fs/afs/file.c

   1 /* AFS filesystem file handling
   2  *
   3  * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
   4  * Written by David Howells (dhowells@redhat.com)
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License
   8  * as published by the Free Software Foundation; either version
   9  * 2 of the License, or (at your option) any later version.
  10  */
  11
  12 #include <linux/kernel.h>
  13 #include <linux/module.h>
  14 #include <linux/init.h>
  15 #include <linux/fs.h>
  16 #include <linux/pagemap.h>
  17 #include <linux/writeback.h>
  18 #include <linux/gfp.h>
  19 #include <linux/task_io_accounting_ops.h>
  20 #include "internal.h"
  21
  22 static int afs_file_mmap(struct file *file, struct vm_area_struct *vma);
  23 static int afs_readpage(struct file *file, struct page *page);
  24 static void afs_invalidatepage(struct page *page, unsigned int offset,
  25                                unsigned int length);
  26 static int afs_releasepage(struct page *page, gfp_t gfp_flags);
  27
  28 static int afs_readpages(struct file *filp, struct address_space *mapping,
  29                          struct list_head *pages, unsigned nr_pages);
  30
  31 const struct file_operations afs_file_operations = {
  32         .open           = afs_open,
  33         .release        = afs_release,
  34         .llseek         = generic_file_llseek,
  35         .read_iter      = generic_file_read_iter,
  36         .write_iter     = afs_file_write,
  37         .mmap           = afs_file_mmap,
  38         .splice_read    = generic_file_splice_read,
  39         .fsync          = afs_fsync,
  40         .lock           = afs_lock,
  41         .flock          = afs_flock,
  42 };
  43
  44 const struct inode_operations afs_file_inode_operations = {
  45         .getattr        = afs_getattr,
  46         .setattr        = afs_setattr,
  47         .permission     = afs_permission,
  48         .listxattr      = afs_listxattr,
  49 };
  50
  51 const struct address_space_operations afs_fs_aops = {
  52         .readpage       = afs_readpage,
  53         .readpages      = afs_readpages,
  54         .set_page_dirty = afs_set_page_dirty,
  55         .launder_page   = afs_launder_page,
  56         .releasepage    = afs_releasepage,
  57         .invalidatepage = afs_invalidatepage,
  58         .write_begin    = afs_write_begin,
  59         .write_end      = afs_write_end,
  60         .writepage      = afs_writepage,
  61         .writepages     = afs_writepages,
  62 };
  63
  64 static const struct vm_operations_struct afs_vm_ops = {
  65         .fault          = filemap_fault,
  66         .map_pages      = filemap_map_pages,
  67         .page_mkwrite   = afs_page_mkwrite,
  68 };
  69
  70 /*
  71  * Discard a pin on a writeback key.
  72  */
  73 void afs_put_wb_key(struct afs_wb_key *wbk)
  74 {
  75         if (refcount_dec_and_test(&wbk->usage)) {
  76                 key_put(wbk->key);
  77                 kfree(wbk);
  78         }
  79 }
  80
  81 /*
  82  * Cache key for writeback.
  83  */
  84 int afs_cache_wb_key(struct afs_vnode *vnode, struct afs_file *af)
  85 {
  86         struct afs_wb_key *wbk, *p;
  87
  88         wbk = kzalloc(sizeof(struct afs_wb_key), GFP_KERNEL);
  89         if (!wbk)
  90                 return -ENOMEM;
  91         refcount_set(&wbk->usage, 2);
  92         wbk->key = af->key;
  93
  94         spin_lock(&vnode->wb_lock);
  95         list_for_each_entry(p, &vnode->wb_keys, vnode_link) {
  96                 if (p->key == wbk->key)
  97                         goto found;
  98         }
  99
 100         key_get(wbk->key);
 101         list_add_tail(&wbk->vnode_link, &vnode->wb_keys);
 102         spin_unlock(&vnode->wb_lock);
 103         af->wb = wbk;
 104         return 0;
 105
 106 found:
 107         refcount_inc(&p->usage);
 108         spin_unlock(&vnode->wb_lock);
 109         af->wb = p;
 110         kfree(wbk);
 111         return 0;
 112 }
 113
 114 /*
 115  * open an AFS file or directory and attach a key to it
 116  */
 117 int afs_open(struct inode *inode, struct file *file)
 118 {
 119         struct afs_vnode *vnode = AFS_FS_I(inode);
 120         struct afs_file *af;
 121         struct key *key;
 122         int ret;
 123
 124         _enter("{%x:%u},", vnode->fid.vid, vnode->fid.vnode);
 125
 126         key = afs_request_key(vnode->volume->cell);
 127         if (IS_ERR(key)) {
 128                 ret = PTR_ERR(key);
 129                 goto error;
 130         }
 131
 132         af = kzalloc(sizeof(*af), GFP_KERNEL);
 133         if (!af) {
 134                 ret = -ENOMEM;
 135                 goto error_key;
 136         }
 137         af->key = key;
 138
 139         ret = afs_validate(vnode, key);
 140         if (ret < 0)
 141                 goto error_af;
 142
 143         if (file->f_mode & FMODE_WRITE) {
 144                 ret = afs_cache_wb_key(vnode, af);
 145                 if (ret < 0)
 146                         goto error_af;
 147         }
 148
 149         if (file->f_flags & O_TRUNC)
 150                 set_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags);
 151
 152         file->private_data = af;
 153         _leave(" = 0");
 154         return 0;
 155
 156 error_af:
 157         kfree(af);
 158 error_key:
 159         key_put(key);
 160 error:
 161         _leave(" = %d", ret);
 162         return ret;
 163 }
 164
 165 /*
 166  * release an AFS file or directory and discard its key
 167  */
 168 int afs_release(struct inode *inode, struct file *file)
 169 {
 170         struct afs_vnode *vnode = AFS_FS_I(inode);
 171         struct afs_file *af = file->private_data;
 172
 173         _enter("{%x:%u},", vnode->fid.vid, vnode->fid.vnode);
 174
 175         if ((file->f_mode & FMODE_WRITE))
 176                 return vfs_fsync(file, 0);
 177
 178         file->private_data = NULL;
 179         if (af->wb)
 180                 afs_put_wb_key(af->wb);
 181         key_put(af->key);
 182         kfree(af);
 183         afs_prune_wb_keys(vnode);
 184         _leave(" = 0");
 185         return 0;
 186 }
 187
 188 /*
 189  * Dispose of a ref to a read record.
 190  */
 191 void afs_put_read(struct afs_read *req)
 192 {
 193         int i;
 194
 195         if (refcount_dec_and_test(&req->usage)) {
 196                 for (i = 0; i < req->nr_pages; i++)
 197                         if (req->pages[i])
 198                                 put_page(req->pages[i]);
 199                 if (req->pages != req->array)
 200                         kfree(req->pages);
 201                 kfree(req);
 202         }
 203 }
 204
 205 #ifdef CONFIG_AFS_FSCACHE
 206 /*
 207  * deal with notification that a page was read from the cache
 208  */
 209 static void afs_file_readpage_read_complete(struct page *page,
 210                                             void *data,
 211                                             int error)
 212 {
 213         _enter("%p,%p,%d", page, data, error);
 214
 215         /* if the read completes with an error, we just unlock the page and let
 216          * the VM reissue the readpage */
 217         if (!error)
 218                 SetPageUptodate(page);
 219         unlock_page(page);
 220 }
 221 #endif
 222
 223 /*
 224  * Fetch file data from the volume.
 225  */
 226 int afs_fetch_data(struct afs_vnode *vnode, struct key *key, struct afs_read *desc)
 227 {
 228         struct afs_fs_cursor fc;
 229         int ret;
 230
 231         _enter("%s{%x:%u.%u},%x,,,",
 232                vnode->volume->name,
 233                vnode->fid.vid,
 234                vnode->fid.vnode,
 235                vnode->fid.unique,
 236                key_serial(key));
 237
 238         ret = -ERESTARTSYS;
 239         if (afs_begin_vnode_operation(&fc, vnode, key)) {
 240                 while (afs_select_fileserver(&fc)) {
 241                         fc.cb_break = afs_calc_vnode_cb_break(vnode);
 242                         afs_fs_fetch_data(&fc, desc);
 243                 }
 244
 245                 afs_check_for_remote_deletion(&fc, fc.vnode);
 246                 afs_vnode_commit_status(&fc, vnode, fc.cb_break);
 247                 ret = afs_end_vnode_operation(&fc);
 248         }
 249
 250         if (ret == 0) {
 251                 afs_stat_v(vnode, n_fetches);
 252                 atomic_long_add(desc->actual_len,
 253                                 &afs_v2net(vnode)->n_fetch_bytes);
 254         }
 255
 256         _leave(" = %d", ret);
 257         return ret;
 258 }
 259
 260 /*
 261  * read page from file, directory or symlink, given a key to use
 262  */
 263 int afs_page_filler(void *data, struct page *page)
 264 {
 265         struct inode *inode = page->mapping->host;
 266         struct afs_vnode *vnode = AFS_FS_I(inode);
 267         struct afs_read *req;
 268         struct key *key = data;
 269         int ret;
 270
 271         _enter("{%x},{%lu},{%lu}", key_serial(key), inode->i_ino, page->index);
 272
 273         BUG_ON(!PageLocked(page));
 274
 275         ret = -ESTALE;
 276         if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
 277                 goto error;
 278
 279         /* is it cached? */
 280 #ifdef CONFIG_AFS_FSCACHE
 281         ret = fscache_read_or_alloc_page(vnode->cache,
 282                                          page,
 283                                          afs_file_readpage_read_complete,
 284                                          NULL,
 285                                          GFP_KERNEL);
 286 #else
 287         ret = -ENOBUFS;
 288 #endif
 289         switch (ret) {
 290                 /* read BIO submitted (page in cache) */
 291         case 0:
 292                 break;
 293
 294                 /* page not yet cached */
 295         case -ENODATA:
 296                 _debug("cache said ENODATA");
 297                 goto go_on;
 298
 299                 /* page will not be cached */
 300         case -ENOBUFS:
 301                 _debug("cache said ENOBUFS");
 302         default:
 303         go_on:
 304                 req = kzalloc(sizeof(struct afs_read) + sizeof(struct page *),
 305                               GFP_KERNEL);
 306                 if (!req)
 307                         goto enomem;
 308
 309                 /* We request a full page.  If the page is a partial one at the
 310                  * end of the file, the server will return a short read and the
 311                  * unmarshalling code will clear the unfilled space.
 312                  */
 313                 refcount_set(&req->usage, 1);
 314                 req->pos = (loff_t)page->index << PAGE_SHIFT;
 315                 req->len = PAGE_SIZE;
 316                 req->nr_pages = 1;
 317                 req->pages = req->array;
 318                 req->pages[0] = page;
 319                 get_page(page);
 320
 321                 /* read the contents of the file from the server into the
 322                  * page */
 323                 ret = afs_fetch_data(vnode, key, req);
 324                 afs_put_read(req);
 325
 326                 if (ret < 0) {
 327                         if (ret == -ENOENT) {
 328                                 _debug("got NOENT from server"
 329                                        " - marking file deleted and stale");
 330                                 set_bit(AFS_VNODE_DELETED, &vnode->flags);
 331                                 ret = -ESTALE;
 332                         }
 333
 334 #ifdef CONFIG_AFS_FSCACHE
 335                         fscache_uncache_page(vnode->cache, page);
 336 #endif
 337                         BUG_ON(PageFsCache(page));
 338
 339                         if (ret == -EINTR ||
 340                             ret == -ENOMEM ||
 341                             ret == -ERESTARTSYS ||
 342                             ret == -EAGAIN)
 343                                 goto error;
 344                         goto io_error;
 345                 }
 346
 347                 SetPageUptodate(page);
 348
 349                 /* send the page to the cache */
 350 #ifdef CONFIG_AFS_FSCACHE
 351                 if (PageFsCache(page) &&
 352                     fscache_write_page(vnode->cache, page, vnode->status.size,
 353                                        GFP_KERNEL) != 0) {
 354                         fscache_uncache_page(vnode->cache, page);
 355                         BUG_ON(PageFsCache(page));
 356                 }
 357 #endif
 358                 unlock_page(page);
 359         }
 360
 361         _leave(" = 0");
 362         return 0;
 363
 364 io_error:
 365         SetPageError(page);
 366         goto error;
 367 enomem:
 368         ret = -ENOMEM;
 369 error:
 370         unlock_page(page);
 371         _leave(" = %d", ret);
 372         return ret;
 373 }
 374
 375 /*
 376  * read page from file, directory or symlink, given a file to nominate the key
 377  * to be used
 378  */
 379 static int afs_readpage(struct file *file, struct page *page)
 380 {
 381         struct key *key;
 382         int ret;
 383
 384         if (file) {
 385                 key = afs_file_key(file);
 386                 ASSERT(key != NULL);
 387                 ret = afs_page_filler(key, page);
 388         } else {
 389                 struct inode *inode = page->mapping->host;
 390                 key = afs_request_key(AFS_FS_S(inode->i_sb)->cell);
 391                 if (IS_ERR(key)) {
 392                         ret = PTR_ERR(key);
 393                 } else {
 394                         ret = afs_page_filler(key, page);
 395                         key_put(key);
 396                 }
 397         }
 398         return ret;
 399 }
 400
 401 /*
 402  * Make pages available as they're filled.
 403  */
 404 static void afs_readpages_page_done(struct afs_call *call, struct afs_read *req)
 405 {
 406 #ifdef CONFIG_AFS_FSCACHE
 407         struct afs_vnode *vnode = call->reply[0];
 408 #endif
 409         struct page *page = req->pages[req->index];
 410
 411         req->pages[req->index] = NULL;
 412         SetPageUptodate(page);
 413
 414         /* send the page to the cache */
 415 #ifdef CONFIG_AFS_FSCACHE
 416         if (PageFsCache(page) &&
 417             fscache_write_page(vnode->cache, page, vnode->status.size,
 418                                GFP_KERNEL) != 0) {
 419                 fscache_uncache_page(vnode->cache, page);
 420                 BUG_ON(PageFsCache(page));
 421         }
 422 #endif
 423         unlock_page(page);
 424         put_page(page);
 425 }
 426
 427 /*
 428  * Read a contiguous set of pages.
 429  */
 430 static int afs_readpages_one(struct file *file, struct address_space *mapping,
 431                              struct list_head *pages)
 432 {
 433         struct afs_vnode *vnode = AFS_FS_I(mapping->host);
 434         struct afs_read *req;
 435         struct list_head *p;
 436         struct page *first, *page;
 437         struct key *key = afs_file_key(file);
 438         pgoff_t index;
 439         int ret, n, i;
 440
 441         /* Count the number of contiguous pages at the front of the list.  Note
 442          * that the list goes prev-wards rather than next-wards.
 443          */
 444         first = list_entry(pages->prev, struct page, lru);
 445         index = first->index + 1;
 446         n = 1;
 447         for (p = first->lru.prev; p != pages; p = p->prev) {
 448                 page = list_entry(p, struct page, lru);
 449                 if (page->index != index)
 450                         break;
 451                 index++;
 452                 n++;
 453         }
 454
 455         req = kzalloc(sizeof(struct afs_read) + sizeof(struct page *) * n,
 456                       GFP_NOFS);
 457         if (!req)
 458                 return -ENOMEM;
 459
 460         refcount_set(&req->usage, 1);
 461         req->page_done = afs_readpages_page_done;
 462         req->pos = first->index;
 463         req->pos <<= PAGE_SHIFT;
 464         req->pages = req->array;
 465
 466         /* Transfer the pages to the request.  We add them in until one fails
 467          * to add to the LRU and then we stop (as that'll make a hole in the
 468          * contiguous run.
 469          *
 470          * Note that it's possible for the file size to change whilst we're
 471          * doing this, but we rely on the server returning less than we asked
 472          * for if the file shrank.  We also rely on this to deal with a partial
 473          * page at the end of the file.
 474          */
 475         do {
 476                 page = list_entry(pages->prev, struct page, lru);
 477                 list_del(&page->lru);
 478                 index = page->index;
 479                 if (add_to_page_cache_lru(page, mapping, index,
 480                                           readahead_gfp_mask(mapping))) {
 481 #ifdef CONFIG_AFS_FSCACHE
 482                         fscache_uncache_page(vnode->cache, page);
 483 #endif
 484                         put_page(page);
 485                         break;
 486                 }
 487
 488                 req->pages[req->nr_pages++] = page;
 489                 req->len += PAGE_SIZE;
 490         } while (req->nr_pages < n);
 491
 492         if (req->nr_pages == 0) {
 493                 kfree(req);
 494                 return 0;
 495         }
 496
 497         ret = afs_fetch_data(vnode, key, req);
 498         if (ret < 0)
 499                 goto error;
 500
 501         task_io_account_read(PAGE_SIZE * req->nr_pages);
 502         afs_put_read(req);
 503         return 0;
 504
 505 error:
 506         if (ret == -ENOENT) {
 507                 _debug("got NOENT from server"
 508                        " - marking file deleted and stale");
 509                 set_bit(AFS_VNODE_DELETED, &vnode->flags);
 510                 ret = -ESTALE;
 511         }
 512
 513         for (i = 0; i < req->nr_pages; i++) {
 514                 page = req->pages[i];
 515                 if (page) {
 516 #ifdef CONFIG_AFS_FSCACHE
 517                         fscache_uncache_page(vnode->cache, page);
 518 #endif
 519                         SetPageError(page);
 520                         unlock_page(page);
 521                 }
 522         }
 523
 524         afs_put_read(req);
 525         return ret;
 526 }
 527
 528 /*
 529  * read a set of pages
 530  */
 531 static int afs_readpages(struct file *file, struct address_space *mapping,
 532                          struct list_head *pages, unsigned nr_pages)
 533 {
 534         struct key *key = afs_file_key(file);
 535         struct afs_vnode *vnode;
 536         int ret = 0;
 537
 538         _enter("{%d},{%lu},,%d",
 539                key_serial(key), mapping->host->i_ino, nr_pages);
 540
 541         ASSERT(key != NULL);
 542
 543         vnode = AFS_FS_I(mapping->host);
 544         if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
 545                 _leave(" = -ESTALE");
 546                 return -ESTALE;
 547         }
 548
 549         /* attempt to read as many of the pages as possible */
 550 #ifdef CONFIG_AFS_FSCACHE
 551         ret = fscache_read_or_alloc_pages(vnode->cache,
 552                                           mapping,
 553                                           pages,
 554                                           &nr_pages,
 555                                           afs_file_readpage_read_complete,
 556                                           NULL,
 557                                           mapping_gfp_mask(mapping));
 558 #else
 559         ret = -ENOBUFS;
 560 #endif
 561
 562         switch (ret) {
 563                 /* all pages are being read from the cache */
 564         case 0:
 565                 BUG_ON(!list_empty(pages));
 566                 BUG_ON(nr_pages != 0);
 567                 _leave(" = 0 [reading all]");
 568                 return 0;
 569
 570                 /* there were pages that couldn't be read from the cache */
 571         case -ENODATA:
 572         case -ENOBUFS:
 573                 break;
 574
 575                 /* other error */
 576         default:
 577                 _leave(" = %d", ret);
 578                 return ret;
 579         }
 580
 581         while (!list_empty(pages)) {
 582                 ret = afs_readpages_one(file, mapping, pages);
 583                 if (ret < 0)
 584                         break;
 585         }
 586
 587         _leave(" = %d [netting]", ret);
 588         return ret;
 589 }
 590
 591 /*
 592  * invalidate part or all of a page
 593  * - release a page and clean up its private data if offset is 0 (indicating
 594  *   the entire page)
 595  */
 596 static void afs_invalidatepage(struct page *page, unsigned int offset,
 597                                unsigned int length)
 598 {
 599         struct afs_vnode *vnode = AFS_FS_I(page->mapping->host);
 600         unsigned long priv;
 601
 602         _enter("{%lu},%u,%u", page->index, offset, length);
 603
 604         BUG_ON(!PageLocked(page));
 605
 606         /* we clean up only if the entire page is being invalidated */
 607         if (offset == 0 && length == PAGE_SIZE) {
 608 #ifdef CONFIG_AFS_FSCACHE
 609                 if (PageFsCache(page)) {
 610                         struct afs_vnode *vnode = AFS_FS_I(page->mapping->host);
 611                         fscache_wait_on_page_write(vnode->cache, page);
 612                         fscache_uncache_page(vnode->cache, page);
 613                 }
 614 #endif
 615
 616                 if (PagePrivate(page)) {
 617                         priv = page_private(page);
 618                         trace_afs_page_dirty(vnode, tracepoint_string("inval"),
 619                                              page->index, priv);
 620                         set_page_private(page, 0);
 621                         ClearPagePrivate(page);
 622                 }
 623         }
 624
 625         _leave("");
 626 }
 627
 628 /*
 629  * release a page and clean up its private state if it's not busy
 630  * - return true if the page can now be released, false if not
 631  */
 632 static int afs_releasepage(struct page *page, gfp_t gfp_flags)
 633 {
 634         struct afs_vnode *vnode = AFS_FS_I(page->mapping->host);
 635         unsigned long priv;
 636
 637         _enter("{{%x:%u}[%lu],%lx},%x",
 638                vnode->fid.vid, vnode->fid.vnode, page->index, page->flags,
 639                gfp_flags);
 640
 641         /* deny if page is being written to the cache and the caller hasn't
 642          * elected to wait */
 643 #ifdef CONFIG_AFS_FSCACHE
 644         if (!fscache_maybe_release_page(vnode->cache, page, gfp_flags)) {
 645                 _leave(" = F [cache busy]");
 646                 return 0;
 647         }
 648 #endif
 649
 650         if (PagePrivate(page)) {
 651                 priv = page_private(page);
 652                 trace_afs_page_dirty(vnode, tracepoint_string("rel"),
 653                                      page->index, priv);
 654                 set_page_private(page, 0);
 655                 ClearPagePrivate(page);
 656         }
 657
 658         /* indicate that the page can be released */
 659         _leave(" = T");
 660         return 1;
 661 }
 662
 663 /*
 664  * Handle setting up a memory mapping on an AFS file.
 665  */
 666 static int afs_file_mmap(struct file *file, struct vm_area_struct *vma)
 667 {
 668         int ret;
 669
 670         ret = generic_file_mmap(file, vma);
 671         if (ret == 0)
 672                 vma->vm_ops = &afs_vm_ops;
 673         return ret;
 674 }