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/sunrpc/clnt.h>
  15 #include <linux/nfs3.h>
  16 #include <linux/nfs4.h>
  17 #include <linux/nfs_page.h>
  18 #include <linux/nfs_fs.h>
  19 #include <linux/nfs_mount.h>
  20
  21 #include "internal.h"
  22
  23 #define NFS_PARANOIA 1
  24
  25 static struct kmem_cache *nfs_page_cachep;
  26
  27 static inline struct nfs_page *
  28 nfs_page_alloc(void)
  29 {
  30         struct nfs_page *p;
  31         p = kmem_cache_alloc(nfs_page_cachep, GFP_KERNEL);
  32         if (p) {
  33                 memset(p, 0, sizeof(*p));
  34                 INIT_LIST_HEAD(&p->wb_list);
  35         }
  36         return p;
  37 }
  38
  39 static inline void
  40 nfs_page_free(struct nfs_page *p)
  41 {
  42         kmem_cache_free(nfs_page_cachep, p);
  43 }
  44
  45 /**
  46  * nfs_create_request - Create an NFS read/write request.
  47  * @file: file descriptor to use
  48  * @inode: inode to which the request is attached
  49  * @page: page to write
  50  * @offset: starting offset within the page for the write
  51  * @count: number of bytes to read/write
  52  *
  53  * The page must be locked by the caller. This makes sure we never
  54  * create two different requests for the same page.
  55  * User should ensure it is safe to sleep in this function.
  56  */
  57 struct nfs_page *
  58 nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
  59                    struct page *page,
  60                    unsigned int offset, unsigned int count)
  61 {
  62         struct nfs_server *server = NFS_SERVER(inode);
  63         struct nfs_page         *req;
  64
  65         for (;;) {
  66                 /* try to allocate the request struct */
  67                 req = nfs_page_alloc();
  68                 if (req != NULL)
  69                         break;
  70
  71                 if (signalled() && (server->flags & NFS_MOUNT_INTR))
  72                         return ERR_PTR(-ERESTARTSYS);
  73                 yield();
  74         }
  75
  76         /* Initialize the request struct. Initially, we assume a
  77          * long write-back delay. This will be adjusted in
  78          * update_nfs_request below if the region is not locked. */
  79         req->wb_page    = page;
  80         atomic_set(&req->wb_complete, 0);
  81         req->wb_index   = page->index;
  82         page_cache_get(page);
  83         BUG_ON(PagePrivate(page));
  84         BUG_ON(!PageLocked(page));
  85         BUG_ON(page->mapping->host != inode);
  86         req->wb_offset  = offset;
  87         req->wb_pgbase  = offset;
  88         req->wb_bytes   = count;
  89         atomic_set(&req->wb_count, 1);
  90         req->wb_context = get_nfs_open_context(ctx);
  91
  92         return req;
  93 }
  94
  95 /**
  96  * nfs_unlock_request - Unlock request and wake up sleepers.
  97  * @req:
  98  */
  99 void nfs_unlock_request(struct nfs_page *req)
 100 {
 101         if (!NFS_WBACK_BUSY(req)) {
 102                 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
 103                 BUG();
 104         }
 105         smp_mb__before_clear_bit();
 106         clear_bit(PG_BUSY, &req->wb_flags);
 107         smp_mb__after_clear_bit();
 108         wake_up_bit(&req->wb_flags, PG_BUSY);
 109         nfs_release_request(req);
 110 }
 111
 112 /**
 113  * nfs_set_page_writeback_locked - Lock a request for writeback
 114  * @req:
 115  */
 116 int nfs_set_page_writeback_locked(struct nfs_page *req)
 117 {
 118         struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);
 119
 120         if (!nfs_lock_request(req))
 121                 return 0;
 122         radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
 123         return 1;
 124 }
 125
 126 /**
 127  * nfs_clear_page_writeback - Unlock request and wake up sleepers
 128  */
 129 void nfs_clear_page_writeback(struct nfs_page *req)
 130 {
 131         struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);
 132
 133         if (req->wb_page != NULL) {
 134                 spin_lock(&nfsi->req_lock);
 135                 radix_tree_tag_clear(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
 136                 spin_unlock(&nfsi->req_lock);
 137         }
 138         nfs_unlock_request(req);
 139 }
 140
 141 /**
 142  * nfs_clear_request - Free up all resources allocated to the request
 143  * @req:
 144  *
 145  * Release page resources associated with a write request after it
 146  * has completed.
 147  */
 148 void nfs_clear_request(struct nfs_page *req)
 149 {
 150         struct page *page = req->wb_page;
 151         if (page != NULL) {
 152                 page_cache_release(page);
 153                 req->wb_page = NULL;
 154         }
 155 }
 156
 157
 158 /**
 159  * nfs_release_request - Release the count on an NFS read/write request
 160  * @req: request to release
 161  *
 162  * Note: Should never be called with the spinlock held!
 163  */
 164 void
 165 nfs_release_request(struct nfs_page *req)
 166 {
 167         if (!atomic_dec_and_test(&req->wb_count))
 168                 return;
 169
 170 #ifdef NFS_PARANOIA
 171         BUG_ON (!list_empty(&req->wb_list));
 172         BUG_ON (NFS_WBACK_BUSY(req));
 173 #endif
 174
 175         /* Release struct file or cached credential */
 176         nfs_clear_request(req);
 177         put_nfs_open_context(req->wb_context);
 178         nfs_page_free(req);
 179 }
 180
 181 static int nfs_wait_bit_interruptible(void *word)
 182 {
 183         int ret = 0;
 184
 185         if (signal_pending(current))
 186                 ret = -ERESTARTSYS;
 187         else
 188                 schedule();
 189         return ret;
 190 }
 191
 192 /**
 193  * nfs_wait_on_request - Wait for a request to complete.
 194  * @req: request to wait upon.
 195  *
 196  * Interruptible by signals only if mounted with intr flag.
 197  * The user is responsible for holding a count on the request.
 198  */
 199 int
 200 nfs_wait_on_request(struct nfs_page *req)
 201 {
 202         struct rpc_clnt *clnt = NFS_CLIENT(req->wb_context->dentry->d_inode);
 203         sigset_t oldmask;
 204         int ret = 0;
 205
 206         if (!test_bit(PG_BUSY, &req->wb_flags))
 207                 goto out;
 208         /*
 209          * Note: the call to rpc_clnt_sigmask() suffices to ensure that we
 210          *       are not interrupted if intr flag is not set
 211          */
 212         rpc_clnt_sigmask(clnt, &oldmask);
 213         ret = out_of_line_wait_on_bit(&req->wb_flags, PG_BUSY,
 214                         nfs_wait_bit_interruptible, TASK_INTERRUPTIBLE);
 215         rpc_clnt_sigunmask(clnt, &oldmask);
 216 out:
 217         return ret;
 218 }
 219
 220 /**
 221  * nfs_pageio_init - initialise a page io descriptor
 222  * @desc: pointer to descriptor
 223  * @inode: pointer to inode
 224  * @doio: pointer to io function
 225  * @bsize: io block size
 226  * @io_flags: extra parameters for the io function
 227  */
 228 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
 229                      struct inode *inode,
 230                      int (*doio)(struct inode *, struct list_head *, unsigned int, size_t, int),
 231                      size_t bsize,
 232                      int io_flags)
 233 {
 234         INIT_LIST_HEAD(&desc->pg_list);
 235         desc->pg_bytes_written = 0;
 236         desc->pg_count = 0;
 237         desc->pg_bsize = bsize;
 238         desc->pg_base = 0;
 239         desc->pg_inode = inode;
 240         desc->pg_doio = doio;
 241         desc->pg_ioflags = io_flags;
 242         desc->pg_error = 0;
 243 }
 244
 245 /**
 246  * nfs_can_coalesce_requests - test two requests for compatibility
 247  * @prev: pointer to nfs_page
 248  * @req: pointer to nfs_page
 249  *
 250  * The nfs_page structures 'prev' and 'req' are compared to ensure that the
 251  * page data area they describe is contiguous, and that their RPC
 252  * credentials, NFSv4 open state, and lockowners are the same.
 253  *
 254  * Return 'true' if this is the case, else return 'false'.
 255  */
 256 static int nfs_can_coalesce_requests(struct nfs_page *prev,
 257                                      struct nfs_page *req)
 258 {
 259         if (req->wb_context->cred != prev->wb_context->cred)
 260                 return 0;
 261         if (req->wb_context->lockowner != prev->wb_context->lockowner)
 262                 return 0;
 263         if (req->wb_context->state != prev->wb_context->state)
 264                 return 0;
 265         if (req->wb_index != (prev->wb_index + 1))
 266                 return 0;
 267         if (req->wb_pgbase != 0)
 268                 return 0;
 269         if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
 270                 return 0;
 271         return 1;
 272 }
 273
 274 /**
 275  * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
 276  * @desc: destination io descriptor
 277  * @req: request
 278  *
 279  * Returns true if the request 'req' was successfully coalesced into the
 280  * existing list of pages 'desc'.
 281  */
 282 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
 283                                      struct nfs_page *req)
 284 {
 285         size_t newlen = req->wb_bytes;
 286
 287         if (desc->pg_count != 0) {
 288                 struct nfs_page *prev;
 289
 290                 /*
 291                  * FIXME: ideally we should be able to coalesce all requests
 292                  * that are not block boundary aligned, but currently this
 293                  * is problematic for the case of bsize < PAGE_CACHE_SIZE,
 294                  * since nfs_flush_multi and nfs_pagein_multi assume you
 295                  * can have only one struct nfs_page.
 296                  */
 297                 if (desc->pg_bsize < PAGE_SIZE)
 298                         return 0;
 299                 newlen += desc->pg_count;
 300                 if (newlen > desc->pg_bsize)
 301                         return 0;
 302                 prev = nfs_list_entry(desc->pg_list.prev);
 303                 if (!nfs_can_coalesce_requests(prev, req))
 304                         return 0;
 305         } else
 306                 desc->pg_base = req->wb_pgbase;
 307         nfs_list_remove_request(req);
 308         nfs_list_add_request(req, &desc->pg_list);
 309         desc->pg_count = newlen;
 310         return 1;
 311 }
 312
 313 /*
 314  * Helper for nfs_pageio_add_request and nfs_pageio_complete
 315  */
 316 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
 317 {
 318         if (!list_empty(&desc->pg_list)) {
 319                 int error = desc->pg_doio(desc->pg_inode,
 320                                           &desc->pg_list,
 321                                           nfs_page_array_len(desc->pg_base,
 322                                                              desc->pg_count),
 323                                           desc->pg_count,
 324                                           desc->pg_ioflags);
 325                 if (error < 0)
 326                         desc->pg_error = error;
 327                 else
 328                         desc->pg_bytes_written += desc->pg_count;
 329         }
 330         if (list_empty(&desc->pg_list)) {
 331                 desc->pg_count = 0;
 332                 desc->pg_base = 0;
 333         }
 334 }
 335
 336 /**
 337  * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
 338  * @desc: destination io descriptor
 339  * @req: request
 340  *
 341  * Returns true if the request 'req' was successfully coalesced into the
 342  * existing list of pages 'desc'.
 343  */
 344 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
 345                            struct nfs_page *req)
 346 {
 347         while (!nfs_pageio_do_add_request(desc, req)) {
 348                 nfs_pageio_doio(desc);
 349                 if (desc->pg_error < 0)
 350                         return 0;
 351         }
 352         return 1;
 353 }
 354
 355 /**
 356  * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
 357  * @desc: pointer to io descriptor
 358  */
 359 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
 360 {
 361         nfs_pageio_doio(desc);
 362 }
 363
 364 #define NFS_SCAN_MAXENTRIES 16
 365 /**
 366  * nfs_scan_list - Scan a list for matching requests
 367  * @nfsi: NFS inode
 368  * @head: One of the NFS inode request lists
 369  * @dst: Destination list
 370  * @idx_start: lower bound of page->index to scan
 371  * @npages: idx_start + npages sets the upper bound to scan.
 372  *
 373  * Moves elements from one of the inode request lists.
 374  * If the number of requests is set to 0, the entire address_space
 375  * starting at index idx_start, is scanned.
 376  * The requests are *not* checked to ensure that they form a contiguous set.
 377  * You must be holding the inode's req_lock when calling this function
 378  */
 379 int nfs_scan_list(struct nfs_inode *nfsi, struct list_head *head,
 380                 struct list_head *dst, pgoff_t idx_start,
 381                 unsigned int npages)
 382 {
 383         struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
 384         struct nfs_page *req;
 385         pgoff_t idx_end;
 386         int found, i;
 387         int res;
 388
 389         res = 0;
 390         if (npages == 0)
 391                 idx_end = ~0;
 392         else
 393                 idx_end = idx_start + npages - 1;
 394
 395         for (;;) {
 396                 found = radix_tree_gang_lookup(&nfsi->nfs_page_tree,
 397                                 (void **)&pgvec[0], idx_start,
 398                                 NFS_SCAN_MAXENTRIES);
 399                 if (found <= 0)
 400                         break;
 401                 for (i = 0; i < found; i++) {
 402                         req = pgvec[i];
 403                         if (req->wb_index > idx_end)
 404                                 goto out;
 405                         idx_start = req->wb_index + 1;
 406                         if (req->wb_list_head != head)
 407                                 continue;
 408                         if (nfs_set_page_writeback_locked(req)) {
 409                                 nfs_list_remove_request(req);
 410                                 nfs_list_add_request(req, dst);
 411                                 res++;
 412                         }
 413                 }
 414
 415         }
 416 out:
 417         return res;
 418 }
 419
 420 int __init nfs_init_nfspagecache(void)
 421 {
 422         nfs_page_cachep = kmem_cache_create("nfs_page",
 423                                             sizeof(struct nfs_page),
 424                                             0, SLAB_HWCACHE_ALIGN,
 425                                             NULL, NULL);
 426         if (nfs_page_cachep == NULL)
 427                 return -ENOMEM;
 428
 429         return 0;
 430 }
 431
 432 void nfs_destroy_nfspagecache(void)
 433 {
 434         kmem_cache_destroy(nfs_page_cachep);
 435 }
 436