fs/nfs/read.c

   1 /*
   2  * linux/fs/nfs/read.c
   3  *
   4  * Block I/O for NFS
   5  *
   6  * Partial copy of Linus' read cache modifications to fs/nfs/file.c
   7  * modified for async RPC by okir@monad.swb.de
   8  */
   9
  10 #include <linux/time.h>
  11 #include <linux/kernel.h>
  12 #include <linux/errno.h>
  13 #include <linux/fcntl.h>
  14 #include <linux/stat.h>
  15 #include <linux/mm.h>
  16 #include <linux/slab.h>
  17 #include <linux/pagemap.h>
  18 #include <linux/sunrpc/clnt.h>
  19 #include <linux/nfs_fs.h>
  20 #include <linux/nfs_page.h>
  21 #include <linux/module.h>
  22
  23 #include "nfs4_fs.h"
  24 #include "internal.h"
  25 #include "iostat.h"
  26 #include "fscache.h"
  27 #include "pnfs.h"
  28
  29 #define NFSDBG_FACILITY         NFSDBG_PAGECACHE
  30
  31 static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops;
  32 static const struct nfs_rw_ops nfs_rw_read_ops;
  33
  34 static struct kmem_cache *nfs_rdata_cachep;
  35
  36 static struct nfs_rw_header *nfs_readhdr_alloc(void)
  37 {
  38         return kmem_cache_zalloc(nfs_rdata_cachep, GFP_KERNEL);
  39 }
  40
  41 static void nfs_readhdr_free(struct nfs_rw_header *rhdr)
  42 {
  43         kmem_cache_free(nfs_rdata_cachep, rhdr);
  44 }
  45
  46 static
  47 int nfs_return_empty_page(struct page *page)
  48 {
  49         zero_user(page, 0, PAGE_CACHE_SIZE);
  50         SetPageUptodate(page);
  51         unlock_page(page);
  52         return 0;
  53 }
  54
  55 void nfs_pageio_init_read(struct nfs_pageio_descriptor *pgio,
  56                               struct inode *inode, bool force_mds,
  57                               const struct nfs_pgio_completion_ops *compl_ops)
  58 {
  59         struct nfs_server *server = NFS_SERVER(inode);
  60         const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops;
  61
  62 #ifdef CONFIG_NFS_V4_1
  63         if (server->pnfs_curr_ld && !force_mds)
  64                 pg_ops = server->pnfs_curr_ld->pg_read_ops;
  65 #endif
  66         nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_read_ops,
  67                         server->rsize, 0);
  68 }
  69 EXPORT_SYMBOL_GPL(nfs_pageio_init_read);
  70
  71 void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio)
  72 {
  73         pgio->pg_ops = &nfs_pgio_rw_ops;
  74         pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->rsize;
  75 }
  76 EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds);
  77
  78 int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
  79                        struct page *page)
  80 {
  81         struct nfs_page *new;
  82         unsigned int len;
  83         struct nfs_pageio_descriptor pgio;
  84
  85         len = nfs_page_length(page);
  86         if (len == 0)
  87                 return nfs_return_empty_page(page);
  88         new = nfs_create_request(ctx, page, NULL, 0, len);
  89         if (IS_ERR(new)) {
  90                 unlock_page(page);
  91                 return PTR_ERR(new);
  92         }
  93         if (len < PAGE_CACHE_SIZE)
  94                 zero_user_segment(page, len, PAGE_CACHE_SIZE);
  95
  96         nfs_pageio_init_read(&pgio, inode, false,
  97                              &nfs_async_read_completion_ops);
  98         nfs_pageio_add_request(&pgio, new);
  99         nfs_pageio_complete(&pgio);
 100         NFS_I(inode)->read_io += pgio.pg_bytes_written;
 101         return 0;
 102 }
 103
 104 static void nfs_readpage_release(struct nfs_page *req)
 105 {
 106         struct inode *d_inode = req->wb_context->dentry->d_inode;
 107
 108         dprintk("NFS: read done (%s/%llu %d@%lld)\n", d_inode->i_sb->s_id,
 109                 (unsigned long long)NFS_FILEID(d_inode), req->wb_bytes,
 110                 (long long)req_offset(req));
 111
 112         if (nfs_page_group_sync_on_bit(req, PG_UNLOCKPAGE)) {
 113                 if (PageUptodate(req->wb_page))
 114                         nfs_readpage_to_fscache(d_inode, req->wb_page, 0);
 115
 116                 unlock_page(req->wb_page);
 117         }
 118
 119         dprintk("NFS: read done (%s/%Lu %d@%Ld)\n",
 120                         req->wb_context->dentry->d_inode->i_sb->s_id,
 121                         (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode),
 122                         req->wb_bytes,
 123                         (long long)req_offset(req));
 124         nfs_release_request(req);
 125 }
 126
 127 static void nfs_page_group_set_uptodate(struct nfs_page *req)
 128 {
 129         if (nfs_page_group_sync_on_bit(req, PG_UPTODATE))
 130                 SetPageUptodate(req->wb_page);
 131 }
 132
 133 static void nfs_read_completion(struct nfs_pgio_header *hdr)
 134 {
 135         unsigned long bytes = 0;
 136
 137         if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
 138                 goto out;
 139         while (!list_empty(&hdr->pages)) {
 140                 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
 141                 struct page *page = req->wb_page;
 142                 unsigned long start = req->wb_pgbase;
 143                 unsigned long end = req->wb_pgbase + req->wb_bytes;
 144
 145                 if (test_bit(NFS_IOHDR_EOF, &hdr->flags)) {
 146                         /* note: regions of the page not covered by a
 147                          * request are zeroed in nfs_readpage_async /
 148                          * readpage_async_filler */
 149                         if (bytes > hdr->good_bytes) {
 150                                 /* nothing in this request was good, so zero
 151                                  * the full extent of the request */
 152                                 zero_user_segment(page, start, end);
 153
 154                         } else if (hdr->good_bytes - bytes < req->wb_bytes) {
 155                                 /* part of this request has good bytes, but
 156                                  * not all. zero the bad bytes */
 157                                 start += hdr->good_bytes - bytes;
 158                                 WARN_ON(start < req->wb_pgbase);
 159                                 zero_user_segment(page, start, end);
 160                         }
 161                 }
 162                 bytes += req->wb_bytes;
 163                 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) {
 164                         if (bytes <= hdr->good_bytes)
 165                                 nfs_page_group_set_uptodate(req);
 166                 } else
 167                         nfs_page_group_set_uptodate(req);
 168                 nfs_list_remove_request(req);
 169                 nfs_readpage_release(req);
 170         }
 171 out:
 172         hdr->release(hdr);
 173 }
 174
 175 static void nfs_initiate_read(struct nfs_pgio_data *data, struct rpc_message *msg,
 176                               struct rpc_task_setup *task_setup_data, int how)
 177 {
 178         struct inode *inode = data->header->inode;
 179         int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0;
 180
 181         task_setup_data->flags |= swap_flags;
 182         NFS_PROTO(inode)->read_setup(data, msg);
 183 }
 184
 185 static void
 186 nfs_async_read_error(struct list_head *head)
 187 {
 188         struct nfs_page *req;
 189
 190         while (!list_empty(head)) {
 191                 req = nfs_list_entry(head->next);
 192                 nfs_list_remove_request(req);
 193                 nfs_readpage_release(req);
 194         }
 195 }
 196
 197 static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops = {
 198         .error_cleanup = nfs_async_read_error,
 199         .completion = nfs_read_completion,
 200 };
 201
 202 /*
 203  * This is the callback from RPC telling us whether a reply was
 204  * received or some error occurred (timeout or socket shutdown).
 205  */
 206 static int nfs_readpage_done(struct rpc_task *task, struct nfs_pgio_data *data,
 207                              struct inode *inode)
 208 {
 209         int status = NFS_PROTO(inode)->read_done(task, data);
 210         if (status != 0)
 211                 return status;
 212
 213         nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, data->res.count);
 214
 215         if (task->tk_status == -ESTALE) {
 216                 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
 217                 nfs_mark_for_revalidate(inode);
 218         }
 219         return 0;
 220 }
 221
 222 static void nfs_readpage_retry(struct rpc_task *task, struct nfs_pgio_data *data)
 223 {
 224         struct nfs_pgio_args *argp = &data->args;
 225         struct nfs_pgio_res  *resp = &data->res;
 226
 227         /* This is a short read! */
 228         nfs_inc_stats(data->header->inode, NFSIOS_SHORTREAD);
 229         /* Has the server at least made some progress? */
 230         if (resp->count == 0) {
 231                 nfs_set_pgio_error(data->header, -EIO, argp->offset);
 232                 return;
 233         }
 234         /* Yes, so retry the read at the end of the data */
 235         data->mds_offset += resp->count;
 236         argp->offset += resp->count;
 237         argp->pgbase += resp->count;
 238         argp->count -= resp->count;
 239         rpc_restart_call_prepare(task);
 240 }
 241
 242 static void nfs_readpage_result(struct rpc_task *task, struct nfs_pgio_data *data)
 243 {
 244         struct nfs_pgio_header *hdr = data->header;
 245
 246         if (data->res.eof) {
 247                 loff_t bound;
 248
 249                 bound = data->args.offset + data->res.count;
 250                 spin_lock(&hdr->lock);
 251                 if (bound < hdr->io_start + hdr->good_bytes) {
 252                         set_bit(NFS_IOHDR_EOF, &hdr->flags);
 253                         clear_bit(NFS_IOHDR_ERROR, &hdr->flags);
 254                         hdr->good_bytes = bound - hdr->io_start;
 255                 }
 256                 spin_unlock(&hdr->lock);
 257         } else if (data->res.count != data->args.count)
 258                 nfs_readpage_retry(task, data);
 259 }
 260
 261 /*
 262  * Read a page over NFS.
 263  * We read the page synchronously in the following case:
 264  *  -   The error flag is set for this page. This happens only when a
 265  *      previous async read operation failed.
 266  */
 267 int nfs_readpage(struct file *file, struct page *page)
 268 {
 269         struct nfs_open_context *ctx;
 270         struct inode *inode = page_file_mapping(page)->host;
 271         int             error;
 272
 273         dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
 274                 page, PAGE_CACHE_SIZE, page_file_index(page));
 275         nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
 276         nfs_add_stats(inode, NFSIOS_READPAGES, 1);
 277
 278         /*
 279          * Try to flush any pending writes to the file..
 280          *
 281          * NOTE! Because we own the page lock, there cannot
 282          * be any new pending writes generated at this point
 283          * for this page (other pages can be written to).
 284          */
 285         error = nfs_wb_page(inode, page);
 286         if (error)
 287                 goto out_unlock;
 288         if (PageUptodate(page))
 289                 goto out_unlock;
 290
 291         error = -ESTALE;
 292         if (NFS_STALE(inode))
 293                 goto out_unlock;
 294
 295         if (file == NULL) {
 296                 error = -EBADF;
 297                 ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
 298                 if (ctx == NULL)
 299                         goto out_unlock;
 300         } else
 301                 ctx = get_nfs_open_context(nfs_file_open_context(file));
 302
 303         if (!IS_SYNC(inode)) {
 304                 error = nfs_readpage_from_fscache(ctx, inode, page);
 305                 if (error == 0)
 306                         goto out;
 307         }
 308
 309         error = nfs_readpage_async(ctx, inode, page);
 310
 311 out:
 312         put_nfs_open_context(ctx);
 313         return error;
 314 out_unlock:
 315         unlock_page(page);
 316         return error;
 317 }
 318
 319 struct nfs_readdesc {
 320         struct nfs_pageio_descriptor *pgio;
 321         struct nfs_open_context *ctx;
 322 };
 323
 324 static int
 325 readpage_async_filler(void *data, struct page *page)
 326 {
 327         struct nfs_readdesc *desc = (struct nfs_readdesc *)data;
 328         struct nfs_page *new;
 329         unsigned int len;
 330         int error;
 331
 332         len = nfs_page_length(page);
 333         if (len == 0)
 334                 return nfs_return_empty_page(page);
 335
 336         new = nfs_create_request(desc->ctx, page, NULL, 0, len);
 337         if (IS_ERR(new))
 338                 goto out_error;
 339
 340         if (len < PAGE_CACHE_SIZE)
 341                 zero_user_segment(page, len, PAGE_CACHE_SIZE);
 342         if (!nfs_pageio_add_request(desc->pgio, new)) {
 343                 error = desc->pgio->pg_error;
 344                 goto out_unlock;
 345         }
 346         return 0;
 347 out_error:
 348         error = PTR_ERR(new);
 349 out_unlock:
 350         unlock_page(page);
 351         return error;
 352 }
 353
 354 int nfs_readpages(struct file *filp, struct address_space *mapping,
 355                 struct list_head *pages, unsigned nr_pages)
 356 {
 357         struct nfs_pageio_descriptor pgio;
 358         struct nfs_readdesc desc = {
 359                 .pgio = &pgio,
 360         };
 361         struct inode *inode = mapping->host;
 362         unsigned long npages;
 363         int ret = -ESTALE;
 364
 365         dprintk("NFS: nfs_readpages (%s/%Lu %d)\n",
 366                         inode->i_sb->s_id,
 367                         (unsigned long long)NFS_FILEID(inode),
 368                         nr_pages);
 369         nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
 370
 371         if (NFS_STALE(inode))
 372                 goto out;
 373
 374         if (filp == NULL) {
 375                 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
 376                 if (desc.ctx == NULL)
 377                         return -EBADF;
 378         } else
 379                 desc.ctx = get_nfs_open_context(nfs_file_open_context(filp));
 380
 381         /* attempt to read as many of the pages as possible from the cache
 382          * - this returns -ENOBUFS immediately if the cookie is negative
 383          */
 384         ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping,
 385                                          pages, &nr_pages);
 386         if (ret == 0)
 387                 goto read_complete; /* all pages were read */
 388
 389         nfs_pageio_init_read(&pgio, inode, false,
 390                              &nfs_async_read_completion_ops);
 391
 392         ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
 393
 394         nfs_pageio_complete(&pgio);
 395         NFS_I(inode)->read_io += pgio.pg_bytes_written;
 396         npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 397         nfs_add_stats(inode, NFSIOS_READPAGES, npages);
 398 read_complete:
 399         put_nfs_open_context(desc.ctx);
 400 out:
 401         return ret;
 402 }
 403
 404 int __init nfs_init_readpagecache(void)
 405 {
 406         nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
 407                                              sizeof(struct nfs_rw_header),
 408                                              0, SLAB_HWCACHE_ALIGN,
 409                                              NULL);
 410         if (nfs_rdata_cachep == NULL)
 411                 return -ENOMEM;
 412
 413         return 0;
 414 }
 415
 416 void nfs_destroy_readpagecache(void)
 417 {
 418         kmem_cache_destroy(nfs_rdata_cachep);
 419 }
 420
 421 static const struct nfs_rw_ops nfs_rw_read_ops = {
 422         .rw_mode                = FMODE_READ,
 423         .rw_alloc_header        = nfs_readhdr_alloc,
 424         .rw_free_header         = nfs_readhdr_free,
 425         .rw_done                = nfs_readpage_done,
 426         .rw_result              = nfs_readpage_result,
 427         .rw_initiate            = nfs_initiate_read,
 428 };