
 134         if (!release_page_wait_timeout(cookie, page))
 135                 _debug("fscache writeout timeout page: %p{%lx}",
 136                         page, page->index);
 137
 138         gfp &= ~__GFP_DIRECT_RECLAIM;
 139         goto try_again;
 140 }
 141 EXPORT_SYMBOL(__fscache_maybe_release_page);
 142
 143 /*
 144  * note that a page has finished being written to the cache
 145  */
 146 static void fscache_end_page_write(struct fscache_object *object,
 147                                    struct page *page)
 148 {
 149         struct fscache_cookie *cookie;
 150         struct page *xpage = NULL, *val;
 151
 152         spin_lock(&object->lock);
 153         cookie = object->cookie;
 154         if (cookie) {
 155                 /* delete the page from the tree if it is now no longer
 156                  * pending */
 157                 spin_lock(&cookie->stores_lock);
 158                 radix_tree_tag_clear(&cookie->stores, page->index,
 159                                      FSCACHE_COOKIE_STORING_TAG);
 160                 trace_fscache_page(cookie, page, fscache_page_radix_clear_store);
 161                 if (!radix_tree_tag_get(&cookie->stores, page->index,
 162                                         FSCACHE_COOKIE_PENDING_TAG)) {
 163                         fscache_stat(&fscache_n_store_radix_deletes);
 164                         xpage = radix_tree_delete(&cookie->stores, page->index);
 165                         trace_fscache_page(cookie, page, fscache_page_radix_delete);
 166                         trace_fscache_page(cookie, page, fscache_page_write_end);
 167
 168                         val = radix_tree_lookup(&cookie->stores, page->index);
 169                         trace_fscache_check_page(cookie, page, val, 1);
 170                 } else {
 171                         trace_fscache_page(cookie, page, fscache_page_write_end_pend);
 172                 }
 173                 spin_unlock(&cookie->stores_lock);
 174                 wake_up_bit(&cookie->flags, 0);
 175                 trace_fscache_wake_cookie(cookie);
 176         } else {
 177                 trace_fscache_page(cookie, page, fscache_page_write_end_noc);
 178         }
 179         spin_unlock(&object->lock);
 180         if (xpage)
 181                 put_page(xpage);
 182 }
 183
 184 /*
 185  * actually apply the changed attributes to a cache object
 186  */
 187 static void fscache_attr_changed_op(struct fscache_operation *op)
 188 {
 189         struct fscache_object *object = op->object;
 190         int ret;
 191
 192         _enter("{OBJ%x OP%x}", object->debug_id, op->debug_id);
 193
 194         fscache_stat(&fscache_n_attr_changed_calls);
 195
 196         if (fscache_object_is_active(object)) {
 197                 fscache_stat(&fscache_n_cop_attr_changed);
 198                 ret = object->cache->ops->attr_changed(object);
 199                 fscache_stat_d(&fscache_n_cop_attr_changed);
 200                 if (ret < 0)
 201                         fscache_abort_object(object);
 202                 fscache_op_complete(op, ret < 0);
 203         } else {
 204                 fscache_op_complete(op, true);
 205         }
 206
 207         _leave("");
 208 }
 209
 210 /*
 211  * notification that the attributes on an object have changed
 212  */
 213 int __fscache_attr_changed(struct fscache_cookie *cookie)
 214 {
 215         struct fscache_operation *op;
 216         struct fscache_object *object;
 217         bool wake_cookie = false;
 218
 219         _enter("%p", cookie);
 220
 221         ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
 222
 223         fscache_stat(&fscache_n_attr_changed);
 224
 225         op = kzalloc(sizeof(*op), GFP_KERNEL);
 226         if (!op) {
 227                 fscache_stat(&fscache_n_attr_changed_nomem);
 228                 _leave(" = -ENOMEM");
 229                 return -ENOMEM;
 230         }
 231
 232         fscache_operation_init(cookie, op, fscache_attr_changed_op, NULL, NULL);
 233         trace_fscache_page_op(cookie, NULL, op, fscache_page_op_attr_changed);
 234         op->flags = FSCACHE_OP_ASYNC |
 235                 (1 << FSCACHE_OP_EXCLUSIVE) |
 236                 (1 << FSCACHE_OP_UNUSE_COOKIE);
 237
 238         spin_lock(&cookie->lock);
 239
 240         if (!fscache_cookie_enabled(cookie) ||
 241             hlist_empty(&cookie->backing_objects))
 242                 goto nobufs;
 243         object = hlist_entry(cookie->backing_objects.first,
 244                              struct fscache_object, cookie_link);
 245
 246         __fscache_use_cookie(cookie);
 247         if (fscache_submit_exclusive_op(object, op) < 0)
 248                 goto nobufs_dec;
 249         spin_unlock(&cookie->lock);
 250         fscache_stat(&fscache_n_attr_changed_ok);
 251         fscache_put_operation(op);
 252         _leave(" = 0");
 253         return 0;
 254
 255 nobufs_dec:
 256         wake_cookie = __fscache_unuse_cookie(cookie);
 257 nobufs:
 258         spin_unlock(&cookie->lock);
 259         fscache_put_operation(op);
 260         if (wake_cookie)
 261                 __fscache_wake_unused_cookie(cookie);
 262         fscache_stat(&fscache_n_attr_changed_nobufs);
 263         _leave(" = %d", -ENOBUFS);
 264         return -ENOBUFS;
 265 }
 266 EXPORT_SYMBOL(__fscache_attr_changed);
 267
 268 /*
 269  * Handle cancellation of a pending retrieval op
 270  */
 271 static void fscache_do_cancel_retrieval(struct fscache_operation *_op)
 272 {
 273         struct fscache_retrieval *op =
 274                 container_of(_op, struct fscache_retrieval, op);
 275
 276         atomic_set(&op->n_pages, 0);
 277 }
 278
 279 /*
 280  * release a retrieval op reference
 281  */
 282 static void fscache_release_retrieval_op(struct fscache_operation *_op)
 283 {
 284         struct fscache_retrieval *op =
 285                 container_of(_op, struct fscache_retrieval, op);
 286
 287         _enter("{OP%x}", op->op.debug_id);
 288
 289         ASSERTIFCMP(op->op.state != FSCACHE_OP_ST_INITIALISED,
 290                     atomic_read(&op->n_pages), ==, 0);
 291
 292         if (op->context)
 293                 fscache_put_context(op->cookie, op->context);
 294
 295         _leave("");
 296 }
 297
 298 /*
 299  * allocate a retrieval op
 300  */
 301 struct fscache_retrieval *fscache_alloc_retrieval(
 302         struct fscache_cookie *cookie,
 303         struct address_space *mapping,
 304         fscache_rw_complete_t end_io_func,
 305         void *context)
 306 {
 307         struct fscache_retrieval *op;
 308
 309         /* allocate a retrieval operation and attempt to submit it */
 310         op = kzalloc(sizeof(*op), GFP_NOIO);
 311         if (!op) {
 312                 fscache_stat(&fscache_n_retrievals_nomem);
 313                 return NULL;
 314         }
 315
 316         fscache_operation_init(cookie, &op->op, NULL,
 317                                fscache_do_cancel_retrieval,
 318                                fscache_release_retrieval_op);
 319         op->op.flags    = FSCACHE_OP_MYTHREAD |
 320                 (1UL << FSCACHE_OP_WAITING) |
 321                 (1UL << FSCACHE_OP_UNUSE_COOKIE);
 322         op->cookie      = cookie;
 323         op->mapping     = mapping;
 324         op->end_io_func = end_io_func;
 325         op->context     = context;
 326         INIT_LIST_HEAD(&op->to_do);
 327
 328         /* Pin the netfs read context in case we need to do the actual netfs
 329          * read because we've encountered a cache read failure.
 330          */
 331         if (context)
 332                 fscache_get_context(op->cookie, context);
 333         return op;
 334 }
 335
 336 /*
 337  * wait for a deferred lookup to complete
 338  */
 339 int fscache_wait_for_deferred_lookup(struct fscache_cookie *cookie)
 340 {
 341         _enter("");
 342
 343         if (!test_bit(FSCACHE_COOKIE_LOOKING_UP, &cookie->flags)) {
 344                 _leave(" = 0 [imm]");
 345                 return 0;
 346         }
 347
 348         fscache_stat(&fscache_n_retrievals_wait);
 349
 350         if (wait_on_bit(&cookie->flags, FSCACHE_COOKIE_LOOKING_UP,
 351                         TASK_INTERRUPTIBLE) != 0) {
 352                 fscache_stat(&fscache_n_retrievals_intr);
 353                 _leave(" = -ERESTARTSYS");
 354                 return -ERESTARTSYS;
 355         }
 356
 357         ASSERT(!test_bit(FSCACHE_COOKIE_LOOKING_UP, &cookie->flags));
 358
 359         smp_rmb();
 360         _leave(" = 0 [dly]");
 361         return 0;
 362 }
 363
 364 /*
 365  * wait for an object to become active (or dead)
 366  */
 367 int fscache_wait_for_operation_activation(struct fscache_object *object,
 368                                           struct fscache_operation *op,
 369                                           atomic_t *stat_op_waits,
 370                                           atomic_t *stat_object_dead)
 371 {
 372         int ret;
 373
 374         if (!test_bit(FSCACHE_OP_WAITING, &op->flags))
 375                 goto check_if_dead;
 376
 377         _debug(">>> WT");
 378         if (stat_op_waits)
 379                 fscache_stat(stat_op_waits);
 380         if (wait_on_bit(&op->flags, FSCACHE_OP_WAITING,
 381                         TASK_INTERRUPTIBLE) != 0) {
 382                 trace_fscache_op(object->cookie, op, fscache_op_signal);
 383                 ret = fscache_cancel_op(op, false);
 384                 if (ret == 0)
 385                         return -ERESTARTSYS;
 386
 387                 /* it's been removed from the pending queue by another party,
 388                  * so we should get to run shortly */
 389                 wait_on_bit(&op->flags, FSCACHE_OP_WAITING,
 390                             TASK_UNINTERRUPTIBLE);
 391         }
 392         _debug("<<< GO");
 393
 394 check_if_dead:
 395         if (op->state == FSCACHE_OP_ST_CANCELLED) {
 396                 if (stat_object_dead)
 397                         fscache_stat(stat_object_dead);
 398                 _leave(" = -ENOBUFS [cancelled]");
 399                 return -ENOBUFS;
 400         }
 401         if (unlikely(fscache_object_is_dying(object) ||
 402                      fscache_cache_is_broken(object))) {
 403                 enum fscache_operation_state state = op->state;
 404                 trace_fscache_op(object->cookie, op, fscache_op_signal);
 405                 fscache_cancel_op(op, true);
 406                 if (stat_object_dead)
 407                         fscache_stat(stat_object_dead);
 408                 _leave(" = -ENOBUFS [obj dead %d]", state);
 409                 return -ENOBUFS;
 410         }
 411         return 0;
 412 }
 413
 414 /*
 415  * read a page from the cache or allocate a block in which to store it
 416  * - we return:
 417  *   -ENOMEM    - out of memory, nothing done
 418  *   -ERESTARTSYS - interrupted
 419  *   -ENOBUFS   - no backing object available in which to cache the block
 420  *   -ENODATA   - no data available in the backing object for this block
 421  *   0          - dispatched a read - it'll call end_io_func() when finished
 422  */
 423 int __fscache_read_or_alloc_page(struct fscache_cookie *cookie,
 424                                  struct page *page,
 425                                  fscache_rw_complete_t end_io_func,
 426                                  void *context,
 427                                  gfp_t gfp)
 428 {
 429         struct fscache_retrieval *op;
 430         struct fscache_object *object;
 431         bool wake_cookie = false;
 432         int ret;
 433
 434         _enter("%p,%p,,,", cookie, page);
 435
 436         fscache_stat(&fscache_n_retrievals);
 437
 438         if (hlist_empty(&cookie->backing_objects))
 439                 goto nobufs;
 440
 441         if (test_bit(FSCACHE_COOKIE_INVALIDATING, &cookie->flags)) {
 442                 _leave(" = -ENOBUFS [invalidating]");
 443                 return -ENOBUFS;
 444         }
 445
 446         ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
 447         ASSERTCMP(page, !=, NULL);
 448
 449         if (fscache_wait_for_deferred_lookup(cookie) < 0)
 450                 return -ERESTARTSYS;
 451
 452         op = fscache_alloc_retrieval(cookie, page->mapping,
 453                                      end_io_func, context);
 454         if (!op) {
 455                 _leave(" = -ENOMEM");
 456                 return -ENOMEM;
 457         }
 458         atomic_set(&op->n_pages, 1);
 459         trace_fscache_page_op(cookie, page, &op->op, fscache_page_op_retr_one);
 460
 461         spin_lock(&cookie->lock);
 462
 463         if (!fscache_cookie_enabled(cookie) ||
 464             hlist_empty(&cookie->backing_objects))
 465                 goto nobufs_unlock;
 466         object = hlist_entry(cookie->backing_objects.first,
 467                              struct fscache_object, cookie_link);
 468
 469         ASSERT(test_bit(FSCACHE_OBJECT_IS_LOOKED_UP, &object->flags));
 470
 471         __fscache_use_cookie(cookie);
 472         atomic_inc(&object->n_reads);
 473         __set_bit(FSCACHE_OP_DEC_READ_CNT, &op->op.flags);
 474
 475         if (fscache_submit_op(object, &op->op) < 0)
 476                 goto nobufs_unlock_dec;
 477         spin_unlock(&cookie->lock);
 478
 479         fscache_stat(&fscache_n_retrieval_ops);
 480
 481         /* we wait for the operation to become active, and then process it
 482          * *here*, in this thread, and not in the thread pool */
 483         ret = fscache_wait_for_operation_activation(
 484                 object, &op->op,
 485                 __fscache_stat(&fscache_n_retrieval_op_waits),
 486                 __fscache_stat(&fscache_n_retrievals_object_dead));
 487         if (ret < 0)
 488                 goto error;
 489
 490         /* ask the cache to honour the operation */
 491         if (test_bit(FSCACHE_COOKIE_NO_DATA_YET, &object->cookie->flags)) {
 492                 fscache_stat(&fscache_n_cop_allocate_page);
 493                 ret = object->cache->ops->allocate_page(op, page, gfp);
 494                 fscache_stat_d(&fscache_n_cop_allocate_page);
 495                 if (ret == 0)
 496                         ret = -ENODATA;
 497         } else {
 498                 fscache_stat(&fscache_n_cop_read_or_alloc_page);
 499                 ret = object->cache->ops->read_or_alloc_page(op, page, gfp);
 500                 fscache_stat_d(&fscache_n_cop_read_or_alloc_page);
 501         }
 502
 503 error:
 504         if (ret == -ENOMEM)
 505                 fscache_stat(&fscache_n_retrievals_nomem);
 506         else if (ret == -ERESTARTSYS)
 507                 fscache_stat(&fscache_n_retrievals_intr);
 508         else if (ret == -ENODATA)
 509                 fscache_stat(&fscache_n_retrievals_nodata);
 510         else if (ret < 0)
 511                 fscache_stat(&fscache_n_retrievals_nobufs);
 512         else
 513                 fscache_stat(&fscache_n_retrievals_ok);
 514
 515         fscache_put_retrieval(op);
 516         _leave(" = %d", ret);
 517         return ret;
 518
 519 nobufs_unlock_dec:
 520         atomic_dec(&object->n_reads);
 521         wake_cookie = __fscache_unuse_cookie(cookie);
 522 nobufs_unlock:
 523         spin_unlock(&cookie->lock);
 524         if (wake_cookie)
 525                 __fscache_wake_unused_cookie(cookie);
 526         fscache_put_retrieval(op);
 527 nobufs:
 528         fscache_stat(&fscache_n_retrievals_nobufs);
 529         _leave(" = -ENOBUFS");
 530         return -ENOBUFS;
 531 }
 532 EXPORT_SYMBOL(__fscache_read_or_alloc_page);
 533
 534 /*
 535  * read a list of page from the cache or allocate a block in which to store
 536  * them
 537  * - we return:
 538  *   -ENOMEM    - out of memory, some pages may be being read
 539  *   -ERESTARTSYS - interrupted, some pages may be being read
 540  *   -ENOBUFS   - no backing object or space available in which to cache any
 541  *                pages not being read
 542  *   -ENODATA   - no data available in the backing object for some or all of
 543  *                the pages
 544  *   0          - dispatched a read on all pages
 545  *
 546  * end_io_func() will be called for each page read from the cache as it is
 547  * finishes being read
 548  *
 549  * any pages for which a read is dispatched will be removed from pages and
 550  * nr_pages
 551  */
 552 int __fscache_read_or_alloc_pages(struct fscache_cookie *cookie,
 553                                   struct address_space *mapping,
 554                                   struct list_head *pages,
 555                                   unsigned *nr_pages,
 556                                   fscache_rw_complete_t end_io_func,
 557                                   void *context,
 558                                   gfp_t gfp)
 559 {
 560         struct fscache_retrieval *op;
 561         struct fscache_object *object;
 562         bool wake_cookie = false;
 563         int ret;
 564
 565         _enter("%p,,%d,,,", cookie, *nr_pages);
 566
 567         fscache_stat(&fscache_n_retrievals);
 568
 569         if (hlist_empty(&cookie->backing_objects))
 570                 goto nobufs;
 571
 572         if (test_bit(FSCACHE_COOKIE_INVALIDATING, &cookie->flags)) {
 573                 _leave(" = -ENOBUFS [invalidating]");
 574                 return -ENOBUFS;
 575         }
 576
 577         ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
 578         ASSERTCMP(*nr_pages, >, 0);
 579         ASSERT(!list_empty(pages));
 580
 581         if (fscache_wait_for_deferred_lookup(cookie) < 0)
 582                 return -ERESTARTSYS;
 583
 584         op = fscache_alloc_retrieval(cookie, mapping, end_io_func, context);
 585         if (!op)
 586                 return -ENOMEM;
 587         atomic_set(&op->n_pages, *nr_pages);
 588         trace_fscache_page_op(cookie, NULL, &op->op, fscache_page_op_retr_multi);
 589
 590         spin_lock(&cookie->lock);
 591
 592         if (!fscache_cookie_enabled(cookie) ||
 593             hlist_empty(&cookie->backing_objects))
 594                 goto nobufs_unlock;
 595         object = hlist_entry(cookie->backing_objects.first,
 596                              struct fscache_object, cookie_link);
 597
 598         __fscache_use_cookie(cookie);
 599         atomic_inc(&object->n_reads);
 600         __set_bit(FSCACHE_OP_DEC_READ_CNT, &op->op.flags);
 601
 602         if (fscache_submit_op(object, &op->op) < 0)
 603                 goto nobufs_unlock_dec;
 604         spin_unlock(&cookie->lock);
 605
 606         fscache_stat(&fscache_n_retrieval_ops);
 607
 608         /* we wait for the operation to become active, and then process it
 609          * *here*, in this thread, and not in the thread pool */
 610         ret = fscache_wait_for_operation_activation(
 611                 object, &op->op,
 612                 __fscache_stat(&fscache_n_retrieval_op_waits),
 613                 __fscache_stat(&fscache_n_retrievals_object_dead));
 614         if (ret < 0)
 615                 goto error;
 616
 617         /* ask the cache to honour the operation */
 618         if (test_bit(FSCACHE_COOKIE_NO_DATA_YET, &object->cookie->flags)) {
 619                 fscache_stat(&fscache_n_cop_allocate_pages);
 620                 ret = object->cache->ops->allocate_pages(
 621                         op, pages, nr_pages, gfp);
 622                 fscache_stat_d(&fscache_n_cop_allocate_pages);
 623         } else {
 624                 fscache_stat(&fscache_n_cop_read_or_alloc_pages);
 625                 ret = object->cache->ops->read_or_alloc_pages(
 626                         op, pages, nr_pages, gfp);
 627                 fscache_stat_d(&fscache_n_cop_read_or_alloc_pages);
 628         }
 629
 630 error:
 631         if (ret == -ENOMEM)
 632                 fscache_stat(&fscache_n_retrievals_nomem);
 633         else if (ret == -ERESTARTSYS)
 634                 fscache_stat(&fscache_n_retrievals_intr);
 635         else if (ret == -ENODATA)
 636                 fscache_stat(&fscache_n_retrievals_nodata);
 637         else if (ret < 0)
 638                 fscache_stat(&fscache_n_retrievals_nobufs);
 639         else
 640                 fscache_stat(&fscache_n_retrievals_ok);
 641
 642         fscache_put_retrieval(op);
 643         _leave(" = %d", ret);
 644         return ret;
 645
 646 nobufs_unlock_dec:
 647         atomic_dec(&object->n_reads);
 648         wake_cookie = __fscache_unuse_cookie(cookie);
 649 nobufs_unlock:
 650         spin_unlock(&cookie->lock);
 651         fscache_put_retrieval(op);
 652         if (wake_cookie)
 653                 __fscache_wake_unused_cookie(cookie);
 654 nobufs:
 655         fscache_stat(&fscache_n_retrievals_nobufs);
 656         _leave(" = -ENOBUFS");
 657         return -ENOBUFS;
 658 }
 659 EXPORT_SYMBOL(__fscache_read_or_alloc_pages);
 660
 661 /*
 662  * allocate a block in the cache on which to store a page
 663  * - we return:
 664  *   -ENOMEM    - out of memory, nothing done
 665  *   -ERESTARTSYS - interrupted
 666  *   -ENOBUFS   - no backing object available in which to cache the block
 667  *   0          - block allocated
 668  */
 669 int __fscache_alloc_page(struct fscache_cookie *cookie,
 670                          struct page *page,
 671                          gfp_t gfp)
 672 {
 673         struct fscache_retrieval *op;
 674         struct fscache_object *object;
 675         bool wake_cookie = false;
 676         int ret;
 677
 678         _enter("%p,%p,,,", cookie, page);
 679
 680         fscache_stat(&fscache_n_allocs);
 681
 682         if (hlist_empty(&cookie->backing_objects))
 683                 goto nobufs;
 684
 685         ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
 686         ASSERTCMP(page, !=, NULL);
 687
 688         if (test_bit(FSCACHE_COOKIE_INVALIDATING, &cookie->flags)) {
 689                 _leave(" = -ENOBUFS [invalidating]");
 690                 return -ENOBUFS;
 691         }
 692
 693         if (fscache_wait_for_deferred_lookup(cookie) < 0)
 694                 return -ERESTARTSYS;
 695
 696         op = fscache_alloc_retrieval(cookie, page->mapping, NULL, NULL);
 697         if (!op)
 698                 return -ENOMEM;
 699         atomic_set(&op->n_pages, 1);
 700         trace_fscache_page_op(cookie, page, &op->op, fscache_page_op_alloc_one);
 701
 702         spin_lock(&cookie->lock);
 703
 704         if (!fscache_cookie_enabled(cookie) ||
 705             hlist_empty(&cookie->backing_objects))
 706                 goto nobufs_unlock;
 707         object = hlist_entry(cookie->backing_objects.first,
 708                              struct fscache_object, cookie_link);
 709
 710         __fscache_use_cookie(cookie);
 711         if (fscache_submit_op(object, &op->op) < 0)
 712                 goto nobufs_unlock_dec;
 713         spin_unlock(&cookie->lock);
 714
 715         fscache_stat(&fscache_n_alloc_ops);
 716
 717         ret = fscache_wait_for_operation_activation(
 718                 object, &op->op,
 719                 __fscache_stat(&fscache_n_alloc_op_waits),
 720                 __fscache_stat(&fscache_n_allocs_object_dead));
 721         if (ret < 0)
 722                 goto error;
 723
 724         /* ask the cache to honour the operation */
 725         fscache_stat(&fscache_n_cop_allocate_page);
 726         ret = object->cache->ops->allocate_page(op, page, gfp);
 727         fscache_stat_d(&fscache_n_cop_allocate_page);
 728
 729 error:
 730         if (ret == -ERESTARTSYS)
 731                 fscache_stat(&fscache_n_allocs_intr);
 732         else if (ret < 0)
 733                 fscache_stat(&fscache_n_allocs_nobufs);
 734         else
 735                 fscache_stat(&fscache_n_allocs_ok);
 736
 737         fscache_put_retrieval(op);
 738         _leave(" = %d", ret);
 739         return ret;
 740
 741 nobufs_unlock_dec:
 742         wake_cookie = __fscache_unuse_cookie(cookie);
 743 nobufs_unlock:
 744         spin_unlock(&cookie->lock);
 745         fscache_put_retrieval(op);
 746         if (wake_cookie)
 747                 __fscache_wake_unused_cookie(cookie);
 748 nobufs:
 749         fscache_stat(&fscache_n_allocs_nobufs);
 750         _leave(" = -ENOBUFS");
 751         return -ENOBUFS;
 752 }
 753 EXPORT_SYMBOL(__fscache_alloc_page);
 754
 755 /*
 756  * Unmark pages allocate in the readahead code path (via:
 757  * fscache_readpages_or_alloc) after delegating to the base filesystem
 758  */
 759 void __fscache_readpages_cancel(struct fscache_cookie *cookie,
 760                                 struct list_head *pages)
 761 {
 762         struct page *page;
 763
 764         list_for_each_entry(page, pages, lru) {
 765                 if (PageFsCache(page))
 766                         __fscache_uncache_page(cookie, page);
 767         }
 768 }
 769 EXPORT_SYMBOL(__fscache_readpages_cancel);
 770
 771 /*
 772  * release a write op reference
 773  */
 774 static void fscache_release_write_op(struct fscache_operation *_op)
 775 {
 776         _enter("{OP%x}", _op->debug_id);
 777 }
 778
 779 /*
 780  * perform the background storage of a page into the cache
 781  */
 782 static void fscache_write_op(struct fscache_operation *_op)
 783 {
 784         struct fscache_storage *op =
 785                 container_of(_op, struct fscache_storage, op);
 786         struct fscache_object *object = op->op.object;
 787         struct fscache_cookie *cookie;
 788         struct page *page;
 789         unsigned n;
 790         void *results[1];
 791         int ret;
 792
 793         _enter("{OP%x,%d}", op->op.debug_id, atomic_read(&op->op.usage));
 794
 795 again:
 796         spin_lock(&object->lock);
 797         cookie = object->cookie;
 798
 799         if (!fscache_object_is_active(object)) {
 800                 /* If we get here, then the on-disk cache object likely no
 801                  * longer exists, so we should just cancel this write
 802                  * operation.
 803                  */
 804                 spin_unlock(&object->lock);
 805                 fscache_op_complete(&op->op, true);
 806                 _leave(" [inactive]");
 807                 return;
 808         }
 809
 810         if (!cookie) {
 811                 /* If we get here, then the cookie belonging to the object was
 812                  * detached, probably by the cookie being withdrawn due to
 813                  * memory pressure, which means that the pages we might write
 814                  * to the cache from no longer exist - therefore, we can just
 815                  * cancel this write operation.
 816                  */
 817                 spin_unlock(&object->lock);
 818                 fscache_op_complete(&op->op, true);
 819                 _leave(" [cancel] op{f=%lx s=%u} obj{s=%s f=%lx}",
 820                        _op->flags, _op->state, object->state->short_name,
 821                        object->flags);
 822                 return;
 823         }
 824
 825         spin_lock(&cookie->stores_lock);
 826
 827         fscache_stat(&fscache_n_store_calls);
 828
 829         /* find a page to store */
 830         results[0] = NULL;
 831         page = NULL;
 832         n = radix_tree_gang_lookup_tag(&cookie->stores, results, 0, 1,
 833                                        FSCACHE_COOKIE_PENDING_TAG);
 834         trace_fscache_gang_lookup(cookie, &op->op, results, n, op->store_limit);
 835         if (n != 1)
 836                 goto superseded;
 837         page = results[0];
 838         _debug("gang %d [%lx]", n, page->index);
 839
 840         radix_tree_tag_set(&cookie->stores, page->index,
 841                            FSCACHE_COOKIE_STORING_TAG);
 842         radix_tree_tag_clear(&cookie->stores, page->index,
 843                              FSCACHE_COOKIE_PENDING_TAG);
 844         trace_fscache_page(cookie, page, fscache_page_radix_pend2store);
 845
 846         spin_unlock(&cookie->stores_lock);
 847         spin_unlock(&object->lock);
 848
 849         if (page->index >= op->store_limit)
 850                 goto discard_page;
 851
 852         fscache_stat(&fscache_n_store_pages);
 853         fscache_stat(&fscache_n_cop_write_page);
 854         ret = object->cache->ops->write_page(op, page);
 855         fscache_stat_d(&fscache_n_cop_write_page);
 856         trace_fscache_wrote_page(cookie, page, &op->op, ret);
 857         fscache_end_page_write(object, page);
 858         if (ret < 0) {
 859                 fscache_abort_object(object);
 860                 fscache_op_complete(&op->op, true);
 861         } else {
 862                 fscache_enqueue_operation(&op->op);
 863         }
 864
 865         _leave("");
 866         return;
 867
 868 discard_page:
 869         fscache_stat(&fscache_n_store_pages_over_limit);
 870         trace_fscache_wrote_page(cookie, page, &op->op, -ENOBUFS);
 871         fscache_end_page_write(object, page);
 872         goto again;
 873
 874 superseded:
 875         /* this writer is going away and there aren't any more things to
 876          * write */
 877         _debug("cease");
 878         spin_unlock(&cookie->stores_lock);
 879         clear_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags);
 880         spin_unlock(&object->lock);
 881         fscache_op_complete(&op->op, false);
 882         _leave("");
 883 }
 884
 885 /*
 886  * Clear the pages pending writing for invalidation
 887  */
 888 void fscache_invalidate_writes(struct fscache_cookie *cookie)
 889 {
 890         struct page *page;
 891         void *results[16];
 892         int n, i;
 893
 894         _enter("");
 895
 896         for (;;) {
 897                 spin_lock(&cookie->stores_lock);
 898                 n = radix_tree_gang_lookup_tag(&cookie->stores, results, 0,
 899                                                ARRAY_SIZE(results),
 900                                                FSCACHE_COOKIE_PENDING_TAG);
 901                 if (n == 0) {
 902                         spin_unlock(&cookie->stores_lock);
 903                         break;
 904                 }
 905
 906                 for (i = n - 1; i >= 0; i--) {
 907                         page = results[i];
 908                         radix_tree_delete(&cookie->stores, page->index);
 909                         trace_fscache_page(cookie, page, fscache_page_radix_delete);
 910                         trace_fscache_page(cookie, page, fscache_page_inval);
 911                 }
 912
 913                 spin_unlock(&cookie->stores_lock);
 914
 915                 for (i = n - 1; i >= 0; i--)
 916                         put_page(results[i]);
 917         }
 918
 919         wake_up_bit(&cookie->flags, 0);
 920         trace_fscache_wake_cookie(cookie);
 921
 922         _leave("");
 923 }
 924
 925 /*
 926  * request a page be stored in the cache
 927  * - returns:
 928  *   -ENOMEM    - out of memory, nothing done
 929  *   -ENOBUFS   - no backing object available in which to cache the page
 930  *   0          - dispatched a write - it'll call end_io_func() when finished
 931  *
 932  * if the cookie still has a backing object at this point, that object can be
 933  * in one of a few states with respect to storage processing:
 934  *
 935  *  (1) negative lookup, object not yet created (FSCACHE_COOKIE_CREATING is
 936  *      set)
 937  *
 938  *      (a) no writes yet
 939  *
 940  *      (b) writes deferred till post-creation (mark page for writing and
 941  *          return immediately)
 942  *
 943  *  (2) negative lookup, object created, initial fill being made from netfs
 944  *
 945  *      (a) fill point not yet reached this page (mark page for writing and
 946  *          return)
 947  *
 948  *      (b) fill point passed this page (queue op to store this page)
 949  *
 950  *  (3) object extant (queue op to store this page)
 951  *
 952  * any other state is invalid
 953  */
 954 int __fscache_write_page(struct fscache_cookie *cookie,
 955                          struct page *page,
 956                          loff_t object_size,
 957                          gfp_t gfp)
 958 {
 959         struct fscache_storage *op;
 960         struct fscache_object *object;
 961         bool wake_cookie = false;
 962         int ret;
 963
 964         _enter("%p,%x,", cookie, (u32) page->flags);
 965
 966         ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
 967         ASSERT(PageFsCache(page));
 968
 969         fscache_stat(&fscache_n_stores);
 970
 971         if (test_bit(FSCACHE_COOKIE_INVALIDATING, &cookie->flags)) {
 972                 _leave(" = -ENOBUFS [invalidating]");
 973                 return -ENOBUFS;
 974         }
 975
 976         op = kzalloc(sizeof(*op), GFP_NOIO | __GFP_NOMEMALLOC | __GFP_NORETRY);
 977         if (!op)
 978                 goto nomem;
 979
 980         fscache_operation_init(cookie, &op->op, fscache_write_op, NULL,
 981                                fscache_release_write_op);
 982         op->op.flags = FSCACHE_OP_ASYNC |
 983                 (1 << FSCACHE_OP_WAITING) |
 984                 (1 << FSCACHE_OP_UNUSE_COOKIE);
 985
 986         ret = radix_tree_maybe_preload(gfp & ~__GFP_HIGHMEM);
 987         if (ret < 0)
 988                 goto nomem_free;
 989
 990         trace_fscache_page_op(cookie, page, &op->op, fscache_page_op_write_one);
 991
 992         ret = -ENOBUFS;
 993         spin_lock(&cookie->lock);
 994
 995         if (!fscache_cookie_enabled(cookie) ||
 996             hlist_empty(&cookie->backing_objects))
 997                 goto nobufs;
 998         object = hlist_entry(cookie->backing_objects.first,
 999                              struct fscache_object, cookie_link);
1000         if (test_bit(FSCACHE_IOERROR, &object->cache->flags))
1001                 goto nobufs;
1002
1003         trace_fscache_page(cookie, page, fscache_page_write);
1004
1005         /* add the page to the pending-storage radix tree on the backing
1006          * object */
1007         spin_lock(&object->lock);
1008
1009         if (object->store_limit_l != object_size)
1010                 fscache_set_store_limit(object, object_size);
1011
1012         spin_lock(&cookie->stores_lock);
1013
1014         _debug("store limit %llx", (unsigned long long) object->store_limit);
1015
1016         ret = radix_tree_insert(&cookie->stores, page->index, page);
1017         if (ret < 0) {
1018                 if (ret == -EEXIST)
1019                         goto already_queued;
1020                 _debug("insert failed %d", ret);
1021                 goto nobufs_unlock_obj;
1022         }
1023
1024         trace_fscache_page(cookie, page, fscache_page_radix_insert);
1025         radix_tree_tag_set(&cookie->stores, page->index,
1026                            FSCACHE_COOKIE_PENDING_TAG);
1027         trace_fscache_page(cookie, page, fscache_page_radix_set_pend);
1028         get_page(page);
1029
1030         /* we only want one writer at a time, but we do need to queue new
1031          * writers after exclusive ops */
1032         if (test_and_set_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags))
1033                 goto already_pending;
1034
1035         spin_unlock(&cookie->stores_lock);
1036         spin_unlock(&object->lock);
1037
1038         op->op.debug_id = atomic_inc_return(&fscache_op_debug_id);
1039         op->store_limit = object->store_limit;
1040
1041         __fscache_use_cookie(cookie);
1042         if (fscache_submit_op(object, &op->op) < 0)
1043                 goto submit_failed;
1044
1045         spin_unlock(&cookie->lock);
1046         radix_tree_preload_end();
1047         fscache_stat(&fscache_n_store_ops);
1048         fscache_stat(&fscache_n_stores_ok);
1049
1050         /* the work queue now carries its own ref on the object */
1051         fscache_put_operation(&op->op);
1052         _leave(" = 0");
1053         return 0;
1054
1055 already_queued:
1056         fscache_stat(&fscache_n_stores_again);
1057 already_pending:
1058         spin_unlock(&cookie->stores_lock);
1059         spin_unlock(&object->lock);
1060         spin_unlock(&cookie->lock);
1061         radix_tree_preload_end();
1062         fscache_put_operation(&op->op);
1063         fscache_stat(&fscache_n_stores_ok);
1064         _leave(" = 0");
1065         return 0;
1066
1067 submit_failed:
1068         spin_lock(&cookie->stores_lock);
1069         radix_tree_delete(&cookie->stores, page->index);
1070         trace_fscache_page(cookie, page, fscache_page_radix_delete);
1071         spin_unlock(&cookie->stores_lock);
1072         wake_cookie = __fscache_unuse_cookie(cookie);
1073         put_page(page);
1074         ret = -ENOBUFS;
1075         goto nobufs;
1076
1077 nobufs_unlock_obj:
1078         spin_unlock(&cookie->stores_lock);
1079         spin_unlock(&object->lock);
1080 nobufs:
1081         spin_unlock(&cookie->lock);
1082         radix_tree_preload_end();
1083         fscache_put_operation(&op->op);
1084         if (wake_cookie)
1085                 __fscache_wake_unused_cookie(cookie);
1086         fscache_stat(&fscache_n_stores_nobufs);
1087         _leave(" = -ENOBUFS");
1088         return -ENOBUFS;
1089
1090 nomem_free:
1091         fscache_put_operation(&op->op);
1092 nomem:
1093         fscache_stat(&fscache_n_stores_oom);
1094         _leave(" = -ENOMEM");
1095         return -ENOMEM;
1096 }
1097 EXPORT_SYMBOL(__fscache_write_page);
1098
1099 /*
1100  * remove a page from the cache
1101  */
1102 void __fscache_uncache_page(struct fscache_cookie *cookie, struct page *page)
1103 {
1104         struct fscache_object *object;
1105
1106         _enter(",%p", page);
1107
1108         ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
1109         ASSERTCMP(page, !=, NULL);
1110
1111         fscache_stat(&fscache_n_uncaches);
1112
1113         /* cache withdrawal may beat us to it */
1114         if (!PageFsCache(page))
1115                 goto done;
1116
1117         trace_fscache_page(cookie, page, fscache_page_uncache);
1118
1119         /* get the object */
1120         spin_lock(&cookie->lock);
1121
1122         if (hlist_empty(&cookie->backing_objects)) {
1123                 ClearPageFsCache(page);
1124                 goto done_unlock;
1125         }
1126
1127         object = hlist_entry(cookie->backing_objects.first,
1128                              struct fscache_object, cookie_link);
1129
1130         /* there might now be stuff on disk we could read */
1131         clear_bit(FSCACHE_COOKIE_NO_DATA_YET, &cookie->flags);
1132
1133         /* only invoke the cache backend if we managed to mark the page
1134          * uncached here; this deals with synchronisation vs withdrawal */
1135         if (TestClearPageFsCache(page) &&
1136             object->cache->ops->uncache_page) {
1137                 /* the cache backend releases the cookie lock */
1138                 fscache_stat(&fscache_n_cop_uncache_page);
1139                 object->cache->ops->uncache_page(object, page);
1140                 fscache_stat_d(&fscache_n_cop_uncache_page);
1141                 goto done;
1142         }
1143
1144 done_unlock:
1145         spin_unlock(&cookie->lock);
1146 done:
1147         _leave("");
1148 }
1149 EXPORT_SYMBOL(__fscache_uncache_page);
1150
1151 /**
1152  * fscache_mark_page_cached - Mark a page as being cached
1153  * @op: The retrieval op pages are being marked for
1154  * @page: The page to be marked
1155  *
1156  * Mark a netfs page as being cached.  After this is called, the netfs
1157  * must call fscache_uncache_page() to remove the mark.
1158  */
1159 void fscache_mark_page_cached(struct fscache_retrieval *op, struct page *page)
1160 {
1161         struct fscache_cookie *cookie = op->op.object->cookie;
1162
1163 #ifdef CONFIG_FSCACHE_STATS
1164         atomic_inc(&fscache_n_marks);
1165 #endif
1166
1167         trace_fscache_page(cookie, page, fscache_page_cached);
1168
1169         _debug("- mark %p{%lx}", page, page->index);
1170         if (TestSetPageFsCache(page)) {
1171                 static bool once_only;
1172                 if (!once_only) {
1173                         once_only = true;
1174                         pr_warn("Cookie type %s marked page %lx multiple times\n",
1175                                 cookie->def->name, page->index);
1176                 }
1177         }
1178
1179         if (cookie->def->mark_page_cached)
1180                 cookie->def->mark_page_cached(cookie->netfs_data,
1181                                               op->mapping, page);
1182 }
1183 EXPORT_SYMBOL(fscache_mark_page_cached);
1184
1185 /**
1186  * fscache_mark_pages_cached - Mark pages as being cached
1187  * @op: The retrieval op pages are being marked for
1188  * @pagevec: The pages to be marked
1189  *
1190  * Mark a bunch of netfs pages as being cached.  After this is called,
1191  * the netfs must call fscache_uncache_page() to remove the mark.
1192  */
1193 void fscache_mark_pages_cached(struct fscache_retrieval *op,
1194                                struct pagevec *pagevec)
1195 {
1196         unsigned long loop;
1197
1198         for (loop = 0; loop < pagevec->nr; loop++)
1199                 fscache_mark_page_cached(op, pagevec->pages[loop]);
1200
1201         pagevec_reinit(pagevec);
1202 }
1203 EXPORT_SYMBOL(fscache_mark_pages_cached);
1204
1205 /*
1206  * Uncache all the pages in an inode that are marked PG_fscache, assuming them
1207  * to be associated with the given cookie.
1208  */
1209 void __fscache_uncache_all_inode_pages(struct fscache_cookie *cookie,
1210                                        struct inode *inode)
1211 {
1212         struct address_space *mapping = inode->i_mapping;
1213         struct pagevec pvec;
1214         pgoff_t next;
1215         int i;
1216
1217         _enter("%p,%p", cookie, inode);
1218
1219         if (!mapping || mapping->nrpages == 0) {
1220                 _leave(" [no pages]");
1221                 return;
1222         }
1223
1224         pagevec_init(&pvec);
1225         next = 0;
1226         do {
1227                 if (!pagevec_lookup(&pvec, mapping, &next))
1228                         break;
1229                 for (i = 0; i < pagevec_count(&pvec); i++) {
1230                         struct page *page = pvec.pages[i];
1231                         if (PageFsCache(page)) {
1232                                 __fscache_wait_on_page_write(cookie, page);
1233                                 __fscache_uncache_page(cookie, page);
1234                         }
1235                 }
1236                 pagevec_release(&pvec);
1237                 cond_resched();
1238         } while (next);
1239
1240         _leave("");
1241 }
1242 EXPORT_SYMBOL(__fscache_uncache_all_inode_pages);