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/config.h>
  13 #include <linux/slab.h>
  14 #include <linux/file.h>
  15 #include <linux/sunrpc/clnt.h>
  16 #include <linux/nfs3.h>
  17 #include <linux/nfs4.h>
  18 #include <linux/nfs_page.h>
  19 #include <linux/nfs_fs.h>
  20 #include <linux/nfs_mount.h>
  21
  22 #define NFS_PARANOIA 1
  23
  24 static kmem_cache_t *nfs_page_cachep;
  25
  26 static inline struct nfs_page *
  27 nfs_page_alloc(void)
  28 {
  29         struct nfs_page *p;
  30         p = kmem_cache_alloc(nfs_page_cachep, SLAB_KERNEL);
  31         if (p) {
  32                 memset(p, 0, sizeof(*p));
  33                 INIT_LIST_HEAD(&p->wb_list);
  34         }
  35         return p;
  36 }
  37
  38 static inline void
  39 nfs_page_free(struct nfs_page *p)
  40 {
  41         kmem_cache_free(nfs_page_cachep, p);
  42 }
  43
  44 /**
  45  * nfs_create_request - Create an NFS read/write request.
  46  * @file: file descriptor to use
  47  * @inode: inode to which the request is attached
  48  * @page: page to write
  49  * @offset: starting offset within the page for the write
  50  * @count: number of bytes to read/write
  51  *
  52  * The page must be locked by the caller. This makes sure we never
  53  * create two different requests for the same page, and avoids
  54  * a possible deadlock when we reach the hard limit on the number
  55  * of dirty pages.
  56  * User should ensure it is safe to sleep in this function.
  57  */
  58 struct nfs_page *
  59 nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
  60                    struct page *page,
  61                    unsigned int offset, unsigned int count)
  62 {
  63         struct nfs_server *server = NFS_SERVER(inode);
  64         struct nfs_page         *req;
  65
  66         /* Deal with hard limits.  */
  67         for (;;) {
  68                 /* try to allocate the request struct */
  69                 req = nfs_page_alloc();
  70                 if (req != NULL)
  71                         break;
  72
  73                 /* Try to free up at least one request in order to stay
  74                  * below the hard limit
  75                  */
  76                 if (signalled() && (server->flags & NFS_MOUNT_INTR))
  77                         return ERR_PTR(-ERESTARTSYS);
  78                 yield();
  79         }
  80
  81         /* Initialize the request struct. Initially, we assume a
  82          * long write-back delay. This will be adjusted in
  83          * update_nfs_request below if the region is not locked. */
  84         req->wb_page    = page;
  85         atomic_set(&req->wb_complete, 0);
  86         req->wb_index   = page->index;
  87         page_cache_get(page);
  88         req->wb_offset  = offset;
  89         req->wb_pgbase  = offset;
  90         req->wb_bytes   = count;
  91         atomic_set(&req->wb_count, 1);
  92         req->wb_context = get_nfs_open_context(ctx);
  93
  94         return req;
  95 }
  96
  97 /**
  98  * nfs_unlock_request - Unlock request and wake up sleepers.
  99  * @req:
 100  */
 101 void nfs_unlock_request(struct nfs_page *req)
 102 {
 103         if (!NFS_WBACK_BUSY(req)) {
 104                 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
 105                 BUG();
 106         }
 107         smp_mb__before_clear_bit();
 108         clear_bit(PG_BUSY, &req->wb_flags);
 109         smp_mb__after_clear_bit();
 110         wake_up_bit(&req->wb_flags, PG_BUSY);
 111         nfs_release_request(req);
 112 }
 113
 114 /**
 115  * nfs_clear_request - Free up all resources allocated to the request
 116  * @req:
 117  *
 118  * Release page resources associated with a write request after it
 119  * has completed.
 120  */
 121 void nfs_clear_request(struct nfs_page *req)
 122 {
 123         if (req->wb_page) {
 124                 page_cache_release(req->wb_page);
 125                 req->wb_page = NULL;
 126         }
 127 }
 128
 129
 130 /**
 131  * nfs_release_request - Release the count on an NFS read/write request
 132  * @req: request to release
 133  *
 134  * Note: Should never be called with the spinlock held!
 135  */
 136 void
 137 nfs_release_request(struct nfs_page *req)
 138 {
 139         if (!atomic_dec_and_test(&req->wb_count))
 140                 return;
 141
 142 #ifdef NFS_PARANOIA
 143         BUG_ON (!list_empty(&req->wb_list));
 144         BUG_ON (NFS_WBACK_BUSY(req));
 145 #endif
 146
 147         /* Release struct file or cached credential */
 148         nfs_clear_request(req);
 149         put_nfs_open_context(req->wb_context);
 150         nfs_page_free(req);
 151 }
 152
 153 /**
 154  * nfs_list_add_request - Insert a request into a sorted list
 155  * @req: request
 156  * @head: head of list into which to insert the request.
 157  *
 158  * Note that the wb_list is sorted by page index in order to facilitate
 159  * coalescing of requests.
 160  * We use an insertion sort that is optimized for the case of appended
 161  * writes.
 162  */
 163 void
 164 nfs_list_add_request(struct nfs_page *req, struct list_head *head)
 165 {
 166         struct list_head *pos;
 167
 168 #ifdef NFS_PARANOIA
 169         if (!list_empty(&req->wb_list)) {
 170                 printk(KERN_ERR "NFS: Add to list failed!\n");
 171                 BUG();
 172         }
 173 #endif
 174         list_for_each_prev(pos, head) {
 175                 struct nfs_page *p = nfs_list_entry(pos);
 176                 if (p->wb_index < req->wb_index)
 177                         break;
 178         }
 179         list_add(&req->wb_list, pos);
 180         req->wb_list_head = head;
 181 }
 182
 183 static int nfs_wait_bit_interruptible(void *word)
 184 {
 185         int ret = 0;
 186
 187         if (signal_pending(current))
 188                 ret = -ERESTARTSYS;
 189         else
 190                 schedule();
 191         return ret;
 192 }
 193
 194 /**
 195  * nfs_wait_on_request - Wait for a request to complete.
 196  * @req: request to wait upon.
 197  *
 198  * Interruptible by signals only if mounted with intr flag.
 199  * The user is responsible for holding a count on the request.
 200  */
 201 int
 202 nfs_wait_on_request(struct nfs_page *req)
 203 {
 204         struct rpc_clnt *clnt = NFS_CLIENT(req->wb_context->dentry->d_inode);
 205         sigset_t oldmask;
 206         int ret = 0;
 207
 208         if (!test_bit(PG_BUSY, &req->wb_flags))
 209                 goto out;
 210         /*
 211          * Note: the call to rpc_clnt_sigmask() suffices to ensure that we
 212          *       are not interrupted if intr flag is not set
 213          */
 214         rpc_clnt_sigmask(clnt, &oldmask);
 215         ret = out_of_line_wait_on_bit(&req->wb_flags, PG_BUSY,
 216                         nfs_wait_bit_interruptible, TASK_INTERRUPTIBLE);
 217         rpc_clnt_sigunmask(clnt, &oldmask);
 218 out:
 219         return ret;
 220 }
 221
 222 /**
 223  * nfs_coalesce_requests - Split coalesced requests out from a list.
 224  * @head: source list
 225  * @dst: destination list
 226  * @nmax: maximum number of requests to coalesce
 227  *
 228  * Moves a maximum of 'nmax' elements from one list to another.
 229  * The elements are checked to ensure that they form a contiguous set
 230  * of pages, and that the RPC credentials are the same.
 231  */
 232 int
 233 nfs_coalesce_requests(struct list_head *head, struct list_head *dst,
 234                       unsigned int nmax)
 235 {
 236         struct nfs_page         *req = NULL;
 237         unsigned int            npages = 0;
 238
 239         while (!list_empty(head)) {
 240                 struct nfs_page *prev = req;
 241
 242                 req = nfs_list_entry(head->next);
 243                 if (prev) {
 244                         if (req->wb_context->cred != prev->wb_context->cred)
 245                                 break;
 246                         if (req->wb_context->lockowner != prev->wb_context->lockowner)
 247                                 break;
 248                         if (req->wb_context->state != prev->wb_context->state)
 249                                 break;
 250                         if (req->wb_index != (prev->wb_index + 1))
 251                                 break;
 252
 253                         if (req->wb_pgbase != 0)
 254                                 break;
 255                 }
 256                 nfs_list_remove_request(req);
 257                 nfs_list_add_request(req, dst);
 258                 npages++;
 259                 if (req->wb_pgbase + req->wb_bytes != PAGE_CACHE_SIZE)
 260                         break;
 261                 if (npages >= nmax)
 262                         break;
 263         }
 264         return npages;
 265 }
 266
 267 /**
 268  * nfs_scan_list - Scan a list for matching requests
 269  * @head: One of the NFS inode request lists
 270  * @dst: Destination list
 271  * @idx_start: lower bound of page->index to scan
 272  * @npages: idx_start + npages sets the upper bound to scan.
 273  *
 274  * Moves elements from one of the inode request lists.
 275  * If the number of requests is set to 0, the entire address_space
 276  * starting at index idx_start, is scanned.
 277  * The requests are *not* checked to ensure that they form a contiguous set.
 278  * You must be holding the inode's req_lock when calling this function
 279  */
 280 int
 281 nfs_scan_list(struct list_head *head, struct list_head *dst,
 282               unsigned long idx_start, unsigned int npages)
 283 {
 284         struct list_head        *pos, *tmp;
 285         struct nfs_page         *req;
 286         unsigned long           idx_end;
 287         int                     res;
 288
 289         res = 0;
 290         if (npages == 0)
 291                 idx_end = ~0;
 292         else
 293                 idx_end = idx_start + npages - 1;
 294
 295         list_for_each_safe(pos, tmp, head) {
 296
 297                 req = nfs_list_entry(pos);
 298
 299                 if (req->wb_index < idx_start)
 300                         continue;
 301                 if (req->wb_index > idx_end)
 302                         break;
 303
 304                 if (!nfs_lock_request(req))
 305                         continue;
 306                 nfs_list_remove_request(req);
 307                 nfs_list_add_request(req, dst);
 308                 res++;
 309         }
 310         return res;
 311 }
 312
 313 int nfs_init_nfspagecache(void)
 314 {
 315         nfs_page_cachep = kmem_cache_create("nfs_page",
 316                                             sizeof(struct nfs_page),
 317                                             0, SLAB_HWCACHE_ALIGN,
 318                                             NULL, NULL);
 319         if (nfs_page_cachep == NULL)
 320                 return -ENOMEM;
 321
 322         return 0;
 323 }
 324
 325 void nfs_destroy_nfspagecache(void)
 326 {
 327         if (kmem_cache_destroy(nfs_page_cachep))
 328                 printk(KERN_INFO "nfs_page: not all structures were freed\n");
 329 }
 330