fs/nfs/pagelist.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * linux/fs/nfs/pagelist.c
   4  *
   5  * A set of helper functions for managing NFS read and write requests.
   6  * The main purpose of these routines is to provide support for the
   7  * coalescing of several requests into a single RPC call.
   8  *
   9  * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
  10  *
  11  */
  12
  13 #include <linux/slab.h>
  14 #include <linux/file.h>
  15 #include <linux/sched.h>
  16 #include <linux/sunrpc/clnt.h>
  17 #include <linux/nfs.h>
  18 #include <linux/nfs3.h>
  19 #include <linux/nfs4.h>
  20 #include <linux/nfs_fs.h>
  21 #include <linux/nfs_page.h>
  22 #include <linux/nfs_mount.h>
  23 #include <linux/export.h>
  24
  25 #include "internal.h"
  26 #include "pnfs.h"
  27
  28 #define NFSDBG_FACILITY         NFSDBG_PAGECACHE
  29
  30 static struct kmem_cache *nfs_page_cachep;
  31 static const struct rpc_call_ops nfs_pgio_common_ops;
  32
  33 struct nfs_pgio_mirror *
  34 nfs_pgio_current_mirror(struct nfs_pageio_descriptor *desc)
  35 {
  36         return nfs_pgio_has_mirroring(desc) ?
  37                 &desc->pg_mirrors[desc->pg_mirror_idx] :
  38                 &desc->pg_mirrors[0];
  39 }
  40 EXPORT_SYMBOL_GPL(nfs_pgio_current_mirror);
  41
  42 void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
  43                        struct nfs_pgio_header *hdr,
  44                        void (*release)(struct nfs_pgio_header *hdr))
  45 {
  46         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
  47
  48
  49         hdr->req = nfs_list_entry(mirror->pg_list.next);
  50         hdr->inode = desc->pg_inode;
  51         hdr->cred = nfs_req_openctx(hdr->req)->cred;
  52         hdr->io_start = req_offset(hdr->req);
  53         hdr->good_bytes = mirror->pg_count;
  54         hdr->io_completion = desc->pg_io_completion;
  55         hdr->dreq = desc->pg_dreq;
  56         hdr->release = release;
  57         hdr->completion_ops = desc->pg_completion_ops;
  58         if (hdr->completion_ops->init_hdr)
  59                 hdr->completion_ops->init_hdr(hdr);
  60
  61         hdr->pgio_mirror_idx = desc->pg_mirror_idx;
  62 }
  63 EXPORT_SYMBOL_GPL(nfs_pgheader_init);
  64
  65 void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
  66 {
  67         unsigned int new = pos - hdr->io_start;
  68
  69         if (hdr->good_bytes > new) {
  70                 hdr->good_bytes = new;
  71                 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
  72                 if (!test_and_set_bit(NFS_IOHDR_ERROR, &hdr->flags))
  73                         hdr->error = error;
  74         }
  75 }
  76
  77 static inline struct nfs_page *
  78 nfs_page_alloc(void)
  79 {
  80         struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_KERNEL);
  81         if (p)
  82                 INIT_LIST_HEAD(&p->wb_list);
  83         return p;
  84 }
  85
  86 static inline void
  87 nfs_page_free(struct nfs_page *p)
  88 {
  89         kmem_cache_free(nfs_page_cachep, p);
  90 }
  91
  92 /**
  93  * nfs_iocounter_wait - wait for i/o to complete
  94  * @l_ctx: nfs_lock_context with io_counter to use
  95  *
  96  * returns -ERESTARTSYS if interrupted by a fatal signal.
  97  * Otherwise returns 0 once the io_count hits 0.
  98  */
  99 int
 100 nfs_iocounter_wait(struct nfs_lock_context *l_ctx)
 101 {
 102         return wait_var_event_killable(&l_ctx->io_count,
 103                                        !atomic_read(&l_ctx->io_count));
 104 }
 105
 106 /**
 107  * nfs_async_iocounter_wait - wait on a rpc_waitqueue for I/O
 108  * to complete
 109  * @task: the rpc_task that should wait
 110  * @l_ctx: nfs_lock_context with io_counter to check
 111  *
 112  * Returns true if there is outstanding I/O to wait on and the
 113  * task has been put to sleep.
 114  */
 115 bool
 116 nfs_async_iocounter_wait(struct rpc_task *task, struct nfs_lock_context *l_ctx)
 117 {
 118         struct inode *inode = d_inode(l_ctx->open_context->dentry);
 119         bool ret = false;
 120
 121         if (atomic_read(&l_ctx->io_count) > 0) {
 122                 rpc_sleep_on(&NFS_SERVER(inode)->uoc_rpcwaitq, task, NULL);
 123                 ret = true;
 124         }
 125
 126         if (atomic_read(&l_ctx->io_count) == 0) {
 127                 rpc_wake_up_queued_task(&NFS_SERVER(inode)->uoc_rpcwaitq, task);
 128                 ret = false;
 129         }
 130
 131         return ret;
 132 }
 133 EXPORT_SYMBOL_GPL(nfs_async_iocounter_wait);
 134
 135 /*
 136  * nfs_page_group_lock - lock the head of the page group
 137  * @req - request in group that is to be locked
 138  *
 139  * this lock must be held when traversing or modifying the page
 140  * group list
 141  *
 142  * return 0 on success, < 0 on error
 143  */
 144 int
 145 nfs_page_group_lock(struct nfs_page *req)
 146 {
 147         struct nfs_page *head = req->wb_head;
 148
 149         WARN_ON_ONCE(head != head->wb_head);
 150
 151         if (!test_and_set_bit(PG_HEADLOCK, &head->wb_flags))
 152                 return 0;
 153
 154         set_bit(PG_CONTENDED1, &head->wb_flags);
 155         smp_mb__after_atomic();
 156         return wait_on_bit_lock(&head->wb_flags, PG_HEADLOCK,
 157                                 TASK_UNINTERRUPTIBLE);
 158 }
 159
 160 /*
 161  * nfs_page_group_unlock - unlock the head of the page group
 162  * @req - request in group that is to be unlocked
 163  */
 164 void
 165 nfs_page_group_unlock(struct nfs_page *req)
 166 {
 167         struct nfs_page *head = req->wb_head;
 168
 169         WARN_ON_ONCE(head != head->wb_head);
 170
 171         smp_mb__before_atomic();
 172         clear_bit(PG_HEADLOCK, &head->wb_flags);
 173         smp_mb__after_atomic();
 174         if (!test_bit(PG_CONTENDED1, &head->wb_flags))
 175                 return;
 176         wake_up_bit(&head->wb_flags, PG_HEADLOCK);
 177 }
 178
 179 /*
 180  * nfs_page_group_sync_on_bit_locked
 181  *
 182  * must be called with page group lock held
 183  */
 184 static bool
 185 nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
 186 {
 187         struct nfs_page *head = req->wb_head;
 188         struct nfs_page *tmp;
 189
 190         WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
 191         WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
 192
 193         tmp = req->wb_this_page;
 194         while (tmp != req) {
 195                 if (!test_bit(bit, &tmp->wb_flags))
 196                         return false;
 197                 tmp = tmp->wb_this_page;
 198         }
 199
 200         /* true! reset all bits */
 201         tmp = req;
 202         do {
 203                 clear_bit(bit, &tmp->wb_flags);
 204                 tmp = tmp->wb_this_page;
 205         } while (tmp != req);
 206
 207         return true;
 208 }
 209
 210 /*
 211  * nfs_page_group_sync_on_bit - set bit on current request, but only
 212  *   return true if the bit is set for all requests in page group
 213  * @req - request in page group
 214  * @bit - PG_* bit that is used to sync page group
 215  */
 216 bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
 217 {
 218         bool ret;
 219
 220         nfs_page_group_lock(req);
 221         ret = nfs_page_group_sync_on_bit_locked(req, bit);
 222         nfs_page_group_unlock(req);
 223
 224         return ret;
 225 }
 226
 227 /*
 228  * nfs_page_group_init - Initialize the page group linkage for @req
 229  * @req - a new nfs request
 230  * @prev - the previous request in page group, or NULL if @req is the first
 231  *         or only request in the group (the head).
 232  */
 233 static inline void
 234 nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
 235 {
 236         struct inode *inode;
 237         WARN_ON_ONCE(prev == req);
 238
 239         if (!prev) {
 240                 /* a head request */
 241                 req->wb_head = req;
 242                 req->wb_this_page = req;
 243         } else {
 244                 /* a subrequest */
 245                 WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
 246                 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
 247                 req->wb_head = prev->wb_head;
 248                 req->wb_this_page = prev->wb_this_page;
 249                 prev->wb_this_page = req;
 250
 251                 /* All subrequests take a ref on the head request until
 252                  * nfs_page_group_destroy is called */
 253                 kref_get(&req->wb_head->wb_kref);
 254
 255                 /* grab extra ref and bump the request count if head request
 256                  * has extra ref from the write/commit path to handle handoff
 257                  * between write and commit lists. */
 258                 if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) {
 259                         inode = page_file_mapping(req->wb_page)->host;
 260                         set_bit(PG_INODE_REF, &req->wb_flags);
 261                         kref_get(&req->wb_kref);
 262                         atomic_long_inc(&NFS_I(inode)->nrequests);
 263                 }
 264         }
 265 }
 266
 267 /*
 268  * nfs_page_group_destroy - sync the destruction of page groups
 269  * @req - request that no longer needs the page group
 270  *
 271  * releases the page group reference from each member once all
 272  * members have called this function.
 273  */
 274 static void
 275 nfs_page_group_destroy(struct kref *kref)
 276 {
 277         struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
 278         struct nfs_page *head = req->wb_head;
 279         struct nfs_page *tmp, *next;
 280
 281         if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
 282                 goto out;
 283
 284         tmp = req;
 285         do {
 286                 next = tmp->wb_this_page;
 287                 /* unlink and free */
 288                 tmp->wb_this_page = tmp;
 289                 tmp->wb_head = tmp;
 290                 nfs_free_request(tmp);
 291                 tmp = next;
 292         } while (tmp != req);
 293 out:
 294         /* subrequests must release the ref on the head request */
 295         if (head != req)
 296                 nfs_release_request(head);
 297 }
 298
 299 static struct nfs_page *
 300 __nfs_create_request(struct nfs_lock_context *l_ctx, struct page *page,
 301                    unsigned int pgbase, unsigned int offset,
 302                    unsigned int count)
 303 {
 304         struct nfs_page         *req;
 305         struct nfs_open_context *ctx = l_ctx->open_context;
 306
 307         if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
 308                 return ERR_PTR(-EBADF);
 309         /* try to allocate the request struct */
 310         req = nfs_page_alloc();
 311         if (req == NULL)
 312                 return ERR_PTR(-ENOMEM);
 313
 314         req->wb_lock_context = l_ctx;
 315         refcount_inc(&l_ctx->count);
 316         atomic_inc(&l_ctx->io_count);
 317
 318         /* Initialize the request struct. Initially, we assume a
 319          * long write-back delay. This will be adjusted in
 320          * update_nfs_request below if the region is not locked. */
 321         req->wb_page    = page;
 322         if (page) {
 323                 req->wb_index = page_index(page);
 324                 get_page(page);
 325         }
 326         req->wb_offset  = offset;
 327         req->wb_pgbase  = pgbase;
 328         req->wb_bytes   = count;
 329         kref_init(&req->wb_kref);
 330         req->wb_nio = 0;
 331         return req;
 332 }
 333
 334 /**
 335  * nfs_create_request - Create an NFS read/write request.
 336  * @ctx: open context to use
 337  * @page: page to write
 338  * @offset: starting offset within the page for the write
 339  * @count: number of bytes to read/write
 340  *
 341  * The page must be locked by the caller. This makes sure we never
 342  * create two different requests for the same page.
 343  * User should ensure it is safe to sleep in this function.
 344  */
 345 struct nfs_page *
 346 nfs_create_request(struct nfs_open_context *ctx, struct page *page,
 347                    unsigned int offset, unsigned int count)
 348 {
 349         struct nfs_lock_context *l_ctx = nfs_get_lock_context(ctx);
 350         struct nfs_page *ret;
 351
 352         if (IS_ERR(l_ctx))
 353                 return ERR_CAST(l_ctx);
 354         ret = __nfs_create_request(l_ctx, page, offset, offset, count);
 355         if (!IS_ERR(ret))
 356                 nfs_page_group_init(ret, NULL);
 357         nfs_put_lock_context(l_ctx);
 358         return ret;
 359 }
 360
 361 static struct nfs_page *
 362 nfs_create_subreq(struct nfs_page *req, struct nfs_page *last,
 363                   unsigned int pgbase, unsigned int offset,
 364                   unsigned int count)
 365 {
 366         struct nfs_page *ret;
 367
 368         ret = __nfs_create_request(req->wb_lock_context, req->wb_page,
 369                         pgbase, offset, count);
 370         if (!IS_ERR(ret)) {
 371                 nfs_lock_request(ret);
 372                 ret->wb_index = req->wb_index;
 373                 nfs_page_group_init(ret, last);
 374                 ret->wb_nio = req->wb_nio;
 375         }
 376         return ret;
 377 }
 378
 379 /**
 380  * nfs_unlock_request - Unlock request and wake up sleepers.
 381  * @req: pointer to request
 382  */
 383 void nfs_unlock_request(struct nfs_page *req)
 384 {
 385         if (!NFS_WBACK_BUSY(req)) {
 386                 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
 387                 BUG();
 388         }
 389         smp_mb__before_atomic();
 390         clear_bit(PG_BUSY, &req->wb_flags);
 391         smp_mb__after_atomic();
 392         if (!test_bit(PG_CONTENDED2, &req->wb_flags))
 393                 return;
 394         wake_up_bit(&req->wb_flags, PG_BUSY);
 395 }
 396
 397 /**
 398  * nfs_unlock_and_release_request - Unlock request and release the nfs_page
 399  * @req: pointer to request
 400  */
 401 void nfs_unlock_and_release_request(struct nfs_page *req)
 402 {
 403         nfs_unlock_request(req);
 404         nfs_release_request(req);
 405 }
 406
 407 /*
 408  * nfs_clear_request - Free up all resources allocated to the request
 409  * @req:
 410  *
 411  * Release page and open context resources associated with a read/write
 412  * request after it has completed.
 413  */
 414 static void nfs_clear_request(struct nfs_page *req)
 415 {
 416         struct page *page = req->wb_page;
 417         struct nfs_lock_context *l_ctx = req->wb_lock_context;
 418         struct nfs_open_context *ctx;
 419
 420         if (page != NULL) {
 421                 put_page(page);
 422                 req->wb_page = NULL;
 423         }
 424         if (l_ctx != NULL) {
 425                 if (atomic_dec_and_test(&l_ctx->io_count)) {
 426                         wake_up_var(&l_ctx->io_count);
 427                         ctx = l_ctx->open_context;
 428                         if (test_bit(NFS_CONTEXT_UNLOCK, &ctx->flags))
 429                                 rpc_wake_up(&NFS_SERVER(d_inode(ctx->dentry))->uoc_rpcwaitq);
 430                 }
 431                 nfs_put_lock_context(l_ctx);
 432                 req->wb_lock_context = NULL;
 433         }
 434 }
 435
 436 /**
 437  * nfs_release_request - Release the count on an NFS read/write request
 438  * @req: request to release
 439  *
 440  * Note: Should never be called with the spinlock held!
 441  */
 442 void nfs_free_request(struct nfs_page *req)
 443 {
 444         WARN_ON_ONCE(req->wb_this_page != req);
 445
 446         /* extra debug: make sure no sync bits are still set */
 447         WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
 448         WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
 449         WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
 450         WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
 451         WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
 452
 453         /* Release struct file and open context */
 454         nfs_clear_request(req);
 455         nfs_page_free(req);
 456 }
 457
 458 void nfs_release_request(struct nfs_page *req)
 459 {
 460         kref_put(&req->wb_kref, nfs_page_group_destroy);
 461 }
 462 EXPORT_SYMBOL_GPL(nfs_release_request);
 463
 464 /**
 465  * nfs_wait_on_request - Wait for a request to complete.
 466  * @req: request to wait upon.
 467  *
 468  * Interruptible by fatal signals only.
 469  * The user is responsible for holding a count on the request.
 470  */
 471 int
 472 nfs_wait_on_request(struct nfs_page *req)
 473 {
 474         if (!test_bit(PG_BUSY, &req->wb_flags))
 475                 return 0;
 476         set_bit(PG_CONTENDED2, &req->wb_flags);
 477         smp_mb__after_atomic();
 478         return wait_on_bit_io(&req->wb_flags, PG_BUSY,
 479                               TASK_UNINTERRUPTIBLE);
 480 }
 481 EXPORT_SYMBOL_GPL(nfs_wait_on_request);
 482
 483 /*
 484  * nfs_generic_pg_test - determine if requests can be coalesced
 485  * @desc: pointer to descriptor
 486  * @prev: previous request in desc, or NULL
 487  * @req: this request
 488  *
 489  * Returns zero if @req cannot be coalesced into @desc, otherwise it returns
 490  * the size of the request.
 491  */
 492 size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
 493                            struct nfs_page *prev, struct nfs_page *req)
 494 {
 495         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
 496
 497
 498         if (mirror->pg_count > mirror->pg_bsize) {
 499                 /* should never happen */
 500                 WARN_ON_ONCE(1);
 501                 return 0;
 502         }
 503
 504         /*
 505          * Limit the request size so that we can still allocate a page array
 506          * for it without upsetting the slab allocator.
 507          */
 508         if (((mirror->pg_count + req->wb_bytes) >> PAGE_SHIFT) *
 509                         sizeof(struct page *) > PAGE_SIZE)
 510                 return 0;
 511
 512         return min(mirror->pg_bsize - mirror->pg_count, (size_t)req->wb_bytes);
 513 }
 514 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
 515
 516 struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops)
 517 {
 518         struct nfs_pgio_header *hdr = ops->rw_alloc_header();
 519
 520         if (hdr) {
 521                 INIT_LIST_HEAD(&hdr->pages);
 522                 hdr->rw_ops = ops;
 523         }
 524         return hdr;
 525 }
 526 EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc);
 527
 528 /**
 529  * nfs_pgio_data_destroy - make @hdr suitable for reuse
 530  *
 531  * Frees memory and releases refs from nfs_generic_pgio, so that it may
 532  * be called again.
 533  *
 534  * @hdr: A header that has had nfs_generic_pgio called
 535  */
 536 static void nfs_pgio_data_destroy(struct nfs_pgio_header *hdr)
 537 {
 538         if (hdr->args.context)
 539                 put_nfs_open_context(hdr->args.context);
 540         if (hdr->page_array.pagevec != hdr->page_array.page_array)
 541                 kfree(hdr->page_array.pagevec);
 542 }
 543
 544 /*
 545  * nfs_pgio_header_free - Free a read or write header
 546  * @hdr: The header to free
 547  */
 548 void nfs_pgio_header_free(struct nfs_pgio_header *hdr)
 549 {
 550         nfs_pgio_data_destroy(hdr);
 551         hdr->rw_ops->rw_free_header(hdr);
 552 }
 553 EXPORT_SYMBOL_GPL(nfs_pgio_header_free);
 554
 555 /**
 556  * nfs_pgio_rpcsetup - Set up arguments for a pageio call
 557  * @hdr: The pageio hdr
 558  * @count: Number of bytes to read
 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(nfs_req_openctx(req));
 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   = count;
 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                       const struct 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  * @hdr: pageio header
 662  */
 663 static void nfs_pgio_error(struct nfs_pgio_header *hdr)
 664 {
 665         set_bit(NFS_IOHDR_REDO, &hdr->flags);
 666         hdr->completion_ops->completion(hdr);
 667 }
 668
 669 /**
 670  * nfs_pgio_release - Release pageio data
 671  * @calldata: The pageio header to release
 672  */
 673 static void nfs_pgio_release(void *calldata)
 674 {
 675         struct nfs_pgio_header *hdr = calldata;
 676         hdr->completion_ops->completion(hdr);
 677 }
 678
 679 static void nfs_pageio_mirror_init(struct nfs_pgio_mirror *mirror,
 680                                    unsigned int bsize)
 681 {
 682         INIT_LIST_HEAD(&mirror->pg_list);
 683         mirror->pg_bytes_written = 0;
 684         mirror->pg_count = 0;
 685         mirror->pg_bsize = bsize;
 686         mirror->pg_base = 0;
 687         mirror->pg_recoalesce = 0;
 688 }
 689
 690 /**
 691  * nfs_pageio_init - initialise a page io descriptor
 692  * @desc: pointer to descriptor
 693  * @inode: pointer to inode
 694  * @pg_ops: pointer to pageio operations
 695  * @compl_ops: pointer to pageio completion operations
 696  * @rw_ops: pointer to nfs read/write operations
 697  * @bsize: io block size
 698  * @io_flags: extra parameters for the io function
 699  */
 700 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
 701                      struct inode *inode,
 702                      const struct nfs_pageio_ops *pg_ops,
 703                      const struct nfs_pgio_completion_ops *compl_ops,
 704                      const struct nfs_rw_ops *rw_ops,
 705                      size_t bsize,
 706                      int io_flags)
 707 {
 708         desc->pg_moreio = 0;
 709         desc->pg_inode = inode;
 710         desc->pg_ops = pg_ops;
 711         desc->pg_completion_ops = compl_ops;
 712         desc->pg_rw_ops = rw_ops;
 713         desc->pg_ioflags = io_flags;
 714         desc->pg_error = 0;
 715         desc->pg_lseg = NULL;
 716         desc->pg_io_completion = NULL;
 717         desc->pg_dreq = NULL;
 718         desc->pg_bsize = bsize;
 719
 720         desc->pg_mirror_count = 1;
 721         desc->pg_mirror_idx = 0;
 722
 723         desc->pg_mirrors_dynamic = NULL;
 724         desc->pg_mirrors = desc->pg_mirrors_static;
 725         nfs_pageio_mirror_init(&desc->pg_mirrors[0], bsize);
 726         desc->pg_maxretrans = 0;
 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                 pg_array->pagevec = kcalloc(pagecount, sizeof(struct page *), gfp_flags);
 779                 if (!pg_array->pagevec) {
 780                         pg_array->npages = 0;
 781                         nfs_pgio_error(hdr);
 782                         desc->pg_error = -ENOMEM;
 783                         return desc->pg_error;
 784                 }
 785         }
 786
 787         nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
 788         pages = hdr->page_array.pagevec;
 789         last_page = NULL;
 790         pageused = 0;
 791         while (!list_empty(head)) {
 792                 req = nfs_list_entry(head->next);
 793                 nfs_list_move_request(req, &hdr->pages);
 794
 795                 if (!last_page || last_page != req->wb_page) {
 796                         pageused++;
 797                         if (pageused > pagecount)
 798                                 break;
 799                         *pages++ = last_page = req->wb_page;
 800                 }
 801         }
 802         if (WARN_ON_ONCE(pageused != pagecount)) {
 803                 nfs_pgio_error(hdr);
 804                 desc->pg_error = -EINVAL;
 805                 return desc->pg_error;
 806         }
 807
 808         if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
 809             (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
 810                 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
 811
 812         /* Set up the argument struct */
 813         nfs_pgio_rpcsetup(hdr, mirror->pg_count, desc->pg_ioflags, &cinfo);
 814         desc->pg_rpc_callops = &nfs_pgio_common_ops;
 815         return 0;
 816 }
 817 EXPORT_SYMBOL_GPL(nfs_generic_pgio);
 818
 819 static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
 820 {
 821         struct nfs_pgio_header *hdr;
 822         int ret;
 823
 824         hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
 825         if (!hdr) {
 826                 desc->pg_error = -ENOMEM;
 827                 return desc->pg_error;
 828         }
 829         nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
 830         ret = nfs_generic_pgio(desc, hdr);
 831         if (ret == 0)
 832                 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
 833                                         hdr,
 834                                         hdr->cred,
 835                                         NFS_PROTO(hdr->inode),
 836                                         desc->pg_rpc_callops,
 837                                         desc->pg_ioflags, 0);
 838         return ret;
 839 }
 840
 841 static struct nfs_pgio_mirror *
 842 nfs_pageio_alloc_mirrors(struct nfs_pageio_descriptor *desc,
 843                 unsigned int mirror_count)
 844 {
 845         struct nfs_pgio_mirror *ret;
 846         unsigned int i;
 847
 848         kfree(desc->pg_mirrors_dynamic);
 849         desc->pg_mirrors_dynamic = NULL;
 850         if (mirror_count == 1)
 851                 return desc->pg_mirrors_static;
 852         ret = kmalloc_array(mirror_count, sizeof(*ret), GFP_KERNEL);
 853         if (ret != NULL) {
 854                 for (i = 0; i < mirror_count; i++)
 855                         nfs_pageio_mirror_init(&ret[i], desc->pg_bsize);
 856                 desc->pg_mirrors_dynamic = ret;
 857         }
 858         return ret;
 859 }
 860
 861 /*
 862  * nfs_pageio_setup_mirroring - determine if mirroring is to be used
 863  *                              by calling the pg_get_mirror_count op
 864  */
 865 static void nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor *pgio,
 866                                        struct nfs_page *req)
 867 {
 868         unsigned int mirror_count = 1;
 869
 870         if (pgio->pg_ops->pg_get_mirror_count)
 871                 mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req);
 872         if (mirror_count == pgio->pg_mirror_count || pgio->pg_error < 0)
 873                 return;
 874
 875         if (!mirror_count || mirror_count > NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX) {
 876                 pgio->pg_error = -EINVAL;
 877                 return;
 878         }
 879
 880         pgio->pg_mirrors = nfs_pageio_alloc_mirrors(pgio, mirror_count);
 881         if (pgio->pg_mirrors == NULL) {
 882                 pgio->pg_error = -ENOMEM;
 883                 pgio->pg_mirrors = pgio->pg_mirrors_static;
 884                 mirror_count = 1;
 885         }
 886         pgio->pg_mirror_count = mirror_count;
 887 }
 888
 889 /*
 890  * nfs_pageio_stop_mirroring - stop using mirroring (set mirror count to 1)
 891  */
 892 void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio)
 893 {
 894         pgio->pg_mirror_count = 1;
 895         pgio->pg_mirror_idx = 0;
 896 }
 897
 898 static void nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor *pgio)
 899 {
 900         pgio->pg_mirror_count = 1;
 901         pgio->pg_mirror_idx = 0;
 902         pgio->pg_mirrors = pgio->pg_mirrors_static;
 903         kfree(pgio->pg_mirrors_dynamic);
 904         pgio->pg_mirrors_dynamic = NULL;
 905 }
 906
 907 static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
 908                 const struct nfs_lock_context *l2)
 909 {
 910         return l1->lockowner == l2->lockowner;
 911 }
 912
 913 /**
 914  * nfs_can_coalesce_requests - test two requests for compatibility
 915  * @prev: pointer to nfs_page
 916  * @req: pointer to nfs_page
 917  * @pgio: pointer to nfs_pagio_descriptor
 918  *
 919  * The nfs_page structures 'prev' and 'req' are compared to ensure that the
 920  * page data area they describe is contiguous, and that their RPC
 921  * credentials, NFSv4 open state, and lockowners are the same.
 922  *
 923  * Return 'true' if this is the case, else return 'false'.
 924  */
 925 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
 926                                       struct nfs_page *req,
 927                                       struct nfs_pageio_descriptor *pgio)
 928 {
 929         size_t size;
 930         struct file_lock_context *flctx;
 931
 932         if (prev) {
 933                 if (!nfs_match_open_context(nfs_req_openctx(req), nfs_req_openctx(prev)))
 934                         return false;
 935                 flctx = d_inode(nfs_req_openctx(req)->dentry)->i_flctx;
 936                 if (flctx != NULL &&
 937                     !(list_empty_careful(&flctx->flc_posix) &&
 938                       list_empty_careful(&flctx->flc_flock)) &&
 939                     !nfs_match_lock_context(req->wb_lock_context,
 940                                             prev->wb_lock_context))
 941                         return false;
 942                 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
 943                         return false;
 944                 if (req->wb_page == prev->wb_page) {
 945                         if (req->wb_pgbase != prev->wb_pgbase + prev->wb_bytes)
 946                                 return false;
 947                 } else {
 948                         if (req->wb_pgbase != 0 ||
 949                             prev->wb_pgbase + prev->wb_bytes != PAGE_SIZE)
 950                                 return false;
 951                 }
 952         }
 953         size = pgio->pg_ops->pg_test(pgio, prev, req);
 954         WARN_ON_ONCE(size > req->wb_bytes);
 955         if (size && size < req->wb_bytes)
 956                 req->wb_bytes = size;
 957         return size > 0;
 958 }
 959
 960 /**
 961  * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
 962  * @desc: destination io descriptor
 963  * @req: request
 964  *
 965  * Returns true if the request 'req' was successfully coalesced into the
 966  * existing list of pages 'desc'.
 967  */
 968 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
 969                                      struct nfs_page *req)
 970 {
 971         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
 972
 973         struct nfs_page *prev = NULL;
 974
 975         if (mirror->pg_count != 0) {
 976                 prev = nfs_list_entry(mirror->pg_list.prev);
 977         } else {
 978                 if (desc->pg_ops->pg_init)
 979                         desc->pg_ops->pg_init(desc, req);
 980                 if (desc->pg_error < 0)
 981                         return 0;
 982                 mirror->pg_base = req->wb_pgbase;
 983         }
 984
 985         if (desc->pg_maxretrans && req->wb_nio > desc->pg_maxretrans) {
 986                 if (NFS_SERVER(desc->pg_inode)->flags & NFS_MOUNT_SOFTERR)
 987                         desc->pg_error = -ETIMEDOUT;
 988                 else
 989                         desc->pg_error = -EIO;
 990                 return 0;
 991         }
 992
 993         if (!nfs_can_coalesce_requests(prev, req, desc))
 994                 return 0;
 995         nfs_list_move_request(req, &mirror->pg_list);
 996         mirror->pg_count += req->wb_bytes;
 997         return 1;
 998 }
 999
1000 /*
1001  * Helper for nfs_pageio_add_request and nfs_pageio_complete
1002  */
1003 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
1004 {
1005         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1006
1007
1008         if (!list_empty(&mirror->pg_list)) {
1009                 int error = desc->pg_ops->pg_doio(desc);
1010                 if (error < 0)
1011                         desc->pg_error = error;
1012                 else
1013                         mirror->pg_bytes_written += mirror->pg_count;
1014         }
1015         if (list_empty(&mirror->pg_list)) {
1016                 mirror->pg_count = 0;
1017                 mirror->pg_base = 0;
1018         }
1019 }
1020
1021 static void
1022 nfs_pageio_cleanup_request(struct nfs_pageio_descriptor *desc,
1023                 struct nfs_page *req)
1024 {
1025         LIST_HEAD(head);
1026
1027         nfs_list_move_request(req, &head);
1028         desc->pg_completion_ops->error_cleanup(&head, desc->pg_error);
1029 }
1030
1031 /**
1032  * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
1033  * @desc: destination io descriptor
1034  * @req: request
1035  *
1036  * This may split a request into subrequests which are all part of the
1037  * same page group.
1038  *
1039  * Returns true if the request 'req' was successfully coalesced into the
1040  * existing list of pages 'desc'.
1041  */
1042 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1043                            struct nfs_page *req)
1044 {
1045         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1046
1047         struct nfs_page *subreq;
1048         unsigned int bytes_left = 0;
1049         unsigned int offset, pgbase;
1050
1051         nfs_page_group_lock(req);
1052
1053         subreq = req;
1054         bytes_left = subreq->wb_bytes;
1055         offset = subreq->wb_offset;
1056         pgbase = subreq->wb_pgbase;
1057
1058         do {
1059                 if (!nfs_pageio_do_add_request(desc, subreq)) {
1060                         /* make sure pg_test call(s) did nothing */
1061                         WARN_ON_ONCE(subreq->wb_bytes != bytes_left);
1062                         WARN_ON_ONCE(subreq->wb_offset != offset);
1063                         WARN_ON_ONCE(subreq->wb_pgbase != pgbase);
1064
1065                         nfs_page_group_unlock(req);
1066                         desc->pg_moreio = 1;
1067                         nfs_pageio_doio(desc);
1068                         if (desc->pg_error < 0 || mirror->pg_recoalesce)
1069                                 goto out_cleanup_subreq;
1070                         /* retry add_request for this subreq */
1071                         nfs_page_group_lock(req);
1072                         continue;
1073                 }
1074
1075                 /* check for buggy pg_test call(s) */
1076                 WARN_ON_ONCE(subreq->wb_bytes + subreq->wb_pgbase > PAGE_SIZE);
1077                 WARN_ON_ONCE(subreq->wb_bytes > bytes_left);
1078                 WARN_ON_ONCE(subreq->wb_bytes == 0);
1079
1080                 bytes_left -= subreq->wb_bytes;
1081                 offset += subreq->wb_bytes;
1082                 pgbase += subreq->wb_bytes;
1083
1084                 if (bytes_left) {
1085                         subreq = nfs_create_subreq(req, subreq, pgbase,
1086                                         offset, bytes_left);
1087                         if (IS_ERR(subreq))
1088                                 goto err_ptr;
1089                 }
1090         } while (bytes_left > 0);
1091
1092         nfs_page_group_unlock(req);
1093         return 1;
1094 err_ptr:
1095         desc->pg_error = PTR_ERR(subreq);
1096         nfs_page_group_unlock(req);
1097         return 0;
1098 out_cleanup_subreq:
1099         if (req != subreq)
1100                 nfs_pageio_cleanup_request(desc, subreq);
1101         return 0;
1102 }
1103
1104 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
1105 {
1106         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1107         LIST_HEAD(head);
1108
1109         do {
1110                 list_splice_init(&mirror->pg_list, &head);
1111                 mirror->pg_bytes_written -= mirror->pg_count;
1112                 mirror->pg_count = 0;
1113                 mirror->pg_base = 0;
1114                 mirror->pg_recoalesce = 0;
1115
1116                 while (!list_empty(&head)) {
1117                         struct nfs_page *req;
1118
1119                         req = list_first_entry(&head, struct nfs_page, wb_list);
1120                         if (__nfs_pageio_add_request(desc, req))
1121                                 continue;
1122                         if (desc->pg_error < 0) {
1123                                 list_splice_tail(&head, &mirror->pg_list);
1124                                 mirror->pg_recoalesce = 1;
1125                                 return 0;
1126                         }
1127                         break;
1128                 }
1129         } while (mirror->pg_recoalesce);
1130         return 1;
1131 }
1132
1133 static int nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor *desc,
1134                 struct nfs_page *req)
1135 {
1136         int ret;
1137
1138         do {
1139                 ret = __nfs_pageio_add_request(desc, req);
1140                 if (ret)
1141                         break;
1142                 if (desc->pg_error < 0)
1143                         break;
1144                 ret = nfs_do_recoalesce(desc);
1145         } while (ret);
1146
1147         return ret;
1148 }
1149
1150 static void nfs_pageio_error_cleanup(struct nfs_pageio_descriptor *desc)
1151 {
1152         u32 midx;
1153         struct nfs_pgio_mirror *mirror;
1154
1155         if (!desc->pg_error)
1156                 return;
1157
1158         for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1159                 mirror = &desc->pg_mirrors[midx];
1160                 desc->pg_completion_ops->error_cleanup(&mirror->pg_list,
1161                                 desc->pg_error);
1162         }
1163 }
1164
1165 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1166                            struct nfs_page *req)
1167 {
1168         u32 midx;
1169         unsigned int pgbase, offset, bytes;
1170         struct nfs_page *dupreq, *lastreq;
1171
1172         pgbase = req->wb_pgbase;
1173         offset = req->wb_offset;
1174         bytes = req->wb_bytes;
1175
1176         nfs_pageio_setup_mirroring(desc, req);
1177         if (desc->pg_error < 0)
1178                 goto out_failed;
1179
1180         for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1181                 if (midx) {
1182                         nfs_page_group_lock(req);
1183
1184                         /* find the last request */
1185                         for (lastreq = req->wb_head;
1186                              lastreq->wb_this_page != req->wb_head;
1187                              lastreq = lastreq->wb_this_page)
1188                                 ;
1189
1190                         dupreq = nfs_create_subreq(req, lastreq,
1191                                         pgbase, offset, bytes);
1192
1193                         nfs_page_group_unlock(req);
1194                         if (IS_ERR(dupreq)) {
1195                                 desc->pg_error = PTR_ERR(dupreq);
1196                                 goto out_failed;
1197                         }
1198                 } else
1199                         dupreq = req;
1200
1201                 if (nfs_pgio_has_mirroring(desc))
1202                         desc->pg_mirror_idx = midx;
1203                 if (!nfs_pageio_add_request_mirror(desc, dupreq))
1204                         goto out_cleanup_subreq;
1205         }
1206
1207         return 1;
1208
1209 out_cleanup_subreq:
1210         if (req != dupreq)
1211                 nfs_pageio_cleanup_request(desc, dupreq);
1212 out_failed:
1213         nfs_pageio_error_cleanup(desc);
1214         return 0;
1215 }
1216
1217 /*
1218  * nfs_pageio_complete_mirror - Complete I/O on the current mirror of an
1219  *                              nfs_pageio_descriptor
1220  * @desc: pointer to io descriptor
1221  * @mirror_idx: pointer to mirror index
1222  */
1223 static void nfs_pageio_complete_mirror(struct nfs_pageio_descriptor *desc,
1224                                        u32 mirror_idx)
1225 {
1226         struct nfs_pgio_mirror *mirror = &desc->pg_mirrors[mirror_idx];
1227         u32 restore_idx = desc->pg_mirror_idx;
1228
1229         if (nfs_pgio_has_mirroring(desc))
1230                 desc->pg_mirror_idx = mirror_idx;
1231         for (;;) {
1232                 nfs_pageio_doio(desc);
1233                 if (desc->pg_error < 0 || !mirror->pg_recoalesce)
1234                         break;
1235                 if (!nfs_do_recoalesce(desc))
1236                         break;
1237         }
1238         desc->pg_mirror_idx = restore_idx;
1239 }
1240
1241 /*
1242  * nfs_pageio_resend - Transfer requests to new descriptor and resend
1243  * @hdr - the pgio header to move request from
1244  * @desc - the pageio descriptor to add requests to
1245  *
1246  * Try to move each request (nfs_page) from @hdr to @desc then attempt
1247  * to send them.
1248  *
1249  * Returns 0 on success and < 0 on error.
1250  */
1251 int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
1252                       struct nfs_pgio_header *hdr)
1253 {
1254         LIST_HEAD(failed);
1255
1256         desc->pg_io_completion = hdr->io_completion;
1257         desc->pg_dreq = hdr->dreq;
1258         while (!list_empty(&hdr->pages)) {
1259                 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
1260
1261                 if (!nfs_pageio_add_request(desc, req))
1262                         nfs_list_move_request(req, &failed);
1263         }
1264         nfs_pageio_complete(desc);
1265         if (!list_empty(&failed)) {
1266                 list_move(&failed, &hdr->pages);
1267                 return desc->pg_error < 0 ? desc->pg_error : -EIO;
1268         }
1269         return 0;
1270 }
1271 EXPORT_SYMBOL_GPL(nfs_pageio_resend);
1272
1273 /**
1274  * nfs_pageio_complete - Complete I/O then cleanup an nfs_pageio_descriptor
1275  * @desc: pointer to io descriptor
1276  */
1277 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1278 {
1279         u32 midx;
1280
1281         for (midx = 0; midx < desc->pg_mirror_count; midx++)
1282                 nfs_pageio_complete_mirror(desc, midx);
1283
1284         if (desc->pg_error < 0)
1285                 nfs_pageio_error_cleanup(desc);
1286         if (desc->pg_ops->pg_cleanup)
1287                 desc->pg_ops->pg_cleanup(desc);
1288         nfs_pageio_cleanup_mirroring(desc);
1289 }
1290
1291 /**
1292  * nfs_pageio_cond_complete - Conditional I/O completion
1293  * @desc: pointer to io descriptor
1294  * @index: page index
1295  *
1296  * It is important to ensure that processes don't try to take locks
1297  * on non-contiguous ranges of pages as that might deadlock. This
1298  * function should be called before attempting to wait on a locked
1299  * nfs_page. It will complete the I/O if the page index 'index'
1300  * is not contiguous with the existing list of pages in 'desc'.
1301  */
1302 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1303 {
1304         struct nfs_pgio_mirror *mirror;
1305         struct nfs_page *prev;
1306         u32 midx;
1307
1308         for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1309                 mirror = &desc->pg_mirrors[midx];
1310                 if (!list_empty(&mirror->pg_list)) {
1311                         prev = nfs_list_entry(mirror->pg_list.prev);
1312                         if (index != prev->wb_index + 1) {
1313                                 nfs_pageio_complete(desc);
1314                                 break;
1315                         }
1316                 }
1317         }
1318 }
1319
1320 int __init nfs_init_nfspagecache(void)
1321 {
1322         nfs_page_cachep = kmem_cache_create("nfs_page",
1323                                             sizeof(struct nfs_page),
1324                                             0, SLAB_HWCACHE_ALIGN,
1325                                             NULL);
1326         if (nfs_page_cachep == NULL)
1327                 return -ENOMEM;
1328
1329         return 0;
1330 }
1331
1332 void nfs_destroy_nfspagecache(void)
1333 {
1334         kmem_cache_destroy(nfs_page_cachep);
1335 }
1336
1337 static const struct rpc_call_ops nfs_pgio_common_ops = {
1338         .rpc_call_prepare = nfs_pgio_prepare,
1339         .rpc_call_done = nfs_pgio_result,
1340         .rpc_release = nfs_pgio_release,
1341 };
1342
1343 const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1344         .pg_test = nfs_generic_pg_test,
1345         .pg_doio = nfs_generic_pg_pgios,
1346 };