fs/gfs2/aops.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/pagemap.h>
  16 #include <linux/pagevec.h>
  17 #include <linux/mpage.h>
  18 #include <linux/fs.h>
  19 #include <linux/writeback.h>
  20 #include <linux/swap.h>
  21 #include <linux/gfs2_ondisk.h>
  22 #include <linux/backing-dev.h>
  23 #include <linux/aio.h>
  24
  25 #include "gfs2.h"
  26 #include "incore.h"
  27 #include "bmap.h"
  28 #include "glock.h"
  29 #include "inode.h"
  30 #include "log.h"
  31 #include "meta_io.h"
  32 #include "quota.h"
  33 #include "trans.h"
  34 #include "rgrp.h"
  35 #include "super.h"
  36 #include "util.h"
  37 #include "glops.h"
  38
  39
  40 static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
  41                                    unsigned int from, unsigned int to)
  42 {
  43         struct buffer_head *head = page_buffers(page);
  44         unsigned int bsize = head->b_size;
  45         struct buffer_head *bh;
  46         unsigned int start, end;
  47
  48         for (bh = head, start = 0; bh != head || !start;
  49              bh = bh->b_this_page, start = end) {
  50                 end = start + bsize;
  51                 if (end <= from || start >= to)
  52                         continue;
  53                 if (gfs2_is_jdata(ip))
  54                         set_buffer_uptodate(bh);
  55                 gfs2_trans_add_data(ip->i_gl, bh);
  56         }
  57 }
  58
  59 /**
  60  * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
  61  * @inode: The inode
  62  * @lblock: The block number to look up
  63  * @bh_result: The buffer head to return the result in
  64  * @create: Non-zero if we may add block to the file
  65  *
  66  * Returns: errno
  67  */
  68
  69 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
  70                                   struct buffer_head *bh_result, int create)
  71 {
  72         int error;
  73
  74         error = gfs2_block_map(inode, lblock, bh_result, 0);
  75         if (error)
  76                 return error;
  77         if (!buffer_mapped(bh_result))
  78                 return -EIO;
  79         return 0;
  80 }
  81
  82 static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
  83                                  struct buffer_head *bh_result, int create)
  84 {
  85         return gfs2_block_map(inode, lblock, bh_result, 0);
  86 }
  87
  88 /**
  89  * gfs2_writepage_common - Common bits of writepage
  90  * @page: The page to be written
  91  * @wbc: The writeback control
  92  *
  93  * Returns: 1 if writepage is ok, otherwise an error code or zero if no error.
  94  */
  95
  96 static int gfs2_writepage_common(struct page *page,
  97                                  struct writeback_control *wbc)
  98 {
  99         struct inode *inode = page->mapping->host;
 100         struct gfs2_inode *ip = GFS2_I(inode);
 101         struct gfs2_sbd *sdp = GFS2_SB(inode);
 102         loff_t i_size = i_size_read(inode);
 103         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
 104         unsigned offset;
 105
 106         if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
 107                 goto out;
 108         if (current->journal_info)
 109                 goto redirty;
 110         /* Is the page fully outside i_size? (truncate in progress) */
 111         offset = i_size & (PAGE_CACHE_SIZE-1);
 112         if (page->index > end_index || (page->index == end_index && !offset)) {
 113                 page->mapping->a_ops->invalidatepage(page, 0);
 114                 goto out;
 115         }
 116         return 1;
 117 redirty:
 118         redirty_page_for_writepage(wbc, page);
 119 out:
 120         unlock_page(page);
 121         return 0;
 122 }
 123
 124 /**
 125  * gfs2_writeback_writepage - Write page for writeback mappings
 126  * @page: The page
 127  * @wbc: The writeback control
 128  *
 129  */
 130
 131 static int gfs2_writeback_writepage(struct page *page,
 132                                     struct writeback_control *wbc)
 133 {
 134         int ret;
 135
 136         ret = gfs2_writepage_common(page, wbc);
 137         if (ret <= 0)
 138                 return ret;
 139
 140         return nobh_writepage(page, gfs2_get_block_noalloc, wbc);
 141 }
 142
 143 /**
 144  * gfs2_ordered_writepage - Write page for ordered data files
 145  * @page: The page to write
 146  * @wbc: The writeback control
 147  *
 148  */
 149
 150 static int gfs2_ordered_writepage(struct page *page,
 151                                   struct writeback_control *wbc)
 152 {
 153         struct inode *inode = page->mapping->host;
 154         struct gfs2_inode *ip = GFS2_I(inode);
 155         int ret;
 156
 157         ret = gfs2_writepage_common(page, wbc);
 158         if (ret <= 0)
 159                 return ret;
 160
 161         if (!page_has_buffers(page)) {
 162                 create_empty_buffers(page, inode->i_sb->s_blocksize,
 163                                      (1 << BH_Dirty)|(1 << BH_Uptodate));
 164         }
 165         gfs2_page_add_databufs(ip, page, 0, inode->i_sb->s_blocksize-1);
 166         return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
 167 }
 168
 169 /**
 170  * __gfs2_jdata_writepage - The core of jdata writepage
 171  * @page: The page to write
 172  * @wbc: The writeback control
 173  *
 174  * This is shared between writepage and writepages and implements the
 175  * core of the writepage operation. If a transaction is required then
 176  * PageChecked will have been set and the transaction will have
 177  * already been started before this is called.
 178  */
 179
 180 static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
 181 {
 182         struct inode *inode = page->mapping->host;
 183         struct gfs2_inode *ip = GFS2_I(inode);
 184         struct gfs2_sbd *sdp = GFS2_SB(inode);
 185
 186         if (PageChecked(page)) {
 187                 ClearPageChecked(page);
 188                 if (!page_has_buffers(page)) {
 189                         create_empty_buffers(page, inode->i_sb->s_blocksize,
 190                                              (1 << BH_Dirty)|(1 << BH_Uptodate));
 191                 }
 192                 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
 193         }
 194         return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
 195 }
 196
 197 /**
 198  * gfs2_jdata_writepage - Write complete page
 199  * @page: Page to write
 200  *
 201  * Returns: errno
 202  *
 203  */
 204
 205 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
 206 {
 207         struct inode *inode = page->mapping->host;
 208         struct gfs2_sbd *sdp = GFS2_SB(inode);
 209         int ret;
 210         int done_trans = 0;
 211
 212         if (PageChecked(page)) {
 213                 if (wbc->sync_mode != WB_SYNC_ALL)
 214                         goto out_ignore;
 215                 ret = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
 216                 if (ret)
 217                         goto out_ignore;
 218                 done_trans = 1;
 219         }
 220         ret = gfs2_writepage_common(page, wbc);
 221         if (ret > 0)
 222                 ret = __gfs2_jdata_writepage(page, wbc);
 223         if (done_trans)
 224                 gfs2_trans_end(sdp);
 225         return ret;
 226
 227 out_ignore:
 228         redirty_page_for_writepage(wbc, page);
 229         unlock_page(page);
 230         return 0;
 231 }
 232
 233 /**
 234  * gfs2_writepages - Write a bunch of dirty pages back to disk
 235  * @mapping: The mapping to write
 236  * @wbc: Write-back control
 237  *
 238  * Used for both ordered and writeback modes.
 239  */
 240 static int gfs2_writepages(struct address_space *mapping,
 241                            struct writeback_control *wbc)
 242 {
 243         return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
 244 }
 245
 246 /**
 247  * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
 248  * @mapping: The mapping
 249  * @wbc: The writeback control
 250  * @writepage: The writepage function to call for each page
 251  * @pvec: The vector of pages
 252  * @nr_pages: The number of pages to write
 253  *
 254  * Returns: non-zero if loop should terminate, zero otherwise
 255  */
 256
 257 static int gfs2_write_jdata_pagevec(struct address_space *mapping,
 258                                     struct writeback_control *wbc,
 259                                     struct pagevec *pvec,
 260                                     int nr_pages, pgoff_t end)
 261 {
 262         struct inode *inode = mapping->host;
 263         struct gfs2_sbd *sdp = GFS2_SB(inode);
 264         loff_t i_size = i_size_read(inode);
 265         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
 266         unsigned offset = i_size & (PAGE_CACHE_SIZE-1);
 267         unsigned nrblocks = nr_pages * (PAGE_CACHE_SIZE/inode->i_sb->s_blocksize);
 268         int i;
 269         int ret;
 270
 271         ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
 272         if (ret < 0)
 273                 return ret;
 274
 275         for(i = 0; i < nr_pages; i++) {
 276                 struct page *page = pvec->pages[i];
 277
 278                 lock_page(page);
 279
 280                 if (unlikely(page->mapping != mapping)) {
 281                         unlock_page(page);
 282                         continue;
 283                 }
 284
 285                 if (!wbc->range_cyclic && page->index > end) {
 286                         ret = 1;
 287                         unlock_page(page);
 288                         continue;
 289                 }
 290
 291                 if (wbc->sync_mode != WB_SYNC_NONE)
 292                         wait_on_page_writeback(page);
 293
 294                 if (PageWriteback(page) ||
 295                     !clear_page_dirty_for_io(page)) {
 296                         unlock_page(page);
 297                         continue;
 298                 }
 299
 300                 /* Is the page fully outside i_size? (truncate in progress) */
 301                 if (page->index > end_index || (page->index == end_index && !offset)) {
 302                         page->mapping->a_ops->invalidatepage(page, 0);
 303                         unlock_page(page);
 304                         continue;
 305                 }
 306
 307                 ret = __gfs2_jdata_writepage(page, wbc);
 308
 309                 if (ret || (--(wbc->nr_to_write) <= 0))
 310                         ret = 1;
 311         }
 312         gfs2_trans_end(sdp);
 313         return ret;
 314 }
 315
 316 /**
 317  * gfs2_write_cache_jdata - Like write_cache_pages but different
 318  * @mapping: The mapping to write
 319  * @wbc: The writeback control
 320  * @writepage: The writepage function to call
 321  * @data: The data to pass to writepage
 322  *
 323  * The reason that we use our own function here is that we need to
 324  * start transactions before we grab page locks. This allows us
 325  * to get the ordering right.
 326  */
 327
 328 static int gfs2_write_cache_jdata(struct address_space *mapping,
 329                                   struct writeback_control *wbc)
 330 {
 331         int ret = 0;
 332         int done = 0;
 333         struct pagevec pvec;
 334         int nr_pages;
 335         pgoff_t index;
 336         pgoff_t end;
 337         int scanned = 0;
 338         int range_whole = 0;
 339
 340         pagevec_init(&pvec, 0);
 341         if (wbc->range_cyclic) {
 342                 index = mapping->writeback_index; /* Start from prev offset */
 343                 end = -1;
 344         } else {
 345                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
 346                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
 347                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
 348                         range_whole = 1;
 349                 scanned = 1;
 350         }
 351
 352 retry:
 353          while (!done && (index <= end) &&
 354                 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 355                                                PAGECACHE_TAG_DIRTY,
 356                                                min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
 357                 scanned = 1;
 358                 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end);
 359                 if (ret)
 360                         done = 1;
 361                 if (ret > 0)
 362                         ret = 0;
 363
 364                 pagevec_release(&pvec);
 365                 cond_resched();
 366         }
 367
 368         if (!scanned && !done) {
 369                 /*
 370                  * We hit the last page and there is more work to be done: wrap
 371                  * back to the start of the file
 372                  */
 373                 scanned = 1;
 374                 index = 0;
 375                 goto retry;
 376         }
 377
 378         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
 379                 mapping->writeback_index = index;
 380         return ret;
 381 }
 382
 383
 384 /**
 385  * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
 386  * @mapping: The mapping to write
 387  * @wbc: The writeback control
 388  *
 389  */
 390
 391 static int gfs2_jdata_writepages(struct address_space *mapping,
 392                                  struct writeback_control *wbc)
 393 {
 394         struct gfs2_inode *ip = GFS2_I(mapping->host);
 395         struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
 396         int ret;
 397
 398         ret = gfs2_write_cache_jdata(mapping, wbc);
 399         if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
 400                 gfs2_log_flush(sdp, ip->i_gl);
 401                 ret = gfs2_write_cache_jdata(mapping, wbc);
 402         }
 403         return ret;
 404 }
 405
 406 /**
 407  * stuffed_readpage - Fill in a Linux page with stuffed file data
 408  * @ip: the inode
 409  * @page: the page
 410  *
 411  * Returns: errno
 412  */
 413
 414 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
 415 {
 416         struct buffer_head *dibh;
 417         u64 dsize = i_size_read(&ip->i_inode);
 418         void *kaddr;
 419         int error;
 420
 421         /*
 422          * Due to the order of unstuffing files and ->fault(), we can be
 423          * asked for a zero page in the case of a stuffed file being extended,
 424          * so we need to supply one here. It doesn't happen often.
 425          */
 426         if (unlikely(page->index)) {
 427                 zero_user(page, 0, PAGE_CACHE_SIZE);
 428                 SetPageUptodate(page);
 429                 return 0;
 430         }
 431
 432         error = gfs2_meta_inode_buffer(ip, &dibh);
 433         if (error)
 434                 return error;
 435
 436         kaddr = kmap_atomic(page);
 437         if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
 438                 dsize = (dibh->b_size - sizeof(struct gfs2_dinode));
 439         memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
 440         memset(kaddr + dsize, 0, PAGE_CACHE_SIZE - dsize);
 441         kunmap_atomic(kaddr);
 442         flush_dcache_page(page);
 443         brelse(dibh);
 444         SetPageUptodate(page);
 445
 446         return 0;
 447 }
 448
 449
 450 /**
 451  * __gfs2_readpage - readpage
 452  * @file: The file to read a page for
 453  * @page: The page to read
 454  *
 455  * This is the core of gfs2's readpage. Its used by the internal file
 456  * reading code as in that case we already hold the glock. Also its
 457  * called by gfs2_readpage() once the required lock has been granted.
 458  *
 459  */
 460
 461 static int __gfs2_readpage(void *file, struct page *page)
 462 {
 463         struct gfs2_inode *ip = GFS2_I(page->mapping->host);
 464         struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
 465         int error;
 466
 467         if (gfs2_is_stuffed(ip)) {
 468                 error = stuffed_readpage(ip, page);
 469                 unlock_page(page);
 470         } else {
 471                 error = mpage_readpage(page, gfs2_block_map);
 472         }
 473
 474         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 475                 return -EIO;
 476
 477         return error;
 478 }
 479
 480 /**
 481  * gfs2_readpage - read a page of a file
 482  * @file: The file to read
 483  * @page: The page of the file
 484  *
 485  * This deals with the locking required. We have to unlock and
 486  * relock the page in order to get the locking in the right
 487  * order.
 488  */
 489
 490 static int gfs2_readpage(struct file *file, struct page *page)
 491 {
 492         struct address_space *mapping = page->mapping;
 493         struct gfs2_inode *ip = GFS2_I(mapping->host);
 494         struct gfs2_holder gh;
 495         int error;
 496
 497         unlock_page(page);
 498         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 499         error = gfs2_glock_nq(&gh);
 500         if (unlikely(error))
 501                 goto out;
 502         error = AOP_TRUNCATED_PAGE;
 503         lock_page(page);
 504         if (page->mapping == mapping && !PageUptodate(page))
 505                 error = __gfs2_readpage(file, page);
 506         else
 507                 unlock_page(page);
 508         gfs2_glock_dq(&gh);
 509 out:
 510         gfs2_holder_uninit(&gh);
 511         if (error && error != AOP_TRUNCATED_PAGE)
 512                 lock_page(page);
 513         return error;
 514 }
 515
 516 /**
 517  * gfs2_internal_read - read an internal file
 518  * @ip: The gfs2 inode
 519  * @buf: The buffer to fill
 520  * @pos: The file position
 521  * @size: The amount to read
 522  *
 523  */
 524
 525 int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
 526                        unsigned size)
 527 {
 528         struct address_space *mapping = ip->i_inode.i_mapping;
 529         unsigned long index = *pos / PAGE_CACHE_SIZE;
 530         unsigned offset = *pos & (PAGE_CACHE_SIZE - 1);
 531         unsigned copied = 0;
 532         unsigned amt;
 533         struct page *page;
 534         void *p;
 535
 536         do {
 537                 amt = size - copied;
 538                 if (offset + size > PAGE_CACHE_SIZE)
 539                         amt = PAGE_CACHE_SIZE - offset;
 540                 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
 541                 if (IS_ERR(page))
 542                         return PTR_ERR(page);
 543                 p = kmap_atomic(page);
 544                 memcpy(buf + copied, p + offset, amt);
 545                 kunmap_atomic(p);
 546                 mark_page_accessed(page);
 547                 page_cache_release(page);
 548                 copied += amt;
 549                 index++;
 550                 offset = 0;
 551         } while(copied < size);
 552         (*pos) += size;
 553         return size;
 554 }
 555
 556 /**
 557  * gfs2_readpages - Read a bunch of pages at once
 558  *
 559  * Some notes:
 560  * 1. This is only for readahead, so we can simply ignore any things
 561  *    which are slightly inconvenient (such as locking conflicts between
 562  *    the page lock and the glock) and return having done no I/O. Its
 563  *    obviously not something we'd want to do on too regular a basis.
 564  *    Any I/O we ignore at this time will be done via readpage later.
 565  * 2. We don't handle stuffed files here we let readpage do the honours.
 566  * 3. mpage_readpages() does most of the heavy lifting in the common case.
 567  * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
 568  */
 569
 570 static int gfs2_readpages(struct file *file, struct address_space *mapping,
 571                           struct list_head *pages, unsigned nr_pages)
 572 {
 573         struct inode *inode = mapping->host;
 574         struct gfs2_inode *ip = GFS2_I(inode);
 575         struct gfs2_sbd *sdp = GFS2_SB(inode);
 576         struct gfs2_holder gh;
 577         int ret;
 578
 579         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 580         ret = gfs2_glock_nq(&gh);
 581         if (unlikely(ret))
 582                 goto out_uninit;
 583         if (!gfs2_is_stuffed(ip))
 584                 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
 585         gfs2_glock_dq(&gh);
 586 out_uninit:
 587         gfs2_holder_uninit(&gh);
 588         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 589                 ret = -EIO;
 590         return ret;
 591 }
 592
 593 /**
 594  * gfs2_write_begin - Begin to write to a file
 595  * @file: The file to write to
 596  * @mapping: The mapping in which to write
 597  * @pos: The file offset at which to start writing
 598  * @len: Length of the write
 599  * @flags: Various flags
 600  * @pagep: Pointer to return the page
 601  * @fsdata: Pointer to return fs data (unused by GFS2)
 602  *
 603  * Returns: errno
 604  */
 605
 606 static int gfs2_write_begin(struct file *file, struct address_space *mapping,
 607                             loff_t pos, unsigned len, unsigned flags,
 608                             struct page **pagep, void **fsdata)
 609 {
 610         struct gfs2_inode *ip = GFS2_I(mapping->host);
 611         struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
 612         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 613         unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
 614         unsigned requested = 0;
 615         int alloc_required;
 616         int error = 0;
 617         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
 618         unsigned from = pos & (PAGE_CACHE_SIZE - 1);
 619         struct page *page;
 620
 621         gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
 622         error = gfs2_glock_nq(&ip->i_gh);
 623         if (unlikely(error))
 624                 goto out_uninit;
 625         if (&ip->i_inode == sdp->sd_rindex) {
 626                 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
 627                                            GL_NOCACHE, &m_ip->i_gh);
 628                 if (unlikely(error)) {
 629                         gfs2_glock_dq(&ip->i_gh);
 630                         goto out_uninit;
 631                 }
 632         }
 633
 634         alloc_required = gfs2_write_alloc_required(ip, pos, len);
 635
 636         if (alloc_required || gfs2_is_jdata(ip))
 637                 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
 638
 639         if (alloc_required) {
 640                 error = gfs2_quota_lock_check(ip);
 641                 if (error)
 642                         goto out_unlock;
 643
 644                 requested = data_blocks + ind_blocks;
 645                 error = gfs2_inplace_reserve(ip, requested, 0);
 646                 if (error)
 647                         goto out_qunlock;
 648         }
 649
 650         rblocks = RES_DINODE + ind_blocks;
 651         if (gfs2_is_jdata(ip))
 652                 rblocks += data_blocks ? data_blocks : 1;
 653         if (ind_blocks || data_blocks)
 654                 rblocks += RES_STATFS + RES_QUOTA;
 655         if (&ip->i_inode == sdp->sd_rindex)
 656                 rblocks += 2 * RES_STATFS;
 657         if (alloc_required)
 658                 rblocks += gfs2_rg_blocks(ip, requested);
 659
 660         error = gfs2_trans_begin(sdp, rblocks,
 661                                  PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
 662         if (error)
 663                 goto out_trans_fail;
 664
 665         error = -ENOMEM;
 666         flags |= AOP_FLAG_NOFS;
 667         page = grab_cache_page_write_begin(mapping, index, flags);
 668         *pagep = page;
 669         if (unlikely(!page))
 670                 goto out_endtrans;
 671
 672         if (gfs2_is_stuffed(ip)) {
 673                 error = 0;
 674                 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
 675                         error = gfs2_unstuff_dinode(ip, page);
 676                         if (error == 0)
 677                                 goto prepare_write;
 678                 } else if (!PageUptodate(page)) {
 679                         error = stuffed_readpage(ip, page);
 680                 }
 681                 goto out;
 682         }
 683
 684 prepare_write:
 685         error = __block_write_begin(page, from, len, gfs2_block_map);
 686 out:
 687         if (error == 0)
 688                 return 0;
 689
 690         unlock_page(page);
 691         page_cache_release(page);
 692
 693         gfs2_trans_end(sdp);
 694         if (pos + len > ip->i_inode.i_size)
 695                 gfs2_trim_blocks(&ip->i_inode);
 696         goto out_trans_fail;
 697
 698 out_endtrans:
 699         gfs2_trans_end(sdp);
 700 out_trans_fail:
 701         if (alloc_required) {
 702                 gfs2_inplace_release(ip);
 703 out_qunlock:
 704                 gfs2_quota_unlock(ip);
 705         }
 706 out_unlock:
 707         if (&ip->i_inode == sdp->sd_rindex) {
 708                 gfs2_glock_dq(&m_ip->i_gh);
 709                 gfs2_holder_uninit(&m_ip->i_gh);
 710         }
 711         gfs2_glock_dq(&ip->i_gh);
 712 out_uninit:
 713         gfs2_holder_uninit(&ip->i_gh);
 714         return error;
 715 }
 716
 717 /**
 718  * adjust_fs_space - Adjusts the free space available due to gfs2_grow
 719  * @inode: the rindex inode
 720  */
 721 static void adjust_fs_space(struct inode *inode)
 722 {
 723         struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
 724         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 725         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 726         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 727         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 728         struct buffer_head *m_bh, *l_bh;
 729         u64 fs_total, new_free;
 730
 731         /* Total up the file system space, according to the latest rindex. */
 732         fs_total = gfs2_ri_total(sdp);
 733         if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
 734                 return;
 735
 736         spin_lock(&sdp->sd_statfs_spin);
 737         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 738                               sizeof(struct gfs2_dinode));
 739         if (fs_total > (m_sc->sc_total + l_sc->sc_total))
 740                 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
 741         else
 742                 new_free = 0;
 743         spin_unlock(&sdp->sd_statfs_spin);
 744         fs_warn(sdp, "File system extended by %llu blocks.\n",
 745                 (unsigned long long)new_free);
 746         gfs2_statfs_change(sdp, new_free, new_free, 0);
 747
 748         if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0)
 749                 goto out;
 750         update_statfs(sdp, m_bh, l_bh);
 751         brelse(l_bh);
 752 out:
 753         brelse(m_bh);
 754 }
 755
 756 /**
 757  * gfs2_stuffed_write_end - Write end for stuffed files
 758  * @inode: The inode
 759  * @dibh: The buffer_head containing the on-disk inode
 760  * @pos: The file position
 761  * @len: The length of the write
 762  * @copied: How much was actually copied by the VFS
 763  * @page: The page
 764  *
 765  * This copies the data from the page into the inode block after
 766  * the inode data structure itself.
 767  *
 768  * Returns: errno
 769  */
 770 static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
 771                                   loff_t pos, unsigned len, unsigned copied,
 772                                   struct page *page)
 773 {
 774         struct gfs2_inode *ip = GFS2_I(inode);
 775         struct gfs2_sbd *sdp = GFS2_SB(inode);
 776         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 777         u64 to = pos + copied;
 778         void *kaddr;
 779         unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
 780
 781         BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
 782         kaddr = kmap_atomic(page);
 783         memcpy(buf + pos, kaddr + pos, copied);
 784         memset(kaddr + pos + copied, 0, len - copied);
 785         flush_dcache_page(page);
 786         kunmap_atomic(kaddr);
 787
 788         if (!PageUptodate(page))
 789                 SetPageUptodate(page);
 790         unlock_page(page);
 791         page_cache_release(page);
 792
 793         if (copied) {
 794                 if (inode->i_size < to)
 795                         i_size_write(inode, to);
 796                 mark_inode_dirty(inode);
 797         }
 798
 799         if (inode == sdp->sd_rindex) {
 800                 adjust_fs_space(inode);
 801                 sdp->sd_rindex_uptodate = 0;
 802         }
 803
 804         brelse(dibh);
 805         gfs2_trans_end(sdp);
 806         if (inode == sdp->sd_rindex) {
 807                 gfs2_glock_dq(&m_ip->i_gh);
 808                 gfs2_holder_uninit(&m_ip->i_gh);
 809         }
 810         gfs2_glock_dq(&ip->i_gh);
 811         gfs2_holder_uninit(&ip->i_gh);
 812         return copied;
 813 }
 814
 815 /**
 816  * gfs2_write_end
 817  * @file: The file to write to
 818  * @mapping: The address space to write to
 819  * @pos: The file position
 820  * @len: The length of the data
 821  * @copied:
 822  * @page: The page that has been written
 823  * @fsdata: The fsdata (unused in GFS2)
 824  *
 825  * The main write_end function for GFS2. We have a separate one for
 826  * stuffed files as they are slightly different, otherwise we just
 827  * put our locking around the VFS provided functions.
 828  *
 829  * Returns: errno
 830  */
 831
 832 static int gfs2_write_end(struct file *file, struct address_space *mapping,
 833                           loff_t pos, unsigned len, unsigned copied,
 834                           struct page *page, void *fsdata)
 835 {
 836         struct inode *inode = page->mapping->host;
 837         struct gfs2_inode *ip = GFS2_I(inode);
 838         struct gfs2_sbd *sdp = GFS2_SB(inode);
 839         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 840         struct buffer_head *dibh;
 841         unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
 842         unsigned int to = from + len;
 843         int ret;
 844
 845         BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
 846
 847         ret = gfs2_meta_inode_buffer(ip, &dibh);
 848         if (unlikely(ret)) {
 849                 unlock_page(page);
 850                 page_cache_release(page);
 851                 goto failed;
 852         }
 853
 854         gfs2_trans_add_meta(ip->i_gl, dibh);
 855
 856         if (gfs2_is_stuffed(ip))
 857                 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
 858
 859         if (!gfs2_is_writeback(ip))
 860                 gfs2_page_add_databufs(ip, page, from, to);
 861
 862         ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
 863
 864         if (inode == sdp->sd_rindex) {
 865                 adjust_fs_space(inode);
 866                 sdp->sd_rindex_uptodate = 0;
 867         }
 868
 869         brelse(dibh);
 870 failed:
 871         gfs2_trans_end(sdp);
 872         gfs2_inplace_release(ip);
 873         if (ip->i_res->rs_qa_qd_num)
 874                 gfs2_quota_unlock(ip);
 875         if (inode == sdp->sd_rindex) {
 876                 gfs2_glock_dq(&m_ip->i_gh);
 877                 gfs2_holder_uninit(&m_ip->i_gh);
 878         }
 879         gfs2_glock_dq(&ip->i_gh);
 880         gfs2_holder_uninit(&ip->i_gh);
 881         return ret;
 882 }
 883
 884 /**
 885  * gfs2_set_page_dirty - Page dirtying function
 886  * @page: The page to dirty
 887  *
 888  * Returns: 1 if it dirtyed the page, or 0 otherwise
 889  */
 890
 891 static int gfs2_set_page_dirty(struct page *page)
 892 {
 893         SetPageChecked(page);
 894         return __set_page_dirty_buffers(page);
 895 }
 896
 897 /**
 898  * gfs2_bmap - Block map function
 899  * @mapping: Address space info
 900  * @lblock: The block to map
 901  *
 902  * Returns: The disk address for the block or 0 on hole or error
 903  */
 904
 905 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
 906 {
 907         struct gfs2_inode *ip = GFS2_I(mapping->host);
 908         struct gfs2_holder i_gh;
 909         sector_t dblock = 0;
 910         int error;
 911
 912         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
 913         if (error)
 914                 return 0;
 915
 916         if (!gfs2_is_stuffed(ip))
 917                 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
 918
 919         gfs2_glock_dq_uninit(&i_gh);
 920
 921         return dblock;
 922 }
 923
 924 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
 925 {
 926         struct gfs2_bufdata *bd;
 927
 928         lock_buffer(bh);
 929         gfs2_log_lock(sdp);
 930         clear_buffer_dirty(bh);
 931         bd = bh->b_private;
 932         if (bd) {
 933                 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
 934                         list_del_init(&bd->bd_list);
 935                 else
 936                         gfs2_remove_from_journal(bh, current->journal_info, 0);
 937         }
 938         bh->b_bdev = NULL;
 939         clear_buffer_mapped(bh);
 940         clear_buffer_req(bh);
 941         clear_buffer_new(bh);
 942         gfs2_log_unlock(sdp);
 943         unlock_buffer(bh);
 944 }
 945
 946 static void gfs2_invalidatepage(struct page *page, unsigned long offset)
 947 {
 948         struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
 949         struct buffer_head *bh, *head;
 950         unsigned long pos = 0;
 951
 952         BUG_ON(!PageLocked(page));
 953         if (offset == 0)
 954                 ClearPageChecked(page);
 955         if (!page_has_buffers(page))
 956                 goto out;
 957
 958         bh = head = page_buffers(page);
 959         do {
 960                 if (offset <= pos)
 961                         gfs2_discard(sdp, bh);
 962                 pos += bh->b_size;
 963                 bh = bh->b_this_page;
 964         } while (bh != head);
 965 out:
 966         if (offset == 0)
 967                 try_to_release_page(page, 0);
 968 }
 969
 970 /**
 971  * gfs2_ok_for_dio - check that dio is valid on this file
 972  * @ip: The inode
 973  * @rw: READ or WRITE
 974  * @offset: The offset at which we are reading or writing
 975  *
 976  * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
 977  *          1 (to accept the i/o request)
 978  */
 979 static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
 980 {
 981         /*
 982          * Should we return an error here? I can't see that O_DIRECT for
 983          * a stuffed file makes any sense. For now we'll silently fall
 984          * back to buffered I/O
 985          */
 986         if (gfs2_is_stuffed(ip))
 987                 return 0;
 988
 989         if (offset >= i_size_read(&ip->i_inode))
 990                 return 0;
 991         return 1;
 992 }
 993
 994
 995
 996 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
 997                               const struct iovec *iov, loff_t offset,
 998                               unsigned long nr_segs)
 999 {
1000         struct file *file = iocb->ki_filp;
1001         struct inode *inode = file->f_mapping->host;
1002         struct gfs2_inode *ip = GFS2_I(inode);
1003         struct gfs2_holder gh;
1004         int rv;
1005
1006         /*
1007          * Deferred lock, even if its a write, since we do no allocation
1008          * on this path. All we need change is atime, and this lock mode
1009          * ensures that other nodes have flushed their buffered read caches
1010          * (i.e. their page cache entries for this inode). We do not,
1011          * unfortunately have the option of only flushing a range like
1012          * the VFS does.
1013          */
1014         gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
1015         rv = gfs2_glock_nq(&gh);
1016         if (rv)
1017                 return rv;
1018         rv = gfs2_ok_for_dio(ip, rw, offset);
1019         if (rv != 1)
1020                 goto out; /* dio not valid, fall back to buffered i/o */
1021
1022         rv = __blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1023                                   offset, nr_segs, gfs2_get_block_direct,
1024                                   NULL, NULL, 0);
1025 out:
1026         gfs2_glock_dq(&gh);
1027         gfs2_holder_uninit(&gh);
1028         return rv;
1029 }
1030
1031 /**
1032  * gfs2_releasepage - free the metadata associated with a page
1033  * @page: the page that's being released
1034  * @gfp_mask: passed from Linux VFS, ignored by us
1035  *
1036  * Call try_to_free_buffers() if the buffers in this page can be
1037  * released.
1038  *
1039  * Returns: 0
1040  */
1041
1042 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1043 {
1044         struct address_space *mapping = page->mapping;
1045         struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
1046         struct buffer_head *bh, *head;
1047         struct gfs2_bufdata *bd;
1048
1049         if (!page_has_buffers(page))
1050                 return 0;
1051
1052         gfs2_log_lock(sdp);
1053         spin_lock(&sdp->sd_ail_lock);
1054         head = bh = page_buffers(page);
1055         do {
1056                 if (atomic_read(&bh->b_count))
1057                         goto cannot_release;
1058                 bd = bh->b_private;
1059                 if (bd && bd->bd_tr)
1060                         goto cannot_release;
1061                 if (buffer_pinned(bh) || buffer_dirty(bh))
1062                         goto not_possible;
1063                 bh = bh->b_this_page;
1064         } while(bh != head);
1065         spin_unlock(&sdp->sd_ail_lock);
1066         gfs2_log_unlock(sdp);
1067
1068         head = bh = page_buffers(page);
1069         do {
1070                 gfs2_log_lock(sdp);
1071                 bd = bh->b_private;
1072                 if (bd) {
1073                         gfs2_assert_warn(sdp, bd->bd_bh == bh);
1074                         if (!list_empty(&bd->bd_list)) {
1075                                 if (!buffer_pinned(bh))
1076                                         list_del_init(&bd->bd_list);
1077                                 else
1078                                         bd = NULL;
1079                         }
1080                         if (bd)
1081                                 bd->bd_bh = NULL;
1082                         bh->b_private = NULL;
1083                 }
1084                 gfs2_log_unlock(sdp);
1085                 if (bd)
1086                         kmem_cache_free(gfs2_bufdata_cachep, bd);
1087
1088                 bh = bh->b_this_page;
1089         } while (bh != head);
1090
1091         return try_to_free_buffers(page);
1092
1093 not_possible: /* Should never happen */
1094         WARN_ON(buffer_dirty(bh));
1095         WARN_ON(buffer_pinned(bh));
1096 cannot_release:
1097         spin_unlock(&sdp->sd_ail_lock);
1098         gfs2_log_unlock(sdp);
1099         return 0;
1100 }
1101
1102 static const struct address_space_operations gfs2_writeback_aops = {
1103         .writepage = gfs2_writeback_writepage,
1104         .writepages = gfs2_writepages,
1105         .readpage = gfs2_readpage,
1106         .readpages = gfs2_readpages,
1107         .write_begin = gfs2_write_begin,
1108         .write_end = gfs2_write_end,
1109         .bmap = gfs2_bmap,
1110         .invalidatepage = gfs2_invalidatepage,
1111         .releasepage = gfs2_releasepage,
1112         .direct_IO = gfs2_direct_IO,
1113         .migratepage = buffer_migrate_page,
1114         .is_partially_uptodate = block_is_partially_uptodate,
1115         .error_remove_page = generic_error_remove_page,
1116 };
1117
1118 static const struct address_space_operations gfs2_ordered_aops = {
1119         .writepage = gfs2_ordered_writepage,
1120         .writepages = gfs2_writepages,
1121         .readpage = gfs2_readpage,
1122         .readpages = gfs2_readpages,
1123         .write_begin = gfs2_write_begin,
1124         .write_end = gfs2_write_end,
1125         .set_page_dirty = gfs2_set_page_dirty,
1126         .bmap = gfs2_bmap,
1127         .invalidatepage = gfs2_invalidatepage,
1128         .releasepage = gfs2_releasepage,
1129         .direct_IO = gfs2_direct_IO,
1130         .migratepage = buffer_migrate_page,
1131         .is_partially_uptodate = block_is_partially_uptodate,
1132         .error_remove_page = generic_error_remove_page,
1133 };
1134
1135 static const struct address_space_operations gfs2_jdata_aops = {
1136         .writepage = gfs2_jdata_writepage,
1137         .writepages = gfs2_jdata_writepages,
1138         .readpage = gfs2_readpage,
1139         .readpages = gfs2_readpages,
1140         .write_begin = gfs2_write_begin,
1141         .write_end = gfs2_write_end,
1142         .set_page_dirty = gfs2_set_page_dirty,
1143         .bmap = gfs2_bmap,
1144         .invalidatepage = gfs2_invalidatepage,
1145         .releasepage = gfs2_releasepage,
1146         .is_partially_uptodate = block_is_partially_uptodate,
1147         .error_remove_page = generic_error_remove_page,
1148 };
1149
1150 void gfs2_set_aops(struct inode *inode)
1151 {
1152         struct gfs2_inode *ip = GFS2_I(inode);
1153
1154         if (gfs2_is_writeback(ip))
1155                 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1156         else if (gfs2_is_ordered(ip))
1157                 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1158         else if (gfs2_is_jdata(ip))
1159                 inode->i_mapping->a_ops = &gfs2_jdata_aops;
1160         else
1161                 BUG();
1162 }
1163