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 #define NFS_PARANOIA 1
  22
  23 static kmem_cache_t *nfs_page_cachep;
  24
  25 static inline struct nfs_page *
  26 nfs_page_alloc(void)
  27 {
  28         struct nfs_page *p;
  29         p = kmem_cache_alloc(nfs_page_cachep, SLAB_KERNEL);
  30         if (p) {
  31                 memset(p, 0, sizeof(*p));
  32                 INIT_LIST_HEAD(&p->wb_list);
  33         }
  34         return p;
  35 }
  36
  37 static inline void
  38 nfs_page_free(struct nfs_page *p)
  39 {
  40         kmem_cache_free(nfs_page_cachep, p);
  41 }
  42
  43 /**
  44  * nfs_create_request - Create an NFS read/write request.
  45  * @file: file descriptor to use
  46  * @inode: inode to which the request is attached
  47  * @page: page to write
  48  * @offset: starting offset within the page for the write
  49  * @count: number of bytes to read/write
  50  *
  51  * The page must be locked by the caller. This makes sure we never
  52  * create two different requests for the same page, and avoids
  53  * a possible deadlock when we reach the hard limit on the number
  54  * of dirty pages.
  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         /* Deal with hard limits.  */
  66         for (;;) {
  67                 /* try to allocate the request struct */
  68                 req = nfs_page_alloc();
  69                 if (req != NULL)
  70                         break;
  71
  72                 /* Try to free up at least one request in order to stay
  73                  * below the hard limit
  74                  */
  75                 if (signalled() && (server->flags & NFS_MOUNT_INTR))
  76                         return ERR_PTR(-ERESTARTSYS);
  77                 yield();
  78         }
  79
  80         /* Initialize the request struct. Initially, we assume a
  81          * long write-back delay. This will be adjusted in
  82          * update_nfs_request below if the region is not locked. */
  83         req->wb_page    = page;
  84         atomic_set(&req->wb_complete, 0);
  85         req->wb_index   = page->index;
  86         page_cache_get(page);
  87         BUG_ON(PagePrivate(page));
  88         BUG_ON(!PageLocked(page));
  89 #if 0   // mask by Victor yu. 02-12-2007
  90         BUG_ON(page->mapping->host != inode);
  91 #else
  92         BUG_ON(page->u.xx.mapping->host != inode);
  93 #endif
  94         req->wb_offset  = offset;
  95         req->wb_pgbase  = offset;
  96         req->wb_bytes   = count;
  97         atomic_set(&req->wb_count, 1);
  98         req->wb_context = get_nfs_open_context(ctx);
  99
 100         return req;
 101 }
 102
 103 /**
 104  * nfs_unlock_request - Unlock request and wake up sleepers.
 105  * @req:
 106  */
 107 void nfs_unlock_request(struct nfs_page *req)
 108 {
 109         if (!NFS_WBACK_BUSY(req)) {
 110                 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
 111                 BUG();
 112         }
 113         smp_mb__before_clear_bit();
 114         clear_bit(PG_BUSY, &req->wb_flags);
 115         smp_mb__after_clear_bit();
 116         wake_up_bit(&req->wb_flags, PG_BUSY);
 117         nfs_release_request(req);
 118 }
 119
 120 /**
 121  * nfs_set_page_writeback_locked - Lock a request for writeback
 122  * @req:
 123  */
 124 int nfs_set_page_writeback_locked(struct nfs_page *req)
 125 {
 126         struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);
 127
 128         if (!nfs_lock_request(req))
 129                 return 0;
 130         radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
 131         return 1;
 132 }
 133
 134 /**
 135  * nfs_clear_page_writeback - Unlock request and wake up sleepers
 136  */
 137 void nfs_clear_page_writeback(struct nfs_page *req)
 138 {
 139         struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);
 140
 141         if (req->wb_page != NULL) {
 142                 spin_lock(&nfsi->req_lock);
 143                 radix_tree_tag_clear(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
 144                 spin_unlock(&nfsi->req_lock);
 145         }
 146         nfs_unlock_request(req);
 147 }
 148
 149 /**
 150  * nfs_clear_request - Free up all resources allocated to the request
 151  * @req:
 152  *
 153  * Release page resources associated with a write request after it
 154  * has completed.
 155  */
 156 void nfs_clear_request(struct nfs_page *req)
 157 {
 158         struct page *page = req->wb_page;
 159         if (page != NULL) {
 160                 page_cache_release(page);
 161                 req->wb_page = NULL;
 162         }
 163 }
 164
 165
 166 /**
 167  * nfs_release_request - Release the count on an NFS read/write request
 168  * @req: request to release
 169  *
 170  * Note: Should never be called with the spinlock held!
 171  */
 172 void
 173 nfs_release_request(struct nfs_page *req)
 174 {
 175         if (!atomic_dec_and_test(&req->wb_count))
 176                 return;
 177
 178 #ifdef NFS_PARANOIA
 179         BUG_ON (!list_empty(&req->wb_list));
 180         BUG_ON (NFS_WBACK_BUSY(req));
 181 #endif
 182
 183         /* Release struct file or cached credential */
 184         nfs_clear_request(req);
 185         put_nfs_open_context(req->wb_context);
 186         nfs_page_free(req);
 187 }
 188
 189 static int nfs_wait_bit_interruptible(void *word)
 190 {
 191         int ret = 0;
 192
 193         if (signal_pending(current))
 194                 ret = -ERESTARTSYS;
 195         else
 196                 schedule();
 197         return ret;
 198 }
 199
 200 /**
 201  * nfs_wait_on_request - Wait for a request to complete.
 202  * @req: request to wait upon.
 203  *
 204  * Interruptible by signals only if mounted with intr flag.
 205  * The user is responsible for holding a count on the request.
 206  */
 207 int
 208 nfs_wait_on_request(struct nfs_page *req)
 209 {
 210         struct rpc_clnt *clnt = NFS_CLIENT(req->wb_context->dentry->d_inode);
 211         sigset_t oldmask;
 212         int ret = 0;
 213
 214         if (!test_bit(PG_BUSY, &req->wb_flags))
 215                 goto out;
 216         /*
 217          * Note: the call to rpc_clnt_sigmask() suffices to ensure that we
 218          *       are not interrupted if intr flag is not set
 219          */
 220         rpc_clnt_sigmask(clnt, &oldmask);
 221         ret = out_of_line_wait_on_bit(&req->wb_flags, PG_BUSY,
 222                         nfs_wait_bit_interruptible, TASK_INTERRUPTIBLE);
 223         rpc_clnt_sigunmask(clnt, &oldmask);
 224 out:
 225         return ret;
 226 }
 227
 228 /**
 229  * nfs_coalesce_requests - Split coalesced requests out from a list.
 230  * @head: source list
 231  * @dst: destination list
 232  * @nmax: maximum number of requests to coalesce
 233  *
 234  * Moves a maximum of 'nmax' elements from one list to another.
 235  * The elements are checked to ensure that they form a contiguous set
 236  * of pages, and that the RPC credentials are the same.
 237  */
 238 int
 239 nfs_coalesce_requests(struct list_head *head, struct list_head *dst,
 240                       unsigned int nmax)
 241 {
 242         struct nfs_page         *req = NULL;
 243         unsigned int            npages = 0;
 244
 245         while (!list_empty(head)) {
 246                 struct nfs_page *prev = req;
 247
 248                 req = nfs_list_entry(head->next);
 249                 if (prev) {
 250                         if (req->wb_context->cred != prev->wb_context->cred)
 251                                 break;
 252                         if (req->wb_context->lockowner != prev->wb_context->lockowner)
 253                                 break;
 254                         if (req->wb_context->state != prev->wb_context->state)
 255                                 break;
 256                         if (req->wb_index != (prev->wb_index + 1))
 257                                 break;
 258
 259                         if (req->wb_pgbase != 0)
 260                                 break;
 261                 }
 262                 nfs_list_remove_request(req);
 263                 nfs_list_add_request(req, dst);
 264                 npages++;
 265                 if (req->wb_pgbase + req->wb_bytes != PAGE_CACHE_SIZE)
 266                         break;
 267                 if (npages >= nmax)
 268                         break;
 269         }
 270         return npages;
 271 }
 272
 273 #define NFS_SCAN_MAXENTRIES 16
 274 /**
 275  * nfs_scan_lock_dirty - Scan the radix tree for dirty requests
 276  * @nfsi: NFS inode
 277  * @dst: Destination list
 278  * @idx_start: lower bound of page->index to scan
 279  * @npages: idx_start + npages sets the upper bound to scan.
 280  *
 281  * Moves elements from one of the inode request lists.
 282  * If the number of requests is set to 0, the entire address_space
 283  * starting at index idx_start, is scanned.
 284  * The requests are *not* checked to ensure that they form a contiguous set.
 285  * You must be holding the inode's req_lock when calling this function
 286  */
 287 int
 288 nfs_scan_lock_dirty(struct nfs_inode *nfsi, struct list_head *dst,
 289               unsigned long idx_start, unsigned int npages)
 290 {
 291         struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
 292         struct nfs_page *req;
 293         unsigned long idx_end;
 294         int found, i;
 295         int res;
 296
 297         res = 0;
 298         if (npages == 0)
 299                 idx_end = ~0;
 300         else
 301                 idx_end = idx_start + npages - 1;
 302
 303         for (;;) {
 304                 found = radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree,
 305                                 (void **)&pgvec[0], idx_start, NFS_SCAN_MAXENTRIES,
 306                                 NFS_PAGE_TAG_DIRTY);
 307                 if (found <= 0)
 308                         break;
 309                 for (i = 0; i < found; i++) {
 310                         req = pgvec[i];
 311                         if (req->wb_index > idx_end)
 312                                 goto out;
 313
 314                         idx_start = req->wb_index + 1;
 315
 316                         if (nfs_set_page_writeback_locked(req)) {
 317                                 radix_tree_tag_clear(&nfsi->nfs_page_tree,
 318                                                 req->wb_index, NFS_PAGE_TAG_DIRTY);
 319                                 nfs_list_remove_request(req);
 320                                 nfs_list_add_request(req, dst);
 321                                 dec_zone_page_state(req->wb_page, NR_FILE_DIRTY);
 322                                 res++;
 323                         }
 324                 }
 325         }
 326 out:
 327         return res;
 328 }
 329
 330 /**
 331  * nfs_scan_list - Scan a list for matching requests
 332  * @nfsi: NFS inode
 333  * @head: One of the NFS inode request lists
 334  * @dst: Destination list
 335  * @idx_start: lower bound of page->index to scan
 336  * @npages: idx_start + npages sets the upper bound to scan.
 337  *
 338  * Moves elements from one of the inode request lists.
 339  * If the number of requests is set to 0, the entire address_space
 340  * starting at index idx_start, is scanned.
 341  * The requests are *not* checked to ensure that they form a contiguous set.
 342  * You must be holding the inode's req_lock when calling this function
 343  */
 344 int nfs_scan_list(struct nfs_inode *nfsi, struct list_head *head,
 345                 struct list_head *dst, unsigned long idx_start,
 346                 unsigned int npages)
 347 {
 348         struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
 349         struct nfs_page *req;
 350         unsigned long idx_end;
 351         int found, i;
 352         int res;
 353
 354         res = 0;
 355         if (npages == 0)
 356                 idx_end = ~0;
 357         else
 358                 idx_end = idx_start + npages - 1;
 359
 360         for (;;) {
 361                 found = radix_tree_gang_lookup(&nfsi->nfs_page_tree,
 362                                 (void **)&pgvec[0], idx_start,
 363                                 NFS_SCAN_MAXENTRIES);
 364                 if (found <= 0)
 365                         break;
 366                 for (i = 0; i < found; i++) {
 367                         req = pgvec[i];
 368                         if (req->wb_index > idx_end)
 369                                 goto out;
 370                         idx_start = req->wb_index + 1;
 371                         if (req->wb_list_head != head)
 372                                 continue;
 373                         if (nfs_set_page_writeback_locked(req)) {
 374                                 nfs_list_remove_request(req);
 375                                 nfs_list_add_request(req, dst);
 376                                 res++;
 377                         }
 378                 }
 379
 380         }
 381 out:
 382         return res;
 383 }
 384
 385 int __init nfs_init_nfspagecache(void)
 386 {
 387         nfs_page_cachep = kmem_cache_create("nfs_page",
 388                                             sizeof(struct nfs_page),
 389                                             0, SLAB_HWCACHE_ALIGN,
 390                                             NULL, NULL);
 391         if (nfs_page_cachep == NULL)
 392                 return -ENOMEM;
 393
 394         return 0;
 395 }
 396
 397 void nfs_destroy_nfspagecache(void)
 398 {
 399         kmem_cache_destroy(nfs_page_cachep);
 400 }
 401