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_page.h>
  20 #include <linux/nfs_fs.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 = hdr->req->wb_context->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         spin_lock(&hdr->lock);
  67         if (!test_and_set_bit(NFS_IOHDR_ERROR, &hdr->flags)
  68             || pos < hdr->io_start + hdr->good_bytes) {
  69                 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
  70                 hdr->good_bytes = pos - hdr->io_start;
  71                 hdr->error = error;
  72         }
  73         spin_unlock(&hdr->lock);
  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_set_headlock - set the request PG_HEADLOCK
 136  * @req: request that is to be locked
 137  *
 138  * this lock must be held when modifying req->wb_head
 139  *
 140  * return 0 on success, < 0 on error
 141  */
 142 int
 143 nfs_page_set_headlock(struct nfs_page *req)
 144 {
 145         if (!test_and_set_bit(PG_HEADLOCK, &req->wb_flags))
 146                 return 0;
 147
 148         set_bit(PG_CONTENDED1, &req->wb_flags);
 149         smp_mb__after_atomic();
 150         return wait_on_bit_lock(&req->wb_flags, PG_HEADLOCK,
 151                                 TASK_UNINTERRUPTIBLE);
 152 }
 153
 154 /*
 155  * nfs_page_clear_headlock - clear the request PG_HEADLOCK
 156  * @req: request that is to be locked
 157  */
 158 void
 159 nfs_page_clear_headlock(struct nfs_page *req)
 160 {
 161         smp_mb__before_atomic();
 162         clear_bit(PG_HEADLOCK, &req->wb_flags);
 163         smp_mb__after_atomic();
 164         if (!test_bit(PG_CONTENDED1, &req->wb_flags))
 165                 return;
 166         wake_up_bit(&req->wb_flags, PG_HEADLOCK);
 167 }
 168
 169 /*
 170  * nfs_page_group_lock - lock the head of the page group
 171  * @req: request in group that is to be locked
 172  *
 173  * this lock must be held when traversing or modifying the page
 174  * group list
 175  *
 176  * return 0 on success, < 0 on error
 177  */
 178 int
 179 nfs_page_group_lock(struct nfs_page *req)
 180 {
 181         int ret;
 182
 183         ret = nfs_page_set_headlock(req);
 184         if (ret || req->wb_head == req)
 185                 return ret;
 186         return nfs_page_set_headlock(req->wb_head);
 187 }
 188
 189 /*
 190  * nfs_page_group_unlock - unlock the head of the page group
 191  * @req: request in group that is to be unlocked
 192  */
 193 void
 194 nfs_page_group_unlock(struct nfs_page *req)
 195 {
 196         if (req != req->wb_head)
 197                 nfs_page_clear_headlock(req->wb_head);
 198         nfs_page_clear_headlock(req);
 199 }
 200
 201 /*
 202  * nfs_page_group_sync_on_bit_locked
 203  *
 204  * must be called with page group lock held
 205  */
 206 static bool
 207 nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
 208 {
 209         struct nfs_page *head = req->wb_head;
 210         struct nfs_page *tmp;
 211
 212         WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
 213         WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
 214
 215         tmp = req->wb_this_page;
 216         while (tmp != req) {
 217                 if (!test_bit(bit, &tmp->wb_flags))
 218                         return false;
 219                 tmp = tmp->wb_this_page;
 220         }
 221
 222         /* true! reset all bits */
 223         tmp = req;
 224         do {
 225                 clear_bit(bit, &tmp->wb_flags);
 226                 tmp = tmp->wb_this_page;
 227         } while (tmp != req);
 228
 229         return true;
 230 }
 231
 232 /*
 233  * nfs_page_group_sync_on_bit - set bit on current request, but only
 234  *   return true if the bit is set for all requests in page group
 235  * @req - request in page group
 236  * @bit - PG_* bit that is used to sync page group
 237  */
 238 bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
 239 {
 240         bool ret;
 241
 242         nfs_page_group_lock(req);
 243         ret = nfs_page_group_sync_on_bit_locked(req, bit);
 244         nfs_page_group_unlock(req);
 245
 246         return ret;
 247 }
 248
 249 /*
 250  * nfs_page_group_init - Initialize the page group linkage for @req
 251  * @req - a new nfs request
 252  * @prev - the previous request in page group, or NULL if @req is the first
 253  *         or only request in the group (the head).
 254  */
 255 static inline void
 256 nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
 257 {
 258         struct inode *inode;
 259         WARN_ON_ONCE(prev == req);
 260
 261         if (!prev) {
 262                 /* a head request */
 263                 req->wb_head = req;
 264                 req->wb_this_page = req;
 265         } else {
 266                 /* a subrequest */
 267                 WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
 268                 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
 269                 req->wb_head = prev->wb_head;
 270                 req->wb_this_page = prev->wb_this_page;
 271                 prev->wb_this_page = req;
 272
 273                 /* All subrequests take a ref on the head request until
 274                  * nfs_page_group_destroy is called */
 275                 kref_get(&req->wb_head->wb_kref);
 276
 277                 /* grab extra ref and bump the request count if head request
 278                  * has extra ref from the write/commit path to handle handoff
 279                  * between write and commit lists. */
 280                 if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) {
 281                         inode = page_file_mapping(req->wb_page)->host;
 282                         set_bit(PG_INODE_REF, &req->wb_flags);
 283                         kref_get(&req->wb_kref);
 284                         atomic_long_inc(&NFS_I(inode)->nrequests);
 285                 }
 286         }
 287 }
 288
 289 /*
 290  * nfs_page_group_destroy - sync the destruction of page groups
 291  * @req - request that no longer needs the page group
 292  *
 293  * releases the page group reference from each member once all
 294  * members have called this function.
 295  */
 296 static void
 297 nfs_page_group_destroy(struct kref *kref)
 298 {
 299         struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
 300         struct nfs_page *head = req->wb_head;
 301         struct nfs_page *tmp, *next;
 302
 303         if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
 304                 goto out;
 305
 306         tmp = req;
 307         do {
 308                 next = tmp->wb_this_page;
 309                 /* unlink and free */
 310                 tmp->wb_this_page = tmp;
 311                 tmp->wb_head = tmp;
 312                 nfs_free_request(tmp);
 313                 tmp = next;
 314         } while (tmp != req);
 315 out:
 316         /* subrequests must release the ref on the head request */
 317         if (head != req)
 318                 nfs_release_request(head);
 319 }
 320
 321 /**
 322  * nfs_create_request - Create an NFS read/write request.
 323  * @ctx: open context to use
 324  * @page: page to write
 325  * @last: last nfs request created for this page group or NULL if head
 326  * @offset: starting offset within the page for the write
 327  * @count: number of bytes to read/write
 328  *
 329  * The page must be locked by the caller. This makes sure we never
 330  * create two different requests for the same page.
 331  * User should ensure it is safe to sleep in this function.
 332  */
 333 struct nfs_page *
 334 nfs_create_request(struct nfs_open_context *ctx, struct page *page,
 335                    struct nfs_page *last, unsigned int offset,
 336                    unsigned int count)
 337 {
 338         struct nfs_page         *req;
 339         struct nfs_lock_context *l_ctx;
 340
 341         if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
 342                 return ERR_PTR(-EBADF);
 343         /* try to allocate the request struct */
 344         req = nfs_page_alloc();
 345         if (req == NULL)
 346                 return ERR_PTR(-ENOMEM);
 347
 348         /* get lock context early so we can deal with alloc failures */
 349         l_ctx = nfs_get_lock_context(ctx);
 350         if (IS_ERR(l_ctx)) {
 351                 nfs_page_free(req);
 352                 return ERR_CAST(l_ctx);
 353         }
 354         req->wb_lock_context = l_ctx;
 355         atomic_inc(&l_ctx->io_count);
 356
 357         /* Initialize the request struct. Initially, we assume a
 358          * long write-back delay. This will be adjusted in
 359          * update_nfs_request below if the region is not locked. */
 360         req->wb_page    = page;
 361         if (page) {
 362                 req->wb_index = page_index(page);
 363                 get_page(page);
 364         }
 365         req->wb_offset  = offset;
 366         req->wb_pgbase  = offset;
 367         req->wb_bytes   = count;
 368         req->wb_context = get_nfs_open_context(ctx);
 369         kref_init(&req->wb_kref);
 370         nfs_page_group_init(req, last);
 371         return req;
 372 }
 373
 374 /**
 375  * nfs_unlock_request - Unlock request and wake up sleepers.
 376  * @req:
 377  */
 378 void nfs_unlock_request(struct nfs_page *req)
 379 {
 380         if (!NFS_WBACK_BUSY(req)) {
 381                 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
 382                 BUG();
 383         }
 384         smp_mb__before_atomic();
 385         clear_bit(PG_BUSY, &req->wb_flags);
 386         smp_mb__after_atomic();
 387         if (!test_bit(PG_CONTENDED2, &req->wb_flags))
 388                 return;
 389         wake_up_bit(&req->wb_flags, PG_BUSY);
 390 }
 391
 392 /**
 393  * nfs_unlock_and_release_request - Unlock request and release the nfs_page
 394  * @req:
 395  */
 396 void nfs_unlock_and_release_request(struct nfs_page *req)
 397 {
 398         nfs_unlock_request(req);
 399         nfs_release_request(req);
 400 }
 401
 402 /*
 403  * nfs_clear_request - Free up all resources allocated to the request
 404  * @req:
 405  *
 406  * Release page and open context resources associated with a read/write
 407  * request after it has completed.
 408  */
 409 static void nfs_clear_request(struct nfs_page *req)
 410 {
 411         struct page *page = req->wb_page;
 412         struct nfs_open_context *ctx = req->wb_context;
 413         struct nfs_lock_context *l_ctx = req->wb_lock_context;
 414
 415         if (page != NULL) {
 416                 put_page(page);
 417                 req->wb_page = NULL;
 418         }
 419         if (l_ctx != NULL) {
 420                 if (atomic_dec_and_test(&l_ctx->io_count)) {
 421                         wake_up_var(&l_ctx->io_count);
 422                         if (test_bit(NFS_CONTEXT_UNLOCK, &ctx->flags))
 423                                 rpc_wake_up(&NFS_SERVER(d_inode(ctx->dentry))->uoc_rpcwaitq);
 424                 }
 425                 nfs_put_lock_context(l_ctx);
 426                 req->wb_lock_context = NULL;
 427         }
 428         if (ctx != NULL) {
 429                 put_nfs_open_context(ctx);
 430                 req->wb_context = NULL;
 431         }
 432 }
 433
 434 /**
 435  * nfs_release_request - Release the count on an NFS read/write request
 436  * @req: request to release
 437  *
 438  * Note: Should never be called with the spinlock held!
 439  */
 440 void nfs_free_request(struct nfs_page *req)
 441 {
 442         WARN_ON_ONCE(req->wb_this_page != req);
 443
 444         /* extra debug: make sure no sync bits are still set */
 445         WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
 446         WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
 447         WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
 448         WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
 449         WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
 450
 451         /* Release struct file and open context */
 452         nfs_clear_request(req);
 453         nfs_page_free(req);
 454 }
 455
 456 void nfs_release_request(struct nfs_page *req)
 457 {
 458         kref_put(&req->wb_kref, nfs_page_group_destroy);
 459 }
 460 EXPORT_SYMBOL_GPL(nfs_release_request);
 461
 462 /**
 463  * nfs_wait_on_request - Wait for a request to complete.
 464  * @req: request to wait upon.
 465  *
 466  * Interruptible by fatal signals only.
 467  * The user is responsible for holding a count on the request.
 468  */
 469 int
 470 nfs_wait_on_request(struct nfs_page *req)
 471 {
 472         if (!test_bit(PG_BUSY, &req->wb_flags))
 473                 return 0;
 474         set_bit(PG_CONTENDED2, &req->wb_flags);
 475         smp_mb__after_atomic();
 476         return wait_on_bit_io(&req->wb_flags, PG_BUSY,
 477                               TASK_UNINTERRUPTIBLE);
 478 }
 479 EXPORT_SYMBOL_GPL(nfs_wait_on_request);
 480
 481 /*
 482  * nfs_generic_pg_test - determine if requests can be coalesced
 483  * @desc: pointer to descriptor
 484  * @prev: previous request in desc, or NULL
 485  * @req: this request
 486  *
 487  * Returns zero if @req can be coalesced into @desc, otherwise it returns
 488  * the size of the request.
 489  */
 490 size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
 491                            struct nfs_page *prev, struct nfs_page *req)
 492 {
 493         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
 494
 495
 496         if (mirror->pg_count > mirror->pg_bsize) {
 497                 /* should never happen */
 498                 WARN_ON_ONCE(1);
 499                 return 0;
 500         }
 501
 502         /*
 503          * Limit the request size so that we can still allocate a page array
 504          * for it without upsetting the slab allocator.
 505          */
 506         if (((mirror->pg_count + req->wb_bytes) >> PAGE_SHIFT) *
 507                         sizeof(struct page *) > PAGE_SIZE)
 508                 return 0;
 509
 510         return min(mirror->pg_bsize - mirror->pg_count, (size_t)req->wb_bytes);
 511 }
 512 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
 513
 514 struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops)
 515 {
 516         struct nfs_pgio_header *hdr = ops->rw_alloc_header();
 517
 518         if (hdr) {
 519                 INIT_LIST_HEAD(&hdr->pages);
 520                 spin_lock_init(&hdr->lock);
 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  * @offset: Initial offset
 559  * @how: How to commit data (writes only)
 560  * @cinfo: Commit information for the call (writes only)
 561  */
 562 static void nfs_pgio_rpcsetup(struct nfs_pgio_header *hdr,
 563                               unsigned int count,
 564                               int how, struct nfs_commit_info *cinfo)
 565 {
 566         struct nfs_page *req = hdr->req;
 567
 568         /* Set up the RPC argument and reply structs
 569          * NB: take care not to mess about with hdr->commit et al. */
 570
 571         hdr->args.fh     = NFS_FH(hdr->inode);
 572         hdr->args.offset = req_offset(req);
 573         /* pnfs_set_layoutcommit needs this */
 574         hdr->mds_offset = hdr->args.offset;
 575         hdr->args.pgbase = req->wb_pgbase;
 576         hdr->args.pages  = hdr->page_array.pagevec;
 577         hdr->args.count  = count;
 578         hdr->args.context = get_nfs_open_context(req->wb_context);
 579         hdr->args.lock_context = req->wb_lock_context;
 580         hdr->args.stable  = NFS_UNSTABLE;
 581         switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
 582         case 0:
 583                 break;
 584         case FLUSH_COND_STABLE:
 585                 if (nfs_reqs_to_commit(cinfo))
 586                         break;
 587                 /* fall through */
 588         default:
 589                 hdr->args.stable = NFS_FILE_SYNC;
 590         }
 591
 592         hdr->res.fattr   = &hdr->fattr;
 593         hdr->res.count   = 0;
 594         hdr->res.eof     = 0;
 595         hdr->res.verf    = &hdr->verf;
 596         nfs_fattr_init(&hdr->fattr);
 597 }
 598
 599 /**
 600  * nfs_pgio_prepare - Prepare pageio hdr to go over the wire
 601  * @task: The current task
 602  * @calldata: pageio header to prepare
 603  */
 604 static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
 605 {
 606         struct nfs_pgio_header *hdr = calldata;
 607         int err;
 608         err = NFS_PROTO(hdr->inode)->pgio_rpc_prepare(task, hdr);
 609         if (err)
 610                 rpc_exit(task, err);
 611 }
 612
 613 int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
 614                       struct rpc_cred *cred, const struct nfs_rpc_ops *rpc_ops,
 615                       const struct rpc_call_ops *call_ops, int how, int flags)
 616 {
 617         struct rpc_task *task;
 618         struct rpc_message msg = {
 619                 .rpc_argp = &hdr->args,
 620                 .rpc_resp = &hdr->res,
 621                 .rpc_cred = cred,
 622         };
 623         struct rpc_task_setup task_setup_data = {
 624                 .rpc_client = clnt,
 625                 .task = &hdr->task,
 626                 .rpc_message = &msg,
 627                 .callback_ops = call_ops,
 628                 .callback_data = hdr,
 629                 .workqueue = nfsiod_workqueue,
 630                 .flags = RPC_TASK_ASYNC | flags,
 631         };
 632         int ret = 0;
 633
 634         hdr->rw_ops->rw_initiate(hdr, &msg, rpc_ops, &task_setup_data, how);
 635
 636         dprintk("NFS: initiated pgio call "
 637                 "(req %s/%llu, %u bytes @ offset %llu)\n",
 638                 hdr->inode->i_sb->s_id,
 639                 (unsigned long long)NFS_FILEID(hdr->inode),
 640                 hdr->args.count,
 641                 (unsigned long long)hdr->args.offset);
 642
 643         task = rpc_run_task(&task_setup_data);
 644         if (IS_ERR(task)) {
 645                 ret = PTR_ERR(task);
 646                 goto out;
 647         }
 648         if (how & FLUSH_SYNC) {
 649                 ret = rpc_wait_for_completion_task(task);
 650                 if (ret == 0)
 651                         ret = task->tk_status;
 652         }
 653         rpc_put_task(task);
 654 out:
 655         return ret;
 656 }
 657 EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
 658
 659 /**
 660  * nfs_pgio_error - Clean up from a pageio error
 661  * @desc: IO descriptor
 662  * @hdr: pageio header
 663  */
 664 static void nfs_pgio_error(struct nfs_pgio_header *hdr)
 665 {
 666         set_bit(NFS_IOHDR_REDO, &hdr->flags);
 667         hdr->completion_ops->completion(hdr);
 668 }
 669
 670 /**
 671  * nfs_pgio_release - Release pageio data
 672  * @calldata: The pageio header to release
 673  */
 674 static void nfs_pgio_release(void *calldata)
 675 {
 676         struct nfs_pgio_header *hdr = calldata;
 677         hdr->completion_ops->completion(hdr);
 678 }
 679
 680 static void nfs_pageio_mirror_init(struct nfs_pgio_mirror *mirror,
 681                                    unsigned int bsize)
 682 {
 683         INIT_LIST_HEAD(&mirror->pg_list);
 684         mirror->pg_bytes_written = 0;
 685         mirror->pg_count = 0;
 686         mirror->pg_bsize = bsize;
 687         mirror->pg_base = 0;
 688         mirror->pg_recoalesce = 0;
 689 }
 690
 691 /**
 692  * nfs_pageio_init - initialise a page io descriptor
 693  * @desc: pointer to descriptor
 694  * @inode: pointer to inode
 695  * @pg_ops: pointer to pageio operations
 696  * @compl_ops: pointer to pageio completion operations
 697  * @rw_ops: pointer to nfs read/write operations
 698  * @bsize: io block size
 699  * @io_flags: extra parameters for the io function
 700  */
 701 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
 702                      struct inode *inode,
 703                      const struct nfs_pageio_ops *pg_ops,
 704                      const struct nfs_pgio_completion_ops *compl_ops,
 705                      const struct nfs_rw_ops *rw_ops,
 706                      size_t bsize,
 707                      int io_flags)
 708 {
 709         desc->pg_moreio = 0;
 710         desc->pg_inode = inode;
 711         desc->pg_ops = pg_ops;
 712         desc->pg_completion_ops = compl_ops;
 713         desc->pg_rw_ops = rw_ops;
 714         desc->pg_ioflags = io_flags;
 715         desc->pg_error = 0;
 716         desc->pg_lseg = NULL;
 717         desc->pg_io_completion = NULL;
 718         desc->pg_dreq = NULL;
 719         desc->pg_bsize = bsize;
 720
 721         desc->pg_mirror_count = 1;
 722         desc->pg_mirror_idx = 0;
 723
 724         desc->pg_mirrors_dynamic = NULL;
 725         desc->pg_mirrors = desc->pg_mirrors_static;
 726         nfs_pageio_mirror_init(&desc->pg_mirrors[0], bsize);
 727 }
 728
 729 /**
 730  * nfs_pgio_result - Basic pageio error handling
 731  * @task: The task that ran
 732  * @calldata: Pageio header to check
 733  */
 734 static void nfs_pgio_result(struct rpc_task *task, void *calldata)
 735 {
 736         struct nfs_pgio_header *hdr = calldata;
 737         struct inode *inode = hdr->inode;
 738
 739         dprintk("NFS: %s: %5u, (status %d)\n", __func__,
 740                 task->tk_pid, task->tk_status);
 741
 742         if (hdr->rw_ops->rw_done(task, hdr, inode) != 0)
 743                 return;
 744         if (task->tk_status < 0)
 745                 nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset);
 746         else
 747                 hdr->rw_ops->rw_result(task, hdr);
 748 }
 749
 750 /*
 751  * Create an RPC task for the given read or write request and kick it.
 752  * The page must have been locked by the caller.
 753  *
 754  * It may happen that the page we're passed is not marked dirty.
 755  * This is the case if nfs_updatepage detects a conflicting request
 756  * that has been written but not committed.
 757  */
 758 int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
 759                      struct nfs_pgio_header *hdr)
 760 {
 761         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
 762
 763         struct nfs_page         *req;
 764         struct page             **pages,
 765                                 *last_page;
 766         struct list_head *head = &mirror->pg_list;
 767         struct nfs_commit_info cinfo;
 768         struct nfs_page_array *pg_array = &hdr->page_array;
 769         unsigned int pagecount, pageused;
 770         gfp_t gfp_flags = GFP_KERNEL;
 771
 772         pagecount = nfs_page_array_len(mirror->pg_base, mirror->pg_count);
 773         pg_array->npages = pagecount;
 774
 775         if (pagecount <= ARRAY_SIZE(pg_array->page_array))
 776                 pg_array->pagevec = pg_array->page_array;
 777         else {
 778                 if (hdr->rw_mode == FMODE_WRITE)
 779                         gfp_flags = GFP_NOIO;
 780                 pg_array->pagevec = kcalloc(pagecount, sizeof(struct page *), gfp_flags);
 781                 if (!pg_array->pagevec) {
 782                         pg_array->npages = 0;
 783                         nfs_pgio_error(hdr);
 784                         desc->pg_error = -ENOMEM;
 785                         return desc->pg_error;
 786                 }
 787         }
 788
 789         nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
 790         pages = hdr->page_array.pagevec;
 791         last_page = NULL;
 792         pageused = 0;
 793         while (!list_empty(head)) {
 794                 req = nfs_list_entry(head->next);
 795                 nfs_list_move_request(req, &hdr->pages);
 796
 797                 if (!last_page || last_page != req->wb_page) {
 798                         pageused++;
 799                         if (pageused > pagecount)
 800                                 break;
 801                         *pages++ = last_page = req->wb_page;
 802                 }
 803         }
 804         if (WARN_ON_ONCE(pageused != pagecount)) {
 805                 nfs_pgio_error(hdr);
 806                 desc->pg_error = -EINVAL;
 807                 return desc->pg_error;
 808         }
 809
 810         if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
 811             (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
 812                 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
 813
 814         /* Set up the argument struct */
 815         nfs_pgio_rpcsetup(hdr, mirror->pg_count, desc->pg_ioflags, &cinfo);
 816         desc->pg_rpc_callops = &nfs_pgio_common_ops;
 817         return 0;
 818 }
 819 EXPORT_SYMBOL_GPL(nfs_generic_pgio);
 820
 821 static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
 822 {
 823         struct nfs_pgio_header *hdr;
 824         int ret;
 825
 826         hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
 827         if (!hdr) {
 828                 desc->pg_error = -ENOMEM;
 829                 return desc->pg_error;
 830         }
 831         nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
 832         ret = nfs_generic_pgio(desc, hdr);
 833         if (ret == 0)
 834                 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
 835                                         hdr,
 836                                         hdr->cred,
 837                                         NFS_PROTO(hdr->inode),
 838                                         desc->pg_rpc_callops,
 839                                         desc->pg_ioflags, 0);
 840         return ret;
 841 }
 842
 843 static struct nfs_pgio_mirror *
 844 nfs_pageio_alloc_mirrors(struct nfs_pageio_descriptor *desc,
 845                 unsigned int mirror_count)
 846 {
 847         struct nfs_pgio_mirror *ret;
 848         unsigned int i;
 849
 850         kfree(desc->pg_mirrors_dynamic);
 851         desc->pg_mirrors_dynamic = NULL;
 852         if (mirror_count == 1)
 853                 return desc->pg_mirrors_static;
 854         ret = kmalloc_array(mirror_count, sizeof(*ret), GFP_NOFS);
 855         if (ret != NULL) {
 856                 for (i = 0; i < mirror_count; i++)
 857                         nfs_pageio_mirror_init(&ret[i], desc->pg_bsize);
 858                 desc->pg_mirrors_dynamic = ret;
 859         }
 860         return ret;
 861 }
 862
 863 /*
 864  * nfs_pageio_setup_mirroring - determine if mirroring is to be used
 865  *                              by calling the pg_get_mirror_count op
 866  */
 867 static void nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor *pgio,
 868                                        struct nfs_page *req)
 869 {
 870         unsigned int mirror_count = 1;
 871
 872         if (pgio->pg_ops->pg_get_mirror_count)
 873                 mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req);
 874         if (mirror_count == pgio->pg_mirror_count || pgio->pg_error < 0)
 875                 return;
 876
 877         if (!mirror_count || mirror_count > NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX) {
 878                 pgio->pg_error = -EINVAL;
 879                 return;
 880         }
 881
 882         pgio->pg_mirrors = nfs_pageio_alloc_mirrors(pgio, mirror_count);
 883         if (pgio->pg_mirrors == NULL) {
 884                 pgio->pg_error = -ENOMEM;
 885                 pgio->pg_mirrors = pgio->pg_mirrors_static;
 886                 mirror_count = 1;
 887         }
 888         pgio->pg_mirror_count = mirror_count;
 889 }
 890
 891 static void nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor *pgio)
 892 {
 893         pgio->pg_mirror_count = 1;
 894         pgio->pg_mirror_idx = 0;
 895         pgio->pg_mirrors = pgio->pg_mirrors_static;
 896         kfree(pgio->pg_mirrors_dynamic);
 897         pgio->pg_mirrors_dynamic = NULL;
 898 }
 899
 900 static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
 901                 const struct nfs_lock_context *l2)
 902 {
 903         return l1->lockowner == l2->lockowner;
 904 }
 905
 906 /**
 907  * nfs_can_coalesce_requests - test two requests for compatibility
 908  * @prev: pointer to nfs_page
 909  * @req: pointer to nfs_page
 910  *
 911  * The nfs_page structures 'prev' and 'req' are compared to ensure that the
 912  * page data area they describe is contiguous, and that their RPC
 913  * credentials, NFSv4 open state, and lockowners are the same.
 914  *
 915  * Return 'true' if this is the case, else return 'false'.
 916  */
 917 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
 918                                       struct nfs_page *req,
 919                                       struct nfs_pageio_descriptor *pgio)
 920 {
 921         size_t size;
 922         struct file_lock_context *flctx;
 923
 924         if (prev) {
 925                 if (!nfs_match_open_context(req->wb_context, prev->wb_context))
 926                         return false;
 927                 flctx = d_inode(req->wb_context->dentry)->i_flctx;
 928                 if (flctx != NULL &&
 929                     !(list_empty_careful(&flctx->flc_posix) &&
 930                       list_empty_careful(&flctx->flc_flock)) &&
 931                     !nfs_match_lock_context(req->wb_lock_context,
 932                                             prev->wb_lock_context))
 933                         return false;
 934                 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
 935                         return false;
 936                 if (req->wb_page == prev->wb_page) {
 937                         if (req->wb_pgbase != prev->wb_pgbase + prev->wb_bytes)
 938                                 return false;
 939                 } else {
 940                         if (req->wb_pgbase != 0 ||
 941                             prev->wb_pgbase + prev->wb_bytes != PAGE_SIZE)
 942                                 return false;
 943                 }
 944         }
 945         size = pgio->pg_ops->pg_test(pgio, prev, req);
 946         WARN_ON_ONCE(size > req->wb_bytes);
 947         if (size && size < req->wb_bytes)
 948                 req->wb_bytes = size;
 949         return size > 0;
 950 }
 951
 952 /**
 953  * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
 954  * @desc: destination io descriptor
 955  * @req: request
 956  *
 957  * Returns true if the request 'req' was successfully coalesced into the
 958  * existing list of pages 'desc'.
 959  */
 960 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
 961                                      struct nfs_page *req)
 962 {
 963         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
 964
 965         struct nfs_page *prev = NULL;
 966
 967         if (mirror->pg_count != 0) {
 968                 prev = nfs_list_entry(mirror->pg_list.prev);
 969         } else {
 970                 if (desc->pg_ops->pg_init)
 971                         desc->pg_ops->pg_init(desc, req);
 972                 if (desc->pg_error < 0)
 973                         return 0;
 974                 mirror->pg_base = req->wb_pgbase;
 975         }
 976         if (!nfs_can_coalesce_requests(prev, req, desc))
 977                 return 0;
 978         nfs_list_move_request(req, &mirror->pg_list);
 979         mirror->pg_count += req->wb_bytes;
 980         return 1;
 981 }
 982
 983 /*
 984  * Helper for nfs_pageio_add_request and nfs_pageio_complete
 985  */
 986 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
 987 {
 988         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
 989
 990         if (!list_empty(&mirror->pg_list)) {
 991                 int error = desc->pg_ops->pg_doio(desc);
 992                 if (error < 0)
 993                         desc->pg_error = error;
 994                 if (list_empty(&mirror->pg_list)) {
 995                         mirror->pg_bytes_written += mirror->pg_count;
 996                         mirror->pg_count = 0;
 997                         mirror->pg_base = 0;
 998                         mirror->pg_recoalesce = 0;
 999                 }
1000         }
1001 }
1002
1003 static void
1004 nfs_pageio_cleanup_request(struct nfs_pageio_descriptor *desc,
1005                 struct nfs_page *req)
1006 {
1007         LIST_HEAD(head);
1008
1009         nfs_list_move_request(req, &head);
1010         desc->pg_completion_ops->error_cleanup(&head, desc->pg_error);
1011 }
1012
1013 /**
1014  * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
1015  * @desc: destination io descriptor
1016  * @req: request
1017  *
1018  * This may split a request into subrequests which are all part of the
1019  * same page group.
1020  *
1021  * Returns true if the request 'req' was successfully coalesced into the
1022  * existing list of pages 'desc'.
1023  */
1024 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1025                            struct nfs_page *req)
1026 {
1027         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1028
1029         struct nfs_page *subreq;
1030         unsigned int bytes_left = 0;
1031         unsigned int offset, pgbase;
1032
1033         nfs_page_group_lock(req);
1034
1035         subreq = req;
1036         bytes_left = subreq->wb_bytes;
1037         offset = subreq->wb_offset;
1038         pgbase = subreq->wb_pgbase;
1039
1040         do {
1041                 if (!nfs_pageio_do_add_request(desc, subreq)) {
1042                         /* make sure pg_test call(s) did nothing */
1043                         WARN_ON_ONCE(subreq->wb_bytes != bytes_left);
1044                         WARN_ON_ONCE(subreq->wb_offset != offset);
1045                         WARN_ON_ONCE(subreq->wb_pgbase != pgbase);
1046
1047                         nfs_page_group_unlock(req);
1048                         desc->pg_moreio = 1;
1049                         nfs_pageio_doio(desc);
1050                         if (desc->pg_error < 0 || mirror->pg_recoalesce)
1051                                 goto out_cleanup_subreq;
1052                         /* retry add_request for this subreq */
1053                         nfs_page_group_lock(req);
1054                         continue;
1055                 }
1056
1057                 /* check for buggy pg_test call(s) */
1058                 WARN_ON_ONCE(subreq->wb_bytes + subreq->wb_pgbase > PAGE_SIZE);
1059                 WARN_ON_ONCE(subreq->wb_bytes > bytes_left);
1060                 WARN_ON_ONCE(subreq->wb_bytes == 0);
1061
1062                 bytes_left -= subreq->wb_bytes;
1063                 offset += subreq->wb_bytes;
1064                 pgbase += subreq->wb_bytes;
1065
1066                 if (bytes_left) {
1067                         subreq = nfs_create_request(req->wb_context,
1068                                         req->wb_page,
1069                                         subreq, pgbase, bytes_left);
1070                         if (IS_ERR(subreq))
1071                                 goto err_ptr;
1072                         nfs_lock_request(subreq);
1073                         subreq->wb_offset  = offset;
1074                         subreq->wb_index = req->wb_index;
1075                 }
1076         } while (bytes_left > 0);
1077
1078         nfs_page_group_unlock(req);
1079         return 1;
1080 err_ptr:
1081         desc->pg_error = PTR_ERR(subreq);
1082         nfs_page_group_unlock(req);
1083         return 0;
1084 out_cleanup_subreq:
1085         if (req != subreq)
1086                 nfs_pageio_cleanup_request(desc, subreq);
1087         return 0;
1088 }
1089
1090 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
1091 {
1092         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1093         LIST_HEAD(head);
1094
1095         do {
1096                 list_splice_init(&mirror->pg_list, &head);
1097                 mirror->pg_count = 0;
1098                 mirror->pg_base = 0;
1099                 mirror->pg_recoalesce = 0;
1100
1101                 while (!list_empty(&head)) {
1102                         struct nfs_page *req;
1103
1104                         req = list_first_entry(&head, struct nfs_page, wb_list);
1105                         if (__nfs_pageio_add_request(desc, req))
1106                                 continue;
1107                         if (desc->pg_error < 0) {
1108                                 list_splice_tail(&head, &mirror->pg_list);
1109                                 mirror->pg_recoalesce = 1;
1110                                 return 0;
1111                         }
1112                         break;
1113                 }
1114         } while (mirror->pg_recoalesce);
1115         return 1;
1116 }
1117
1118 static int nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor *desc,
1119                 struct nfs_page *req)
1120 {
1121         int ret;
1122
1123         do {
1124                 ret = __nfs_pageio_add_request(desc, req);
1125                 if (ret)
1126                         break;
1127                 if (desc->pg_error < 0)
1128                         break;
1129                 ret = nfs_do_recoalesce(desc);
1130         } while (ret);
1131
1132         return ret;
1133 }
1134
1135 static void nfs_pageio_error_cleanup(struct nfs_pageio_descriptor *desc)
1136 {
1137         u32 midx;
1138         struct nfs_pgio_mirror *mirror;
1139
1140         if (!desc->pg_error)
1141                 return;
1142
1143         for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1144                 mirror = &desc->pg_mirrors[midx];
1145                 desc->pg_completion_ops->error_cleanup(&mirror->pg_list,
1146                                 desc->pg_error);
1147         }
1148 }
1149
1150 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1151                            struct nfs_page *req)
1152 {
1153         u32 midx;
1154         unsigned int pgbase, offset, bytes;
1155         struct nfs_page *dupreq, *lastreq;
1156
1157         pgbase = req->wb_pgbase;
1158         offset = req->wb_offset;
1159         bytes = req->wb_bytes;
1160
1161         nfs_pageio_setup_mirroring(desc, req);
1162         if (desc->pg_error < 0)
1163                 goto out_failed;
1164
1165         for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1166                 if (midx) {
1167                         nfs_page_group_lock(req);
1168
1169                         /* find the last request */
1170                         for (lastreq = req->wb_head;
1171                              lastreq->wb_this_page != req->wb_head;
1172                              lastreq = lastreq->wb_this_page)
1173                                 ;
1174
1175                         dupreq = nfs_create_request(req->wb_context,
1176                                         req->wb_page, lastreq, pgbase, bytes);
1177
1178                         if (IS_ERR(dupreq)) {
1179                                 nfs_page_group_unlock(req);
1180                                 desc->pg_error = PTR_ERR(dupreq);
1181                                 goto out_failed;
1182                         }
1183
1184                         nfs_lock_request(dupreq);
1185                         nfs_page_group_unlock(req);
1186                         dupreq->wb_offset = offset;
1187                         dupreq->wb_index = req->wb_index;
1188                 } else
1189                         dupreq = req;
1190
1191                 if (nfs_pgio_has_mirroring(desc))
1192                         desc->pg_mirror_idx = midx;
1193                 if (!nfs_pageio_add_request_mirror(desc, dupreq))
1194                         goto out_cleanup_subreq;
1195         }
1196
1197         return 1;
1198
1199 out_cleanup_subreq:
1200         if (req != dupreq)
1201                 nfs_pageio_cleanup_request(desc, dupreq);
1202 out_failed:
1203         /* remember fatal errors */
1204         if (nfs_error_is_fatal(desc->pg_error))
1205                 nfs_context_set_write_error(req->wb_context,
1206                                                 desc->pg_error);
1207         nfs_pageio_error_cleanup(desc);
1208         return 0;
1209 }
1210
1211 /*
1212  * nfs_pageio_complete_mirror - Complete I/O on the current mirror of an
1213  *                              nfs_pageio_descriptor
1214  * @desc: pointer to io descriptor
1215  * @mirror_idx: pointer to mirror index
1216  */
1217 static void nfs_pageio_complete_mirror(struct nfs_pageio_descriptor *desc,
1218                                        u32 mirror_idx)
1219 {
1220         struct nfs_pgio_mirror *mirror = &desc->pg_mirrors[mirror_idx];
1221         u32 restore_idx = desc->pg_mirror_idx;
1222
1223         if (nfs_pgio_has_mirroring(desc))
1224                 desc->pg_mirror_idx = mirror_idx;
1225         for (;;) {
1226                 nfs_pageio_doio(desc);
1227                 if (desc->pg_error < 0 || !mirror->pg_recoalesce)
1228                         break;
1229                 if (!nfs_do_recoalesce(desc))
1230                         break;
1231         }
1232         desc->pg_mirror_idx = restore_idx;
1233 }
1234
1235 /*
1236  * nfs_pageio_resend - Transfer requests to new descriptor and resend
1237  * @hdr - the pgio header to move request from
1238  * @desc - the pageio descriptor to add requests to
1239  *
1240  * Try to move each request (nfs_page) from @hdr to @desc then attempt
1241  * to send them.
1242  *
1243  * Returns 0 on success and < 0 on error.
1244  */
1245 int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
1246                       struct nfs_pgio_header *hdr)
1247 {
1248         LIST_HEAD(pages);
1249
1250         desc->pg_io_completion = hdr->io_completion;
1251         desc->pg_dreq = hdr->dreq;
1252         list_splice_init(&hdr->pages, &pages);
1253         while (!list_empty(&pages)) {
1254                 struct nfs_page *req = nfs_list_entry(pages.next);
1255
1256                 if (!nfs_pageio_add_request(desc, req))
1257                         break;
1258         }
1259         nfs_pageio_complete(desc);
1260         if (!list_empty(&pages)) {
1261                 int err = desc->pg_error < 0 ? desc->pg_error : -EIO;
1262                 hdr->completion_ops->error_cleanup(&pages, err);
1263                 nfs_set_pgio_error(hdr, err, hdr->io_start);
1264                 return err;
1265         }
1266         return 0;
1267 }
1268 EXPORT_SYMBOL_GPL(nfs_pageio_resend);
1269
1270 /**
1271  * nfs_pageio_complete - Complete I/O then cleanup an nfs_pageio_descriptor
1272  * @desc: pointer to io descriptor
1273  */
1274 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1275 {
1276         u32 midx;
1277
1278         for (midx = 0; midx < desc->pg_mirror_count; midx++)
1279                 nfs_pageio_complete_mirror(desc, midx);
1280
1281         if (desc->pg_error < 0)
1282                 nfs_pageio_error_cleanup(desc);
1283         if (desc->pg_ops->pg_cleanup)
1284                 desc->pg_ops->pg_cleanup(desc);
1285         nfs_pageio_cleanup_mirroring(desc);
1286 }
1287
1288 /**
1289  * nfs_pageio_cond_complete - Conditional I/O completion
1290  * @desc: pointer to io descriptor
1291  * @index: page index
1292  *
1293  * It is important to ensure that processes don't try to take locks
1294  * on non-contiguous ranges of pages as that might deadlock. This
1295  * function should be called before attempting to wait on a locked
1296  * nfs_page. It will complete the I/O if the page index 'index'
1297  * is not contiguous with the existing list of pages in 'desc'.
1298  */
1299 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1300 {
1301         struct nfs_pgio_mirror *mirror;
1302         struct nfs_page *prev;
1303         u32 midx;
1304
1305         for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1306                 mirror = &desc->pg_mirrors[midx];
1307                 if (!list_empty(&mirror->pg_list)) {
1308                         prev = nfs_list_entry(mirror->pg_list.prev);
1309                         if (index != prev->wb_index + 1) {
1310                                 nfs_pageio_complete(desc);
1311                                 break;
1312                         }
1313                 }
1314         }
1315 }
1316
1317 /*
1318  * nfs_pageio_stop_mirroring - stop using mirroring (set mirror count to 1)
1319  */
1320 void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio)
1321 {
1322         nfs_pageio_complete(pgio);
1323 }
1324
1325 int __init nfs_init_nfspagecache(void)
1326 {
1327         nfs_page_cachep = kmem_cache_create("nfs_page",
1328                                             sizeof(struct nfs_page),
1329                                             0, SLAB_HWCACHE_ALIGN,
1330                                             NULL);
1331         if (nfs_page_cachep == NULL)
1332                 return -ENOMEM;
1333
1334         return 0;
1335 }
1336
1337 void nfs_destroy_nfspagecache(void)
1338 {
1339         kmem_cache_destroy(nfs_page_cachep);
1340 }
1341
1342 static const struct rpc_call_ops nfs_pgio_common_ops = {
1343         .rpc_call_prepare = nfs_pgio_prepare,
1344         .rpc_call_done = nfs_pgio_result,
1345         .rpc_release = nfs_pgio_release,
1346 };
1347
1348 const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1349         .pg_test = nfs_generic_pg_test,
1350         .pg_doio = nfs_generic_pg_pgios,
1351 };