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 static struct kmem_cache *nfs_page_cachep;
  28
  29 bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount)
  30 {
  31         p->npages = pagecount;
  32         if (pagecount <= ARRAY_SIZE(p->page_array))
  33                 p->pagevec = p->page_array;
  34         else {
  35                 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
  36                 if (!p->pagevec)
  37                         p->npages = 0;
  38         }
  39         return p->pagevec != NULL;
  40 }
  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         hdr->req = nfs_list_entry(desc->pg_list.next);
  47         hdr->inode = desc->pg_inode;
  48         hdr->cred = hdr->req->wb_context->cred;
  49         hdr->io_start = req_offset(hdr->req);
  50         hdr->good_bytes = desc->pg_count;
  51         hdr->dreq = desc->pg_dreq;
  52         hdr->layout_private = desc->pg_layout_private;
  53         hdr->release = release;
  54         hdr->completion_ops = desc->pg_completion_ops;
  55         if (hdr->completion_ops->init_hdr)
  56                 hdr->completion_ops->init_hdr(hdr);
  57 }
  58 EXPORT_SYMBOL_GPL(nfs_pgheader_init);
  59
  60 void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
  61 {
  62         spin_lock(&hdr->lock);
  63         if (pos < hdr->io_start + hdr->good_bytes) {
  64                 set_bit(NFS_IOHDR_ERROR, &hdr->flags);
  65                 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
  66                 hdr->good_bytes = pos - hdr->io_start;
  67                 hdr->error = error;
  68         }
  69         spin_unlock(&hdr->lock);
  70 }
  71
  72 static inline struct nfs_page *
  73 nfs_page_alloc(void)
  74 {
  75         struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_NOIO);
  76         if (p)
  77                 INIT_LIST_HEAD(&p->wb_list);
  78         return p;
  79 }
  80
  81 static inline void
  82 nfs_page_free(struct nfs_page *p)
  83 {
  84         kmem_cache_free(nfs_page_cachep, p);
  85 }
  86
  87 /**
  88  * nfs_create_request - Create an NFS read/write request.
  89  * @ctx: open context to use
  90  * @inode: inode to which the request is attached
  91  * @page: page to write
  92  * @offset: starting offset within the page for the write
  93  * @count: number of bytes to read/write
  94  *
  95  * The page must be locked by the caller. This makes sure we never
  96  * create two different requests for the same page.
  97  * User should ensure it is safe to sleep in this function.
  98  */
  99 struct nfs_page *
 100 nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
 101                    struct page *page,
 102                    unsigned int offset, unsigned int count)
 103 {
 104         struct nfs_page         *req;
 105         struct nfs_lock_context *l_ctx;
 106
 107         /* try to allocate the request struct */
 108         req = nfs_page_alloc();
 109         if (req == NULL)
 110                 return ERR_PTR(-ENOMEM);
 111
 112         /* get lock context early so we can deal with alloc failures */
 113         l_ctx = nfs_get_lock_context(ctx);
 114         if (IS_ERR(l_ctx)) {
 115                 nfs_page_free(req);
 116                 return ERR_CAST(l_ctx);
 117         }
 118         req->wb_lock_context = l_ctx;
 119
 120         /* Initialize the request struct. Initially, we assume a
 121          * long write-back delay. This will be adjusted in
 122          * update_nfs_request below if the region is not locked. */
 123         req->wb_page    = page;
 124         req->wb_index   = page_file_index(page);
 125         page_cache_get(page);
 126         req->wb_offset  = offset;
 127         req->wb_pgbase  = offset;
 128         req->wb_bytes   = count;
 129         req->wb_context = get_nfs_open_context(ctx);
 130         kref_init(&req->wb_kref);
 131         return req;
 132 }
 133
 134 /**
 135  * nfs_unlock_request - Unlock request and wake up sleepers.
 136  * @req:
 137  */
 138 void nfs_unlock_request(struct nfs_page *req)
 139 {
 140         if (!NFS_WBACK_BUSY(req)) {
 141                 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
 142                 BUG();
 143         }
 144         smp_mb__before_clear_bit();
 145         clear_bit(PG_BUSY, &req->wb_flags);
 146         smp_mb__after_clear_bit();
 147         wake_up_bit(&req->wb_flags, PG_BUSY);
 148 }
 149
 150 /**
 151  * nfs_unlock_and_release_request - Unlock request and release the nfs_page
 152  * @req:
 153  */
 154 void nfs_unlock_and_release_request(struct nfs_page *req)
 155 {
 156         nfs_unlock_request(req);
 157         nfs_release_request(req);
 158 }
 159
 160 /*
 161  * nfs_clear_request - Free up all resources allocated to the request
 162  * @req:
 163  *
 164  * Release page and open context resources associated with a read/write
 165  * request after it has completed.
 166  */
 167 static void nfs_clear_request(struct nfs_page *req)
 168 {
 169         struct page *page = req->wb_page;
 170         struct nfs_open_context *ctx = req->wb_context;
 171         struct nfs_lock_context *l_ctx = req->wb_lock_context;
 172
 173         if (page != NULL) {
 174                 page_cache_release(page);
 175                 req->wb_page = NULL;
 176         }
 177         if (l_ctx != NULL) {
 178                 nfs_put_lock_context(l_ctx);
 179                 req->wb_lock_context = NULL;
 180         }
 181         if (ctx != NULL) {
 182                 put_nfs_open_context(ctx);
 183                 req->wb_context = NULL;
 184         }
 185 }
 186
 187
 188 /**
 189  * nfs_release_request - Release the count on an NFS read/write request
 190  * @req: request to release
 191  *
 192  * Note: Should never be called with the spinlock held!
 193  */
 194 static void nfs_free_request(struct kref *kref)
 195 {
 196         struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
 197
 198         /* Release struct file and open context */
 199         nfs_clear_request(req);
 200         nfs_page_free(req);
 201 }
 202
 203 void nfs_release_request(struct nfs_page *req)
 204 {
 205         kref_put(&req->wb_kref, nfs_free_request);
 206 }
 207
 208 static int nfs_wait_bit_uninterruptible(void *word)
 209 {
 210         io_schedule();
 211         return 0;
 212 }
 213
 214 /**
 215  * nfs_wait_on_request - Wait for a request to complete.
 216  * @req: request to wait upon.
 217  *
 218  * Interruptible by fatal signals only.
 219  * The user is responsible for holding a count on the request.
 220  */
 221 int
 222 nfs_wait_on_request(struct nfs_page *req)
 223 {
 224         return wait_on_bit(&req->wb_flags, PG_BUSY,
 225                         nfs_wait_bit_uninterruptible,
 226                         TASK_UNINTERRUPTIBLE);
 227 }
 228
 229 bool nfs_generic_pg_test(struct nfs_pageio_descriptor *desc, struct nfs_page *prev, struct nfs_page *req)
 230 {
 231         /*
 232          * FIXME: ideally we should be able to coalesce all requests
 233          * that are not block boundary aligned, but currently this
 234          * is problematic for the case of bsize < PAGE_CACHE_SIZE,
 235          * since nfs_flush_multi and nfs_pagein_multi assume you
 236          * can have only one struct nfs_page.
 237          */
 238         if (desc->pg_bsize < PAGE_SIZE)
 239                 return 0;
 240
 241         return desc->pg_count + req->wb_bytes <= desc->pg_bsize;
 242 }
 243 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
 244
 245 /**
 246  * nfs_pageio_init - initialise a page io descriptor
 247  * @desc: pointer to descriptor
 248  * @inode: pointer to inode
 249  * @doio: pointer to io function
 250  * @bsize: io block size
 251  * @io_flags: extra parameters for the io function
 252  */
 253 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
 254                      struct inode *inode,
 255                      const struct nfs_pageio_ops *pg_ops,
 256                      const struct nfs_pgio_completion_ops *compl_ops,
 257                      size_t bsize,
 258                      int io_flags)
 259 {
 260         INIT_LIST_HEAD(&desc->pg_list);
 261         desc->pg_bytes_written = 0;
 262         desc->pg_count = 0;
 263         desc->pg_bsize = bsize;
 264         desc->pg_base = 0;
 265         desc->pg_moreio = 0;
 266         desc->pg_recoalesce = 0;
 267         desc->pg_inode = inode;
 268         desc->pg_ops = pg_ops;
 269         desc->pg_completion_ops = compl_ops;
 270         desc->pg_ioflags = io_flags;
 271         desc->pg_error = 0;
 272         desc->pg_lseg = NULL;
 273         desc->pg_dreq = NULL;
 274         desc->pg_layout_private = NULL;
 275 }
 276 EXPORT_SYMBOL_GPL(nfs_pageio_init);
 277
 278 /**
 279  * nfs_can_coalesce_requests - test two requests for compatibility
 280  * @prev: pointer to nfs_page
 281  * @req: pointer to nfs_page
 282  *
 283  * The nfs_page structures 'prev' and 'req' are compared to ensure that the
 284  * page data area they describe is contiguous, and that their RPC
 285  * credentials, NFSv4 open state, and lockowners are the same.
 286  *
 287  * Return 'true' if this is the case, else return 'false'.
 288  */
 289 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
 290                                       struct nfs_page *req,
 291                                       struct nfs_pageio_descriptor *pgio)
 292 {
 293         if (req->wb_context->cred != prev->wb_context->cred)
 294                 return false;
 295         if (req->wb_lock_context->lockowner.l_owner != prev->wb_lock_context->lockowner.l_owner)
 296                 return false;
 297         if (req->wb_lock_context->lockowner.l_pid != prev->wb_lock_context->lockowner.l_pid)
 298                 return false;
 299         if (req->wb_context->state != prev->wb_context->state)
 300                 return false;
 301         if (req->wb_pgbase != 0)
 302                 return false;
 303         if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
 304                 return false;
 305         if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
 306                 return false;
 307         return pgio->pg_ops->pg_test(pgio, prev, req);
 308 }
 309
 310 /**
 311  * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
 312  * @desc: destination io descriptor
 313  * @req: request
 314  *
 315  * Returns true if the request 'req' was successfully coalesced into the
 316  * existing list of pages 'desc'.
 317  */
 318 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
 319                                      struct nfs_page *req)
 320 {
 321         if (desc->pg_count != 0) {
 322                 struct nfs_page *prev;
 323
 324                 prev = nfs_list_entry(desc->pg_list.prev);
 325                 if (!nfs_can_coalesce_requests(prev, req, desc))
 326                         return 0;
 327         } else {
 328                 if (desc->pg_ops->pg_init)
 329                         desc->pg_ops->pg_init(desc, req);
 330                 desc->pg_base = req->wb_pgbase;
 331         }
 332         nfs_list_remove_request(req);
 333         nfs_list_add_request(req, &desc->pg_list);
 334         desc->pg_count += req->wb_bytes;
 335         return 1;
 336 }
 337
 338 /*
 339  * Helper for nfs_pageio_add_request and nfs_pageio_complete
 340  */
 341 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
 342 {
 343         if (!list_empty(&desc->pg_list)) {
 344                 int error = desc->pg_ops->pg_doio(desc);
 345                 if (error < 0)
 346                         desc->pg_error = error;
 347                 else
 348                         desc->pg_bytes_written += desc->pg_count;
 349         }
 350         if (list_empty(&desc->pg_list)) {
 351                 desc->pg_count = 0;
 352                 desc->pg_base = 0;
 353         }
 354 }
 355
 356 /**
 357  * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
 358  * @desc: destination io descriptor
 359  * @req: request
 360  *
 361  * Returns true if the request 'req' was successfully coalesced into the
 362  * existing list of pages 'desc'.
 363  */
 364 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
 365                            struct nfs_page *req)
 366 {
 367         while (!nfs_pageio_do_add_request(desc, req)) {
 368                 desc->pg_moreio = 1;
 369                 nfs_pageio_doio(desc);
 370                 if (desc->pg_error < 0)
 371                         return 0;
 372                 desc->pg_moreio = 0;
 373                 if (desc->pg_recoalesce)
 374                         return 0;
 375         }
 376         return 1;
 377 }
 378
 379 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
 380 {
 381         LIST_HEAD(head);
 382
 383         do {
 384                 list_splice_init(&desc->pg_list, &head);
 385                 desc->pg_bytes_written -= desc->pg_count;
 386                 desc->pg_count = 0;
 387                 desc->pg_base = 0;
 388                 desc->pg_recoalesce = 0;
 389
 390                 while (!list_empty(&head)) {
 391                         struct nfs_page *req;
 392
 393                         req = list_first_entry(&head, struct nfs_page, wb_list);
 394                         nfs_list_remove_request(req);
 395                         if (__nfs_pageio_add_request(desc, req))
 396                                 continue;
 397                         if (desc->pg_error < 0)
 398                                 return 0;
 399                         break;
 400                 }
 401         } while (desc->pg_recoalesce);
 402         return 1;
 403 }
 404
 405 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
 406                 struct nfs_page *req)
 407 {
 408         int ret;
 409
 410         do {
 411                 ret = __nfs_pageio_add_request(desc, req);
 412                 if (ret)
 413                         break;
 414                 if (desc->pg_error < 0)
 415                         break;
 416                 ret = nfs_do_recoalesce(desc);
 417         } while (ret);
 418         return ret;
 419 }
 420 EXPORT_SYMBOL_GPL(nfs_pageio_add_request);
 421
 422 /**
 423  * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
 424  * @desc: pointer to io descriptor
 425  */
 426 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
 427 {
 428         for (;;) {
 429                 nfs_pageio_doio(desc);
 430                 if (!desc->pg_recoalesce)
 431                         break;
 432                 if (!nfs_do_recoalesce(desc))
 433                         break;
 434         }
 435 }
 436 EXPORT_SYMBOL_GPL(nfs_pageio_complete);
 437
 438 /**
 439  * nfs_pageio_cond_complete - Conditional I/O completion
 440  * @desc: pointer to io descriptor
 441  * @index: page index
 442  *
 443  * It is important to ensure that processes don't try to take locks
 444  * on non-contiguous ranges of pages as that might deadlock. This
 445  * function should be called before attempting to wait on a locked
 446  * nfs_page. It will complete the I/O if the page index 'index'
 447  * is not contiguous with the existing list of pages in 'desc'.
 448  */
 449 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
 450 {
 451         if (!list_empty(&desc->pg_list)) {
 452                 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
 453                 if (index != prev->wb_index + 1)
 454                         nfs_pageio_complete(desc);
 455         }
 456 }
 457
 458 int __init nfs_init_nfspagecache(void)
 459 {
 460         nfs_page_cachep = kmem_cache_create("nfs_page",
 461                                             sizeof(struct nfs_page),
 462                                             0, SLAB_HWCACHE_ALIGN,
 463                                             NULL);
 464         if (nfs_page_cachep == NULL)
 465                 return -ENOMEM;
 466
 467         return 0;
 468 }
 469
 470 void nfs_destroy_nfspagecache(void)
 471 {
 472         kmem_cache_destroy(nfs_page_cachep);
 473 }
 474