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