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_clear_bit();
 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 /*
 142  * nfs_page_group_lock - lock the head of the page group
 143  * @req - request in group that is to be locked
 144  *
 145  * this lock must be held if modifying the page group list
 146  */
 147 void
 148 nfs_page_group_lock(struct nfs_page *req)
 149 {
 150         struct nfs_page *head = req->wb_head;
 151         int err = -EAGAIN;
 152
 153         WARN_ON_ONCE(head != head->wb_head);
 154
 155         while (err)
 156                 err = wait_on_bit_lock(&head->wb_flags, PG_HEADLOCK,
 157                         nfs_wait_bit_killable, TASK_KILLABLE);
 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_clear_bit();
 172         clear_bit(PG_HEADLOCK, &head->wb_flags);
 173         smp_mb__after_clear_bit();
 174         wake_up_bit(&head->wb_flags, PG_HEADLOCK);
 175 }
 176
 177 /*
 178  * nfs_page_group_sync_on_bit_locked
 179  *
 180  * must be called with page group lock held
 181  */
 182 static bool
 183 nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
 184 {
 185         struct nfs_page *head = req->wb_head;
 186         struct nfs_page *tmp;
 187
 188         WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
 189         WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
 190
 191         tmp = req->wb_this_page;
 192         while (tmp != req) {
 193                 if (!test_bit(bit, &tmp->wb_flags))
 194                         return false;
 195                 tmp = tmp->wb_this_page;
 196         }
 197
 198         /* true! reset all bits */
 199         tmp = req;
 200         do {
 201                 clear_bit(bit, &tmp->wb_flags);
 202                 tmp = tmp->wb_this_page;
 203         } while (tmp != req);
 204
 205         return true;
 206 }
 207
 208 /*
 209  * nfs_page_group_sync_on_bit - set bit on current request, but only
 210  *   return true if the bit is set for all requests in page group
 211  * @req - request in page group
 212  * @bit - PG_* bit that is used to sync page group
 213  */
 214 bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
 215 {
 216         bool ret;
 217
 218         nfs_page_group_lock(req);
 219         ret = nfs_page_group_sync_on_bit_locked(req, bit);
 220         nfs_page_group_unlock(req);
 221
 222         return ret;
 223 }
 224
 225 /*
 226  * nfs_page_group_init - Initialize the page group linkage for @req
 227  * @req - a new nfs request
 228  * @prev - the previous request in page group, or NULL if @req is the first
 229  *         or only request in the group (the head).
 230  */
 231 static inline void
 232 nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
 233 {
 234         WARN_ON_ONCE(prev == req);
 235
 236         if (!prev) {
 237                 req->wb_head = req;
 238                 req->wb_this_page = req;
 239         } else {
 240                 WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
 241                 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
 242                 req->wb_head = prev->wb_head;
 243                 req->wb_this_page = prev->wb_this_page;
 244                 prev->wb_this_page = req;
 245
 246                 /* grab extra ref if head request has extra ref from
 247                  * the write/commit path to handle handoff between write
 248                  * and commit lists */
 249                 if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags))
 250                         kref_get(&req->wb_kref);
 251         }
 252 }
 253
 254 /*
 255  * nfs_page_group_destroy - sync the destruction of page groups
 256  * @req - request that no longer needs the page group
 257  *
 258  * releases the page group reference from each member once all
 259  * members have called this function.
 260  */
 261 static void
 262 nfs_page_group_destroy(struct kref *kref)
 263 {
 264         struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
 265         struct nfs_page *tmp, *next;
 266
 267         if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
 268                 return;
 269
 270         tmp = req;
 271         do {
 272                 next = tmp->wb_this_page;
 273                 /* unlink and free */
 274                 tmp->wb_this_page = tmp;
 275                 tmp->wb_head = tmp;
 276                 nfs_free_request(tmp);
 277                 tmp = next;
 278         } while (tmp != req);
 279 }
 280
 281 /**
 282  * nfs_create_request - Create an NFS read/write request.
 283  * @ctx: open context to use
 284  * @page: page to write
 285  * @last: last nfs request created for this page group or NULL if head
 286  * @offset: starting offset within the page for the write
 287  * @count: number of bytes to read/write
 288  *
 289  * The page must be locked by the caller. This makes sure we never
 290  * create two different requests for the same page.
 291  * User should ensure it is safe to sleep in this function.
 292  */
 293 struct nfs_page *
 294 nfs_create_request(struct nfs_open_context *ctx, struct page *page,
 295                    struct nfs_page *last, unsigned int offset,
 296                    unsigned int count)
 297 {
 298         struct nfs_page         *req;
 299         struct nfs_lock_context *l_ctx;
 300
 301         if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
 302                 return ERR_PTR(-EBADF);
 303         /* try to allocate the request struct */
 304         req = nfs_page_alloc();
 305         if (req == NULL)
 306                 return ERR_PTR(-ENOMEM);
 307
 308         /* get lock context early so we can deal with alloc failures */
 309         l_ctx = nfs_get_lock_context(ctx);
 310         if (IS_ERR(l_ctx)) {
 311                 nfs_page_free(req);
 312                 return ERR_CAST(l_ctx);
 313         }
 314         req->wb_lock_context = l_ctx;
 315         nfs_iocounter_inc(&l_ctx->io_count);
 316
 317         /* Initialize the request struct. Initially, we assume a
 318          * long write-back delay. This will be adjusted in
 319          * update_nfs_request below if the region is not locked. */
 320         req->wb_page    = page;
 321         req->wb_index   = page_file_index(page);
 322         page_cache_get(page);
 323         req->wb_offset  = offset;
 324         req->wb_pgbase  = offset;
 325         req->wb_bytes   = count;
 326         req->wb_context = get_nfs_open_context(ctx);
 327         kref_init(&req->wb_kref);
 328         nfs_page_group_init(req, last);
 329         return req;
 330 }
 331
 332 /**
 333  * nfs_unlock_request - Unlock request and wake up sleepers.
 334  * @req:
 335  */
 336 void nfs_unlock_request(struct nfs_page *req)
 337 {
 338         if (!NFS_WBACK_BUSY(req)) {
 339                 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
 340                 BUG();
 341         }
 342         smp_mb__before_clear_bit();
 343         clear_bit(PG_BUSY, &req->wb_flags);
 344         smp_mb__after_clear_bit();
 345         wake_up_bit(&req->wb_flags, PG_BUSY);
 346 }
 347
 348 /**
 349  * nfs_unlock_and_release_request - Unlock request and release the nfs_page
 350  * @req:
 351  */
 352 void nfs_unlock_and_release_request(struct nfs_page *req)
 353 {
 354         nfs_unlock_request(req);
 355         nfs_release_request(req);
 356 }
 357
 358 /*
 359  * nfs_clear_request - Free up all resources allocated to the request
 360  * @req:
 361  *
 362  * Release page and open context resources associated with a read/write
 363  * request after it has completed.
 364  */
 365 static void nfs_clear_request(struct nfs_page *req)
 366 {
 367         struct page *page = req->wb_page;
 368         struct nfs_open_context *ctx = req->wb_context;
 369         struct nfs_lock_context *l_ctx = req->wb_lock_context;
 370
 371         if (page != NULL) {
 372                 page_cache_release(page);
 373                 req->wb_page = NULL;
 374         }
 375         if (l_ctx != NULL) {
 376                 nfs_iocounter_dec(&l_ctx->io_count);
 377                 nfs_put_lock_context(l_ctx);
 378                 req->wb_lock_context = NULL;
 379         }
 380         if (ctx != NULL) {
 381                 put_nfs_open_context(ctx);
 382                 req->wb_context = NULL;
 383         }
 384 }
 385
 386 /**
 387  * nfs_release_request - Release the count on an NFS read/write request
 388  * @req: request to release
 389  *
 390  * Note: Should never be called with the spinlock held!
 391  */
 392 static void nfs_free_request(struct nfs_page *req)
 393 {
 394         WARN_ON_ONCE(req->wb_this_page != req);
 395
 396         /* extra debug: make sure no sync bits are still set */
 397         WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
 398         WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
 399         WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
 400         WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
 401         WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
 402
 403         /* Release struct file and open context */
 404         nfs_clear_request(req);
 405         nfs_page_free(req);
 406 }
 407
 408 void nfs_release_request(struct nfs_page *req)
 409 {
 410         kref_put(&req->wb_kref, nfs_page_group_destroy);
 411 }
 412
 413 static int nfs_wait_bit_uninterruptible(void *word)
 414 {
 415         io_schedule();
 416         return 0;
 417 }
 418
 419 /**
 420  * nfs_wait_on_request - Wait for a request to complete.
 421  * @req: request to wait upon.
 422  *
 423  * Interruptible by fatal signals only.
 424  * The user is responsible for holding a count on the request.
 425  */
 426 int
 427 nfs_wait_on_request(struct nfs_page *req)
 428 {
 429         return wait_on_bit(&req->wb_flags, PG_BUSY,
 430                         nfs_wait_bit_uninterruptible,
 431                         TASK_UNINTERRUPTIBLE);
 432 }
 433
 434 /*
 435  * nfs_generic_pg_test - determine if requests can be coalesced
 436  * @desc: pointer to descriptor
 437  * @prev: previous request in desc, or NULL
 438  * @req: this request
 439  *
 440  * Returns zero if @req can be coalesced into @desc, otherwise it returns
 441  * the size of the request.
 442  */
 443 size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
 444                            struct nfs_page *prev, struct nfs_page *req)
 445 {
 446         if (desc->pg_count > desc->pg_bsize) {
 447                 /* should never happen */
 448                 WARN_ON_ONCE(1);
 449                 return 0;
 450         }
 451
 452         return min(desc->pg_bsize - desc->pg_count, (size_t)req->wb_bytes);
 453 }
 454 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
 455
 456 static inline struct nfs_rw_header *NFS_RW_HEADER(struct nfs_pgio_header *hdr)
 457 {
 458         return container_of(hdr, struct nfs_rw_header, header);
 459 }
 460
 461 /**
 462  * nfs_rw_header_alloc - Allocate a header for a read or write
 463  * @ops: Read or write function vector
 464  */
 465 struct nfs_rw_header *nfs_rw_header_alloc(const struct nfs_rw_ops *ops)
 466 {
 467         struct nfs_rw_header *header = ops->rw_alloc_header();
 468
 469         if (header) {
 470                 struct nfs_pgio_header *hdr = &header->header;
 471
 472                 INIT_LIST_HEAD(&hdr->pages);
 473                 INIT_LIST_HEAD(&hdr->rpc_list);
 474                 spin_lock_init(&hdr->lock);
 475                 atomic_set(&hdr->refcnt, 0);
 476                 hdr->rw_ops = ops;
 477         }
 478         return header;
 479 }
 480 EXPORT_SYMBOL_GPL(nfs_rw_header_alloc);
 481
 482 /*
 483  * nfs_rw_header_free - Free a read or write header
 484  * @hdr: The header to free
 485  */
 486 void nfs_rw_header_free(struct nfs_pgio_header *hdr)
 487 {
 488         hdr->rw_ops->rw_free_header(NFS_RW_HEADER(hdr));
 489 }
 490 EXPORT_SYMBOL_GPL(nfs_rw_header_free);
 491
 492 /**
 493  * nfs_pgio_data_alloc - Allocate pageio data
 494  * @hdr: The header making a request
 495  * @pagecount: Number of pages to create
 496  */
 497 static struct nfs_pgio_data *nfs_pgio_data_alloc(struct nfs_pgio_header *hdr,
 498                                                  unsigned int pagecount)
 499 {
 500         struct nfs_pgio_data *data, *prealloc;
 501
 502         prealloc = &NFS_RW_HEADER(hdr)->rpc_data;
 503         if (prealloc->header == NULL)
 504                 data = prealloc;
 505         else
 506                 data = kzalloc(sizeof(*data), GFP_KERNEL);
 507         if (!data)
 508                 goto out;
 509
 510         if (nfs_pgarray_set(&data->pages, pagecount)) {
 511                 data->header = hdr;
 512                 atomic_inc(&hdr->refcnt);
 513         } else {
 514                 if (data != prealloc)
 515                         kfree(data);
 516                 data = NULL;
 517         }
 518 out:
 519         return data;
 520 }
 521
 522 /**
 523  * nfs_pgio_data_release - Properly free pageio data
 524  * @data: The data to release
 525  */
 526 void nfs_pgio_data_release(struct nfs_pgio_data *data)
 527 {
 528         struct nfs_pgio_header *hdr = data->header;
 529         struct nfs_rw_header *pageio_header = NFS_RW_HEADER(hdr);
 530
 531         put_nfs_open_context(data->args.context);
 532         if (data->pages.pagevec != data->pages.page_array)
 533                 kfree(data->pages.pagevec);
 534         if (data == &pageio_header->rpc_data) {
 535                 data->header = NULL;
 536                 data = NULL;
 537         }
 538         if (atomic_dec_and_test(&hdr->refcnt))
 539                 hdr->completion_ops->completion(hdr);
 540         /* Note: we only free the rpc_task after callbacks are done.
 541          * See the comment in rpc_free_task() for why
 542          */
 543         kfree(data);
 544 }
 545 EXPORT_SYMBOL_GPL(nfs_pgio_data_release);
 546
 547 /**
 548  * nfs_pgio_rpcsetup - Set up arguments for a pageio call
 549  * @data: The pageio data
 550  * @count: Number of bytes to read
 551  * @offset: Initial offset
 552  * @how: How to commit data (writes only)
 553  * @cinfo: Commit information for the call (writes only)
 554  */
 555 static void nfs_pgio_rpcsetup(struct nfs_pgio_data *data,
 556                               unsigned int count, unsigned int offset,
 557                               int how, struct nfs_commit_info *cinfo)
 558 {
 559         struct nfs_page *req = data->header->req;
 560
 561         /* Set up the RPC argument and reply structs
 562          * NB: take care not to mess about with data->commit et al. */
 563
 564         data->args.fh     = NFS_FH(data->header->inode);
 565         data->args.offset = req_offset(req) + offset;
 566         /* pnfs_set_layoutcommit needs this */
 567         data->mds_offset = data->args.offset;
 568         data->args.pgbase = req->wb_pgbase + offset;
 569         data->args.pages  = data->pages.pagevec;
 570         data->args.count  = count;
 571         data->args.context = get_nfs_open_context(req->wb_context);
 572         data->args.lock_context = req->wb_lock_context;
 573         data->args.stable  = NFS_UNSTABLE;
 574         switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
 575         case 0:
 576                 break;
 577         case FLUSH_COND_STABLE:
 578                 if (nfs_reqs_to_commit(cinfo))
 579                         break;
 580         default:
 581                 data->args.stable = NFS_FILE_SYNC;
 582         }
 583
 584         data->res.fattr   = &data->fattr;
 585         data->res.count   = count;
 586         data->res.eof     = 0;
 587         data->res.verf    = &data->verf;
 588         nfs_fattr_init(&data->fattr);
 589 }
 590
 591 /**
 592  * nfs_pgio_prepare - Prepare pageio data to go over the wire
 593  * @task: The current task
 594  * @calldata: pageio data to prepare
 595  */
 596 static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
 597 {
 598         struct nfs_pgio_data *data = calldata;
 599         int err;
 600         err = NFS_PROTO(data->header->inode)->pgio_rpc_prepare(task, data);
 601         if (err)
 602                 rpc_exit(task, err);
 603 }
 604
 605 int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_data *data,
 606                       const struct rpc_call_ops *call_ops, int how, int flags)
 607 {
 608         struct rpc_task *task;
 609         struct rpc_message msg = {
 610                 .rpc_argp = &data->args,
 611                 .rpc_resp = &data->res,
 612                 .rpc_cred = data->header->cred,
 613         };
 614         struct rpc_task_setup task_setup_data = {
 615                 .rpc_client = clnt,
 616                 .task = &data->task,
 617                 .rpc_message = &msg,
 618                 .callback_ops = call_ops,
 619                 .callback_data = data,
 620                 .workqueue = nfsiod_workqueue,
 621                 .flags = RPC_TASK_ASYNC | flags,
 622         };
 623         int ret = 0;
 624
 625         data->header->rw_ops->rw_initiate(data, &msg, &task_setup_data, how);
 626
 627         dprintk("NFS: %5u initiated pgio call "
 628                 "(req %s/%llu, %u bytes @ offset %llu)\n",
 629                 data->task.tk_pid,
 630                 data->header->inode->i_sb->s_id,
 631                 (unsigned long long)NFS_FILEID(data->header->inode),
 632                 data->args.count,
 633                 (unsigned long long)data->args.offset);
 634
 635         task = rpc_run_task(&task_setup_data);
 636         if (IS_ERR(task)) {
 637                 ret = PTR_ERR(task);
 638                 goto out;
 639         }
 640         if (how & FLUSH_SYNC) {
 641                 ret = rpc_wait_for_completion_task(task);
 642                 if (ret == 0)
 643                         ret = task->tk_status;
 644         }
 645         rpc_put_task(task);
 646 out:
 647         return ret;
 648 }
 649 EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
 650
 651 static int nfs_do_multiple_pgios(struct list_head *head,
 652                                  const struct rpc_call_ops *call_ops,
 653                                  int how)
 654 {
 655         struct nfs_pgio_data *data;
 656         int ret = 0;
 657
 658         while (!list_empty(head)) {
 659                 int ret2;
 660
 661                 data = list_first_entry(head, struct nfs_pgio_data, list);
 662                 list_del_init(&data->list);
 663
 664                 ret2 = nfs_initiate_pgio(NFS_CLIENT(data->header->inode),
 665                                          data, call_ops, how, 0);
 666                 if (ret == 0)
 667                          ret = ret2;
 668         }
 669         return ret;
 670 }
 671
 672 /**
 673  * nfs_pgio_error - Clean up from a pageio error
 674  * @desc: IO descriptor
 675  * @hdr: pageio header
 676  */
 677 static int nfs_pgio_error(struct nfs_pageio_descriptor *desc,
 678                           struct nfs_pgio_header *hdr)
 679 {
 680         struct nfs_pgio_data *data;
 681
 682         set_bit(NFS_IOHDR_REDO, &hdr->flags);
 683         while (!list_empty(&hdr->rpc_list)) {
 684                 data = list_first_entry(&hdr->rpc_list, struct nfs_pgio_data, list);
 685                 list_del(&data->list);
 686                 nfs_pgio_data_release(data);
 687         }
 688         desc->pg_completion_ops->error_cleanup(&desc->pg_list);
 689         return -ENOMEM;
 690 }
 691
 692 /**
 693  * nfs_pgio_release - Release pageio data
 694  * @calldata: The pageio data to release
 695  */
 696 static void nfs_pgio_release(void *calldata)
 697 {
 698         struct nfs_pgio_data *data = calldata;
 699         if (data->header->rw_ops->rw_release)
 700                 data->header->rw_ops->rw_release(data);
 701         nfs_pgio_data_release(data);
 702 }
 703
 704 /**
 705  * nfs_pageio_init - initialise a page io descriptor
 706  * @desc: pointer to descriptor
 707  * @inode: pointer to inode
 708  * @doio: pointer to io function
 709  * @bsize: io block size
 710  * @io_flags: extra parameters for the io function
 711  */
 712 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
 713                      struct inode *inode,
 714                      const struct nfs_pageio_ops *pg_ops,
 715                      const struct nfs_pgio_completion_ops *compl_ops,
 716                      const struct nfs_rw_ops *rw_ops,
 717                      size_t bsize,
 718                      int io_flags)
 719 {
 720         INIT_LIST_HEAD(&desc->pg_list);
 721         desc->pg_bytes_written = 0;
 722         desc->pg_count = 0;
 723         desc->pg_bsize = bsize;
 724         desc->pg_base = 0;
 725         desc->pg_moreio = 0;
 726         desc->pg_recoalesce = 0;
 727         desc->pg_inode = inode;
 728         desc->pg_ops = pg_ops;
 729         desc->pg_completion_ops = compl_ops;
 730         desc->pg_rw_ops = rw_ops;
 731         desc->pg_ioflags = io_flags;
 732         desc->pg_error = 0;
 733         desc->pg_lseg = NULL;
 734         desc->pg_dreq = NULL;
 735         desc->pg_layout_private = NULL;
 736 }
 737 EXPORT_SYMBOL_GPL(nfs_pageio_init);
 738
 739 /**
 740  * nfs_pgio_result - Basic pageio error handling
 741  * @task: The task that ran
 742  * @calldata: Pageio data to check
 743  */
 744 static void nfs_pgio_result(struct rpc_task *task, void *calldata)
 745 {
 746         struct nfs_pgio_data *data = calldata;
 747         struct inode *inode = data->header->inode;
 748
 749         dprintk("NFS: %s: %5u, (status %d)\n", __func__,
 750                 task->tk_pid, task->tk_status);
 751
 752         if (data->header->rw_ops->rw_done(task, data, inode) != 0)
 753                 return;
 754         if (task->tk_status < 0)
 755                 nfs_set_pgio_error(data->header, task->tk_status, data->args.offset);
 756         else
 757                 data->header->rw_ops->rw_result(task, data);
 758 }
 759
 760 /*
 761  * Create an RPC task for the given read or write request and kick it.
 762  * The page must have been locked by the caller.
 763  *
 764  * It may happen that the page we're passed is not marked dirty.
 765  * This is the case if nfs_updatepage detects a conflicting request
 766  * that has been written but not committed.
 767  */
 768 int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
 769                      struct nfs_pgio_header *hdr)
 770 {
 771         struct nfs_page         *req;
 772         struct page             **pages;
 773         struct nfs_pgio_data    *data;
 774         struct list_head *head = &desc->pg_list;
 775         struct nfs_commit_info cinfo;
 776
 777         data = nfs_pgio_data_alloc(hdr, nfs_page_array_len(desc->pg_base,
 778                                                            desc->pg_count));
 779         if (!data)
 780                 return nfs_pgio_error(desc, hdr);
 781
 782         nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
 783         pages = data->pages.pagevec;
 784         while (!list_empty(head)) {
 785                 req = nfs_list_entry(head->next);
 786                 nfs_list_remove_request(req);
 787                 nfs_list_add_request(req, &hdr->pages);
 788                 *pages++ = req->wb_page;
 789         }
 790
 791         if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
 792             (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
 793                 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
 794
 795         /* Set up the argument struct */
 796         nfs_pgio_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags, &cinfo);
 797         list_add(&data->list, &hdr->rpc_list);
 798         desc->pg_rpc_callops = &nfs_pgio_common_ops;
 799         return 0;
 800 }
 801 EXPORT_SYMBOL_GPL(nfs_generic_pgio);
 802
 803 static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
 804 {
 805         struct nfs_rw_header *rw_hdr;
 806         struct nfs_pgio_header *hdr;
 807         int ret;
 808
 809         rw_hdr = nfs_rw_header_alloc(desc->pg_rw_ops);
 810         if (!rw_hdr) {
 811                 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
 812                 return -ENOMEM;
 813         }
 814         hdr = &rw_hdr->header;
 815         nfs_pgheader_init(desc, hdr, nfs_rw_header_free);
 816         atomic_inc(&hdr->refcnt);
 817         ret = nfs_generic_pgio(desc, hdr);
 818         if (ret == 0)
 819                 ret = nfs_do_multiple_pgios(&hdr->rpc_list,
 820                                             desc->pg_rpc_callops,
 821                                             desc->pg_ioflags);
 822         if (atomic_dec_and_test(&hdr->refcnt))
 823                 hdr->completion_ops->completion(hdr);
 824         return ret;
 825 }
 826
 827 static bool nfs_match_open_context(const struct nfs_open_context *ctx1,
 828                 const struct nfs_open_context *ctx2)
 829 {
 830         return ctx1->cred == ctx2->cred && ctx1->state == ctx2->state;
 831 }
 832
 833 static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
 834                 const struct nfs_lock_context *l2)
 835 {
 836         return l1->lockowner.l_owner == l2->lockowner.l_owner
 837                 && l1->lockowner.l_pid == l2->lockowner.l_pid;
 838 }
 839
 840 /**
 841  * nfs_can_coalesce_requests - test two requests for compatibility
 842  * @prev: pointer to nfs_page
 843  * @req: pointer to nfs_page
 844  *
 845  * The nfs_page structures 'prev' and 'req' are compared to ensure that the
 846  * page data area they describe is contiguous, and that their RPC
 847  * credentials, NFSv4 open state, and lockowners are the same.
 848  *
 849  * Return 'true' if this is the case, else return 'false'.
 850  */
 851 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
 852                                       struct nfs_page *req,
 853                                       struct nfs_pageio_descriptor *pgio)
 854 {
 855         size_t size;
 856
 857         if (prev) {
 858                 if (!nfs_match_open_context(req->wb_context, prev->wb_context))
 859                         return false;
 860                 if (req->wb_context->dentry->d_inode->i_flock != NULL &&
 861                     !nfs_match_lock_context(req->wb_lock_context,
 862                                             prev->wb_lock_context))
 863                         return false;
 864                 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
 865                         return false;
 866         }
 867         size = pgio->pg_ops->pg_test(pgio, prev, req);
 868         WARN_ON_ONCE(size > req->wb_bytes);
 869         if (size && size < req->wb_bytes)
 870                 req->wb_bytes = size;
 871         return size > 0;
 872 }
 873
 874 /**
 875  * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
 876  * @desc: destination io descriptor
 877  * @req: request
 878  *
 879  * Returns true if the request 'req' was successfully coalesced into the
 880  * existing list of pages 'desc'.
 881  */
 882 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
 883                                      struct nfs_page *req)
 884 {
 885         struct nfs_page *prev = NULL;
 886         if (desc->pg_count != 0) {
 887                 prev = nfs_list_entry(desc->pg_list.prev);
 888         } else {
 889                 if (desc->pg_ops->pg_init)
 890                         desc->pg_ops->pg_init(desc, req);
 891                 desc->pg_base = req->wb_pgbase;
 892         }
 893         if (!nfs_can_coalesce_requests(prev, req, desc))
 894                 return 0;
 895         nfs_list_remove_request(req);
 896         nfs_list_add_request(req, &desc->pg_list);
 897         desc->pg_count += req->wb_bytes;
 898         return 1;
 899 }
 900
 901 /*
 902  * Helper for nfs_pageio_add_request and nfs_pageio_complete
 903  */
 904 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
 905 {
 906         if (!list_empty(&desc->pg_list)) {
 907                 int error = desc->pg_ops->pg_doio(desc);
 908                 if (error < 0)
 909                         desc->pg_error = error;
 910                 else
 911                         desc->pg_bytes_written += desc->pg_count;
 912         }
 913         if (list_empty(&desc->pg_list)) {
 914                 desc->pg_count = 0;
 915                 desc->pg_base = 0;
 916         }
 917 }
 918
 919 /**
 920  * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
 921  * @desc: destination io descriptor
 922  * @req: request
 923  *
 924  * This may split a request into subrequests which are all part of the
 925  * same page group.
 926  *
 927  * Returns true if the request 'req' was successfully coalesced into the
 928  * existing list of pages 'desc'.
 929  */
 930 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
 931                            struct nfs_page *req)
 932 {
 933         struct nfs_page *subreq;
 934         unsigned int bytes_left = 0;
 935         unsigned int offset, pgbase;
 936
 937         nfs_page_group_lock(req);
 938
 939         subreq = req;
 940         bytes_left = subreq->wb_bytes;
 941         offset = subreq->wb_offset;
 942         pgbase = subreq->wb_pgbase;
 943
 944         do {
 945                 if (!nfs_pageio_do_add_request(desc, subreq)) {
 946                         /* make sure pg_test call(s) did nothing */
 947                         WARN_ON_ONCE(subreq->wb_bytes != bytes_left);
 948                         WARN_ON_ONCE(subreq->wb_offset != offset);
 949                         WARN_ON_ONCE(subreq->wb_pgbase != pgbase);
 950
 951                         nfs_page_group_unlock(req);
 952                         desc->pg_moreio = 1;
 953                         nfs_pageio_doio(desc);
 954                         if (desc->pg_error < 0)
 955                                 return 0;
 956                         desc->pg_moreio = 0;
 957                         if (desc->pg_recoalesce)
 958                                 return 0;
 959                         /* retry add_request for this subreq */
 960                         nfs_page_group_lock(req);
 961                         continue;
 962                 }
 963
 964                 /* check for buggy pg_test call(s) */
 965                 WARN_ON_ONCE(subreq->wb_bytes + subreq->wb_pgbase > PAGE_SIZE);
 966                 WARN_ON_ONCE(subreq->wb_bytes > bytes_left);
 967                 WARN_ON_ONCE(subreq->wb_bytes == 0);
 968
 969                 bytes_left -= subreq->wb_bytes;
 970                 offset += subreq->wb_bytes;
 971                 pgbase += subreq->wb_bytes;
 972
 973                 if (bytes_left) {
 974                         subreq = nfs_create_request(req->wb_context,
 975                                         req->wb_page,
 976                                         subreq, pgbase, bytes_left);
 977                         nfs_lock_request(subreq);
 978                         subreq->wb_offset  = offset;
 979                         subreq->wb_index = req->wb_index;
 980                 }
 981         } while (bytes_left > 0);
 982
 983         nfs_page_group_unlock(req);
 984         return 1;
 985 }
 986
 987 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
 988 {
 989         LIST_HEAD(head);
 990
 991         do {
 992                 list_splice_init(&desc->pg_list, &head);
 993                 desc->pg_bytes_written -= desc->pg_count;
 994                 desc->pg_count = 0;
 995                 desc->pg_base = 0;
 996                 desc->pg_recoalesce = 0;
 997
 998                 while (!list_empty(&head)) {
 999                         struct nfs_page *req;
1000
1001                         req = list_first_entry(&head, struct nfs_page, wb_list);
1002                         nfs_list_remove_request(req);
1003                         if (__nfs_pageio_add_request(desc, req))
1004                                 continue;
1005                         if (desc->pg_error < 0)
1006                                 return 0;
1007                         break;
1008                 }
1009         } while (desc->pg_recoalesce);
1010         return 1;
1011 }
1012
1013 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1014                 struct nfs_page *req)
1015 {
1016         int ret;
1017
1018         do {
1019                 ret = __nfs_pageio_add_request(desc, req);
1020                 if (ret)
1021                         break;
1022                 if (desc->pg_error < 0)
1023                         break;
1024                 ret = nfs_do_recoalesce(desc);
1025         } while (ret);
1026         return ret;
1027 }
1028 EXPORT_SYMBOL_GPL(nfs_pageio_add_request);
1029
1030 /**
1031  * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
1032  * @desc: pointer to io descriptor
1033  */
1034 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1035 {
1036         for (;;) {
1037                 nfs_pageio_doio(desc);
1038                 if (!desc->pg_recoalesce)
1039                         break;
1040                 if (!nfs_do_recoalesce(desc))
1041                         break;
1042         }
1043 }
1044 EXPORT_SYMBOL_GPL(nfs_pageio_complete);
1045
1046 /**
1047  * nfs_pageio_cond_complete - Conditional I/O completion
1048  * @desc: pointer to io descriptor
1049  * @index: page index
1050  *
1051  * It is important to ensure that processes don't try to take locks
1052  * on non-contiguous ranges of pages as that might deadlock. This
1053  * function should be called before attempting to wait on a locked
1054  * nfs_page. It will complete the I/O if the page index 'index'
1055  * is not contiguous with the existing list of pages in 'desc'.
1056  */
1057 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1058 {
1059         if (!list_empty(&desc->pg_list)) {
1060                 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
1061                 if (index != prev->wb_index + 1)
1062                         nfs_pageio_complete(desc);
1063         }
1064 }
1065
1066 int __init nfs_init_nfspagecache(void)
1067 {
1068         nfs_page_cachep = kmem_cache_create("nfs_page",
1069                                             sizeof(struct nfs_page),
1070                                             0, SLAB_HWCACHE_ALIGN,
1071                                             NULL);
1072         if (nfs_page_cachep == NULL)
1073                 return -ENOMEM;
1074
1075         return 0;
1076 }
1077
1078 void nfs_destroy_nfspagecache(void)
1079 {
1080         kmem_cache_destroy(nfs_page_cachep);
1081 }
1082
1083 static const struct rpc_call_ops nfs_pgio_common_ops = {
1084         .rpc_call_prepare = nfs_pgio_prepare,
1085         .rpc_call_done = nfs_pgio_result,
1086         .rpc_release = nfs_pgio_release,
1087 };
1088
1089 const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1090         .pg_test = nfs_generic_pg_test,
1091         .pg_doio = nfs_generic_pg_pgios,
1092 };