fs/nfs/pagelist.c

   1 /*
   2  * linux/fs/nfs/pagelist.c
   3  *
   4  * A set of helper functions for managing NFS read and write requests.
   5  * The main purpose of these routines is to provide support for the
   6  * coalescing of several requests into a single RPC call.
   7  *
   8  * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
   9  *
  10  */
  11
  12 #include <linux/slab.h>
  13 #include <linux/file.h>
  14 #include <linux/sched.h>
  15 #include <linux/sunrpc/clnt.h>
  16 #include <linux/nfs.h>
  17 #include <linux/nfs3.h>
  18 #include <linux/nfs4.h>
  19 #include <linux/nfs_fs.h>
  20 #include <linux/nfs_page.h>
  21 #include <linux/nfs_mount.h>
  22 #include <linux/export.h>
  23
  24 #include "internal.h"
  25 #include "pnfs.h"
  26
  27 #define NFSDBG_FACILITY         NFSDBG_PAGECACHE
  28
  29 static struct kmem_cache *nfs_page_cachep;
  30 static const struct rpc_call_ops nfs_pgio_common_ops;
  31
  32 struct nfs_pgio_mirror *
  33 nfs_pgio_current_mirror(struct nfs_pageio_descriptor *desc)
  34 {
  35         return nfs_pgio_has_mirroring(desc) ?
  36                 &desc->pg_mirrors[desc->pg_mirror_idx] :
  37                 &desc->pg_mirrors[0];
  38 }
  39 EXPORT_SYMBOL_GPL(nfs_pgio_current_mirror);
  40
  41 void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
  42                        struct nfs_pgio_header *hdr,
  43                        void (*release)(struct nfs_pgio_header *hdr))
  44 {
  45         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
  46
  47
  48         hdr->req = nfs_list_entry(mirror->pg_list.next);
  49         hdr->inode = desc->pg_inode;
  50         hdr->cred = nfs_req_openctx(hdr->req)->cred;
  51         hdr->io_start = req_offset(hdr->req);
  52         hdr->good_bytes = mirror->pg_count;
  53         hdr->io_completion = desc->pg_io_completion;
  54         hdr->dreq = desc->pg_dreq;
  55         hdr->release = release;
  56         hdr->completion_ops = desc->pg_completion_ops;
  57         if (hdr->completion_ops->init_hdr)
  58                 hdr->completion_ops->init_hdr(hdr);
  59
  60         hdr->pgio_mirror_idx = desc->pg_mirror_idx;
  61 }
  62 EXPORT_SYMBOL_GPL(nfs_pgheader_init);
  63
  64 void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
  65 {
  66         unsigned int new = pos - hdr->io_start;
  67
  68         if (hdr->good_bytes > new) {
  69                 hdr->good_bytes = new;
  70                 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
  71                 if (!test_and_set_bit(NFS_IOHDR_ERROR, &hdr->flags))
  72                         hdr->error = error;
  73         }
  74 }
  75
  76 static inline struct nfs_page *
  77 nfs_page_alloc(void)
  78 {
  79         struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_NOIO);
  80         if (p)
  81                 INIT_LIST_HEAD(&p->wb_list);
  82         return p;
  83 }
  84
  85 static inline void
  86 nfs_page_free(struct nfs_page *p)
  87 {
  88         kmem_cache_free(nfs_page_cachep, p);
  89 }
  90
  91 /**
  92  * nfs_iocounter_wait - wait for i/o to complete
  93  * @l_ctx: nfs_lock_context with io_counter to use
  94  *
  95  * returns -ERESTARTSYS if interrupted by a fatal signal.
  96  * Otherwise returns 0 once the io_count hits 0.
  97  */
  98 int
  99 nfs_iocounter_wait(struct nfs_lock_context *l_ctx)
 100 {
 101         return wait_var_event_killable(&l_ctx->io_count,
 102                                        !atomic_read(&l_ctx->io_count));
 103 }
 104
 105 /**
 106  * nfs_async_iocounter_wait - wait on a rpc_waitqueue for I/O
 107  * to complete
 108  * @task: the rpc_task that should wait
 109  * @l_ctx: nfs_lock_context with io_counter to check
 110  *
 111  * Returns true if there is outstanding I/O to wait on and the
 112  * task has been put to sleep.
 113  */
 114 bool
 115 nfs_async_iocounter_wait(struct rpc_task *task, struct nfs_lock_context *l_ctx)
 116 {
 117         struct inode *inode = d_inode(l_ctx->open_context->dentry);
 118         bool ret = false;
 119
 120         if (atomic_read(&l_ctx->io_count) > 0) {
 121                 rpc_sleep_on(&NFS_SERVER(inode)->uoc_rpcwaitq, task, NULL);
 122                 ret = true;
 123         }
 124
 125         if (atomic_read(&l_ctx->io_count) == 0) {
 126                 rpc_wake_up_queued_task(&NFS_SERVER(inode)->uoc_rpcwaitq, task);
 127                 ret = false;
 128         }
 129
 130         return ret;
 131 }
 132 EXPORT_SYMBOL_GPL(nfs_async_iocounter_wait);
 133
 134 /*
 135  * nfs_page_group_lock - lock the head of the page group
 136  * @req - request in group that is to be locked
 137  *
 138  * this lock must be held when traversing or modifying the page
 139  * group list
 140  *
 141  * return 0 on success, < 0 on error
 142  */
 143 int
 144 nfs_page_group_lock(struct nfs_page *req)
 145 {
 146         struct nfs_page *head = req->wb_head;
 147
 148         WARN_ON_ONCE(head != head->wb_head);
 149
 150         if (!test_and_set_bit(PG_HEADLOCK, &head->wb_flags))
 151                 return 0;
 152
 153         set_bit(PG_CONTENDED1, &head->wb_flags);
 154         smp_mb__after_atomic();
 155         return wait_on_bit_lock(&head->wb_flags, PG_HEADLOCK,
 156                                 TASK_UNINTERRUPTIBLE);
 157 }
 158
 159 /*
 160  * nfs_page_group_unlock - unlock the head of the page group
 161  * @req - request in group that is to be unlocked
 162  */
 163 void
 164 nfs_page_group_unlock(struct nfs_page *req)
 165 {
 166         struct nfs_page *head = req->wb_head;
 167
 168         WARN_ON_ONCE(head != head->wb_head);
 169
 170         smp_mb__before_atomic();
 171         clear_bit(PG_HEADLOCK, &head->wb_flags);
 172         smp_mb__after_atomic();
 173         if (!test_bit(PG_CONTENDED1, &head->wb_flags))
 174                 return;
 175         wake_up_bit(&head->wb_flags, PG_HEADLOCK);
 176 }
 177
 178 /*
 179  * nfs_page_group_sync_on_bit_locked
 180  *
 181  * must be called with page group lock held
 182  */
 183 static bool
 184 nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
 185 {
 186         struct nfs_page *head = req->wb_head;
 187         struct nfs_page *tmp;
 188
 189         WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
 190         WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
 191
 192         tmp = req->wb_this_page;
 193         while (tmp != req) {
 194                 if (!test_bit(bit, &tmp->wb_flags))
 195                         return false;
 196                 tmp = tmp->wb_this_page;
 197         }
 198
 199         /* true! reset all bits */
 200         tmp = req;
 201         do {
 202                 clear_bit(bit, &tmp->wb_flags);
 203                 tmp = tmp->wb_this_page;
 204         } while (tmp != req);
 205
 206         return true;
 207 }
 208
 209 /*
 210  * nfs_page_group_sync_on_bit - set bit on current request, but only
 211  *   return true if the bit is set for all requests in page group
 212  * @req - request in page group
 213  * @bit - PG_* bit that is used to sync page group
 214  */
 215 bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
 216 {
 217         bool ret;
 218
 219         nfs_page_group_lock(req);
 220         ret = nfs_page_group_sync_on_bit_locked(req, bit);
 221         nfs_page_group_unlock(req);
 222
 223         return ret;
 224 }
 225
 226 /*
 227  * nfs_page_group_init - Initialize the page group linkage for @req
 228  * @req - a new nfs request
 229  * @prev - the previous request in page group, or NULL if @req is the first
 230  *         or only request in the group (the head).
 231  */
 232 static inline void
 233 nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
 234 {
 235         struct inode *inode;
 236         WARN_ON_ONCE(prev == req);
 237
 238         if (!prev) {
 239                 /* a head request */
 240                 req->wb_head = req;
 241                 req->wb_this_page = req;
 242         } else {
 243                 /* a subrequest */
 244                 WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
 245                 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
 246                 req->wb_head = prev->wb_head;
 247                 req->wb_this_page = prev->wb_this_page;
 248                 prev->wb_this_page = req;
 249
 250                 /* All subrequests take a ref on the head request until
 251                  * nfs_page_group_destroy is called */
 252                 kref_get(&req->wb_head->wb_kref);
 253
 254                 /* grab extra ref and bump the request count if head request
 255                  * has extra ref from the write/commit path to handle handoff
 256                  * between write and commit lists. */
 257                 if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) {
 258                         inode = page_file_mapping(req->wb_page)->host;
 259                         set_bit(PG_INODE_REF, &req->wb_flags);
 260                         kref_get(&req->wb_kref);
 261                         atomic_long_inc(&NFS_I(inode)->nrequests);
 262                 }
 263         }
 264 }
 265
 266 /*
 267  * nfs_page_group_destroy - sync the destruction of page groups
 268  * @req - request that no longer needs the page group
 269  *
 270  * releases the page group reference from each member once all
 271  * members have called this function.
 272  */
 273 static void
 274 nfs_page_group_destroy(struct kref *kref)
 275 {
 276         struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
 277         struct nfs_page *head = req->wb_head;
 278         struct nfs_page *tmp, *next;
 279
 280         if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
 281                 goto out;
 282
 283         tmp = req;
 284         do {
 285                 next = tmp->wb_this_page;
 286                 /* unlink and free */
 287                 tmp->wb_this_page = tmp;
 288                 tmp->wb_head = tmp;
 289                 nfs_free_request(tmp);
 290                 tmp = next;
 291         } while (tmp != req);
 292 out:
 293         /* subrequests must release the ref on the head request */
 294         if (head != req)
 295                 nfs_release_request(head);
 296 }
 297
 298 static struct nfs_page *
 299 __nfs_create_request(struct nfs_lock_context *l_ctx, struct page *page,
 300                    unsigned int pgbase, unsigned int offset,
 301                    unsigned int count)
 302 {
 303         struct nfs_page         *req;
 304         struct nfs_open_context *ctx = l_ctx->open_context;
 305
 306         if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
 307                 return ERR_PTR(-EBADF);
 308         /* try to allocate the request struct */
 309         req = nfs_page_alloc();
 310         if (req == NULL)
 311                 return ERR_PTR(-ENOMEM);
 312
 313         req->wb_lock_context = l_ctx;
 314         refcount_inc(&l_ctx->count);
 315         atomic_inc(&l_ctx->io_count);
 316
 317         /* Initialize the request struct. Initially, we assume a
 318          * long write-back delay. This will be adjusted in
 319          * update_nfs_request below if the region is not locked. */
 320         req->wb_page    = page;
 321         if (page) {
 322                 req->wb_index = page_index(page);
 323                 get_page(page);
 324         }
 325         req->wb_offset  = offset;
 326         req->wb_pgbase  = pgbase;
 327         req->wb_bytes   = count;
 328         kref_init(&req->wb_kref);
 329         req->wb_nio = 0;
 330         return req;
 331 }
 332
 333 /**
 334  * nfs_create_request - Create an NFS read/write request.
 335  * @ctx: open context to use
 336  * @page: page to write
 337  * @offset: starting offset within the page for the write
 338  * @count: number of bytes to read/write
 339  *
 340  * The page must be locked by the caller. This makes sure we never
 341  * create two different requests for the same page.
 342  * User should ensure it is safe to sleep in this function.
 343  */
 344 struct nfs_page *
 345 nfs_create_request(struct nfs_open_context *ctx, struct page *page,
 346                    unsigned int offset, unsigned int count)
 347 {
 348         struct nfs_lock_context *l_ctx = nfs_get_lock_context(ctx);
 349         struct nfs_page *ret;
 350
 351         if (IS_ERR(l_ctx))
 352                 return ERR_CAST(l_ctx);
 353         ret = __nfs_create_request(l_ctx, page, offset, offset, count);
 354         if (!IS_ERR(ret))
 355                 nfs_page_group_init(ret, NULL);
 356         nfs_put_lock_context(l_ctx);
 357         return ret;
 358 }
 359
 360 static struct nfs_page *
 361 nfs_create_subreq(struct nfs_page *req, struct nfs_page *last,
 362                   unsigned int pgbase, unsigned int offset,
 363                   unsigned int count)
 364 {
 365         struct nfs_page *ret;
 366
 367         ret = __nfs_create_request(req->wb_lock_context, req->wb_page,
 368                         pgbase, offset, count);
 369         if (!IS_ERR(ret)) {
 370                 nfs_lock_request(ret);
 371                 ret->wb_index = req->wb_index;
 372                 nfs_page_group_init(ret, last);
 373                 ret->wb_nio = req->wb_nio;
 374         }
 375         return ret;
 376 }
 377
 378 /**
 379  * nfs_unlock_request - Unlock request and wake up sleepers.
 380  * @req: pointer to request
 381  */
 382 void nfs_unlock_request(struct nfs_page *req)
 383 {
 384         if (!NFS_WBACK_BUSY(req)) {
 385                 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
 386                 BUG();
 387         }
 388         smp_mb__before_atomic();
 389         clear_bit(PG_BUSY, &req->wb_flags);
 390         smp_mb__after_atomic();
 391         if (!test_bit(PG_CONTENDED2, &req->wb_flags))
 392                 return;
 393         wake_up_bit(&req->wb_flags, PG_BUSY);
 394 }
 395
 396 /**
 397  * nfs_unlock_and_release_request - Unlock request and release the nfs_page
 398  * @req: pointer to request
 399  */
 400 void nfs_unlock_and_release_request(struct nfs_page *req)
 401 {
 402         nfs_unlock_request(req);
 403         nfs_release_request(req);
 404 }
 405
 406 /*
 407  * nfs_clear_request - Free up all resources allocated to the request
 408  * @req:
 409  *
 410  * Release page and open context resources associated with a read/write
 411  * request after it has completed.
 412  */
 413 static void nfs_clear_request(struct nfs_page *req)
 414 {
 415         struct page *page = req->wb_page;
 416         struct nfs_lock_context *l_ctx = req->wb_lock_context;
 417         struct nfs_open_context *ctx;
 418
 419         if (page != NULL) {
 420                 put_page(page);
 421                 req->wb_page = NULL;
 422         }
 423         if (l_ctx != NULL) {
 424                 if (atomic_dec_and_test(&l_ctx->io_count)) {
 425                         wake_up_var(&l_ctx->io_count);
 426                         ctx = l_ctx->open_context;
 427                         if (test_bit(NFS_CONTEXT_UNLOCK, &ctx->flags))
 428                                 rpc_wake_up(&NFS_SERVER(d_inode(ctx->dentry))->uoc_rpcwaitq);
 429                 }
 430                 nfs_put_lock_context(l_ctx);
 431                 req->wb_lock_context = NULL;
 432         }
 433 }
 434
 435 /**
 436  * nfs_release_request - Release the count on an NFS read/write request
 437  * @req: request to release
 438  *
 439  * Note: Should never be called with the spinlock held!
 440  */
 441 void nfs_free_request(struct nfs_page *req)
 442 {
 443         WARN_ON_ONCE(req->wb_this_page != req);
 444
 445         /* extra debug: make sure no sync bits are still set */
 446         WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
 447         WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
 448         WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
 449         WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
 450         WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
 451
 452         /* Release struct file and open context */
 453         nfs_clear_request(req);
 454         nfs_page_free(req);
 455 }
 456
 457 void nfs_release_request(struct nfs_page *req)
 458 {
 459         kref_put(&req->wb_kref, nfs_page_group_destroy);
 460 }
 461 EXPORT_SYMBOL_GPL(nfs_release_request);
 462
 463 /**
 464  * nfs_wait_on_request - Wait for a request to complete.
 465  * @req: request to wait upon.
 466  *
 467  * Interruptible by fatal signals only.
 468  * The user is responsible for holding a count on the request.
 469  */
 470 int
 471 nfs_wait_on_request(struct nfs_page *req)
 472 {
 473         if (!test_bit(PG_BUSY, &req->wb_flags))
 474                 return 0;
 475         set_bit(PG_CONTENDED2, &req->wb_flags);
 476         smp_mb__after_atomic();
 477         return wait_on_bit_io(&req->wb_flags, PG_BUSY,
 478                               TASK_UNINTERRUPTIBLE);
 479 }
 480 EXPORT_SYMBOL_GPL(nfs_wait_on_request);
 481
 482 /*
 483  * nfs_generic_pg_test - determine if requests can be coalesced
 484  * @desc: pointer to descriptor
 485  * @prev: previous request in desc, or NULL
 486  * @req: this request
 487  *
 488  * Returns zero if @req cannot be coalesced into @desc, otherwise it returns
 489  * the size of the request.
 490  */
 491 size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
 492                            struct nfs_page *prev, struct nfs_page *req)
 493 {
 494         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
 495
 496
 497         if (mirror->pg_count > mirror->pg_bsize) {
 498                 /* should never happen */
 499                 WARN_ON_ONCE(1);
 500                 return 0;
 501         }
 502
 503         /*
 504          * Limit the request size so that we can still allocate a page array
 505          * for it without upsetting the slab allocator.
 506          */
 507         if (((mirror->pg_count + req->wb_bytes) >> PAGE_SHIFT) *
 508                         sizeof(struct page *) > PAGE_SIZE)
 509                 return 0;
 510
 511         return min(mirror->pg_bsize - mirror->pg_count, (size_t)req->wb_bytes);
 512 }
 513 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
 514
 515 struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops)
 516 {
 517         struct nfs_pgio_header *hdr = ops->rw_alloc_header();
 518
 519         if (hdr) {
 520                 INIT_LIST_HEAD(&hdr->pages);
 521                 hdr->rw_ops = ops;
 522         }
 523         return hdr;
 524 }
 525 EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc);
 526
 527 /**
 528  * nfs_pgio_data_destroy - make @hdr suitable for reuse
 529  *
 530  * Frees memory and releases refs from nfs_generic_pgio, so that it may
 531  * be called again.
 532  *
 533  * @hdr: A header that has had nfs_generic_pgio called
 534  */
 535 static void nfs_pgio_data_destroy(struct nfs_pgio_header *hdr)
 536 {
 537         if (hdr->args.context)
 538                 put_nfs_open_context(hdr->args.context);
 539         if (hdr->page_array.pagevec != hdr->page_array.page_array)
 540                 kfree(hdr->page_array.pagevec);
 541 }
 542
 543 /*
 544  * nfs_pgio_header_free - Free a read or write header
 545  * @hdr: The header to free
 546  */
 547 void nfs_pgio_header_free(struct nfs_pgio_header *hdr)
 548 {
 549         nfs_pgio_data_destroy(hdr);
 550         hdr->rw_ops->rw_free_header(hdr);
 551 }
 552 EXPORT_SYMBOL_GPL(nfs_pgio_header_free);
 553
 554 /**
 555  * nfs_pgio_rpcsetup - Set up arguments for a pageio call
 556  * @hdr: The pageio hdr
 557  * @count: Number of bytes to read
 558  * @how: How to commit data (writes only)
 559  * @cinfo: Commit information for the call (writes only)
 560  */
 561 static void nfs_pgio_rpcsetup(struct nfs_pgio_header *hdr,
 562                               unsigned int count,
 563                               int how, struct nfs_commit_info *cinfo)
 564 {
 565         struct nfs_page *req = hdr->req;
 566
 567         /* Set up the RPC argument and reply structs
 568          * NB: take care not to mess about with hdr->commit et al. */
 569
 570         hdr->args.fh     = NFS_FH(hdr->inode);
 571         hdr->args.offset = req_offset(req);
 572         /* pnfs_set_layoutcommit needs this */
 573         hdr->mds_offset = hdr->args.offset;
 574         hdr->args.pgbase = req->wb_pgbase;
 575         hdr->args.pages  = hdr->page_array.pagevec;
 576         hdr->args.count  = count;
 577         hdr->args.context = get_nfs_open_context(nfs_req_openctx(req));
 578         hdr->args.lock_context = req->wb_lock_context;
 579         hdr->args.stable  = NFS_UNSTABLE;
 580         switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
 581         case 0:
 582                 break;
 583         case FLUSH_COND_STABLE:
 584                 if (nfs_reqs_to_commit(cinfo))
 585                         break;
 586                 /* fall through */
 587         default:
 588                 hdr->args.stable = NFS_FILE_SYNC;
 589         }
 590
 591         hdr->res.fattr   = &hdr->fattr;
 592         hdr->res.count   = count;
 593         hdr->res.eof     = 0;
 594         hdr->res.verf    = &hdr->verf;
 595         nfs_fattr_init(&hdr->fattr);
 596 }
 597
 598 /**
 599  * nfs_pgio_prepare - Prepare pageio hdr to go over the wire
 600  * @task: The current task
 601  * @calldata: pageio header to prepare
 602  */
 603 static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
 604 {
 605         struct nfs_pgio_header *hdr = calldata;
 606         int err;
 607         err = NFS_PROTO(hdr->inode)->pgio_rpc_prepare(task, hdr);
 608         if (err)
 609                 rpc_exit(task, err);
 610 }
 611
 612 int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
 613                       const struct cred *cred, const struct nfs_rpc_ops *rpc_ops,
 614                       const struct rpc_call_ops *call_ops, int how, int flags)
 615 {
 616         struct rpc_task *task;
 617         struct rpc_message msg = {
 618                 .rpc_argp = &hdr->args,
 619                 .rpc_resp = &hdr->res,
 620                 .rpc_cred = cred,
 621         };
 622         struct rpc_task_setup task_setup_data = {
 623                 .rpc_client = clnt,
 624                 .task = &hdr->task,
 625                 .rpc_message = &msg,
 626                 .callback_ops = call_ops,
 627                 .callback_data = hdr,
 628                 .workqueue = nfsiod_workqueue,
 629                 .flags = RPC_TASK_ASYNC | flags,
 630         };
 631         int ret = 0;
 632
 633         hdr->rw_ops->rw_initiate(hdr, &msg, rpc_ops, &task_setup_data, how);
 634
 635         dprintk("NFS: initiated pgio call "
 636                 "(req %s/%llu, %u bytes @ offset %llu)\n",
 637                 hdr->inode->i_sb->s_id,
 638                 (unsigned long long)NFS_FILEID(hdr->inode),
 639                 hdr->args.count,
 640                 (unsigned long long)hdr->args.offset);
 641
 642         task = rpc_run_task(&task_setup_data);
 643         if (IS_ERR(task)) {
 644                 ret = PTR_ERR(task);
 645                 goto out;
 646         }
 647         if (how & FLUSH_SYNC) {
 648                 ret = rpc_wait_for_completion_task(task);
 649                 if (ret == 0)
 650                         ret = task->tk_status;
 651         }
 652         rpc_put_task(task);
 653 out:
 654         return ret;
 655 }
 656 EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
 657
 658 /**
 659  * nfs_pgio_error - Clean up from a pageio error
 660  * @hdr: pageio header
 661  */
 662 static void nfs_pgio_error(struct nfs_pgio_header *hdr)
 663 {
 664         set_bit(NFS_IOHDR_REDO, &hdr->flags);
 665         hdr->completion_ops->completion(hdr);
 666 }
 667
 668 /**
 669  * nfs_pgio_release - Release pageio data
 670  * @calldata: The pageio header to release
 671  */
 672 static void nfs_pgio_release(void *calldata)
 673 {
 674         struct nfs_pgio_header *hdr = calldata;
 675         hdr->completion_ops->completion(hdr);
 676 }
 677
 678 static void nfs_pageio_mirror_init(struct nfs_pgio_mirror *mirror,
 679                                    unsigned int bsize)
 680 {
 681         INIT_LIST_HEAD(&mirror->pg_list);
 682         mirror->pg_bytes_written = 0;
 683         mirror->pg_count = 0;
 684         mirror->pg_bsize = bsize;
 685         mirror->pg_base = 0;
 686         mirror->pg_recoalesce = 0;
 687 }
 688
 689 /**
 690  * nfs_pageio_init - initialise a page io descriptor
 691  * @desc: pointer to descriptor
 692  * @inode: pointer to inode
 693  * @pg_ops: pointer to pageio operations
 694  * @compl_ops: pointer to pageio completion operations
 695  * @rw_ops: pointer to nfs read/write operations
 696  * @bsize: io block size
 697  * @io_flags: extra parameters for the io function
 698  */
 699 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
 700                      struct inode *inode,
 701                      const struct nfs_pageio_ops *pg_ops,
 702                      const struct nfs_pgio_completion_ops *compl_ops,
 703                      const struct nfs_rw_ops *rw_ops,
 704                      size_t bsize,
 705                      int io_flags)
 706 {
 707         desc->pg_moreio = 0;
 708         desc->pg_inode = inode;
 709         desc->pg_ops = pg_ops;
 710         desc->pg_completion_ops = compl_ops;
 711         desc->pg_rw_ops = rw_ops;
 712         desc->pg_ioflags = io_flags;
 713         desc->pg_error = 0;
 714         desc->pg_lseg = NULL;
 715         desc->pg_io_completion = NULL;
 716         desc->pg_dreq = NULL;
 717         desc->pg_bsize = bsize;
 718
 719         desc->pg_mirror_count = 1;
 720         desc->pg_mirror_idx = 0;
 721
 722         desc->pg_mirrors_dynamic = NULL;
 723         desc->pg_mirrors = desc->pg_mirrors_static;
 724         nfs_pageio_mirror_init(&desc->pg_mirrors[0], bsize);
 725         desc->pg_maxretrans = 0;
 726 }
 727
 728 /**
 729  * nfs_pgio_result - Basic pageio error handling
 730  * @task: The task that ran
 731  * @calldata: Pageio header to check
 732  */
 733 static void nfs_pgio_result(struct rpc_task *task, void *calldata)
 734 {
 735         struct nfs_pgio_header *hdr = calldata;
 736         struct inode *inode = hdr->inode;
 737
 738         dprintk("NFS: %s: %5u, (status %d)\n", __func__,
 739                 task->tk_pid, task->tk_status);
 740
 741         if (hdr->rw_ops->rw_done(task, hdr, inode) != 0)
 742                 return;
 743         if (task->tk_status < 0)
 744                 nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset);
 745         else
 746                 hdr->rw_ops->rw_result(task, hdr);
 747 }
 748
 749 /*
 750  * Create an RPC task for the given read or write request and kick it.
 751  * The page must have been locked by the caller.
 752  *
 753  * It may happen that the page we're passed is not marked dirty.
 754  * This is the case if nfs_updatepage detects a conflicting request
 755  * that has been written but not committed.
 756  */
 757 int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
 758                      struct nfs_pgio_header *hdr)
 759 {
 760         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
 761
 762         struct nfs_page         *req;
 763         struct page             **pages,
 764                                 *last_page;
 765         struct list_head *head = &mirror->pg_list;
 766         struct nfs_commit_info cinfo;
 767         struct nfs_page_array *pg_array = &hdr->page_array;
 768         unsigned int pagecount, pageused;
 769         gfp_t gfp_flags = GFP_KERNEL;
 770
 771         pagecount = nfs_page_array_len(mirror->pg_base, mirror->pg_count);
 772         pg_array->npages = pagecount;
 773
 774         if (pagecount <= ARRAY_SIZE(pg_array->page_array))
 775                 pg_array->pagevec = pg_array->page_array;
 776         else {
 777                 if (hdr->rw_mode == FMODE_WRITE)
 778                         gfp_flags = GFP_NOIO;
 779                 pg_array->pagevec = kcalloc(pagecount, sizeof(struct page *), gfp_flags);
 780                 if (!pg_array->pagevec) {
 781                         pg_array->npages = 0;
 782                         nfs_pgio_error(hdr);
 783                         desc->pg_error = -ENOMEM;
 784                         return desc->pg_error;
 785                 }
 786         }
 787
 788         nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
 789         pages = hdr->page_array.pagevec;
 790         last_page = NULL;
 791         pageused = 0;
 792         while (!list_empty(head)) {
 793                 req = nfs_list_entry(head->next);
 794                 nfs_list_move_request(req, &hdr->pages);
 795
 796                 if (!last_page || last_page != req->wb_page) {
 797                         pageused++;
 798                         if (pageused > pagecount)
 799                                 break;
 800                         *pages++ = last_page = req->wb_page;
 801                 }
 802         }
 803         if (WARN_ON_ONCE(pageused != pagecount)) {
 804                 nfs_pgio_error(hdr);
 805                 desc->pg_error = -EINVAL;
 806                 return desc->pg_error;
 807         }
 808
 809         if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
 810             (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
 811                 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
 812
 813         /* Set up the argument struct */
 814         nfs_pgio_rpcsetup(hdr, mirror->pg_count, desc->pg_ioflags, &cinfo);
 815         desc->pg_rpc_callops = &nfs_pgio_common_ops;
 816         return 0;
 817 }
 818 EXPORT_SYMBOL_GPL(nfs_generic_pgio);
 819
 820 static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
 821 {
 822         struct nfs_pgio_header *hdr;
 823         int ret;
 824
 825         hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
 826         if (!hdr) {
 827                 desc->pg_error = -ENOMEM;
 828                 return desc->pg_error;
 829         }
 830         nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
 831         ret = nfs_generic_pgio(desc, hdr);
 832         if (ret == 0)
 833                 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
 834                                         hdr,
 835                                         hdr->cred,
 836                                         NFS_PROTO(hdr->inode),
 837                                         desc->pg_rpc_callops,
 838                                         desc->pg_ioflags, 0);
 839         return ret;
 840 }
 841
 842 static struct nfs_pgio_mirror *
 843 nfs_pageio_alloc_mirrors(struct nfs_pageio_descriptor *desc,
 844                 unsigned int mirror_count)
 845 {
 846         struct nfs_pgio_mirror *ret;
 847         unsigned int i;
 848
 849         kfree(desc->pg_mirrors_dynamic);
 850         desc->pg_mirrors_dynamic = NULL;
 851         if (mirror_count == 1)
 852                 return desc->pg_mirrors_static;
 853         ret = kmalloc_array(mirror_count, sizeof(*ret), GFP_NOFS);
 854         if (ret != NULL) {
 855                 for (i = 0; i < mirror_count; i++)
 856                         nfs_pageio_mirror_init(&ret[i], desc->pg_bsize);
 857                 desc->pg_mirrors_dynamic = ret;
 858         }
 859         return ret;
 860 }
 861
 862 /*
 863  * nfs_pageio_setup_mirroring - determine if mirroring is to be used
 864  *                              by calling the pg_get_mirror_count op
 865  */
 866 static void nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor *pgio,
 867                                        struct nfs_page *req)
 868 {
 869         unsigned int mirror_count = 1;
 870
 871         if (pgio->pg_ops->pg_get_mirror_count)
 872                 mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req);
 873         if (mirror_count == pgio->pg_mirror_count || pgio->pg_error < 0)
 874                 return;
 875
 876         if (!mirror_count || mirror_count > NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX) {
 877                 pgio->pg_error = -EINVAL;
 878                 return;
 879         }
 880
 881         pgio->pg_mirrors = nfs_pageio_alloc_mirrors(pgio, mirror_count);
 882         if (pgio->pg_mirrors == NULL) {
 883                 pgio->pg_error = -ENOMEM;
 884                 pgio->pg_mirrors = pgio->pg_mirrors_static;
 885                 mirror_count = 1;
 886         }
 887         pgio->pg_mirror_count = mirror_count;
 888 }
 889
 890 /*
 891  * nfs_pageio_stop_mirroring - stop using mirroring (set mirror count to 1)
 892  */
 893 void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio)
 894 {
 895         pgio->pg_mirror_count = 1;
 896         pgio->pg_mirror_idx = 0;
 897 }
 898
 899 static void nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor *pgio)
 900 {
 901         pgio->pg_mirror_count = 1;
 902         pgio->pg_mirror_idx = 0;
 903         pgio->pg_mirrors = pgio->pg_mirrors_static;
 904         kfree(pgio->pg_mirrors_dynamic);
 905         pgio->pg_mirrors_dynamic = NULL;
 906 }
 907
 908 static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
 909                 const struct nfs_lock_context *l2)
 910 {
 911         return l1->lockowner == l2->lockowner;
 912 }
 913
 914 /**
 915  * nfs_can_coalesce_requests - test two requests for compatibility
 916  * @prev: pointer to nfs_page
 917  * @req: pointer to nfs_page
 918  * @pgio: pointer to nfs_pagio_descriptor
 919  *
 920  * The nfs_page structures 'prev' and 'req' are compared to ensure that the
 921  * page data area they describe is contiguous, and that their RPC
 922  * credentials, NFSv4 open state, and lockowners are the same.
 923  *
 924  * Return 'true' if this is the case, else return 'false'.
 925  */
 926 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
 927                                       struct nfs_page *req,
 928                                       struct nfs_pageio_descriptor *pgio)
 929 {
 930         size_t size;
 931         struct file_lock_context *flctx;
 932
 933         if (prev) {
 934                 if (!nfs_match_open_context(nfs_req_openctx(req), nfs_req_openctx(prev)))
 935                         return false;
 936                 flctx = d_inode(nfs_req_openctx(req)->dentry)->i_flctx;
 937                 if (flctx != NULL &&
 938                     !(list_empty_careful(&flctx->flc_posix) &&
 939                       list_empty_careful(&flctx->flc_flock)) &&
 940                     !nfs_match_lock_context(req->wb_lock_context,
 941                                             prev->wb_lock_context))
 942                         return false;
 943                 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
 944                         return false;
 945                 if (req->wb_page == prev->wb_page) {
 946                         if (req->wb_pgbase != prev->wb_pgbase + prev->wb_bytes)
 947                                 return false;
 948                 } else {
 949                         if (req->wb_pgbase != 0 ||
 950                             prev->wb_pgbase + prev->wb_bytes != PAGE_SIZE)
 951                                 return false;
 952                 }
 953         }
 954         size = pgio->pg_ops->pg_test(pgio, prev, req);
 955         WARN_ON_ONCE(size > req->wb_bytes);
 956         if (size && size < req->wb_bytes)
 957                 req->wb_bytes = size;
 958         return size > 0;
 959 }
 960
 961 /**
 962  * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
 963  * @desc: destination io descriptor
 964  * @req: request
 965  *
 966  * Returns true if the request 'req' was successfully coalesced into the
 967  * existing list of pages 'desc'.
 968  */
 969 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
 970                                      struct nfs_page *req)
 971 {
 972         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
 973
 974         struct nfs_page *prev = NULL;
 975
 976         if (mirror->pg_count != 0) {
 977                 prev = nfs_list_entry(mirror->pg_list.prev);
 978         } else {
 979                 if (desc->pg_ops->pg_init)
 980                         desc->pg_ops->pg_init(desc, req);
 981                 if (desc->pg_error < 0)
 982                         return 0;
 983                 mirror->pg_base = req->wb_pgbase;
 984         }
 985
 986         if (desc->pg_maxretrans && req->wb_nio > desc->pg_maxretrans) {
 987                 if (NFS_SERVER(desc->pg_inode)->flags & NFS_MOUNT_SOFTERR)
 988                         desc->pg_error = -ETIMEDOUT;
 989                 else
 990                         desc->pg_error = -EIO;
 991                 return 0;
 992         }
 993
 994         if (!nfs_can_coalesce_requests(prev, req, desc))
 995                 return 0;
 996         nfs_list_move_request(req, &mirror->pg_list);
 997         mirror->pg_count += req->wb_bytes;
 998         return 1;
 999 }
1000
1001 /*
1002  * Helper for nfs_pageio_add_request and nfs_pageio_complete
1003  */
1004 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
1005 {
1006         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1007
1008
1009         if (!list_empty(&mirror->pg_list)) {
1010                 int error = desc->pg_ops->pg_doio(desc);
1011                 if (error < 0)
1012                         desc->pg_error = error;
1013                 else
1014                         mirror->pg_bytes_written += mirror->pg_count;
1015         }
1016         if (list_empty(&mirror->pg_list)) {
1017                 mirror->pg_count = 0;
1018                 mirror->pg_base = 0;
1019         }
1020 }
1021
1022 static void
1023 nfs_pageio_cleanup_request(struct nfs_pageio_descriptor *desc,
1024                 struct nfs_page *req)
1025 {
1026         LIST_HEAD(head);
1027
1028         nfs_list_move_request(req, &head);
1029         desc->pg_completion_ops->error_cleanup(&head, desc->pg_error);
1030 }
1031
1032 /**
1033  * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
1034  * @desc: destination io descriptor
1035  * @req: request
1036  *
1037  * This may split a request into subrequests which are all part of the
1038  * same page group.
1039  *
1040  * Returns true if the request 'req' was successfully coalesced into the
1041  * existing list of pages 'desc'.
1042  */
1043 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1044                            struct nfs_page *req)
1045 {
1046         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1047
1048         struct nfs_page *subreq;
1049         unsigned int bytes_left = 0;
1050         unsigned int offset, pgbase;
1051
1052         nfs_page_group_lock(req);
1053
1054         subreq = req;
1055         bytes_left = subreq->wb_bytes;
1056         offset = subreq->wb_offset;
1057         pgbase = subreq->wb_pgbase;
1058
1059         do {
1060                 if (!nfs_pageio_do_add_request(desc, subreq)) {
1061                         /* make sure pg_test call(s) did nothing */
1062                         WARN_ON_ONCE(subreq->wb_bytes != bytes_left);
1063                         WARN_ON_ONCE(subreq->wb_offset != offset);
1064                         WARN_ON_ONCE(subreq->wb_pgbase != pgbase);
1065
1066                         nfs_page_group_unlock(req);
1067                         desc->pg_moreio = 1;
1068                         nfs_pageio_doio(desc);
1069                         if (desc->pg_error < 0 || mirror->pg_recoalesce)
1070                                 goto out_cleanup_subreq;
1071                         /* retry add_request for this subreq */
1072                         nfs_page_group_lock(req);
1073                         continue;
1074                 }
1075
1076                 /* check for buggy pg_test call(s) */
1077                 WARN_ON_ONCE(subreq->wb_bytes + subreq->wb_pgbase > PAGE_SIZE);
1078                 WARN_ON_ONCE(subreq->wb_bytes > bytes_left);
1079                 WARN_ON_ONCE(subreq->wb_bytes == 0);
1080
1081                 bytes_left -= subreq->wb_bytes;
1082                 offset += subreq->wb_bytes;
1083                 pgbase += subreq->wb_bytes;
1084
1085                 if (bytes_left) {
1086                         subreq = nfs_create_subreq(req, subreq, pgbase,
1087                                         offset, bytes_left);
1088                         if (IS_ERR(subreq))
1089                                 goto err_ptr;
1090                 }
1091         } while (bytes_left > 0);
1092
1093         nfs_page_group_unlock(req);
1094         return 1;
1095 err_ptr:
1096         desc->pg_error = PTR_ERR(subreq);
1097         nfs_page_group_unlock(req);
1098         return 0;
1099 out_cleanup_subreq:
1100         if (req != subreq)
1101                 nfs_pageio_cleanup_request(desc, subreq);
1102         return 0;
1103 }
1104
1105 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
1106 {
1107         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1108         LIST_HEAD(head);
1109
1110         do {
1111                 list_splice_init(&mirror->pg_list, &head);
1112                 mirror->pg_bytes_written -= mirror->pg_count;
1113                 mirror->pg_count = 0;
1114                 mirror->pg_base = 0;
1115                 mirror->pg_recoalesce = 0;
1116
1117                 while (!list_empty(&head)) {
1118                         struct nfs_page *req;
1119
1120                         req = list_first_entry(&head, struct nfs_page, wb_list);
1121                         if (__nfs_pageio_add_request(desc, req))
1122                                 continue;
1123                         if (desc->pg_error < 0) {
1124                                 list_splice_tail(&head, &mirror->pg_list);
1125                                 mirror->pg_recoalesce = 1;
1126                                 return 0;
1127                         }
1128                         break;
1129                 }
1130         } while (mirror->pg_recoalesce);
1131         return 1;
1132 }
1133
1134 static int nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor *desc,
1135                 struct nfs_page *req)
1136 {
1137         int ret;
1138
1139         do {
1140                 ret = __nfs_pageio_add_request(desc, req);
1141                 if (ret)
1142                         break;
1143                 if (desc->pg_error < 0)
1144                         break;
1145                 ret = nfs_do_recoalesce(desc);
1146         } while (ret);
1147
1148         return ret;
1149 }
1150
1151 static void nfs_pageio_error_cleanup(struct nfs_pageio_descriptor *desc)
1152 {
1153         u32 midx;
1154         struct nfs_pgio_mirror *mirror;
1155
1156         if (!desc->pg_error)
1157                 return;
1158
1159         for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1160                 mirror = &desc->pg_mirrors[midx];
1161                 desc->pg_completion_ops->error_cleanup(&mirror->pg_list,
1162                                 desc->pg_error);
1163         }
1164 }
1165
1166 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1167                            struct nfs_page *req)
1168 {
1169         u32 midx;
1170         unsigned int pgbase, offset, bytes;
1171         struct nfs_page *dupreq, *lastreq;
1172
1173         pgbase = req->wb_pgbase;
1174         offset = req->wb_offset;
1175         bytes = req->wb_bytes;
1176
1177         nfs_pageio_setup_mirroring(desc, req);
1178         if (desc->pg_error < 0)
1179                 goto out_failed;
1180
1181         for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1182                 if (midx) {
1183                         nfs_page_group_lock(req);
1184
1185                         /* find the last request */
1186                         for (lastreq = req->wb_head;
1187                              lastreq->wb_this_page != req->wb_head;
1188                              lastreq = lastreq->wb_this_page)
1189                                 ;
1190
1191                         dupreq = nfs_create_subreq(req, lastreq,
1192                                         pgbase, offset, bytes);
1193
1194                         nfs_page_group_unlock(req);
1195                         if (IS_ERR(dupreq)) {
1196                                 desc->pg_error = PTR_ERR(dupreq);
1197                                 goto out_failed;
1198                         }
1199                 } else
1200                         dupreq = req;
1201
1202                 if (nfs_pgio_has_mirroring(desc))
1203                         desc->pg_mirror_idx = midx;
1204                 if (!nfs_pageio_add_request_mirror(desc, dupreq))
1205                         goto out_cleanup_subreq;
1206         }
1207
1208         return 1;
1209
1210 out_cleanup_subreq:
1211         if (req != dupreq)
1212                 nfs_pageio_cleanup_request(desc, dupreq);
1213 out_failed:
1214         nfs_pageio_error_cleanup(desc);
1215         return 0;
1216 }
1217
1218 /*
1219  * nfs_pageio_complete_mirror - Complete I/O on the current mirror of an
1220  *                              nfs_pageio_descriptor
1221  * @desc: pointer to io descriptor
1222  * @mirror_idx: pointer to mirror index
1223  */
1224 static void nfs_pageio_complete_mirror(struct nfs_pageio_descriptor *desc,
1225                                        u32 mirror_idx)
1226 {
1227         struct nfs_pgio_mirror *mirror = &desc->pg_mirrors[mirror_idx];
1228         u32 restore_idx = desc->pg_mirror_idx;
1229
1230         if (nfs_pgio_has_mirroring(desc))
1231                 desc->pg_mirror_idx = mirror_idx;
1232         for (;;) {
1233                 nfs_pageio_doio(desc);
1234                 if (desc->pg_error < 0 || !mirror->pg_recoalesce)
1235                         break;
1236                 if (!nfs_do_recoalesce(desc))
1237                         break;
1238         }
1239         desc->pg_mirror_idx = restore_idx;
1240 }
1241
1242 /*
1243  * nfs_pageio_resend - Transfer requests to new descriptor and resend
1244  * @hdr - the pgio header to move request from
1245  * @desc - the pageio descriptor to add requests to
1246  *
1247  * Try to move each request (nfs_page) from @hdr to @desc then attempt
1248  * to send them.
1249  *
1250  * Returns 0 on success and < 0 on error.
1251  */
1252 int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
1253                       struct nfs_pgio_header *hdr)
1254 {
1255         LIST_HEAD(failed);
1256
1257         desc->pg_io_completion = hdr->io_completion;
1258         desc->pg_dreq = hdr->dreq;
1259         while (!list_empty(&hdr->pages)) {
1260                 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
1261
1262                 if (!nfs_pageio_add_request(desc, req))
1263                         nfs_list_move_request(req, &failed);
1264         }
1265         nfs_pageio_complete(desc);
1266         if (!list_empty(&failed)) {
1267                 list_move(&failed, &hdr->pages);
1268                 return desc->pg_error < 0 ? desc->pg_error : -EIO;
1269         }
1270         return 0;
1271 }
1272 EXPORT_SYMBOL_GPL(nfs_pageio_resend);
1273
1274 /**
1275  * nfs_pageio_complete - Complete I/O then cleanup an nfs_pageio_descriptor
1276  * @desc: pointer to io descriptor
1277  */
1278 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1279 {
1280         u32 midx;
1281
1282         for (midx = 0; midx < desc->pg_mirror_count; midx++)
1283                 nfs_pageio_complete_mirror(desc, midx);
1284
1285         if (desc->pg_error < 0)
1286                 nfs_pageio_error_cleanup(desc);
1287         if (desc->pg_ops->pg_cleanup)
1288                 desc->pg_ops->pg_cleanup(desc);
1289         nfs_pageio_cleanup_mirroring(desc);
1290 }
1291
1292 /**
1293  * nfs_pageio_cond_complete - Conditional I/O completion
1294  * @desc: pointer to io descriptor
1295  * @index: page index
1296  *
1297  * It is important to ensure that processes don't try to take locks
1298  * on non-contiguous ranges of pages as that might deadlock. This
1299  * function should be called before attempting to wait on a locked
1300  * nfs_page. It will complete the I/O if the page index 'index'
1301  * is not contiguous with the existing list of pages in 'desc'.
1302  */
1303 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1304 {
1305         struct nfs_pgio_mirror *mirror;
1306         struct nfs_page *prev;
1307         u32 midx;
1308
1309         for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1310                 mirror = &desc->pg_mirrors[midx];
1311                 if (!list_empty(&mirror->pg_list)) {
1312                         prev = nfs_list_entry(mirror->pg_list.prev);
1313                         if (index != prev->wb_index + 1) {
1314                                 nfs_pageio_complete(desc);
1315                                 break;
1316                         }
1317                 }
1318         }
1319 }
1320
1321 int __init nfs_init_nfspagecache(void)
1322 {
1323         nfs_page_cachep = kmem_cache_create("nfs_page",
1324                                             sizeof(struct nfs_page),
1325                                             0, SLAB_HWCACHE_ALIGN,
1326                                             NULL);
1327         if (nfs_page_cachep == NULL)
1328                 return -ENOMEM;
1329
1330         return 0;
1331 }
1332
1333 void nfs_destroy_nfspagecache(void)
1334 {
1335         kmem_cache_destroy(nfs_page_cachep);
1336 }
1337
1338 static const struct rpc_call_ops nfs_pgio_common_ops = {
1339         .rpc_call_prepare = nfs_pgio_prepare,
1340         .rpc_call_done = nfs_pgio_result,
1341         .rpc_release = nfs_pgio_release,
1342 };
1343
1344 const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1345         .pg_test = nfs_generic_pg_test,
1346         .pg_doio = nfs_generic_pg_pgios,
1347 };