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_pageio_ops nfs_pageio_read_ops;
  32 static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops;
  33 static const struct nfs_rw_ops nfs_rw_read_ops;
  34
  35 static struct kmem_cache *nfs_rdata_cachep;
  36
  37 static struct nfs_rw_header *nfs_readhdr_alloc(void)
  38 {
  39         return kmem_cache_zalloc(nfs_rdata_cachep, GFP_KERNEL);
  40 }
  41
  42 static void nfs_readhdr_free(struct nfs_rw_header *rhdr)
  43 {
  44         kmem_cache_free(nfs_rdata_cachep, rhdr);
  45 }
  46
  47 static
  48 int nfs_return_empty_page(struct page *page)
  49 {
  50         zero_user(page, 0, PAGE_CACHE_SIZE);
  51         SetPageUptodate(page);
  52         unlock_page(page);
  53         return 0;
  54 }
  55
  56 void nfs_pageio_init_read(struct nfs_pageio_descriptor *pgio,
  57                               struct inode *inode, bool force_mds,
  58                               const struct nfs_pgio_completion_ops *compl_ops)
  59 {
  60         struct nfs_server *server = NFS_SERVER(inode);
  61         const struct nfs_pageio_ops *pg_ops = &nfs_pageio_read_ops;
  62
  63 #ifdef CONFIG_NFS_V4_1
  64         if (server->pnfs_curr_ld && !force_mds)
  65                 pg_ops = server->pnfs_curr_ld->pg_read_ops;
  66 #endif
  67         nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_read_ops,
  68                         server->rsize, 0);
  69 }
  70 EXPORT_SYMBOL_GPL(nfs_pageio_init_read);
  71
  72 void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio)
  73 {
  74         pgio->pg_ops = &nfs_pageio_read_ops;
  75         pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->rsize;
  76 }
  77 EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds);
  78
  79 int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
  80                        struct page *page)
  81 {
  82         struct nfs_page *new;
  83         unsigned int len;
  84         struct nfs_pageio_descriptor pgio;
  85
  86         len = nfs_page_length(page);
  87         if (len == 0)
  88                 return nfs_return_empty_page(page);
  89         new = nfs_create_request(ctx, inode, page, 0, len);
  90         if (IS_ERR(new)) {
  91                 unlock_page(page);
  92                 return PTR_ERR(new);
  93         }
  94         if (len < PAGE_CACHE_SIZE)
  95                 zero_user_segment(page, len, PAGE_CACHE_SIZE);
  96
  97         nfs_pageio_init_read(&pgio, inode, false,
  98                              &nfs_async_read_completion_ops);
  99         nfs_pageio_add_request(&pgio, new);
 100         nfs_pageio_complete(&pgio);
 101         NFS_I(inode)->read_io += pgio.pg_bytes_written;
 102         return 0;
 103 }
 104
 105 static void nfs_readpage_release(struct nfs_page *req)
 106 {
 107         struct inode *d_inode = req->wb_context->dentry->d_inode;
 108
 109         if (PageUptodate(req->wb_page))
 110                 nfs_readpage_to_fscache(d_inode, req->wb_page, 0);
 111
 112         unlock_page(req->wb_page);
 113
 114         dprintk("NFS: read done (%s/%Lu %d@%Ld)\n",
 115                         req->wb_context->dentry->d_inode->i_sb->s_id,
 116                         (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode),
 117                         req->wb_bytes,
 118                         (long long)req_offset(req));
 119         nfs_release_request(req);
 120 }
 121
 122 /* Note io was page aligned */
 123 static void nfs_read_completion(struct nfs_pgio_header *hdr)
 124 {
 125         unsigned long bytes = 0;
 126
 127         if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
 128                 goto out;
 129         while (!list_empty(&hdr->pages)) {
 130                 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
 131                 struct page *page = req->wb_page;
 132
 133                 if (test_bit(NFS_IOHDR_EOF, &hdr->flags)) {
 134                         if (bytes > hdr->good_bytes)
 135                                 zero_user(page, 0, PAGE_SIZE);
 136                         else if (hdr->good_bytes - bytes < PAGE_SIZE)
 137                                 zero_user_segment(page,
 138                                         hdr->good_bytes & ~PAGE_MASK,
 139                                         PAGE_SIZE);
 140                 }
 141                 bytes += req->wb_bytes;
 142                 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) {
 143                         if (bytes <= hdr->good_bytes)
 144                                 SetPageUptodate(page);
 145                 } else
 146                         SetPageUptodate(page);
 147                 nfs_list_remove_request(req);
 148                 nfs_readpage_release(req);
 149         }
 150 out:
 151         hdr->release(hdr);
 152 }
 153
 154 static void nfs_initiate_read(struct nfs_pgio_data *data, struct rpc_message *msg,
 155                               struct rpc_task_setup *task_setup_data, int how)
 156 {
 157         struct inode *inode = data->header->inode;
 158         int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0;
 159
 160         task_setup_data->flags |= swap_flags;
 161         NFS_PROTO(inode)->read_setup(data, msg);
 162 }
 163
 164 static int nfs_do_read(struct nfs_pgio_data *data,
 165                 const struct rpc_call_ops *call_ops)
 166 {
 167         struct inode *inode = data->header->inode;
 168
 169         return nfs_initiate_pgio(NFS_CLIENT(inode), data, call_ops, 0, 0);
 170 }
 171
 172 static int
 173 nfs_do_multiple_reads(struct list_head *head,
 174                 const struct rpc_call_ops *call_ops)
 175 {
 176         struct nfs_pgio_data *data;
 177         int ret = 0;
 178
 179         while (!list_empty(head)) {
 180                 int ret2;
 181
 182                 data = list_first_entry(head, struct nfs_pgio_data, list);
 183                 list_del_init(&data->list);
 184
 185                 ret2 = nfs_do_read(data, call_ops);
 186                 if (ret == 0)
 187                         ret = ret2;
 188         }
 189         return ret;
 190 }
 191
 192 static void
 193 nfs_async_read_error(struct list_head *head)
 194 {
 195         struct nfs_page *req;
 196
 197         while (!list_empty(head)) {
 198                 req = nfs_list_entry(head->next);
 199                 nfs_list_remove_request(req);
 200                 nfs_readpage_release(req);
 201         }
 202 }
 203
 204 static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops = {
 205         .error_cleanup = nfs_async_read_error,
 206         .completion = nfs_read_completion,
 207 };
 208
 209 static int nfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
 210 {
 211         struct nfs_rw_header *rhdr;
 212         struct nfs_pgio_header *hdr;
 213         int ret;
 214
 215         rhdr = nfs_rw_header_alloc(desc->pg_rw_ops);
 216         if (!rhdr) {
 217                 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
 218                 return -ENOMEM;
 219         }
 220         hdr = &rhdr->header;
 221         nfs_pgheader_init(desc, hdr, nfs_rw_header_free);
 222         atomic_inc(&hdr->refcnt);
 223         ret = nfs_generic_pgio(desc, hdr);
 224         if (ret == 0)
 225                 ret = nfs_do_multiple_reads(&hdr->rpc_list,
 226                                             desc->pg_rpc_callops);
 227         if (atomic_dec_and_test(&hdr->refcnt))
 228                 hdr->completion_ops->completion(hdr);
 229         return ret;
 230 }
 231
 232 static const struct nfs_pageio_ops nfs_pageio_read_ops = {
 233         .pg_test = nfs_generic_pg_test,
 234         .pg_doio = nfs_generic_pg_readpages,
 235 };
 236
 237 /*
 238  * This is the callback from RPC telling us whether a reply was
 239  * received or some error occurred (timeout or socket shutdown).
 240  */
 241 static int nfs_readpage_done(struct rpc_task *task, struct nfs_pgio_data *data,
 242                              struct inode *inode)
 243 {
 244         int status = NFS_PROTO(inode)->read_done(task, data);
 245         if (status != 0)
 246                 return status;
 247
 248         nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, data->res.count);
 249
 250         if (task->tk_status == -ESTALE) {
 251                 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
 252                 nfs_mark_for_revalidate(inode);
 253         }
 254         return 0;
 255 }
 256
 257 static void nfs_readpage_retry(struct rpc_task *task, struct nfs_pgio_data *data)
 258 {
 259         struct nfs_pgio_args *argp = &data->args;
 260         struct nfs_pgio_res  *resp = &data->res;
 261
 262         /* This is a short read! */
 263         nfs_inc_stats(data->header->inode, NFSIOS_SHORTREAD);
 264         /* Has the server at least made some progress? */
 265         if (resp->count == 0) {
 266                 nfs_set_pgio_error(data->header, -EIO, argp->offset);
 267                 return;
 268         }
 269         /* Yes, so retry the read at the end of the data */
 270         data->mds_offset += resp->count;
 271         argp->offset += resp->count;
 272         argp->pgbase += resp->count;
 273         argp->count -= resp->count;
 274         rpc_restart_call_prepare(task);
 275 }
 276
 277 static void nfs_readpage_result(struct rpc_task *task, struct nfs_pgio_data *data)
 278 {
 279         struct nfs_pgio_header *hdr = data->header;
 280
 281         if (data->res.eof) {
 282                 loff_t bound;
 283
 284                 bound = data->args.offset + data->res.count;
 285                 spin_lock(&hdr->lock);
 286                 if (bound < hdr->io_start + hdr->good_bytes) {
 287                         set_bit(NFS_IOHDR_EOF, &hdr->flags);
 288                         clear_bit(NFS_IOHDR_ERROR, &hdr->flags);
 289                         hdr->good_bytes = bound - hdr->io_start;
 290                 }
 291                 spin_unlock(&hdr->lock);
 292         } else if (data->res.count != data->args.count)
 293                 nfs_readpage_retry(task, data);
 294 }
 295
 296 /*
 297  * Read a page over NFS.
 298  * We read the page synchronously in the following case:
 299  *  -   The error flag is set for this page. This happens only when a
 300  *      previous async read operation failed.
 301  */
 302 int nfs_readpage(struct file *file, struct page *page)
 303 {
 304         struct nfs_open_context *ctx;
 305         struct inode *inode = page_file_mapping(page)->host;
 306         int             error;
 307
 308         dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
 309                 page, PAGE_CACHE_SIZE, page_file_index(page));
 310         nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
 311         nfs_add_stats(inode, NFSIOS_READPAGES, 1);
 312
 313         /*
 314          * Try to flush any pending writes to the file..
 315          *
 316          * NOTE! Because we own the page lock, there cannot
 317          * be any new pending writes generated at this point
 318          * for this page (other pages can be written to).
 319          */
 320         error = nfs_wb_page(inode, page);
 321         if (error)
 322                 goto out_unlock;
 323         if (PageUptodate(page))
 324                 goto out_unlock;
 325
 326         error = -ESTALE;
 327         if (NFS_STALE(inode))
 328                 goto out_unlock;
 329
 330         if (file == NULL) {
 331                 error = -EBADF;
 332                 ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
 333                 if (ctx == NULL)
 334                         goto out_unlock;
 335         } else
 336                 ctx = get_nfs_open_context(nfs_file_open_context(file));
 337
 338         if (!IS_SYNC(inode)) {
 339                 error = nfs_readpage_from_fscache(ctx, inode, page);
 340                 if (error == 0)
 341                         goto out;
 342         }
 343
 344         error = nfs_readpage_async(ctx, inode, page);
 345
 346 out:
 347         put_nfs_open_context(ctx);
 348         return error;
 349 out_unlock:
 350         unlock_page(page);
 351         return error;
 352 }
 353
 354 struct nfs_readdesc {
 355         struct nfs_pageio_descriptor *pgio;
 356         struct nfs_open_context *ctx;
 357 };
 358
 359 static int
 360 readpage_async_filler(void *data, struct page *page)
 361 {
 362         struct nfs_readdesc *desc = (struct nfs_readdesc *)data;
 363         struct inode *inode = page_file_mapping(page)->host;
 364         struct nfs_page *new;
 365         unsigned int len;
 366         int error;
 367
 368         len = nfs_page_length(page);
 369         if (len == 0)
 370                 return nfs_return_empty_page(page);
 371
 372         new = nfs_create_request(desc->ctx, inode, page, 0, len);
 373         if (IS_ERR(new))
 374                 goto out_error;
 375
 376         if (len < PAGE_CACHE_SIZE)
 377                 zero_user_segment(page, len, PAGE_CACHE_SIZE);
 378         if (!nfs_pageio_add_request(desc->pgio, new)) {
 379                 error = desc->pgio->pg_error;
 380                 goto out_unlock;
 381         }
 382         return 0;
 383 out_error:
 384         error = PTR_ERR(new);
 385 out_unlock:
 386         unlock_page(page);
 387         return error;
 388 }
 389
 390 int nfs_readpages(struct file *filp, struct address_space *mapping,
 391                 struct list_head *pages, unsigned nr_pages)
 392 {
 393         struct nfs_pageio_descriptor pgio;
 394         struct nfs_readdesc desc = {
 395                 .pgio = &pgio,
 396         };
 397         struct inode *inode = mapping->host;
 398         unsigned long npages;
 399         int ret = -ESTALE;
 400
 401         dprintk("NFS: nfs_readpages (%s/%Lu %d)\n",
 402                         inode->i_sb->s_id,
 403                         (unsigned long long)NFS_FILEID(inode),
 404                         nr_pages);
 405         nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
 406
 407         if (NFS_STALE(inode))
 408                 goto out;
 409
 410         if (filp == NULL) {
 411                 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
 412                 if (desc.ctx == NULL)
 413                         return -EBADF;
 414         } else
 415                 desc.ctx = get_nfs_open_context(nfs_file_open_context(filp));
 416
 417         /* attempt to read as many of the pages as possible from the cache
 418          * - this returns -ENOBUFS immediately if the cookie is negative
 419          */
 420         ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping,
 421                                          pages, &nr_pages);
 422         if (ret == 0)
 423                 goto read_complete; /* all pages were read */
 424
 425         nfs_pageio_init_read(&pgio, inode, false,
 426                              &nfs_async_read_completion_ops);
 427
 428         ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
 429
 430         nfs_pageio_complete(&pgio);
 431         NFS_I(inode)->read_io += pgio.pg_bytes_written;
 432         npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 433         nfs_add_stats(inode, NFSIOS_READPAGES, npages);
 434 read_complete:
 435         put_nfs_open_context(desc.ctx);
 436 out:
 437         return ret;
 438 }
 439
 440 int __init nfs_init_readpagecache(void)
 441 {
 442         nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
 443                                              sizeof(struct nfs_rw_header),
 444                                              0, SLAB_HWCACHE_ALIGN,
 445                                              NULL);
 446         if (nfs_rdata_cachep == NULL)
 447                 return -ENOMEM;
 448
 449         return 0;
 450 }
 451
 452 void nfs_destroy_readpagecache(void)
 453 {
 454         kmem_cache_destroy(nfs_rdata_cachep);
 455 }
 456
 457 static const struct nfs_rw_ops nfs_rw_read_ops = {
 458         .rw_mode                = FMODE_READ,
 459         .rw_alloc_header        = nfs_readhdr_alloc,
 460         .rw_free_header         = nfs_readhdr_free,
 461         .rw_done                = nfs_readpage_done,
 462         .rw_result              = nfs_readpage_result,
 463         .rw_initiate            = nfs_initiate_read,
 464 };