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, PAGE_CACHE_SIZE);
 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                                                              PAGE_CACHE_SIZE);
 304                         unlock_page(page);
 305                         continue;
 306                 }
 307
 308                 ret = __gfs2_jdata_writepage(page, wbc);
 309
 310                 if (ret || (--(wbc->nr_to_write) <= 0))
 311                         ret = 1;
 312         }
 313         gfs2_trans_end(sdp);
 314         return ret;
 315 }
 316
 317 /**
 318  * gfs2_write_cache_jdata - Like write_cache_pages but different
 319  * @mapping: The mapping to write
 320  * @wbc: The writeback control
 321  * @writepage: The writepage function to call
 322  * @data: The data to pass to writepage
 323  *
 324  * The reason that we use our own function here is that we need to
 325  * start transactions before we grab page locks. This allows us
 326  * to get the ordering right.
 327  */
 328
 329 static int gfs2_write_cache_jdata(struct address_space *mapping,
 330                                   struct writeback_control *wbc)
 331 {
 332         int ret = 0;
 333         int done = 0;
 334         struct pagevec pvec;
 335         int nr_pages;
 336         pgoff_t index;
 337         pgoff_t end;
 338         int scanned = 0;
 339         int range_whole = 0;
 340
 341         pagevec_init(&pvec, 0);
 342         if (wbc->range_cyclic) {
 343                 index = mapping->writeback_index; /* Start from prev offset */
 344                 end = -1;
 345         } else {
 346                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
 347                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
 348                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
 349                         range_whole = 1;
 350                 scanned = 1;
 351         }
 352
 353 retry:
 354          while (!done && (index <= end) &&
 355                 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 356                                                PAGECACHE_TAG_DIRTY,
 357                                                min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
 358                 scanned = 1;
 359                 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end);
 360                 if (ret)
 361                         done = 1;
 362                 if (ret > 0)
 363                         ret = 0;
 364
 365                 pagevec_release(&pvec);
 366                 cond_resched();
 367         }
 368
 369         if (!scanned && !done) {
 370                 /*
 371                  * We hit the last page and there is more work to be done: wrap
 372                  * back to the start of the file
 373                  */
 374                 scanned = 1;
 375                 index = 0;
 376                 goto retry;
 377         }
 378
 379         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
 380                 mapping->writeback_index = index;
 381         return ret;
 382 }
 383
 384
 385 /**
 386  * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
 387  * @mapping: The mapping to write
 388  * @wbc: The writeback control
 389  *
 390  */
 391
 392 static int gfs2_jdata_writepages(struct address_space *mapping,
 393                                  struct writeback_control *wbc)
 394 {
 395         struct gfs2_inode *ip = GFS2_I(mapping->host);
 396         struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
 397         int ret;
 398
 399         ret = gfs2_write_cache_jdata(mapping, wbc);
 400         if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
 401                 gfs2_log_flush(sdp, ip->i_gl);
 402                 ret = gfs2_write_cache_jdata(mapping, wbc);
 403         }
 404         return ret;
 405 }
 406
 407 /**
 408  * stuffed_readpage - Fill in a Linux page with stuffed file data
 409  * @ip: the inode
 410  * @page: the page
 411  *
 412  * Returns: errno
 413  */
 414
 415 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
 416 {
 417         struct buffer_head *dibh;
 418         u64 dsize = i_size_read(&ip->i_inode);
 419         void *kaddr;
 420         int error;
 421
 422         /*
 423          * Due to the order of unstuffing files and ->fault(), we can be
 424          * asked for a zero page in the case of a stuffed file being extended,
 425          * so we need to supply one here. It doesn't happen often.
 426          */
 427         if (unlikely(page->index)) {
 428                 zero_user(page, 0, PAGE_CACHE_SIZE);
 429                 SetPageUptodate(page);
 430                 return 0;
 431         }
 432
 433         error = gfs2_meta_inode_buffer(ip, &dibh);
 434         if (error)
 435                 return error;
 436
 437         kaddr = kmap_atomic(page);
 438         if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
 439                 dsize = (dibh->b_size - sizeof(struct gfs2_dinode));
 440         memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
 441         memset(kaddr + dsize, 0, PAGE_CACHE_SIZE - dsize);
 442         kunmap_atomic(kaddr);
 443         flush_dcache_page(page);
 444         brelse(dibh);
 445         SetPageUptodate(page);
 446
 447         return 0;
 448 }
 449
 450
 451 /**
 452  * __gfs2_readpage - readpage
 453  * @file: The file to read a page for
 454  * @page: The page to read
 455  *
 456  * This is the core of gfs2's readpage. Its used by the internal file
 457  * reading code as in that case we already hold the glock. Also its
 458  * called by gfs2_readpage() once the required lock has been granted.
 459  *
 460  */
 461
 462 static int __gfs2_readpage(void *file, struct page *page)
 463 {
 464         struct gfs2_inode *ip = GFS2_I(page->mapping->host);
 465         struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
 466         int error;
 467
 468         if (gfs2_is_stuffed(ip)) {
 469                 error = stuffed_readpage(ip, page);
 470                 unlock_page(page);
 471         } else {
 472                 error = mpage_readpage(page, gfs2_block_map);
 473         }
 474
 475         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 476                 return -EIO;
 477
 478         return error;
 479 }
 480
 481 /**
 482  * gfs2_readpage - read a page of a file
 483  * @file: The file to read
 484  * @page: The page of the file
 485  *
 486  * This deals with the locking required. We have to unlock and
 487  * relock the page in order to get the locking in the right
 488  * order.
 489  */
 490
 491 static int gfs2_readpage(struct file *file, struct page *page)
 492 {
 493         struct address_space *mapping = page->mapping;
 494         struct gfs2_inode *ip = GFS2_I(mapping->host);
 495         struct gfs2_holder gh;
 496         int error;
 497
 498         unlock_page(page);
 499         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 500         error = gfs2_glock_nq(&gh);
 501         if (unlikely(error))
 502                 goto out;
 503         error = AOP_TRUNCATED_PAGE;
 504         lock_page(page);
 505         if (page->mapping == mapping && !PageUptodate(page))
 506                 error = __gfs2_readpage(file, page);
 507         else
 508                 unlock_page(page);
 509         gfs2_glock_dq(&gh);
 510 out:
 511         gfs2_holder_uninit(&gh);
 512         if (error && error != AOP_TRUNCATED_PAGE)
 513                 lock_page(page);
 514         return error;
 515 }
 516
 517 /**
 518  * gfs2_internal_read - read an internal file
 519  * @ip: The gfs2 inode
 520  * @buf: The buffer to fill
 521  * @pos: The file position
 522  * @size: The amount to read
 523  *
 524  */
 525
 526 int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
 527                        unsigned size)
 528 {
 529         struct address_space *mapping = ip->i_inode.i_mapping;
 530         unsigned long index = *pos / PAGE_CACHE_SIZE;
 531         unsigned offset = *pos & (PAGE_CACHE_SIZE - 1);
 532         unsigned copied = 0;
 533         unsigned amt;
 534         struct page *page;
 535         void *p;
 536
 537         do {
 538                 amt = size - copied;
 539                 if (offset + size > PAGE_CACHE_SIZE)
 540                         amt = PAGE_CACHE_SIZE - offset;
 541                 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
 542                 if (IS_ERR(page))
 543                         return PTR_ERR(page);
 544                 p = kmap_atomic(page);
 545                 memcpy(buf + copied, p + offset, amt);
 546                 kunmap_atomic(p);
 547                 mark_page_accessed(page);
 548                 page_cache_release(page);
 549                 copied += amt;
 550                 index++;
 551                 offset = 0;
 552         } while(copied < size);
 553         (*pos) += size;
 554         return size;
 555 }
 556
 557 /**
 558  * gfs2_readpages - Read a bunch of pages at once
 559  *
 560  * Some notes:
 561  * 1. This is only for readahead, so we can simply ignore any things
 562  *    which are slightly inconvenient (such as locking conflicts between
 563  *    the page lock and the glock) and return having done no I/O. Its
 564  *    obviously not something we'd want to do on too regular a basis.
 565  *    Any I/O we ignore at this time will be done via readpage later.
 566  * 2. We don't handle stuffed files here we let readpage do the honours.
 567  * 3. mpage_readpages() does most of the heavy lifting in the common case.
 568  * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
 569  */
 570
 571 static int gfs2_readpages(struct file *file, struct address_space *mapping,
 572                           struct list_head *pages, unsigned nr_pages)
 573 {
 574         struct inode *inode = mapping->host;
 575         struct gfs2_inode *ip = GFS2_I(inode);
 576         struct gfs2_sbd *sdp = GFS2_SB(inode);
 577         struct gfs2_holder gh;
 578         int ret;
 579
 580         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 581         ret = gfs2_glock_nq(&gh);
 582         if (unlikely(ret))
 583                 goto out_uninit;
 584         if (!gfs2_is_stuffed(ip))
 585                 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
 586         gfs2_glock_dq(&gh);
 587 out_uninit:
 588         gfs2_holder_uninit(&gh);
 589         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 590                 ret = -EIO;
 591         return ret;
 592 }
 593
 594 /**
 595  * gfs2_write_begin - Begin to write to a file
 596  * @file: The file to write to
 597  * @mapping: The mapping in which to write
 598  * @pos: The file offset at which to start writing
 599  * @len: Length of the write
 600  * @flags: Various flags
 601  * @pagep: Pointer to return the page
 602  * @fsdata: Pointer to return fs data (unused by GFS2)
 603  *
 604  * Returns: errno
 605  */
 606
 607 static int gfs2_write_begin(struct file *file, struct address_space *mapping,
 608                             loff_t pos, unsigned len, unsigned flags,
 609                             struct page **pagep, void **fsdata)
 610 {
 611         struct gfs2_inode *ip = GFS2_I(mapping->host);
 612         struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
 613         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 614         unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
 615         unsigned requested = 0;
 616         int alloc_required;
 617         int error = 0;
 618         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
 619         unsigned from = pos & (PAGE_CACHE_SIZE - 1);
 620         struct page *page;
 621
 622         gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
 623         error = gfs2_glock_nq(&ip->i_gh);
 624         if (unlikely(error))
 625                 goto out_uninit;
 626         if (&ip->i_inode == sdp->sd_rindex) {
 627                 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
 628                                            GL_NOCACHE, &m_ip->i_gh);
 629                 if (unlikely(error)) {
 630                         gfs2_glock_dq(&ip->i_gh);
 631                         goto out_uninit;
 632                 }
 633         }
 634
 635         alloc_required = gfs2_write_alloc_required(ip, pos, len);
 636
 637         if (alloc_required || gfs2_is_jdata(ip))
 638                 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
 639
 640         if (alloc_required) {
 641                 error = gfs2_quota_lock_check(ip);
 642                 if (error)
 643                         goto out_unlock;
 644
 645                 requested = data_blocks + ind_blocks;
 646                 error = gfs2_inplace_reserve(ip, requested, 0);
 647                 if (error)
 648                         goto out_qunlock;
 649         }
 650
 651         rblocks = RES_DINODE + ind_blocks;
 652         if (gfs2_is_jdata(ip))
 653                 rblocks += data_blocks ? data_blocks : 1;
 654         if (ind_blocks || data_blocks)
 655                 rblocks += RES_STATFS + RES_QUOTA;
 656         if (&ip->i_inode == sdp->sd_rindex)
 657                 rblocks += 2 * RES_STATFS;
 658         if (alloc_required)
 659                 rblocks += gfs2_rg_blocks(ip, requested);
 660
 661         error = gfs2_trans_begin(sdp, rblocks,
 662                                  PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
 663         if (error)
 664                 goto out_trans_fail;
 665
 666         error = -ENOMEM;
 667         flags |= AOP_FLAG_NOFS;
 668         page = grab_cache_page_write_begin(mapping, index, flags);
 669         *pagep = page;
 670         if (unlikely(!page))
 671                 goto out_endtrans;
 672
 673         if (gfs2_is_stuffed(ip)) {
 674                 error = 0;
 675                 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
 676                         error = gfs2_unstuff_dinode(ip, page);
 677                         if (error == 0)
 678                                 goto prepare_write;
 679                 } else if (!PageUptodate(page)) {
 680                         error = stuffed_readpage(ip, page);
 681                 }
 682                 goto out;
 683         }
 684
 685 prepare_write:
 686         error = __block_write_begin(page, from, len, gfs2_block_map);
 687 out:
 688         if (error == 0)
 689                 return 0;
 690
 691         unlock_page(page);
 692         page_cache_release(page);
 693
 694         gfs2_trans_end(sdp);
 695         if (pos + len > ip->i_inode.i_size)
 696                 gfs2_trim_blocks(&ip->i_inode);
 697         goto out_trans_fail;
 698
 699 out_endtrans:
 700         gfs2_trans_end(sdp);
 701 out_trans_fail:
 702         if (alloc_required) {
 703                 gfs2_inplace_release(ip);
 704 out_qunlock:
 705                 gfs2_quota_unlock(ip);
 706         }
 707 out_unlock:
 708         if (&ip->i_inode == sdp->sd_rindex) {
 709                 gfs2_glock_dq(&m_ip->i_gh);
 710                 gfs2_holder_uninit(&m_ip->i_gh);
 711         }
 712         gfs2_glock_dq(&ip->i_gh);
 713 out_uninit:
 714         gfs2_holder_uninit(&ip->i_gh);
 715         return error;
 716 }
 717
 718 /**
 719  * adjust_fs_space - Adjusts the free space available due to gfs2_grow
 720  * @inode: the rindex inode
 721  */
 722 static void adjust_fs_space(struct inode *inode)
 723 {
 724         struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
 725         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 726         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 727         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 728         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 729         struct buffer_head *m_bh, *l_bh;
 730         u64 fs_total, new_free;
 731
 732         /* Total up the file system space, according to the latest rindex. */
 733         fs_total = gfs2_ri_total(sdp);
 734         if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
 735                 return;
 736
 737         spin_lock(&sdp->sd_statfs_spin);
 738         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 739                               sizeof(struct gfs2_dinode));
 740         if (fs_total > (m_sc->sc_total + l_sc->sc_total))
 741                 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
 742         else
 743                 new_free = 0;
 744         spin_unlock(&sdp->sd_statfs_spin);
 745         fs_warn(sdp, "File system extended by %llu blocks.\n",
 746                 (unsigned long long)new_free);
 747         gfs2_statfs_change(sdp, new_free, new_free, 0);
 748
 749         if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0)
 750                 goto out;
 751         update_statfs(sdp, m_bh, l_bh);
 752         brelse(l_bh);
 753 out:
 754         brelse(m_bh);
 755 }
 756
 757 /**
 758  * gfs2_stuffed_write_end - Write end for stuffed files
 759  * @inode: The inode
 760  * @dibh: The buffer_head containing the on-disk inode
 761  * @pos: The file position
 762  * @len: The length of the write
 763  * @copied: How much was actually copied by the VFS
 764  * @page: The page
 765  *
 766  * This copies the data from the page into the inode block after
 767  * the inode data structure itself.
 768  *
 769  * Returns: errno
 770  */
 771 static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
 772                                   loff_t pos, unsigned len, unsigned copied,
 773                                   struct page *page)
 774 {
 775         struct gfs2_inode *ip = GFS2_I(inode);
 776         struct gfs2_sbd *sdp = GFS2_SB(inode);
 777         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 778         u64 to = pos + copied;
 779         void *kaddr;
 780         unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
 781
 782         BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
 783         kaddr = kmap_atomic(page);
 784         memcpy(buf + pos, kaddr + pos, copied);
 785         memset(kaddr + pos + copied, 0, len - copied);
 786         flush_dcache_page(page);
 787         kunmap_atomic(kaddr);
 788
 789         if (!PageUptodate(page))
 790                 SetPageUptodate(page);
 791         unlock_page(page);
 792         page_cache_release(page);
 793
 794         if (copied) {
 795                 if (inode->i_size < to)
 796                         i_size_write(inode, to);
 797                 mark_inode_dirty(inode);
 798         }
 799
 800         if (inode == sdp->sd_rindex) {
 801                 adjust_fs_space(inode);
 802                 sdp->sd_rindex_uptodate = 0;
 803         }
 804
 805         brelse(dibh);
 806         gfs2_trans_end(sdp);
 807         if (inode == sdp->sd_rindex) {
 808                 gfs2_glock_dq(&m_ip->i_gh);
 809                 gfs2_holder_uninit(&m_ip->i_gh);
 810         }
 811         gfs2_glock_dq(&ip->i_gh);
 812         gfs2_holder_uninit(&ip->i_gh);
 813         return copied;
 814 }
 815
 816 /**
 817  * gfs2_write_end
 818  * @file: The file to write to
 819  * @mapping: The address space to write to
 820  * @pos: The file position
 821  * @len: The length of the data
 822  * @copied:
 823  * @page: The page that has been written
 824  * @fsdata: The fsdata (unused in GFS2)
 825  *
 826  * The main write_end function for GFS2. We have a separate one for
 827  * stuffed files as they are slightly different, otherwise we just
 828  * put our locking around the VFS provided functions.
 829  *
 830  * Returns: errno
 831  */
 832
 833 static int gfs2_write_end(struct file *file, struct address_space *mapping,
 834                           loff_t pos, unsigned len, unsigned copied,
 835                           struct page *page, void *fsdata)
 836 {
 837         struct inode *inode = page->mapping->host;
 838         struct gfs2_inode *ip = GFS2_I(inode);
 839         struct gfs2_sbd *sdp = GFS2_SB(inode);
 840         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 841         struct buffer_head *dibh;
 842         unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
 843         unsigned int to = from + len;
 844         int ret;
 845
 846         BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
 847
 848         ret = gfs2_meta_inode_buffer(ip, &dibh);
 849         if (unlikely(ret)) {
 850                 unlock_page(page);
 851                 page_cache_release(page);
 852                 goto failed;
 853         }
 854
 855         gfs2_trans_add_meta(ip->i_gl, dibh);
 856
 857         if (gfs2_is_stuffed(ip))
 858                 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
 859
 860         if (!gfs2_is_writeback(ip))
 861                 gfs2_page_add_databufs(ip, page, from, to);
 862
 863         ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
 864
 865         if (inode == sdp->sd_rindex) {
 866                 adjust_fs_space(inode);
 867                 sdp->sd_rindex_uptodate = 0;
 868         }
 869
 870         brelse(dibh);
 871 failed:
 872         gfs2_trans_end(sdp);
 873         gfs2_inplace_release(ip);
 874         if (ip->i_res->rs_qa_qd_num)
 875                 gfs2_quota_unlock(ip);
 876         if (inode == sdp->sd_rindex) {
 877                 gfs2_glock_dq(&m_ip->i_gh);
 878                 gfs2_holder_uninit(&m_ip->i_gh);
 879         }
 880         gfs2_glock_dq(&ip->i_gh);
 881         gfs2_holder_uninit(&ip->i_gh);
 882         return ret;
 883 }
 884
 885 /**
 886  * gfs2_set_page_dirty - Page dirtying function
 887  * @page: The page to dirty
 888  *
 889  * Returns: 1 if it dirtyed the page, or 0 otherwise
 890  */
 891
 892 static int gfs2_set_page_dirty(struct page *page)
 893 {
 894         SetPageChecked(page);
 895         return __set_page_dirty_buffers(page);
 896 }
 897
 898 /**
 899  * gfs2_bmap - Block map function
 900  * @mapping: Address space info
 901  * @lblock: The block to map
 902  *
 903  * Returns: The disk address for the block or 0 on hole or error
 904  */
 905
 906 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
 907 {
 908         struct gfs2_inode *ip = GFS2_I(mapping->host);
 909         struct gfs2_holder i_gh;
 910         sector_t dblock = 0;
 911         int error;
 912
 913         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
 914         if (error)
 915                 return 0;
 916
 917         if (!gfs2_is_stuffed(ip))
 918                 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
 919
 920         gfs2_glock_dq_uninit(&i_gh);
 921
 922         return dblock;
 923 }
 924
 925 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
 926 {
 927         struct gfs2_bufdata *bd;
 928
 929         lock_buffer(bh);
 930         gfs2_log_lock(sdp);
 931         clear_buffer_dirty(bh);
 932         bd = bh->b_private;
 933         if (bd) {
 934                 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
 935                         list_del_init(&bd->bd_list);
 936                 else
 937                         gfs2_remove_from_journal(bh, current->journal_info, 0);
 938         }
 939         bh->b_bdev = NULL;
 940         clear_buffer_mapped(bh);
 941         clear_buffer_req(bh);
 942         clear_buffer_new(bh);
 943         gfs2_log_unlock(sdp);
 944         unlock_buffer(bh);
 945 }
 946
 947 static void gfs2_invalidatepage(struct page *page, unsigned int offset,
 948                                 unsigned int length)
 949 {
 950         struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
 951         unsigned int stop = offset + length;
 952         int partial_page = (offset || length < PAGE_CACHE_SIZE);
 953         struct buffer_head *bh, *head;
 954         unsigned long pos = 0;
 955
 956         BUG_ON(!PageLocked(page));
 957         if (!partial_page)
 958                 ClearPageChecked(page);
 959         if (!page_has_buffers(page))
 960                 goto out;
 961
 962         bh = head = page_buffers(page);
 963         do {
 964                 if (pos + bh->b_size > stop)
 965                         return;
 966
 967                 if (offset <= pos)
 968                         gfs2_discard(sdp, bh);
 969                 pos += bh->b_size;
 970                 bh = bh->b_this_page;
 971         } while (bh != head);
 972 out:
 973         if (!partial_page)
 974                 try_to_release_page(page, 0);
 975 }
 976
 977 /**
 978  * gfs2_ok_for_dio - check that dio is valid on this file
 979  * @ip: The inode
 980  * @rw: READ or WRITE
 981  * @offset: The offset at which we are reading or writing
 982  *
 983  * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
 984  *          1 (to accept the i/o request)
 985  */
 986 static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
 987 {
 988         /*
 989          * Should we return an error here? I can't see that O_DIRECT for
 990          * a stuffed file makes any sense. For now we'll silently fall
 991          * back to buffered I/O
 992          */
 993         if (gfs2_is_stuffed(ip))
 994                 return 0;
 995
 996         if (offset >= i_size_read(&ip->i_inode))
 997                 return 0;
 998         return 1;
 999 }
1000
1001
1002
1003 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
1004                               const struct iovec *iov, loff_t offset,
1005                               unsigned long nr_segs)
1006 {
1007         struct file *file = iocb->ki_filp;
1008         struct inode *inode = file->f_mapping->host;
1009         struct gfs2_inode *ip = GFS2_I(inode);
1010         struct gfs2_holder gh;
1011         int rv;
1012
1013         /*
1014          * Deferred lock, even if its a write, since we do no allocation
1015          * on this path. All we need change is atime, and this lock mode
1016          * ensures that other nodes have flushed their buffered read caches
1017          * (i.e. their page cache entries for this inode). We do not,
1018          * unfortunately have the option of only flushing a range like
1019          * the VFS does.
1020          */
1021         gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
1022         rv = gfs2_glock_nq(&gh);
1023         if (rv)
1024                 return rv;
1025         rv = gfs2_ok_for_dio(ip, rw, offset);
1026         if (rv != 1)
1027                 goto out; /* dio not valid, fall back to buffered i/o */
1028
1029         rv = __blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1030                                   offset, nr_segs, gfs2_get_block_direct,
1031                                   NULL, NULL, 0);
1032 out:
1033         gfs2_glock_dq(&gh);
1034         gfs2_holder_uninit(&gh);
1035         return rv;
1036 }
1037
1038 /**
1039  * gfs2_releasepage - free the metadata associated with a page
1040  * @page: the page that's being released
1041  * @gfp_mask: passed from Linux VFS, ignored by us
1042  *
1043  * Call try_to_free_buffers() if the buffers in this page can be
1044  * released.
1045  *
1046  * Returns: 0
1047  */
1048
1049 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1050 {
1051         struct address_space *mapping = page->mapping;
1052         struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
1053         struct buffer_head *bh, *head;
1054         struct gfs2_bufdata *bd;
1055
1056         if (!page_has_buffers(page))
1057                 return 0;
1058
1059         gfs2_log_lock(sdp);
1060         spin_lock(&sdp->sd_ail_lock);
1061         head = bh = page_buffers(page);
1062         do {
1063                 if (atomic_read(&bh->b_count))
1064                         goto cannot_release;
1065                 bd = bh->b_private;
1066                 if (bd && bd->bd_tr)
1067                         goto cannot_release;
1068                 if (buffer_pinned(bh) || buffer_dirty(bh))
1069                         goto not_possible;
1070                 bh = bh->b_this_page;
1071         } while(bh != head);
1072         spin_unlock(&sdp->sd_ail_lock);
1073         gfs2_log_unlock(sdp);
1074
1075         head = bh = page_buffers(page);
1076         do {
1077                 gfs2_log_lock(sdp);
1078                 bd = bh->b_private;
1079                 if (bd) {
1080                         gfs2_assert_warn(sdp, bd->bd_bh == bh);
1081                         if (!list_empty(&bd->bd_list)) {
1082                                 if (!buffer_pinned(bh))
1083                                         list_del_init(&bd->bd_list);
1084                                 else
1085                                         bd = NULL;
1086                         }
1087                         if (bd)
1088                                 bd->bd_bh = NULL;
1089                         bh->b_private = NULL;
1090                 }
1091                 gfs2_log_unlock(sdp);
1092                 if (bd)
1093                         kmem_cache_free(gfs2_bufdata_cachep, bd);
1094
1095                 bh = bh->b_this_page;
1096         } while (bh != head);
1097
1098         return try_to_free_buffers(page);
1099
1100 not_possible: /* Should never happen */
1101         WARN_ON(buffer_dirty(bh));
1102         WARN_ON(buffer_pinned(bh));
1103 cannot_release:
1104         spin_unlock(&sdp->sd_ail_lock);
1105         gfs2_log_unlock(sdp);
1106         return 0;
1107 }
1108
1109 static const struct address_space_operations gfs2_writeback_aops = {
1110         .writepage = gfs2_writeback_writepage,
1111         .writepages = gfs2_writepages,
1112         .readpage = gfs2_readpage,
1113         .readpages = gfs2_readpages,
1114         .write_begin = gfs2_write_begin,
1115         .write_end = gfs2_write_end,
1116         .bmap = gfs2_bmap,
1117         .invalidatepage = gfs2_invalidatepage,
1118         .releasepage = gfs2_releasepage,
1119         .direct_IO = gfs2_direct_IO,
1120         .migratepage = buffer_migrate_page,
1121         .is_partially_uptodate = block_is_partially_uptodate,
1122         .error_remove_page = generic_error_remove_page,
1123 };
1124
1125 static const struct address_space_operations gfs2_ordered_aops = {
1126         .writepage = gfs2_ordered_writepage,
1127         .writepages = gfs2_writepages,
1128         .readpage = gfs2_readpage,
1129         .readpages = gfs2_readpages,
1130         .write_begin = gfs2_write_begin,
1131         .write_end = gfs2_write_end,
1132         .set_page_dirty = gfs2_set_page_dirty,
1133         .bmap = gfs2_bmap,
1134         .invalidatepage = gfs2_invalidatepage,
1135         .releasepage = gfs2_releasepage,
1136         .direct_IO = gfs2_direct_IO,
1137         .migratepage = buffer_migrate_page,
1138         .is_partially_uptodate = block_is_partially_uptodate,
1139         .error_remove_page = generic_error_remove_page,
1140 };
1141
1142 static const struct address_space_operations gfs2_jdata_aops = {
1143         .writepage = gfs2_jdata_writepage,
1144         .writepages = gfs2_jdata_writepages,
1145         .readpage = gfs2_readpage,
1146         .readpages = gfs2_readpages,
1147         .write_begin = gfs2_write_begin,
1148         .write_end = gfs2_write_end,
1149         .set_page_dirty = gfs2_set_page_dirty,
1150         .bmap = gfs2_bmap,
1151         .invalidatepage = gfs2_invalidatepage,
1152         .releasepage = gfs2_releasepage,
1153         .is_partially_uptodate = block_is_partially_uptodate,
1154         .error_remove_page = generic_error_remove_page,
1155 };
1156
1157 void gfs2_set_aops(struct inode *inode)
1158 {
1159         struct gfs2_inode *ip = GFS2_I(inode);
1160
1161         if (gfs2_is_writeback(ip))
1162                 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1163         else if (gfs2_is_ordered(ip))
1164                 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1165         else if (gfs2_is_jdata(ip))
1166                 inode->i_mapping->a_ops = &gfs2_jdata_aops;
1167         else
1168                 BUG();
1169 }
1170