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 static void nfs_free_request(struct nfs_page *);
  33
  34 static bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount)
  35 {
  36         p->npages = pagecount;
  37         if (pagecount <= ARRAY_SIZE(p->page_array))
  38                 p->pagevec = p->page_array;
  39         else {
  40                 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
  41                 if (!p->pagevec)
  42                         p->npages = 0;
  43         }
  44         return p->pagevec != NULL;
  45 }
  46
  47 void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
  48                        struct nfs_pgio_header *hdr,
  49                        void (*release)(struct nfs_pgio_header *hdr))
  50 {
  51         hdr->req = nfs_list_entry(desc->pg_list.next);
  52         hdr->inode = desc->pg_inode;
  53         hdr->cred = hdr->req->wb_context->cred;
  54         hdr->io_start = req_offset(hdr->req);
  55         hdr->good_bytes = desc->pg_count;
  56         hdr->dreq = desc->pg_dreq;
  57         hdr->layout_private = desc->pg_layout_private;
  58         hdr->release = release;
  59         hdr->completion_ops = desc->pg_completion_ops;
  60         if (hdr->completion_ops->init_hdr)
  61                 hdr->completion_ops->init_hdr(hdr);
  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         spin_lock(&hdr->lock);
  68         if (pos < hdr->io_start + hdr->good_bytes) {
  69                 set_bit(NFS_IOHDR_ERROR, &hdr->flags);
  70                 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
  71                 hdr->good_bytes = pos - hdr->io_start;
  72                 hdr->error = error;
  73         }
  74         spin_unlock(&hdr->lock);
  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_NOIO);
  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 static void
  93 nfs_iocounter_inc(struct nfs_io_counter *c)
  94 {
  95         atomic_inc(&c->io_count);
  96 }
  97
  98 static void
  99 nfs_iocounter_dec(struct nfs_io_counter *c)
 100 {
 101         if (atomic_dec_and_test(&c->io_count)) {
 102                 clear_bit(NFS_IO_INPROGRESS, &c->flags);
 103                 smp_mb__after_atomic();
 104                 wake_up_bit(&c->flags, NFS_IO_INPROGRESS);
 105         }
 106 }
 107
 108 static int
 109 __nfs_iocounter_wait(struct nfs_io_counter *c)
 110 {
 111         wait_queue_head_t *wq = bit_waitqueue(&c->flags, NFS_IO_INPROGRESS);
 112         DEFINE_WAIT_BIT(q, &c->flags, NFS_IO_INPROGRESS);
 113         int ret = 0;
 114
 115         do {
 116                 prepare_to_wait(wq, &q.wait, TASK_KILLABLE);
 117                 set_bit(NFS_IO_INPROGRESS, &c->flags);
 118                 if (atomic_read(&c->io_count) == 0)
 119                         break;
 120                 ret = nfs_wait_bit_killable(&c->flags);
 121         } while (atomic_read(&c->io_count) != 0);
 122         finish_wait(wq, &q.wait);
 123         return ret;
 124 }
 125
 126 /**
 127  * nfs_iocounter_wait - wait for i/o to complete
 128  * @c: nfs_io_counter to use
 129  *
 130  * returns -ERESTARTSYS if interrupted by a fatal signal.
 131  * Otherwise returns 0 once the io_count hits 0.
 132  */
 133 int
 134 nfs_iocounter_wait(struct nfs_io_counter *c)
 135 {
 136         if (atomic_read(&c->io_count) == 0)
 137                 return 0;
 138         return __nfs_iocounter_wait(c);
 139 }
 140
 141 static int nfs_wait_bit_uninterruptible(void *word)
 142 {
 143         io_schedule();
 144         return 0;
 145 }
 146
 147 /*
 148  * nfs_page_group_lock - lock the head of the page group
 149  * @req - request in group that is to be locked
 150  *
 151  * this lock must be held if modifying the page group list
 152  */
 153 void
 154 nfs_page_group_lock(struct nfs_page *req)
 155 {
 156         struct nfs_page *head = req->wb_head;
 157
 158         WARN_ON_ONCE(head != head->wb_head);
 159
 160         wait_on_bit_lock(&head->wb_flags, PG_HEADLOCK,
 161                         nfs_wait_bit_uninterruptible,
 162                         TASK_UNINTERRUPTIBLE);
 163 }
 164
 165 /*
 166  * nfs_page_group_unlock - unlock the head of the page group
 167  * @req - request in group that is to be unlocked
 168  */
 169 void
 170 nfs_page_group_unlock(struct nfs_page *req)
 171 {
 172         struct nfs_page *head = req->wb_head;
 173
 174         WARN_ON_ONCE(head != head->wb_head);
 175
 176         smp_mb__before_atomic();
 177         clear_bit(PG_HEADLOCK, &head->wb_flags);
 178         smp_mb__after_atomic();
 179         wake_up_bit(&head->wb_flags, PG_HEADLOCK);
 180 }
 181
 182 /*
 183  * nfs_page_group_sync_on_bit_locked
 184  *
 185  * must be called with page group lock held
 186  */
 187 static bool
 188 nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
 189 {
 190         struct nfs_page *head = req->wb_head;
 191         struct nfs_page *tmp;
 192
 193         WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
 194         WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
 195
 196         tmp = req->wb_this_page;
 197         while (tmp != req) {
 198                 if (!test_bit(bit, &tmp->wb_flags))
 199                         return false;
 200                 tmp = tmp->wb_this_page;
 201         }
 202
 203         /* true! reset all bits */
 204         tmp = req;
 205         do {
 206                 clear_bit(bit, &tmp->wb_flags);
 207                 tmp = tmp->wb_this_page;
 208         } while (tmp != req);
 209
 210         return true;
 211 }
 212
 213 /*
 214  * nfs_page_group_sync_on_bit - set bit on current request, but only
 215  *   return true if the bit is set for all requests in page group
 216  * @req - request in page group
 217  * @bit - PG_* bit that is used to sync page group
 218  */
 219 bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
 220 {
 221         bool ret;
 222
 223         nfs_page_group_lock(req);
 224         ret = nfs_page_group_sync_on_bit_locked(req, bit);
 225         nfs_page_group_unlock(req);
 226
 227         return ret;
 228 }
 229
 230 /*
 231  * nfs_page_group_init - Initialize the page group linkage for @req
 232  * @req - a new nfs request
 233  * @prev - the previous request in page group, or NULL if @req is the first
 234  *         or only request in the group (the head).
 235  */
 236 static inline void
 237 nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
 238 {
 239         WARN_ON_ONCE(prev == req);
 240
 241         if (!prev) {
 242                 req->wb_head = req;
 243                 req->wb_this_page = req;
 244         } else {
 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                 /* grab extra ref if head request has extra ref from
 252                  * the write/commit path to handle handoff between write
 253                  * and commit lists */
 254                 if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags))
 255                         kref_get(&req->wb_kref);
 256         }
 257 }
 258
 259 /*
 260  * nfs_page_group_destroy - sync the destruction of page groups
 261  * @req - request that no longer needs the page group
 262  *
 263  * releases the page group reference from each member once all
 264  * members have called this function.
 265  */
 266 static void
 267 nfs_page_group_destroy(struct kref *kref)
 268 {
 269         struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
 270         struct nfs_page *tmp, *next;
 271
 272         if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
 273                 return;
 274
 275         tmp = req;
 276         do {
 277                 next = tmp->wb_this_page;
 278                 /* unlink and free */
 279                 tmp->wb_this_page = tmp;
 280                 tmp->wb_head = tmp;
 281                 nfs_free_request(tmp);
 282                 tmp = next;
 283         } while (tmp != req);
 284 }
 285
 286 /**
 287  * nfs_create_request - Create an NFS read/write request.
 288  * @ctx: open context to use
 289  * @page: page to write
 290  * @last: last nfs request created for this page group or NULL if head
 291  * @offset: starting offset within the page for the write
 292  * @count: number of bytes to read/write
 293  *
 294  * The page must be locked by the caller. This makes sure we never
 295  * create two different requests for the same page.
 296  * User should ensure it is safe to sleep in this function.
 297  */
 298 struct nfs_page *
 299 nfs_create_request(struct nfs_open_context *ctx, struct page *page,
 300                    struct nfs_page *last, unsigned int offset,
 301                    unsigned int count)
 302 {
 303         struct nfs_page         *req;
 304         struct nfs_lock_context *l_ctx;
 305
 306         if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
 307                 return ERR_PTR(-EBADF);
 308         /* try to allocate the request struct */
 309         req = nfs_page_alloc();
 310         if (req == NULL)
 311                 return ERR_PTR(-ENOMEM);
 312
 313         /* get lock context early so we can deal with alloc failures */
 314         l_ctx = nfs_get_lock_context(ctx);
 315         if (IS_ERR(l_ctx)) {
 316                 nfs_page_free(req);
 317                 return ERR_CAST(l_ctx);
 318         }
 319         req->wb_lock_context = l_ctx;
 320         nfs_iocounter_inc(&l_ctx->io_count);
 321
 322         /* Initialize the request struct. Initially, we assume a
 323          * long write-back delay. This will be adjusted in
 324          * update_nfs_request below if the region is not locked. */
 325         req->wb_page    = page;
 326         req->wb_index   = page_file_index(page);
 327         page_cache_get(page);
 328         req->wb_offset  = offset;
 329         req->wb_pgbase  = offset;
 330         req->wb_bytes   = count;
 331         req->wb_context = get_nfs_open_context(ctx);
 332         kref_init(&req->wb_kref);
 333         nfs_page_group_init(req, last);
 334         return req;
 335 }
 336
 337 /**
 338  * nfs_unlock_request - Unlock request and wake up sleepers.
 339  * @req:
 340  */
 341 void nfs_unlock_request(struct nfs_page *req)
 342 {
 343         if (!NFS_WBACK_BUSY(req)) {
 344                 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
 345                 BUG();
 346         }
 347         smp_mb__before_atomic();
 348         clear_bit(PG_BUSY, &req->wb_flags);
 349         smp_mb__after_atomic();
 350         wake_up_bit(&req->wb_flags, PG_BUSY);
 351 }
 352
 353 /**
 354  * nfs_unlock_and_release_request - Unlock request and release the nfs_page
 355  * @req:
 356  */
 357 void nfs_unlock_and_release_request(struct nfs_page *req)
 358 {
 359         nfs_unlock_request(req);
 360         nfs_release_request(req);
 361 }
 362
 363 /*
 364  * nfs_clear_request - Free up all resources allocated to the request
 365  * @req:
 366  *
 367  * Release page and open context resources associated with a read/write
 368  * request after it has completed.
 369  */
 370 static void nfs_clear_request(struct nfs_page *req)
 371 {
 372         struct page *page = req->wb_page;
 373         struct nfs_open_context *ctx = req->wb_context;
 374         struct nfs_lock_context *l_ctx = req->wb_lock_context;
 375
 376         if (page != NULL) {
 377                 page_cache_release(page);
 378                 req->wb_page = NULL;
 379         }
 380         if (l_ctx != NULL) {
 381                 nfs_iocounter_dec(&l_ctx->io_count);
 382                 nfs_put_lock_context(l_ctx);
 383                 req->wb_lock_context = NULL;
 384         }
 385         if (ctx != NULL) {
 386                 put_nfs_open_context(ctx);
 387                 req->wb_context = NULL;
 388         }
 389 }
 390
 391 /**
 392  * nfs_release_request - Release the count on an NFS read/write request
 393  * @req: request to release
 394  *
 395  * Note: Should never be called with the spinlock held!
 396  */
 397 static void nfs_free_request(struct nfs_page *req)
 398 {
 399         WARN_ON_ONCE(req->wb_this_page != req);
 400
 401         /* extra debug: make sure no sync bits are still set */
 402         WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
 403         WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
 404         WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
 405         WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
 406         WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
 407
 408         /* Release struct file and open context */
 409         nfs_clear_request(req);
 410         nfs_page_free(req);
 411 }
 412
 413 void nfs_release_request(struct nfs_page *req)
 414 {
 415         kref_put(&req->wb_kref, nfs_page_group_destroy);
 416 }
 417
 418 /**
 419  * nfs_wait_on_request - Wait for a request to complete.
 420  * @req: request to wait upon.
 421  *
 422  * Interruptible by fatal signals only.
 423  * The user is responsible for holding a count on the request.
 424  */
 425 int
 426 nfs_wait_on_request(struct nfs_page *req)
 427 {
 428         return wait_on_bit(&req->wb_flags, PG_BUSY,
 429                         nfs_wait_bit_uninterruptible,
 430                         TASK_UNINTERRUPTIBLE);
 431 }
 432
 433 /*
 434  * nfs_generic_pg_test - determine if requests can be coalesced
 435  * @desc: pointer to descriptor
 436  * @prev: previous request in desc, or NULL
 437  * @req: this request
 438  *
 439  * Returns zero if @req can be coalesced into @desc, otherwise it returns
 440  * the size of the request.
 441  */
 442 size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
 443                            struct nfs_page *prev, struct nfs_page *req)
 444 {
 445         if (desc->pg_count > desc->pg_bsize) {
 446                 /* should never happen */
 447                 WARN_ON_ONCE(1);
 448                 return 0;
 449         }
 450
 451         return min(desc->pg_bsize - desc->pg_count, (size_t)req->wb_bytes);
 452 }
 453 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
 454
 455 static inline struct nfs_rw_header *NFS_RW_HEADER(struct nfs_pgio_header *hdr)
 456 {
 457         return container_of(hdr, struct nfs_rw_header, header);
 458 }
 459
 460 /**
 461  * nfs_rw_header_alloc - Allocate a header for a read or write
 462  * @ops: Read or write function vector
 463  */
 464 struct nfs_rw_header *nfs_rw_header_alloc(const struct nfs_rw_ops *ops)
 465 {
 466         struct nfs_rw_header *header = ops->rw_alloc_header();
 467
 468         if (header) {
 469                 struct nfs_pgio_header *hdr = &header->header;
 470
 471                 INIT_LIST_HEAD(&hdr->pages);
 472                 spin_lock_init(&hdr->lock);
 473                 atomic_set(&hdr->refcnt, 0);
 474                 hdr->rw_ops = ops;
 475         }
 476         return header;
 477 }
 478 EXPORT_SYMBOL_GPL(nfs_rw_header_alloc);
 479
 480 /*
 481  * nfs_rw_header_free - Free a read or write header
 482  * @hdr: The header to free
 483  */
 484 void nfs_rw_header_free(struct nfs_pgio_header *hdr)
 485 {
 486         hdr->rw_ops->rw_free_header(NFS_RW_HEADER(hdr));
 487 }
 488 EXPORT_SYMBOL_GPL(nfs_rw_header_free);
 489
 490 /**
 491  * nfs_pgio_data_alloc - Allocate pageio data
 492  * @hdr: The header making a request
 493  * @pagecount: Number of pages to create
 494  */
 495 static struct nfs_pgio_data *nfs_pgio_data_alloc(struct nfs_pgio_header *hdr,
 496                                                  unsigned int pagecount)
 497 {
 498         struct nfs_pgio_data *data, *prealloc;
 499
 500         prealloc = &NFS_RW_HEADER(hdr)->rpc_data;
 501         if (prealloc->header == NULL)
 502                 data = prealloc;
 503         else
 504                 data = kzalloc(sizeof(*data), GFP_KERNEL);
 505         if (!data)
 506                 goto out;
 507
 508         if (nfs_pgarray_set(&data->pages, pagecount)) {
 509                 data->header = hdr;
 510                 atomic_inc(&hdr->refcnt);
 511         } else {
 512                 if (data != prealloc)
 513                         kfree(data);
 514                 data = NULL;
 515         }
 516 out:
 517         return data;
 518 }
 519
 520 /**
 521  * nfs_pgio_data_release - Properly free pageio data
 522  * @data: The data to release
 523  */
 524 void nfs_pgio_data_release(struct nfs_pgio_data *data)
 525 {
 526         struct nfs_pgio_header *hdr = data->header;
 527         struct nfs_rw_header *pageio_header = NFS_RW_HEADER(hdr);
 528
 529         put_nfs_open_context(data->args.context);
 530         if (data->pages.pagevec != data->pages.page_array)
 531                 kfree(data->pages.pagevec);
 532         if (data == &pageio_header->rpc_data) {
 533                 data->header = NULL;
 534                 data = NULL;
 535         }
 536         if (atomic_dec_and_test(&hdr->refcnt))
 537                 hdr->completion_ops->completion(hdr);
 538         /* Note: we only free the rpc_task after callbacks are done.
 539          * See the comment in rpc_free_task() for why
 540          */
 541         kfree(data);
 542 }
 543 EXPORT_SYMBOL_GPL(nfs_pgio_data_release);
 544
 545 /**
 546  * nfs_pgio_rpcsetup - Set up arguments for a pageio call
 547  * @data: The pageio data
 548  * @count: Number of bytes to read
 549  * @offset: Initial offset
 550  * @how: How to commit data (writes only)
 551  * @cinfo: Commit information for the call (writes only)
 552  */
 553 static void nfs_pgio_rpcsetup(struct nfs_pgio_data *data,
 554                               unsigned int count, unsigned int offset,
 555                               int how, struct nfs_commit_info *cinfo)
 556 {
 557         struct nfs_page *req = data->header->req;
 558
 559         /* Set up the RPC argument and reply structs
 560          * NB: take care not to mess about with data->commit et al. */
 561
 562         data->args.fh     = NFS_FH(data->header->inode);
 563         data->args.offset = req_offset(req) + offset;
 564         /* pnfs_set_layoutcommit needs this */
 565         data->mds_offset = data->args.offset;
 566         data->args.pgbase = req->wb_pgbase + offset;
 567         data->args.pages  = data->pages.pagevec;
 568         data->args.count  = count;
 569         data->args.context = get_nfs_open_context(req->wb_context);
 570         data->args.lock_context = req->wb_lock_context;
 571         data->args.stable  = NFS_UNSTABLE;
 572         switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
 573         case 0:
 574                 break;
 575         case FLUSH_COND_STABLE:
 576                 if (nfs_reqs_to_commit(cinfo))
 577                         break;
 578         default:
 579                 data->args.stable = NFS_FILE_SYNC;
 580         }
 581
 582         data->res.fattr   = &data->fattr;
 583         data->res.count   = count;
 584         data->res.eof     = 0;
 585         data->res.verf    = &data->verf;
 586         nfs_fattr_init(&data->fattr);
 587 }
 588
 589 /**
 590  * nfs_pgio_prepare - Prepare pageio data to go over the wire
 591  * @task: The current task
 592  * @calldata: pageio data to prepare
 593  */
 594 static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
 595 {
 596         struct nfs_pgio_data *data = calldata;
 597         int err;
 598         err = NFS_PROTO(data->header->inode)->pgio_rpc_prepare(task, data);
 599         if (err)
 600                 rpc_exit(task, err);
 601 }
 602
 603 int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_data *data,
 604                       const struct rpc_call_ops *call_ops, int how, int flags)
 605 {
 606         struct rpc_task *task;
 607         struct rpc_message msg = {
 608                 .rpc_argp = &data->args,
 609                 .rpc_resp = &data->res,
 610                 .rpc_cred = data->header->cred,
 611         };
 612         struct rpc_task_setup task_setup_data = {
 613                 .rpc_client = clnt,
 614                 .task = &data->task,
 615                 .rpc_message = &msg,
 616                 .callback_ops = call_ops,
 617                 .callback_data = data,
 618                 .workqueue = nfsiod_workqueue,
 619                 .flags = RPC_TASK_ASYNC | flags,
 620         };
 621         int ret = 0;
 622
 623         data->header->rw_ops->rw_initiate(data, &msg, &task_setup_data, how);
 624
 625         dprintk("NFS: %5u initiated pgio call "
 626                 "(req %s/%llu, %u bytes @ offset %llu)\n",
 627                 data->task.tk_pid,
 628                 data->header->inode->i_sb->s_id,
 629                 (unsigned long long)NFS_FILEID(data->header->inode),
 630                 data->args.count,
 631                 (unsigned long long)data->args.offset);
 632
 633         task = rpc_run_task(&task_setup_data);
 634         if (IS_ERR(task)) {
 635                 ret = PTR_ERR(task);
 636                 goto out;
 637         }
 638         if (how & FLUSH_SYNC) {
 639                 ret = rpc_wait_for_completion_task(task);
 640                 if (ret == 0)
 641                         ret = task->tk_status;
 642         }
 643         rpc_put_task(task);
 644 out:
 645         return ret;
 646 }
 647 EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
 648
 649 /**
 650  * nfs_pgio_error - Clean up from a pageio error
 651  * @desc: IO descriptor
 652  * @hdr: pageio header
 653  */
 654 static int nfs_pgio_error(struct nfs_pageio_descriptor *desc,
 655                           struct nfs_pgio_header *hdr)
 656 {
 657         set_bit(NFS_IOHDR_REDO, &hdr->flags);
 658         nfs_pgio_data_release(hdr->data);
 659         hdr->data = NULL;
 660         desc->pg_completion_ops->error_cleanup(&desc->pg_list);
 661         return -ENOMEM;
 662 }
 663
 664 /**
 665  * nfs_pgio_release - Release pageio data
 666  * @calldata: The pageio data to release
 667  */
 668 static void nfs_pgio_release(void *calldata)
 669 {
 670         struct nfs_pgio_data *data = calldata;
 671         if (data->header->rw_ops->rw_release)
 672                 data->header->rw_ops->rw_release(data);
 673         nfs_pgio_data_release(data);
 674 }
 675
 676 /**
 677  * nfs_pageio_init - initialise a page io descriptor
 678  * @desc: pointer to descriptor
 679  * @inode: pointer to inode
 680  * @doio: pointer to io function
 681  * @bsize: io block size
 682  * @io_flags: extra parameters for the io function
 683  */
 684 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
 685                      struct inode *inode,
 686                      const struct nfs_pageio_ops *pg_ops,
 687                      const struct nfs_pgio_completion_ops *compl_ops,
 688                      const struct nfs_rw_ops *rw_ops,
 689                      size_t bsize,
 690                      int io_flags)
 691 {
 692         INIT_LIST_HEAD(&desc->pg_list);
 693         desc->pg_bytes_written = 0;
 694         desc->pg_count = 0;
 695         desc->pg_bsize = bsize;
 696         desc->pg_base = 0;
 697         desc->pg_moreio = 0;
 698         desc->pg_recoalesce = 0;
 699         desc->pg_inode = inode;
 700         desc->pg_ops = pg_ops;
 701         desc->pg_completion_ops = compl_ops;
 702         desc->pg_rw_ops = rw_ops;
 703         desc->pg_ioflags = io_flags;
 704         desc->pg_error = 0;
 705         desc->pg_lseg = NULL;
 706         desc->pg_dreq = NULL;
 707         desc->pg_layout_private = NULL;
 708 }
 709 EXPORT_SYMBOL_GPL(nfs_pageio_init);
 710
 711 /**
 712  * nfs_pgio_result - Basic pageio error handling
 713  * @task: The task that ran
 714  * @calldata: Pageio data to check
 715  */
 716 static void nfs_pgio_result(struct rpc_task *task, void *calldata)
 717 {
 718         struct nfs_pgio_data *data = calldata;
 719         struct inode *inode = data->header->inode;
 720
 721         dprintk("NFS: %s: %5u, (status %d)\n", __func__,
 722                 task->tk_pid, task->tk_status);
 723
 724         if (data->header->rw_ops->rw_done(task, data, inode) != 0)
 725                 return;
 726         if (task->tk_status < 0)
 727                 nfs_set_pgio_error(data->header, task->tk_status, data->args.offset);
 728         else
 729                 data->header->rw_ops->rw_result(task, data);
 730 }
 731
 732 /*
 733  * Create an RPC task for the given read or write request and kick it.
 734  * The page must have been locked by the caller.
 735  *
 736  * It may happen that the page we're passed is not marked dirty.
 737  * This is the case if nfs_updatepage detects a conflicting request
 738  * that has been written but not committed.
 739  */
 740 int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
 741                      struct nfs_pgio_header *hdr)
 742 {
 743         struct nfs_page         *req;
 744         struct page             **pages;
 745         struct nfs_pgio_data    *data;
 746         struct list_head *head = &desc->pg_list;
 747         struct nfs_commit_info cinfo;
 748
 749         data = nfs_pgio_data_alloc(hdr, nfs_page_array_len(desc->pg_base,
 750                                                            desc->pg_count));
 751         if (!data)
 752                 return nfs_pgio_error(desc, hdr);
 753
 754         nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
 755         pages = data->pages.pagevec;
 756         while (!list_empty(head)) {
 757                 req = nfs_list_entry(head->next);
 758                 nfs_list_remove_request(req);
 759                 nfs_list_add_request(req, &hdr->pages);
 760                 *pages++ = req->wb_page;
 761         }
 762
 763         if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
 764             (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
 765                 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
 766
 767         /* Set up the argument struct */
 768         nfs_pgio_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags, &cinfo);
 769         hdr->data = data;
 770         desc->pg_rpc_callops = &nfs_pgio_common_ops;
 771         return 0;
 772 }
 773 EXPORT_SYMBOL_GPL(nfs_generic_pgio);
 774
 775 static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
 776 {
 777         struct nfs_rw_header *rw_hdr;
 778         struct nfs_pgio_header *hdr;
 779         int ret;
 780
 781         rw_hdr = nfs_rw_header_alloc(desc->pg_rw_ops);
 782         if (!rw_hdr) {
 783                 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
 784                 return -ENOMEM;
 785         }
 786         hdr = &rw_hdr->header;
 787         nfs_pgheader_init(desc, hdr, nfs_rw_header_free);
 788         atomic_inc(&hdr->refcnt);
 789         ret = nfs_generic_pgio(desc, hdr);
 790         if (ret == 0)
 791                 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
 792                                         hdr->data, desc->pg_rpc_callops,
 793                                         desc->pg_ioflags, 0);
 794         if (atomic_dec_and_test(&hdr->refcnt))
 795                 hdr->completion_ops->completion(hdr);
 796         return ret;
 797 }
 798
 799 static bool nfs_match_open_context(const struct nfs_open_context *ctx1,
 800                 const struct nfs_open_context *ctx2)
 801 {
 802         return ctx1->cred == ctx2->cred && ctx1->state == ctx2->state;
 803 }
 804
 805 static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
 806                 const struct nfs_lock_context *l2)
 807 {
 808         return l1->lockowner.l_owner == l2->lockowner.l_owner
 809                 && l1->lockowner.l_pid == l2->lockowner.l_pid;
 810 }
 811
 812 /**
 813  * nfs_can_coalesce_requests - test two requests for compatibility
 814  * @prev: pointer to nfs_page
 815  * @req: pointer to nfs_page
 816  *
 817  * The nfs_page structures 'prev' and 'req' are compared to ensure that the
 818  * page data area they describe is contiguous, and that their RPC
 819  * credentials, NFSv4 open state, and lockowners are the same.
 820  *
 821  * Return 'true' if this is the case, else return 'false'.
 822  */
 823 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
 824                                       struct nfs_page *req,
 825                                       struct nfs_pageio_descriptor *pgio)
 826 {
 827         size_t size;
 828
 829         if (prev) {
 830                 if (!nfs_match_open_context(req->wb_context, prev->wb_context))
 831                         return false;
 832                 if (req->wb_context->dentry->d_inode->i_flock != NULL &&
 833                     !nfs_match_lock_context(req->wb_lock_context,
 834                                             prev->wb_lock_context))
 835                         return false;
 836                 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
 837                         return false;
 838         }
 839         size = pgio->pg_ops->pg_test(pgio, prev, req);
 840         WARN_ON_ONCE(size > req->wb_bytes);
 841         if (size && size < req->wb_bytes)
 842                 req->wb_bytes = size;
 843         return size > 0;
 844 }
 845
 846 /**
 847  * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
 848  * @desc: destination io descriptor
 849  * @req: request
 850  *
 851  * Returns true if the request 'req' was successfully coalesced into the
 852  * existing list of pages 'desc'.
 853  */
 854 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
 855                                      struct nfs_page *req)
 856 {
 857         struct nfs_page *prev = NULL;
 858         if (desc->pg_count != 0) {
 859                 prev = nfs_list_entry(desc->pg_list.prev);
 860         } else {
 861                 if (desc->pg_ops->pg_init)
 862                         desc->pg_ops->pg_init(desc, req);
 863                 desc->pg_base = req->wb_pgbase;
 864         }
 865         if (!nfs_can_coalesce_requests(prev, req, desc))
 866                 return 0;
 867         nfs_list_remove_request(req);
 868         nfs_list_add_request(req, &desc->pg_list);
 869         desc->pg_count += req->wb_bytes;
 870         return 1;
 871 }
 872
 873 /*
 874  * Helper for nfs_pageio_add_request and nfs_pageio_complete
 875  */
 876 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
 877 {
 878         if (!list_empty(&desc->pg_list)) {
 879                 int error = desc->pg_ops->pg_doio(desc);
 880                 if (error < 0)
 881                         desc->pg_error = error;
 882                 else
 883                         desc->pg_bytes_written += desc->pg_count;
 884         }
 885         if (list_empty(&desc->pg_list)) {
 886                 desc->pg_count = 0;
 887                 desc->pg_base = 0;
 888         }
 889 }
 890
 891 /**
 892  * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
 893  * @desc: destination io descriptor
 894  * @req: request
 895  *
 896  * This may split a request into subrequests which are all part of the
 897  * same page group.
 898  *
 899  * Returns true if the request 'req' was successfully coalesced into the
 900  * existing list of pages 'desc'.
 901  */
 902 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
 903                            struct nfs_page *req)
 904 {
 905         struct nfs_page *subreq;
 906         unsigned int bytes_left = 0;
 907         unsigned int offset, pgbase;
 908
 909         nfs_page_group_lock(req);
 910
 911         subreq = req;
 912         bytes_left = subreq->wb_bytes;
 913         offset = subreq->wb_offset;
 914         pgbase = subreq->wb_pgbase;
 915
 916         do {
 917                 if (!nfs_pageio_do_add_request(desc, subreq)) {
 918                         /* make sure pg_test call(s) did nothing */
 919                         WARN_ON_ONCE(subreq->wb_bytes != bytes_left);
 920                         WARN_ON_ONCE(subreq->wb_offset != offset);
 921                         WARN_ON_ONCE(subreq->wb_pgbase != pgbase);
 922
 923                         nfs_page_group_unlock(req);
 924                         desc->pg_moreio = 1;
 925                         nfs_pageio_doio(desc);
 926                         if (desc->pg_error < 0)
 927                                 return 0;
 928                         desc->pg_moreio = 0;
 929                         if (desc->pg_recoalesce)
 930                                 return 0;
 931                         /* retry add_request for this subreq */
 932                         nfs_page_group_lock(req);
 933                         continue;
 934                 }
 935
 936                 /* check for buggy pg_test call(s) */
 937                 WARN_ON_ONCE(subreq->wb_bytes + subreq->wb_pgbase > PAGE_SIZE);
 938                 WARN_ON_ONCE(subreq->wb_bytes > bytes_left);
 939                 WARN_ON_ONCE(subreq->wb_bytes == 0);
 940
 941                 bytes_left -= subreq->wb_bytes;
 942                 offset += subreq->wb_bytes;
 943                 pgbase += subreq->wb_bytes;
 944
 945                 if (bytes_left) {
 946                         subreq = nfs_create_request(req->wb_context,
 947                                         req->wb_page,
 948                                         subreq, pgbase, bytes_left);
 949                         if (IS_ERR(subreq))
 950                                 goto err_ptr;
 951                         nfs_lock_request(subreq);
 952                         subreq->wb_offset  = offset;
 953                         subreq->wb_index = req->wb_index;
 954                 }
 955         } while (bytes_left > 0);
 956
 957         nfs_page_group_unlock(req);
 958         return 1;
 959 err_ptr:
 960         desc->pg_error = PTR_ERR(subreq);
 961         nfs_page_group_unlock(req);
 962         return 0;
 963 }
 964
 965 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
 966 {
 967         LIST_HEAD(head);
 968
 969         do {
 970                 list_splice_init(&desc->pg_list, &head);
 971                 desc->pg_bytes_written -= desc->pg_count;
 972                 desc->pg_count = 0;
 973                 desc->pg_base = 0;
 974                 desc->pg_recoalesce = 0;
 975
 976                 while (!list_empty(&head)) {
 977                         struct nfs_page *req;
 978
 979                         req = list_first_entry(&head, struct nfs_page, wb_list);
 980                         nfs_list_remove_request(req);
 981                         if (__nfs_pageio_add_request(desc, req))
 982                                 continue;
 983                         if (desc->pg_error < 0)
 984                                 return 0;
 985                         break;
 986                 }
 987         } while (desc->pg_recoalesce);
 988         return 1;
 989 }
 990
 991 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
 992                 struct nfs_page *req)
 993 {
 994         int ret;
 995
 996         do {
 997                 ret = __nfs_pageio_add_request(desc, req);
 998                 if (ret)
 999                         break;
1000                 if (desc->pg_error < 0)
1001                         break;
1002                 ret = nfs_do_recoalesce(desc);
1003         } while (ret);
1004         return ret;
1005 }
1006 EXPORT_SYMBOL_GPL(nfs_pageio_add_request);
1007
1008 /**
1009  * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
1010  * @desc: pointer to io descriptor
1011  */
1012 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1013 {
1014         for (;;) {
1015                 nfs_pageio_doio(desc);
1016                 if (!desc->pg_recoalesce)
1017                         break;
1018                 if (!nfs_do_recoalesce(desc))
1019                         break;
1020         }
1021 }
1022 EXPORT_SYMBOL_GPL(nfs_pageio_complete);
1023
1024 /**
1025  * nfs_pageio_cond_complete - Conditional I/O completion
1026  * @desc: pointer to io descriptor
1027  * @index: page index
1028  *
1029  * It is important to ensure that processes don't try to take locks
1030  * on non-contiguous ranges of pages as that might deadlock. This
1031  * function should be called before attempting to wait on a locked
1032  * nfs_page. It will complete the I/O if the page index 'index'
1033  * is not contiguous with the existing list of pages in 'desc'.
1034  */
1035 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1036 {
1037         if (!list_empty(&desc->pg_list)) {
1038                 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
1039                 if (index != prev->wb_index + 1)
1040                         nfs_pageio_complete(desc);
1041         }
1042 }
1043
1044 int __init nfs_init_nfspagecache(void)
1045 {
1046         nfs_page_cachep = kmem_cache_create("nfs_page",
1047                                             sizeof(struct nfs_page),
1048                                             0, SLAB_HWCACHE_ALIGN,
1049                                             NULL);
1050         if (nfs_page_cachep == NULL)
1051                 return -ENOMEM;
1052
1053         return 0;
1054 }
1055
1056 void nfs_destroy_nfspagecache(void)
1057 {
1058         kmem_cache_destroy(nfs_page_cachep);
1059 }
1060
1061 static const struct rpc_call_ops nfs_pgio_common_ops = {
1062         .rpc_call_prepare = nfs_pgio_prepare,
1063         .rpc_call_done = nfs_pgio_result,
1064         .rpc_release = nfs_pgio_release,
1065 };
1066
1067 const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1068         .pg_test = nfs_generic_pg_test,
1069         .pg_doio = nfs_generic_pg_pgios,
1070 };