fs/gfs2/ops_address.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/lm_interface.h>
  23 #include <linux/backing-dev.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 "ops_address.h"
  33 #include "quota.h"
  34 #include "trans.h"
  35 #include "rgrp.h"
  36 #include "super.h"
  37 #include "util.h"
  38 #include "glops.h"
  39
  40
  41 static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
  42                                    unsigned int from, unsigned int to)
  43 {
  44         struct buffer_head *head = page_buffers(page);
  45         unsigned int bsize = head->b_size;
  46         struct buffer_head *bh;
  47         unsigned int start, end;
  48
  49         for (bh = head, start = 0; bh != head || !start;
  50              bh = bh->b_this_page, start = end) {
  51                 end = start + bsize;
  52                 if (end <= from || start >= to)
  53                         continue;
  54                 if (gfs2_is_jdata(ip))
  55                         set_buffer_uptodate(bh);
  56                 gfs2_trans_add_bh(ip->i_gl, bh, 0);
  57         }
  58 }
  59
  60 /**
  61  * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
  62  * @inode: The inode
  63  * @lblock: The block number to look up
  64  * @bh_result: The buffer head to return the result in
  65  * @create: Non-zero if we may add block to the file
  66  *
  67  * Returns: errno
  68  */
  69
  70 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
  71                                   struct buffer_head *bh_result, int create)
  72 {
  73         int error;
  74
  75         error = gfs2_block_map(inode, lblock, bh_result, 0);
  76         if (error)
  77                 return error;
  78         if (!buffer_mapped(bh_result))
  79                 return -EIO;
  80         return 0;
  81 }
  82
  83 static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
  84                                  struct buffer_head *bh_result, int create)
  85 {
  86         return gfs2_block_map(inode, lblock, bh_result, 0);
  87 }
  88
  89 /**
  90  * gfs2_writepage_common - Common bits of writepage
  91  * @page: The page to be written
  92  * @wbc: The writeback control
  93  *
  94  * Returns: 1 if writepage is ok, otherwise an error code or zero if no error.
  95  */
  96
  97 static int gfs2_writepage_common(struct page *page,
  98                                  struct writeback_control *wbc)
  99 {
 100         struct inode *inode = page->mapping->host;
 101         struct gfs2_inode *ip = GFS2_I(inode);
 102         struct gfs2_sbd *sdp = GFS2_SB(inode);
 103         loff_t i_size = i_size_read(inode);
 104         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
 105         unsigned offset;
 106
 107         if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
 108                 goto out;
 109         if (current->journal_info)
 110                 goto redirty;
 111         /* Is the page fully outside i_size? (truncate in progress) */
 112         offset = i_size & (PAGE_CACHE_SIZE-1);
 113         if (page->index > end_index || (page->index == end_index && !offset)) {
 114                 page->mapping->a_ops->invalidatepage(page, 0);
 115                 goto out;
 116         }
 117         return 1;
 118 redirty:
 119         redirty_page_for_writepage(wbc, page);
 120 out:
 121         unlock_page(page);
 122         return 0;
 123 }
 124
 125 /**
 126  * gfs2_writeback_writepage - Write page for writeback mappings
 127  * @page: The page
 128  * @wbc: The writeback control
 129  *
 130  */
 131
 132 static int gfs2_writeback_writepage(struct page *page,
 133                                     struct writeback_control *wbc)
 134 {
 135         int ret;
 136
 137         ret = gfs2_writepage_common(page, wbc);
 138         if (ret <= 0)
 139                 return ret;
 140
 141         ret = mpage_writepage(page, gfs2_get_block_noalloc, wbc);
 142         if (ret == -EAGAIN)
 143                 ret = block_write_full_page(page, gfs2_get_block_noalloc, wbc);
 144         return ret;
 145 }
 146
 147 /**
 148  * gfs2_ordered_writepage - Write page for ordered data files
 149  * @page: The page to write
 150  * @wbc: The writeback control
 151  *
 152  */
 153
 154 static int gfs2_ordered_writepage(struct page *page,
 155                                   struct writeback_control *wbc)
 156 {
 157         struct inode *inode = page->mapping->host;
 158         struct gfs2_inode *ip = GFS2_I(inode);
 159         int ret;
 160
 161         ret = gfs2_writepage_common(page, wbc);
 162         if (ret <= 0)
 163                 return ret;
 164
 165         if (!page_has_buffers(page)) {
 166                 create_empty_buffers(page, inode->i_sb->s_blocksize,
 167                                      (1 << BH_Dirty)|(1 << BH_Uptodate));
 168         }
 169         gfs2_page_add_databufs(ip, page, 0, inode->i_sb->s_blocksize-1);
 170         return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
 171 }
 172
 173 /**
 174  * __gfs2_jdata_writepage - The core of jdata writepage
 175  * @page: The page to write
 176  * @wbc: The writeback control
 177  *
 178  * This is shared between writepage and writepages and implements the
 179  * core of the writepage operation. If a transaction is required then
 180  * PageChecked will have been set and the transaction will have
 181  * already been started before this is called.
 182  */
 183
 184 static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
 185 {
 186         struct inode *inode = page->mapping->host;
 187         struct gfs2_inode *ip = GFS2_I(inode);
 188         struct gfs2_sbd *sdp = GFS2_SB(inode);
 189
 190         if (PageChecked(page)) {
 191                 ClearPageChecked(page);
 192                 if (!page_has_buffers(page)) {
 193                         create_empty_buffers(page, inode->i_sb->s_blocksize,
 194                                              (1 << BH_Dirty)|(1 << BH_Uptodate));
 195                 }
 196                 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
 197         }
 198         return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
 199 }
 200
 201 /**
 202  * gfs2_jdata_writepage - Write complete page
 203  * @page: Page to write
 204  *
 205  * Returns: errno
 206  *
 207  */
 208
 209 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
 210 {
 211         struct inode *inode = page->mapping->host;
 212         struct gfs2_sbd *sdp = GFS2_SB(inode);
 213         int error;
 214         int done_trans = 0;
 215
 216         error = gfs2_writepage_common(page, wbc);
 217         if (error <= 0)
 218                 return error;
 219
 220         if (PageChecked(page)) {
 221                 if (wbc->sync_mode != WB_SYNC_ALL)
 222                         goto out_ignore;
 223                 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
 224                 if (error)
 225                         goto out_ignore;
 226                 done_trans = 1;
 227         }
 228         error = __gfs2_jdata_writepage(page, wbc);
 229         if (done_trans)
 230                 gfs2_trans_end(sdp);
 231         return error;
 232
 233 out_ignore:
 234         redirty_page_for_writepage(wbc, page);
 235         unlock_page(page);
 236         return 0;
 237 }
 238
 239 /**
 240  * gfs2_writeback_writepages - Write a bunch of dirty pages back to disk
 241  * @mapping: The mapping to write
 242  * @wbc: Write-back control
 243  *
 244  * For the data=writeback case we can already ignore buffer heads
 245  * and write whole extents at once. This is a big reduction in the
 246  * number of I/O requests we send and the bmap calls we make in this case.
 247  */
 248 static int gfs2_writeback_writepages(struct address_space *mapping,
 249                                      struct writeback_control *wbc)
 250 {
 251         return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
 252 }
 253
 254 /**
 255  * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
 256  * @mapping: The mapping
 257  * @wbc: The writeback control
 258  * @writepage: The writepage function to call for each page
 259  * @pvec: The vector of pages
 260  * @nr_pages: The number of pages to write
 261  *
 262  * Returns: non-zero if loop should terminate, zero otherwise
 263  */
 264
 265 static int gfs2_write_jdata_pagevec(struct address_space *mapping,
 266                                     struct writeback_control *wbc,
 267                                     struct pagevec *pvec,
 268                                     int nr_pages, pgoff_t end)
 269 {
 270         struct inode *inode = mapping->host;
 271         struct gfs2_sbd *sdp = GFS2_SB(inode);
 272         loff_t i_size = i_size_read(inode);
 273         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
 274         unsigned offset = i_size & (PAGE_CACHE_SIZE-1);
 275         unsigned nrblocks = nr_pages * (PAGE_CACHE_SIZE/inode->i_sb->s_blocksize);
 276         struct backing_dev_info *bdi = mapping->backing_dev_info;
 277         int i;
 278         int ret;
 279
 280         ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
 281         if (ret < 0)
 282                 return ret;
 283
 284         for(i = 0; i < nr_pages; i++) {
 285                 struct page *page = pvec->pages[i];
 286
 287                 lock_page(page);
 288
 289                 if (unlikely(page->mapping != mapping)) {
 290                         unlock_page(page);
 291                         continue;
 292                 }
 293
 294                 if (!wbc->range_cyclic && page->index > end) {
 295                         ret = 1;
 296                         unlock_page(page);
 297                         continue;
 298                 }
 299
 300                 if (wbc->sync_mode != WB_SYNC_NONE)
 301                         wait_on_page_writeback(page);
 302
 303                 if (PageWriteback(page) ||
 304                     !clear_page_dirty_for_io(page)) {
 305                         unlock_page(page);
 306                         continue;
 307                 }
 308
 309                 /* Is the page fully outside i_size? (truncate in progress) */
 310                 if (page->index > end_index || (page->index == end_index && !offset)) {
 311                         page->mapping->a_ops->invalidatepage(page, 0);
 312                         unlock_page(page);
 313                         continue;
 314                 }
 315
 316                 ret = __gfs2_jdata_writepage(page, wbc);
 317
 318                 if (ret || (--(wbc->nr_to_write) <= 0))
 319                         ret = 1;
 320                 if (wbc->nonblocking && bdi_write_congested(bdi)) {
 321                         wbc->encountered_congestion = 1;
 322                         ret = 1;
 323                 }
 324
 325         }
 326         gfs2_trans_end(sdp);
 327         return ret;
 328 }
 329
 330 /**
 331  * gfs2_write_cache_jdata - Like write_cache_pages but different
 332  * @mapping: The mapping to write
 333  * @wbc: The writeback control
 334  * @writepage: The writepage function to call
 335  * @data: The data to pass to writepage
 336  *
 337  * The reason that we use our own function here is that we need to
 338  * start transactions before we grab page locks. This allows us
 339  * to get the ordering right.
 340  */
 341
 342 static int gfs2_write_cache_jdata(struct address_space *mapping,
 343                                   struct writeback_control *wbc)
 344 {
 345         struct backing_dev_info *bdi = mapping->backing_dev_info;
 346         int ret = 0;
 347         int done = 0;
 348         struct pagevec pvec;
 349         int nr_pages;
 350         pgoff_t index;
 351         pgoff_t end;
 352         int scanned = 0;
 353         int range_whole = 0;
 354
 355         if (wbc->nonblocking && bdi_write_congested(bdi)) {
 356                 wbc->encountered_congestion = 1;
 357                 return 0;
 358         }
 359
 360         pagevec_init(&pvec, 0);
 361         if (wbc->range_cyclic) {
 362                 index = mapping->writeback_index; /* Start from prev offset */
 363                 end = -1;
 364         } else {
 365                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
 366                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
 367                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
 368                         range_whole = 1;
 369                 scanned = 1;
 370         }
 371
 372 retry:
 373          while (!done && (index <= end) &&
 374                 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 375                                                PAGECACHE_TAG_DIRTY,
 376                                                min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
 377                 scanned = 1;
 378                 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end);
 379                 if (ret)
 380                         done = 1;
 381                 if (ret > 0)
 382                         ret = 0;
 383
 384                 pagevec_release(&pvec);
 385                 cond_resched();
 386         }
 387
 388         if (!scanned && !done) {
 389                 /*
 390                  * We hit the last page and there is more work to be done: wrap
 391                  * back to the start of the file
 392                  */
 393                 scanned = 1;
 394                 index = 0;
 395                 goto retry;
 396         }
 397
 398         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
 399                 mapping->writeback_index = index;
 400         return ret;
 401 }
 402
 403
 404 /**
 405  * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
 406  * @mapping: The mapping to write
 407  * @wbc: The writeback control
 408  *
 409  */
 410
 411 static int gfs2_jdata_writepages(struct address_space *mapping,
 412                                  struct writeback_control *wbc)
 413 {
 414         struct gfs2_inode *ip = GFS2_I(mapping->host);
 415         struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
 416         int ret;
 417
 418         ret = gfs2_write_cache_jdata(mapping, wbc);
 419         if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
 420                 gfs2_log_flush(sdp, ip->i_gl);
 421                 ret = gfs2_write_cache_jdata(mapping, wbc);
 422         }
 423         return ret;
 424 }
 425
 426 /**
 427  * stuffed_readpage - Fill in a Linux page with stuffed file data
 428  * @ip: the inode
 429  * @page: the page
 430  *
 431  * Returns: errno
 432  */
 433
 434 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
 435 {
 436         struct buffer_head *dibh;
 437         void *kaddr;
 438         int error;
 439
 440         /*
 441          * Due to the order of unstuffing files and ->fault(), we can be
 442          * asked for a zero page in the case of a stuffed file being extended,
 443          * so we need to supply one here. It doesn't happen often.
 444          */
 445         if (unlikely(page->index)) {
 446                 zero_user(page, 0, PAGE_CACHE_SIZE);
 447                 return 0;
 448         }
 449
 450         error = gfs2_meta_inode_buffer(ip, &dibh);
 451         if (error)
 452                 return error;
 453
 454         kaddr = kmap_atomic(page, KM_USER0);
 455         memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
 456                ip->i_di.di_size);
 457         memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
 458         kunmap_atomic(kaddr, KM_USER0);
 459         flush_dcache_page(page);
 460         brelse(dibh);
 461         SetPageUptodate(page);
 462
 463         return 0;
 464 }
 465
 466
 467 /**
 468  * __gfs2_readpage - readpage
 469  * @file: The file to read a page for
 470  * @page: The page to read
 471  *
 472  * This is the core of gfs2's readpage. Its used by the internal file
 473  * reading code as in that case we already hold the glock. Also its
 474  * called by gfs2_readpage() once the required lock has been granted.
 475  *
 476  */
 477
 478 static int __gfs2_readpage(void *file, struct page *page)
 479 {
 480         struct gfs2_inode *ip = GFS2_I(page->mapping->host);
 481         struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
 482         int error;
 483
 484         if (gfs2_is_stuffed(ip)) {
 485                 error = stuffed_readpage(ip, page);
 486                 unlock_page(page);
 487         } else {
 488                 error = mpage_readpage(page, gfs2_block_map);
 489         }
 490
 491         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 492                 return -EIO;
 493
 494         return error;
 495 }
 496
 497 /**
 498  * gfs2_readpage - read a page of a file
 499  * @file: The file to read
 500  * @page: The page of the file
 501  *
 502  * This deals with the locking required. We use a trylock in order to
 503  * avoid the page lock / glock ordering problems returning AOP_TRUNCATED_PAGE
 504  * in the event that we are unable to get the lock.
 505  */
 506
 507 static int gfs2_readpage(struct file *file, struct page *page)
 508 {
 509         struct gfs2_inode *ip = GFS2_I(page->mapping->host);
 510         struct gfs2_holder gh;
 511         int error;
 512
 513         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh);
 514         error = gfs2_glock_nq_atime(&gh);
 515         if (unlikely(error)) {
 516                 unlock_page(page);
 517                 goto out;
 518         }
 519         error = __gfs2_readpage(file, page);
 520         gfs2_glock_dq(&gh);
 521 out:
 522         gfs2_holder_uninit(&gh);
 523         if (error == GLR_TRYFAILED) {
 524                 yield();
 525                 return AOP_TRUNCATED_PAGE;
 526         }
 527         return error;
 528 }
 529
 530 /**
 531  * gfs2_internal_read - read an internal file
 532  * @ip: The gfs2 inode
 533  * @ra_state: The readahead state (or NULL for no readahead)
 534  * @buf: The buffer to fill
 535  * @pos: The file position
 536  * @size: The amount to read
 537  *
 538  */
 539
 540 int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state,
 541                        char *buf, loff_t *pos, unsigned size)
 542 {
 543         struct address_space *mapping = ip->i_inode.i_mapping;
 544         unsigned long index = *pos / PAGE_CACHE_SIZE;
 545         unsigned offset = *pos & (PAGE_CACHE_SIZE - 1);
 546         unsigned copied = 0;
 547         unsigned amt;
 548         struct page *page;
 549         void *p;
 550
 551         do {
 552                 amt = size - copied;
 553                 if (offset + size > PAGE_CACHE_SIZE)
 554                         amt = PAGE_CACHE_SIZE - offset;
 555                 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
 556                 if (IS_ERR(page))
 557                         return PTR_ERR(page);
 558                 p = kmap_atomic(page, KM_USER0);
 559                 memcpy(buf + copied, p + offset, amt);
 560                 kunmap_atomic(p, KM_USER0);
 561                 mark_page_accessed(page);
 562                 page_cache_release(page);
 563                 copied += amt;
 564                 index++;
 565                 offset = 0;
 566         } while(copied < size);
 567         (*pos) += size;
 568         return size;
 569 }
 570
 571 /**
 572  * gfs2_readpages - Read a bunch of pages at once
 573  *
 574  * Some notes:
 575  * 1. This is only for readahead, so we can simply ignore any things
 576  *    which are slightly inconvenient (such as locking conflicts between
 577  *    the page lock and the glock) and return having done no I/O. Its
 578  *    obviously not something we'd want to do on too regular a basis.
 579  *    Any I/O we ignore at this time will be done via readpage later.
 580  * 2. We don't handle stuffed files here we let readpage do the honours.
 581  * 3. mpage_readpages() does most of the heavy lifting in the common case.
 582  * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
 583  */
 584
 585 static int gfs2_readpages(struct file *file, struct address_space *mapping,
 586                           struct list_head *pages, unsigned nr_pages)
 587 {
 588         struct inode *inode = mapping->host;
 589         struct gfs2_inode *ip = GFS2_I(inode);
 590         struct gfs2_sbd *sdp = GFS2_SB(inode);
 591         struct gfs2_holder gh;
 592         int ret;
 593
 594         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
 595         ret = gfs2_glock_nq_atime(&gh);
 596         if (unlikely(ret))
 597                 goto out_uninit;
 598         if (!gfs2_is_stuffed(ip))
 599                 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
 600         gfs2_glock_dq(&gh);
 601 out_uninit:
 602         gfs2_holder_uninit(&gh);
 603         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 604                 ret = -EIO;
 605         return ret;
 606 }
 607
 608 /**
 609  * gfs2_write_begin - Begin to write to a file
 610  * @file: The file to write to
 611  * @mapping: The mapping in which to write
 612  * @pos: The file offset at which to start writing
 613  * @len: Length of the write
 614  * @flags: Various flags
 615  * @pagep: Pointer to return the page
 616  * @fsdata: Pointer to return fs data (unused by GFS2)
 617  *
 618  * Returns: errno
 619  */
 620
 621 static int gfs2_write_begin(struct file *file, struct address_space *mapping,
 622                             loff_t pos, unsigned len, unsigned flags,
 623                             struct page **pagep, void **fsdata)
 624 {
 625         struct gfs2_inode *ip = GFS2_I(mapping->host);
 626         struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
 627         unsigned int data_blocks, ind_blocks, rblocks;
 628         int alloc_required;
 629         int error = 0;
 630         struct gfs2_alloc *al;
 631         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
 632         unsigned from = pos & (PAGE_CACHE_SIZE - 1);
 633         unsigned to = from + len;
 634         struct page *page;
 635
 636         gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME, &ip->i_gh);
 637         error = gfs2_glock_nq_atime(&ip->i_gh);
 638         if (unlikely(error))
 639                 goto out_uninit;
 640
 641         gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
 642         error = gfs2_write_alloc_required(ip, pos, len, &alloc_required);
 643         if (error)
 644                 goto out_unlock;
 645
 646         if (alloc_required) {
 647                 al = gfs2_alloc_get(ip);
 648                 if (!al) {
 649                         error = -ENOMEM;
 650                         goto out_unlock;
 651                 }
 652
 653                 error = gfs2_quota_lock_check(ip);
 654                 if (error)
 655                         goto out_alloc_put;
 656
 657                 al->al_requested = data_blocks + ind_blocks;
 658                 error = gfs2_inplace_reserve(ip);
 659                 if (error)
 660                         goto out_qunlock;
 661         }
 662
 663         rblocks = RES_DINODE + ind_blocks;
 664         if (gfs2_is_jdata(ip))
 665                 rblocks += data_blocks ? data_blocks : 1;
 666         if (ind_blocks || data_blocks)
 667                 rblocks += RES_STATFS + RES_QUOTA;
 668
 669         error = gfs2_trans_begin(sdp, rblocks,
 670                                  PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
 671         if (error)
 672                 goto out_trans_fail;
 673
 674         error = -ENOMEM;
 675         page = __grab_cache_page(mapping, index);
 676         *pagep = page;
 677         if (unlikely(!page))
 678                 goto out_endtrans;
 679
 680         if (gfs2_is_stuffed(ip)) {
 681                 error = 0;
 682                 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
 683                         error = gfs2_unstuff_dinode(ip, page);
 684                         if (error == 0)
 685                                 goto prepare_write;
 686                 } else if (!PageUptodate(page)) {
 687                         error = stuffed_readpage(ip, page);
 688                 }
 689                 goto out;
 690         }
 691
 692 prepare_write:
 693         error = block_prepare_write(page, from, to, gfs2_block_map);
 694 out:
 695         if (error == 0)
 696                 return 0;
 697
 698         page_cache_release(page);
 699         if (pos + len > ip->i_inode.i_size)
 700                 vmtruncate(&ip->i_inode, ip->i_inode.i_size);
 701 out_endtrans:
 702         gfs2_trans_end(sdp);
 703 out_trans_fail:
 704         if (alloc_required) {
 705                 gfs2_inplace_release(ip);
 706 out_qunlock:
 707                 gfs2_quota_unlock(ip);
 708 out_alloc_put:
 709                 gfs2_alloc_put(ip);
 710         }
 711 out_unlock:
 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_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 726         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 727         u64 fs_total, new_free;
 728
 729         /* Total up the file system space, according to the latest rindex. */
 730         fs_total = gfs2_ri_total(sdp);
 731
 732         spin_lock(&sdp->sd_statfs_spin);
 733         if (fs_total > (m_sc->sc_total + l_sc->sc_total))
 734                 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
 735         else
 736                 new_free = 0;
 737         spin_unlock(&sdp->sd_statfs_spin);
 738         fs_warn(sdp, "File system extended by %llu blocks.\n",
 739                 (unsigned long long)new_free);
 740         gfs2_statfs_change(sdp, new_free, new_free, 0);
 741 }
 742
 743 /**
 744  * gfs2_stuffed_write_end - Write end for stuffed files
 745  * @inode: The inode
 746  * @dibh: The buffer_head containing the on-disk inode
 747  * @pos: The file position
 748  * @len: The length of the write
 749  * @copied: How much was actually copied by the VFS
 750  * @page: The page
 751  *
 752  * This copies the data from the page into the inode block after
 753  * the inode data structure itself.
 754  *
 755  * Returns: errno
 756  */
 757 static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
 758                                   loff_t pos, unsigned len, unsigned copied,
 759                                   struct page *page)
 760 {
 761         struct gfs2_inode *ip = GFS2_I(inode);
 762         struct gfs2_sbd *sdp = GFS2_SB(inode);
 763         u64 to = pos + copied;
 764         void *kaddr;
 765         unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
 766         struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
 767
 768         BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
 769         kaddr = kmap_atomic(page, KM_USER0);
 770         memcpy(buf + pos, kaddr + pos, copied);
 771         memset(kaddr + pos + copied, 0, len - copied);
 772         flush_dcache_page(page);
 773         kunmap_atomic(kaddr, KM_USER0);
 774
 775         if (!PageUptodate(page))
 776                 SetPageUptodate(page);
 777         unlock_page(page);
 778         page_cache_release(page);
 779
 780         if (inode->i_size < to) {
 781                 i_size_write(inode, to);
 782                 ip->i_di.di_size = inode->i_size;
 783                 di->di_size = cpu_to_be64(inode->i_size);
 784                 mark_inode_dirty(inode);
 785         }
 786
 787         if (inode == sdp->sd_rindex)
 788                 adjust_fs_space(inode);
 789
 790         brelse(dibh);
 791         gfs2_trans_end(sdp);
 792         gfs2_glock_dq(&ip->i_gh);
 793         gfs2_holder_uninit(&ip->i_gh);
 794         return copied;
 795 }
 796
 797 /**
 798  * gfs2_write_end
 799  * @file: The file to write to
 800  * @mapping: The address space to write to
 801  * @pos: The file position
 802  * @len: The length of the data
 803  * @copied:
 804  * @page: The page that has been written
 805  * @fsdata: The fsdata (unused in GFS2)
 806  *
 807  * The main write_end function for GFS2. We have a separate one for
 808  * stuffed files as they are slightly different, otherwise we just
 809  * put our locking around the VFS provided functions.
 810  *
 811  * Returns: errno
 812  */
 813
 814 static int gfs2_write_end(struct file *file, struct address_space *mapping,
 815                           loff_t pos, unsigned len, unsigned copied,
 816                           struct page *page, void *fsdata)
 817 {
 818         struct inode *inode = page->mapping->host;
 819         struct gfs2_inode *ip = GFS2_I(inode);
 820         struct gfs2_sbd *sdp = GFS2_SB(inode);
 821         struct buffer_head *dibh;
 822         struct gfs2_alloc *al = ip->i_alloc;
 823         struct gfs2_dinode *di;
 824         unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
 825         unsigned int to = from + len;
 826         int ret;
 827
 828         BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
 829
 830         ret = gfs2_meta_inode_buffer(ip, &dibh);
 831         if (unlikely(ret)) {
 832                 unlock_page(page);
 833                 page_cache_release(page);
 834                 goto failed;
 835         }
 836
 837         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
 838
 839         if (gfs2_is_stuffed(ip))
 840                 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
 841
 842         if (!gfs2_is_writeback(ip))
 843                 gfs2_page_add_databufs(ip, page, from, to);
 844
 845         ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
 846
 847         if (likely(ret >= 0) && (inode->i_size > ip->i_di.di_size)) {
 848                 di = (struct gfs2_dinode *)dibh->b_data;
 849                 ip->i_di.di_size = inode->i_size;
 850                 di->di_size = cpu_to_be64(inode->i_size);
 851                 mark_inode_dirty(inode);
 852         }
 853
 854         if (inode == sdp->sd_rindex)
 855                 adjust_fs_space(inode);
 856
 857         brelse(dibh);
 858         gfs2_trans_end(sdp);
 859 failed:
 860         if (al) {
 861                 gfs2_inplace_release(ip);
 862                 gfs2_quota_unlock(ip);
 863                 gfs2_alloc_put(ip);
 864         }
 865         gfs2_glock_dq(&ip->i_gh);
 866         gfs2_holder_uninit(&ip->i_gh);
 867         return ret;
 868 }
 869
 870 /**
 871  * gfs2_set_page_dirty - Page dirtying function
 872  * @page: The page to dirty
 873  *
 874  * Returns: 1 if it dirtyed the page, or 0 otherwise
 875  */
 876
 877 static int gfs2_set_page_dirty(struct page *page)
 878 {
 879         SetPageChecked(page);
 880         return __set_page_dirty_buffers(page);
 881 }
 882
 883 /**
 884  * gfs2_bmap - Block map function
 885  * @mapping: Address space info
 886  * @lblock: The block to map
 887  *
 888  * Returns: The disk address for the block or 0 on hole or error
 889  */
 890
 891 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
 892 {
 893         struct gfs2_inode *ip = GFS2_I(mapping->host);
 894         struct gfs2_holder i_gh;
 895         sector_t dblock = 0;
 896         int error;
 897
 898         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
 899         if (error)
 900                 return 0;
 901
 902         if (!gfs2_is_stuffed(ip))
 903                 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
 904
 905         gfs2_glock_dq_uninit(&i_gh);
 906
 907         return dblock;
 908 }
 909
 910 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
 911 {
 912         struct gfs2_bufdata *bd;
 913
 914         lock_buffer(bh);
 915         gfs2_log_lock(sdp);
 916         clear_buffer_dirty(bh);
 917         bd = bh->b_private;
 918         if (bd) {
 919                 if (!list_empty(&bd->bd_le.le_list) && !buffer_pinned(bh))
 920                         list_del_init(&bd->bd_le.le_list);
 921                 else
 922                         gfs2_remove_from_journal(bh, current->journal_info, 0);
 923         }
 924         bh->b_bdev = NULL;
 925         clear_buffer_mapped(bh);
 926         clear_buffer_req(bh);
 927         clear_buffer_new(bh);
 928         gfs2_log_unlock(sdp);
 929         unlock_buffer(bh);
 930 }
 931
 932 static void gfs2_invalidatepage(struct page *page, unsigned long offset)
 933 {
 934         struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
 935         struct buffer_head *bh, *head;
 936         unsigned long pos = 0;
 937
 938         BUG_ON(!PageLocked(page));
 939         if (offset == 0)
 940                 ClearPageChecked(page);
 941         if (!page_has_buffers(page))
 942                 goto out;
 943
 944         bh = head = page_buffers(page);
 945         do {
 946                 if (offset <= pos)
 947                         gfs2_discard(sdp, bh);
 948                 pos += bh->b_size;
 949                 bh = bh->b_this_page;
 950         } while (bh != head);
 951 out:
 952         if (offset == 0)
 953                 try_to_release_page(page, 0);
 954 }
 955
 956 /**
 957  * gfs2_ok_for_dio - check that dio is valid on this file
 958  * @ip: The inode
 959  * @rw: READ or WRITE
 960  * @offset: The offset at which we are reading or writing
 961  *
 962  * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
 963  *          1 (to accept the i/o request)
 964  */
 965 static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
 966 {
 967         /*
 968          * Should we return an error here? I can't see that O_DIRECT for
 969          * a stuffed file makes any sense. For now we'll silently fall
 970          * back to buffered I/O
 971          */
 972         if (gfs2_is_stuffed(ip))
 973                 return 0;
 974
 975         if (offset > i_size_read(&ip->i_inode))
 976                 return 0;
 977         return 1;
 978 }
 979
 980
 981
 982 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
 983                               const struct iovec *iov, loff_t offset,
 984                               unsigned long nr_segs)
 985 {
 986         struct file *file = iocb->ki_filp;
 987         struct inode *inode = file->f_mapping->host;
 988         struct gfs2_inode *ip = GFS2_I(inode);
 989         struct gfs2_holder gh;
 990         int rv;
 991
 992         /*
 993          * Deferred lock, even if its a write, since we do no allocation
 994          * on this path. All we need change is atime, and this lock mode
 995          * ensures that other nodes have flushed their buffered read caches
 996          * (i.e. their page cache entries for this inode). We do not,
 997          * unfortunately have the option of only flushing a range like
 998          * the VFS does.
 999          */
1000         gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh);
1001         rv = gfs2_glock_nq_atime(&gh);
1002         if (rv)
1003                 return rv;
1004         rv = gfs2_ok_for_dio(ip, rw, offset);
1005         if (rv != 1)
1006                 goto out; /* dio not valid, fall back to buffered i/o */
1007
1008         rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
1009                                            iov, offset, nr_segs,
1010                                            gfs2_get_block_direct, NULL);
1011 out:
1012         gfs2_glock_dq_m(1, &gh);
1013         gfs2_holder_uninit(&gh);
1014         return rv;
1015 }
1016
1017 /**
1018  * gfs2_releasepage - free the metadata associated with a page
1019  * @page: the page that's being released
1020  * @gfp_mask: passed from Linux VFS, ignored by us
1021  *
1022  * Call try_to_free_buffers() if the buffers in this page can be
1023  * released.
1024  *
1025  * Returns: 0
1026  */
1027
1028 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1029 {
1030         struct inode *aspace = page->mapping->host;
1031         struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
1032         struct buffer_head *bh, *head;
1033         struct gfs2_bufdata *bd;
1034
1035         if (!page_has_buffers(page))
1036                 return 0;
1037
1038         gfs2_log_lock(sdp);
1039         head = bh = page_buffers(page);
1040         do {
1041                 if (atomic_read(&bh->b_count))
1042                         goto cannot_release;
1043                 bd = bh->b_private;
1044                 if (bd && bd->bd_ail)
1045                         goto cannot_release;
1046                 gfs2_assert_warn(sdp, !buffer_pinned(bh));
1047                 gfs2_assert_warn(sdp, !buffer_dirty(bh));
1048                 bh = bh->b_this_page;
1049         } while(bh != head);
1050         gfs2_log_unlock(sdp);
1051
1052         head = bh = page_buffers(page);
1053         do {
1054                 gfs2_log_lock(sdp);
1055                 bd = bh->b_private;
1056                 if (bd) {
1057                         gfs2_assert_warn(sdp, bd->bd_bh == bh);
1058                         gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
1059                         if (!list_empty(&bd->bd_le.le_list)) {
1060                                 if (!buffer_pinned(bh))
1061                                         list_del_init(&bd->bd_le.le_list);
1062                                 else
1063                                         bd = NULL;
1064                         }
1065                         if (bd)
1066                                 bd->bd_bh = NULL;
1067                         bh->b_private = NULL;
1068                 }
1069                 gfs2_log_unlock(sdp);
1070                 if (bd)
1071                         kmem_cache_free(gfs2_bufdata_cachep, bd);
1072
1073                 bh = bh->b_this_page;
1074         } while (bh != head);
1075
1076         return try_to_free_buffers(page);
1077 cannot_release:
1078         gfs2_log_unlock(sdp);
1079         return 0;
1080 }
1081
1082 static const struct address_space_operations gfs2_writeback_aops = {
1083         .writepage = gfs2_writeback_writepage,
1084         .writepages = gfs2_writeback_writepages,
1085         .readpage = gfs2_readpage,
1086         .readpages = gfs2_readpages,
1087         .sync_page = block_sync_page,
1088         .write_begin = gfs2_write_begin,
1089         .write_end = gfs2_write_end,
1090         .bmap = gfs2_bmap,
1091         .invalidatepage = gfs2_invalidatepage,
1092         .releasepage = gfs2_releasepage,
1093         .direct_IO = gfs2_direct_IO,
1094         .migratepage = buffer_migrate_page,
1095 };
1096
1097 static const struct address_space_operations gfs2_ordered_aops = {
1098         .writepage = gfs2_ordered_writepage,
1099         .readpage = gfs2_readpage,
1100         .readpages = gfs2_readpages,
1101         .sync_page = block_sync_page,
1102         .write_begin = gfs2_write_begin,
1103         .write_end = gfs2_write_end,
1104         .set_page_dirty = gfs2_set_page_dirty,
1105         .bmap = gfs2_bmap,
1106         .invalidatepage = gfs2_invalidatepage,
1107         .releasepage = gfs2_releasepage,
1108         .direct_IO = gfs2_direct_IO,
1109         .migratepage = buffer_migrate_page,
1110 };
1111
1112 static const struct address_space_operations gfs2_jdata_aops = {
1113         .writepage = gfs2_jdata_writepage,
1114         .writepages = gfs2_jdata_writepages,
1115         .readpage = gfs2_readpage,
1116         .readpages = gfs2_readpages,
1117         .sync_page = block_sync_page,
1118         .write_begin = gfs2_write_begin,
1119         .write_end = gfs2_write_end,
1120         .set_page_dirty = gfs2_set_page_dirty,
1121         .bmap = gfs2_bmap,
1122         .invalidatepage = gfs2_invalidatepage,
1123         .releasepage = gfs2_releasepage,
1124 };
1125
1126 void gfs2_set_aops(struct inode *inode)
1127 {
1128         struct gfs2_inode *ip = GFS2_I(inode);
1129
1130         if (gfs2_is_writeback(ip))
1131                 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1132         else if (gfs2_is_ordered(ip))
1133                 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1134         else if (gfs2_is_jdata(ip))
1135                 inode->i_mapping->a_ops = &gfs2_jdata_aops;
1136         else
1137                 BUG();
1138 }
1139