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/uio.h> 24 #include <trace/events/writeback.h> 25 #include <linux/sched/signal.h> 26 27 #include "gfs2.h" 28 #include "incore.h" 29 #include "bmap.h" 30 #include "glock.h" 31 #include "inode.h" 32 #include "log.h" 33 #include "meta_io.h" 34 #include "quota.h" 35 #include "trans.h" 36 #include "rgrp.h" 37 #include "super.h" 38 #include "util.h" 39 #include "glops.h" 40 #include "aops.h" 41 42 43 void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page, 44 unsigned int from, unsigned int len) 45 { 46 struct buffer_head *head = page_buffers(page); 47 unsigned int bsize = head->b_size; 48 struct buffer_head *bh; 49 unsigned int to = from + len; 50 unsigned int start, end; 51 52 for (bh = head, start = 0; bh != head || !start; 53 bh = bh->b_this_page, start = end) { 54 end = start + bsize; 55 if (end <= from) 56 continue; 57 if (start >= to) 58 break; 59 set_buffer_uptodate(bh); 60 gfs2_trans_add_data(ip->i_gl, bh); 61 } 62 } 63 64 /** 65 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block 66 * @inode: The inode 67 * @lblock: The block number to look up 68 * @bh_result: The buffer head to return the result in 69 * @create: Non-zero if we may add block to the file 70 * 71 * Returns: errno 72 */ 73 74 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock, 75 struct buffer_head *bh_result, int create) 76 { 77 int error; 78 79 error = gfs2_block_map(inode, lblock, bh_result, 0); 80 if (error) 81 return error; 82 if (!buffer_mapped(bh_result)) 83 return -EIO; 84 return 0; 85 } 86 87 static int gfs2_get_block_direct(struct inode *inode, sector_t lblock, 88 struct buffer_head *bh_result, int create) 89 { 90 return gfs2_block_map(inode, lblock, bh_result, 0); 91 } 92 93 /** 94 * gfs2_writepage_common - Common bits of writepage 95 * @page: The page to be written 96 * @wbc: The writeback control 97 * 98 * Returns: 1 if writepage is ok, otherwise an error code or zero if no error. 99 */ 100 101 static int gfs2_writepage_common(struct page *page, 102 struct writeback_control *wbc) 103 { 104 struct inode *inode = page->mapping->host; 105 struct gfs2_inode *ip = GFS2_I(inode); 106 struct gfs2_sbd *sdp = GFS2_SB(inode); 107 loff_t i_size = i_size_read(inode); 108 pgoff_t end_index = i_size >> PAGE_SHIFT; 109 unsigned offset; 110 111 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) 112 goto out; 113 if (current->journal_info) 114 goto redirty; 115 /* Is the page fully outside i_size? (truncate in progress) */ 116 offset = i_size & (PAGE_SIZE-1); 117 if (page->index > end_index || (page->index == end_index && !offset)) { 118 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); 119 goto out; 120 } 121 return 1; 122 redirty: 123 redirty_page_for_writepage(wbc, page); 124 out: 125 unlock_page(page); 126 return 0; 127 } 128 129 /** 130 * gfs2_writepage - Write page for writeback mappings 131 * @page: The page 132 * @wbc: The writeback control 133 * 134 */ 135 136 static int gfs2_writepage(struct page *page, struct writeback_control *wbc) 137 { 138 int ret; 139 140 ret = gfs2_writepage_common(page, wbc); 141 if (ret <= 0) 142 return ret; 143 144 return nobh_writepage(page, gfs2_get_block_noalloc, wbc); 145 } 146 147 /* This is the same as calling block_write_full_page, but it also 148 * writes pages outside of i_size 149 */ 150 static int gfs2_write_full_page(struct page *page, get_block_t *get_block, 151 struct writeback_control *wbc) 152 { 153 struct inode * const inode = page->mapping->host; 154 loff_t i_size = i_size_read(inode); 155 const pgoff_t end_index = i_size >> PAGE_SHIFT; 156 unsigned offset; 157 158 /* 159 * The page straddles i_size. It must be zeroed out on each and every 160 * writepage invocation because it may be mmapped. "A file is mapped 161 * in multiples of the page size. For a file that is not a multiple of 162 * the page size, the remaining memory is zeroed when mapped, and 163 * writes to that region are not written out to the file." 164 */ 165 offset = i_size & (PAGE_SIZE-1); 166 if (page->index == end_index && offset) 167 zero_user_segment(page, offset, PAGE_SIZE); 168 169 return __block_write_full_page(inode, page, get_block, wbc, 170 end_buffer_async_write); 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 BIT(BH_Dirty)|BIT(BH_Uptodate)); 195 } 196 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize); 197 } 198 return gfs2_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 * @wbc: The writeback control 205 * 206 * Returns: errno 207 * 208 */ 209 210 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc) 211 { 212 struct inode *inode = page->mapping->host; 213 struct gfs2_inode *ip = GFS2_I(inode); 214 struct gfs2_sbd *sdp = GFS2_SB(inode); 215 int ret; 216 217 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) 218 goto out; 219 if (PageChecked(page) || current->journal_info) 220 goto out_ignore; 221 ret = __gfs2_jdata_writepage(page, wbc); 222 return ret; 223 224 out_ignore: 225 redirty_page_for_writepage(wbc, page); 226 out: 227 unlock_page(page); 228 return 0; 229 } 230 231 /** 232 * gfs2_writepages - Write a bunch of dirty pages back to disk 233 * @mapping: The mapping to write 234 * @wbc: Write-back control 235 * 236 * Used for both ordered and writeback modes. 237 */ 238 static int gfs2_writepages(struct address_space *mapping, 239 struct writeback_control *wbc) 240 { 241 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); 242 int ret = mpage_writepages(mapping, wbc, gfs2_get_block_noalloc); 243 244 /* 245 * Even if we didn't write any pages here, we might still be holding 246 * dirty pages in the ail. We forcibly flush the ail because we don't 247 * want balance_dirty_pages() to loop indefinitely trying to write out 248 * pages held in the ail that it can't find. 249 */ 250 if (ret == 0) 251 set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags); 252 253 return ret; 254 } 255 256 /** 257 * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages 258 * @mapping: The mapping 259 * @wbc: The writeback control 260 * @pvec: The vector of pages 261 * @nr_pages: The number of pages to write 262 * @done_index: Page index 263 * 264 * Returns: non-zero if loop should terminate, zero otherwise 265 */ 266 267 static int gfs2_write_jdata_pagevec(struct address_space *mapping, 268 struct writeback_control *wbc, 269 struct pagevec *pvec, 270 int nr_pages, 271 pgoff_t *done_index) 272 { 273 struct inode *inode = mapping->host; 274 struct gfs2_sbd *sdp = GFS2_SB(inode); 275 unsigned nrblocks = nr_pages * (PAGE_SIZE/inode->i_sb->s_blocksize); 276 int i; 277 int ret; 278 279 ret = gfs2_trans_begin(sdp, nrblocks, nrblocks); 280 if (ret < 0) 281 return ret; 282 283 for(i = 0; i < nr_pages; i++) { 284 struct page *page = pvec->pages[i]; 285 286 *done_index = page->index; 287 288 lock_page(page); 289 290 if (unlikely(page->mapping != mapping)) { 291 continue_unlock: 292 unlock_page(page); 293 continue; 294 } 295 296 if (!PageDirty(page)) { 297 /* someone wrote it for us */ 298 goto continue_unlock; 299 } 300 301 if (PageWriteback(page)) { 302 if (wbc->sync_mode != WB_SYNC_NONE) 303 wait_on_page_writeback(page); 304 else 305 goto continue_unlock; 306 } 307 308 BUG_ON(PageWriteback(page)); 309 if (!clear_page_dirty_for_io(page)) 310 goto continue_unlock; 311 312 trace_wbc_writepage(wbc, inode_to_bdi(inode)); 313 314 ret = __gfs2_jdata_writepage(page, wbc); 315 if (unlikely(ret)) { 316 if (ret == AOP_WRITEPAGE_ACTIVATE) { 317 unlock_page(page); 318 ret = 0; 319 } else { 320 321 /* 322 * done_index is set past this page, 323 * so media errors will not choke 324 * background writeout for the entire 325 * file. This has consequences for 326 * range_cyclic semantics (ie. it may 327 * not be suitable for data integrity 328 * writeout). 329 */ 330 *done_index = page->index + 1; 331 ret = 1; 332 break; 333 } 334 } 335 336 /* 337 * We stop writing back only if we are not doing 338 * integrity sync. In case of integrity sync we have to 339 * keep going until we have written all the pages 340 * we tagged for writeback prior to entering this loop. 341 */ 342 if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) { 343 ret = 1; 344 break; 345 } 346 347 } 348 gfs2_trans_end(sdp); 349 return ret; 350 } 351 352 /** 353 * gfs2_write_cache_jdata - Like write_cache_pages but different 354 * @mapping: The mapping to write 355 * @wbc: The writeback control 356 * 357 * The reason that we use our own function here is that we need to 358 * start transactions before we grab page locks. This allows us 359 * to get the ordering right. 360 */ 361 362 static int gfs2_write_cache_jdata(struct address_space *mapping, 363 struct writeback_control *wbc) 364 { 365 int ret = 0; 366 int done = 0; 367 struct pagevec pvec; 368 int nr_pages; 369 pgoff_t uninitialized_var(writeback_index); 370 pgoff_t index; 371 pgoff_t end; 372 pgoff_t done_index; 373 int cycled; 374 int range_whole = 0; 375 int tag; 376 377 pagevec_init(&pvec); 378 if (wbc->range_cyclic) { 379 writeback_index = mapping->writeback_index; /* prev offset */ 380 index = writeback_index; 381 if (index == 0) 382 cycled = 1; 383 else 384 cycled = 0; 385 end = -1; 386 } else { 387 index = wbc->range_start >> PAGE_SHIFT; 388 end = wbc->range_end >> PAGE_SHIFT; 389 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 390 range_whole = 1; 391 cycled = 1; /* ignore range_cyclic tests */ 392 } 393 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 394 tag = PAGECACHE_TAG_TOWRITE; 395 else 396 tag = PAGECACHE_TAG_DIRTY; 397 398 retry: 399 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 400 tag_pages_for_writeback(mapping, index, end); 401 done_index = index; 402 while (!done && (index <= end)) { 403 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, 404 tag); 405 if (nr_pages == 0) 406 break; 407 408 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, &done_index); 409 if (ret) 410 done = 1; 411 if (ret > 0) 412 ret = 0; 413 pagevec_release(&pvec); 414 cond_resched(); 415 } 416 417 if (!cycled && !done) { 418 /* 419 * range_cyclic: 420 * We hit the last page and there is more work to be done: wrap 421 * back to the start of the file 422 */ 423 cycled = 1; 424 index = 0; 425 end = writeback_index - 1; 426 goto retry; 427 } 428 429 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 430 mapping->writeback_index = done_index; 431 432 return ret; 433 } 434 435 436 /** 437 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk 438 * @mapping: The mapping to write 439 * @wbc: The writeback control 440 * 441 */ 442 443 static int gfs2_jdata_writepages(struct address_space *mapping, 444 struct writeback_control *wbc) 445 { 446 struct gfs2_inode *ip = GFS2_I(mapping->host); 447 struct gfs2_sbd *sdp = GFS2_SB(mapping->host); 448 int ret; 449 450 ret = gfs2_write_cache_jdata(mapping, wbc); 451 if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) { 452 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL | 453 GFS2_LFC_JDATA_WPAGES); 454 ret = gfs2_write_cache_jdata(mapping, wbc); 455 } 456 return ret; 457 } 458 459 /** 460 * stuffed_readpage - Fill in a Linux page with stuffed file data 461 * @ip: the inode 462 * @page: the page 463 * 464 * Returns: errno 465 */ 466 467 int stuffed_readpage(struct gfs2_inode *ip, struct page *page) 468 { 469 struct buffer_head *dibh; 470 u64 dsize = i_size_read(&ip->i_inode); 471 void *kaddr; 472 int error; 473 474 /* 475 * Due to the order of unstuffing files and ->fault(), we can be 476 * asked for a zero page in the case of a stuffed file being extended, 477 * so we need to supply one here. It doesn't happen often. 478 */ 479 if (unlikely(page->index)) { 480 zero_user(page, 0, PAGE_SIZE); 481 SetPageUptodate(page); 482 return 0; 483 } 484 485 error = gfs2_meta_inode_buffer(ip, &dibh); 486 if (error) 487 return error; 488 489 kaddr = kmap_atomic(page); 490 if (dsize > gfs2_max_stuffed_size(ip)) 491 dsize = gfs2_max_stuffed_size(ip); 492 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize); 493 memset(kaddr + dsize, 0, PAGE_SIZE - dsize); 494 kunmap_atomic(kaddr); 495 flush_dcache_page(page); 496 brelse(dibh); 497 SetPageUptodate(page); 498 499 return 0; 500 } 501 502 503 /** 504 * __gfs2_readpage - readpage 505 * @file: The file to read a page for 506 * @page: The page to read 507 * 508 * This is the core of gfs2's readpage. It's used by the internal file 509 * reading code as in that case we already hold the glock. Also it's 510 * called by gfs2_readpage() once the required lock has been granted. 511 */ 512 513 static int __gfs2_readpage(void *file, struct page *page) 514 { 515 struct gfs2_inode *ip = GFS2_I(page->mapping->host); 516 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); 517 int error; 518 519 if (gfs2_is_stuffed(ip)) { 520 error = stuffed_readpage(ip, page); 521 unlock_page(page); 522 } else { 523 error = mpage_readpage(page, gfs2_block_map); 524 } 525 526 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 527 return -EIO; 528 529 return error; 530 } 531 532 /** 533 * gfs2_readpage - read a page of a file 534 * @file: The file to read 535 * @page: The page of the file 536 * 537 * This deals with the locking required. We have to unlock and 538 * relock the page in order to get the locking in the right 539 * order. 540 */ 541 542 static int gfs2_readpage(struct file *file, struct page *page) 543 { 544 struct address_space *mapping = page->mapping; 545 struct gfs2_inode *ip = GFS2_I(mapping->host); 546 struct gfs2_holder gh; 547 int error; 548 549 unlock_page(page); 550 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 551 error = gfs2_glock_nq(&gh); 552 if (unlikely(error)) 553 goto out; 554 error = AOP_TRUNCATED_PAGE; 555 lock_page(page); 556 if (page->mapping == mapping && !PageUptodate(page)) 557 error = __gfs2_readpage(file, page); 558 else 559 unlock_page(page); 560 gfs2_glock_dq(&gh); 561 out: 562 gfs2_holder_uninit(&gh); 563 if (error && error != AOP_TRUNCATED_PAGE) 564 lock_page(page); 565 return error; 566 } 567 568 /** 569 * gfs2_internal_read - read an internal file 570 * @ip: The gfs2 inode 571 * @buf: The buffer to fill 572 * @pos: The file position 573 * @size: The amount to read 574 * 575 */ 576 577 int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos, 578 unsigned size) 579 { 580 struct address_space *mapping = ip->i_inode.i_mapping; 581 unsigned long index = *pos / PAGE_SIZE; 582 unsigned offset = *pos & (PAGE_SIZE - 1); 583 unsigned copied = 0; 584 unsigned amt; 585 struct page *page; 586 void *p; 587 588 do { 589 amt = size - copied; 590 if (offset + size > PAGE_SIZE) 591 amt = PAGE_SIZE - offset; 592 page = read_cache_page(mapping, index, __gfs2_readpage, NULL); 593 if (IS_ERR(page)) 594 return PTR_ERR(page); 595 p = kmap_atomic(page); 596 memcpy(buf + copied, p + offset, amt); 597 kunmap_atomic(p); 598 put_page(page); 599 copied += amt; 600 index++; 601 offset = 0; 602 } while(copied < size); 603 (*pos) += size; 604 return size; 605 } 606 607 /** 608 * gfs2_readpages - Read a bunch of pages at once 609 * @file: The file to read from 610 * @mapping: Address space info 611 * @pages: List of pages to read 612 * @nr_pages: Number of pages to read 613 * 614 * Some notes: 615 * 1. This is only for readahead, so we can simply ignore any things 616 * which are slightly inconvenient (such as locking conflicts between 617 * the page lock and the glock) and return having done no I/O. Its 618 * obviously not something we'd want to do on too regular a basis. 619 * Any I/O we ignore at this time will be done via readpage later. 620 * 2. We don't handle stuffed files here we let readpage do the honours. 621 * 3. mpage_readpages() does most of the heavy lifting in the common case. 622 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places. 623 */ 624 625 static int gfs2_readpages(struct file *file, struct address_space *mapping, 626 struct list_head *pages, unsigned nr_pages) 627 { 628 struct inode *inode = mapping->host; 629 struct gfs2_inode *ip = GFS2_I(inode); 630 struct gfs2_sbd *sdp = GFS2_SB(inode); 631 struct gfs2_holder gh; 632 int ret; 633 634 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 635 ret = gfs2_glock_nq(&gh); 636 if (unlikely(ret)) 637 goto out_uninit; 638 if (!gfs2_is_stuffed(ip)) 639 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map); 640 gfs2_glock_dq(&gh); 641 out_uninit: 642 gfs2_holder_uninit(&gh); 643 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 644 ret = -EIO; 645 return ret; 646 } 647 648 /** 649 * gfs2_write_begin - Begin to write to a file 650 * @file: The file to write to 651 * @mapping: The mapping in which to write 652 * @pos: The file offset at which to start writing 653 * @len: Length of the write 654 * @flags: Various flags 655 * @pagep: Pointer to return the page 656 * @fsdata: Pointer to return fs data (unused by GFS2) 657 * 658 * Returns: errno 659 */ 660 661 static int gfs2_write_begin(struct file *file, struct address_space *mapping, 662 loff_t pos, unsigned len, unsigned flags, 663 struct page **pagep, void **fsdata) 664 { 665 struct gfs2_inode *ip = GFS2_I(mapping->host); 666 struct gfs2_sbd *sdp = GFS2_SB(mapping->host); 667 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); 668 unsigned int data_blocks = 0, ind_blocks = 0, rblocks; 669 unsigned requested = 0; 670 int alloc_required; 671 int error = 0; 672 pgoff_t index = pos >> PAGE_SHIFT; 673 unsigned from = pos & (PAGE_SIZE - 1); 674 struct page *page; 675 676 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh); 677 error = gfs2_glock_nq(&ip->i_gh); 678 if (unlikely(error)) 679 goto out_uninit; 680 if (&ip->i_inode == sdp->sd_rindex) { 681 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 682 GL_NOCACHE, &m_ip->i_gh); 683 if (unlikely(error)) { 684 gfs2_glock_dq(&ip->i_gh); 685 goto out_uninit; 686 } 687 } 688 689 alloc_required = gfs2_write_alloc_required(ip, pos, len); 690 691 if (alloc_required || gfs2_is_jdata(ip)) 692 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks); 693 694 if (alloc_required) { 695 struct gfs2_alloc_parms ap = { .aflags = 0, }; 696 requested = data_blocks + ind_blocks; 697 ap.target = requested; 698 error = gfs2_quota_lock_check(ip, &ap); 699 if (error) 700 goto out_unlock; 701 702 error = gfs2_inplace_reserve(ip, &ap); 703 if (error) 704 goto out_qunlock; 705 } 706 707 rblocks = RES_DINODE + ind_blocks; 708 if (gfs2_is_jdata(ip)) 709 rblocks += data_blocks ? data_blocks : 1; 710 if (ind_blocks || data_blocks) 711 rblocks += RES_STATFS + RES_QUOTA; 712 if (&ip->i_inode == sdp->sd_rindex) 713 rblocks += 2 * RES_STATFS; 714 if (alloc_required) 715 rblocks += gfs2_rg_blocks(ip, requested); 716 717 error = gfs2_trans_begin(sdp, rblocks, 718 PAGE_SIZE/sdp->sd_sb.sb_bsize); 719 if (error) 720 goto out_trans_fail; 721 722 error = -ENOMEM; 723 flags |= AOP_FLAG_NOFS; 724 page = grab_cache_page_write_begin(mapping, index, flags); 725 *pagep = page; 726 if (unlikely(!page)) 727 goto out_endtrans; 728 729 if (gfs2_is_stuffed(ip)) { 730 error = 0; 731 if (pos + len > gfs2_max_stuffed_size(ip)) { 732 error = gfs2_unstuff_dinode(ip, page); 733 if (error == 0) 734 goto prepare_write; 735 } else if (!PageUptodate(page)) { 736 error = stuffed_readpage(ip, page); 737 } 738 goto out; 739 } 740 741 prepare_write: 742 error = __block_write_begin(page, from, len, gfs2_block_map); 743 out: 744 if (error == 0) 745 return 0; 746 747 unlock_page(page); 748 put_page(page); 749 750 gfs2_trans_end(sdp); 751 if (alloc_required) { 752 gfs2_inplace_release(ip); 753 if (pos + len > ip->i_inode.i_size) 754 gfs2_trim_blocks(&ip->i_inode); 755 } 756 goto out_qunlock; 757 758 out_endtrans: 759 gfs2_trans_end(sdp); 760 out_trans_fail: 761 if (alloc_required) 762 gfs2_inplace_release(ip); 763 out_qunlock: 764 if (alloc_required) 765 gfs2_quota_unlock(ip); 766 out_unlock: 767 if (&ip->i_inode == sdp->sd_rindex) { 768 gfs2_glock_dq(&m_ip->i_gh); 769 gfs2_holder_uninit(&m_ip->i_gh); 770 } 771 gfs2_glock_dq(&ip->i_gh); 772 out_uninit: 773 gfs2_holder_uninit(&ip->i_gh); 774 return error; 775 } 776 777 /** 778 * adjust_fs_space - Adjusts the free space available due to gfs2_grow 779 * @inode: the rindex inode 780 */ 781 void adjust_fs_space(struct inode *inode) 782 { 783 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; 784 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); 785 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); 786 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; 787 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 788 struct buffer_head *m_bh, *l_bh; 789 u64 fs_total, new_free; 790 791 /* Total up the file system space, according to the latest rindex. */ 792 fs_total = gfs2_ri_total(sdp); 793 if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0) 794 return; 795 796 spin_lock(&sdp->sd_statfs_spin); 797 gfs2_statfs_change_in(m_sc, m_bh->b_data + 798 sizeof(struct gfs2_dinode)); 799 if (fs_total > (m_sc->sc_total + l_sc->sc_total)) 800 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total); 801 else 802 new_free = 0; 803 spin_unlock(&sdp->sd_statfs_spin); 804 fs_warn(sdp, "File system extended by %llu blocks.\n", 805 (unsigned long long)new_free); 806 gfs2_statfs_change(sdp, new_free, new_free, 0); 807 808 if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0) 809 goto out; 810 update_statfs(sdp, m_bh, l_bh); 811 brelse(l_bh); 812 out: 813 brelse(m_bh); 814 } 815 816 /** 817 * gfs2_stuffed_write_end - Write end for stuffed files 818 * @inode: The inode 819 * @dibh: The buffer_head containing the on-disk inode 820 * @pos: The file position 821 * @copied: How much was actually copied by the VFS 822 * @page: The page 823 * 824 * This copies the data from the page into the inode block after 825 * the inode data structure itself. 826 * 827 * Returns: copied bytes or errno 828 */ 829 int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh, 830 loff_t pos, unsigned copied, 831 struct page *page) 832 { 833 struct gfs2_inode *ip = GFS2_I(inode); 834 u64 to = pos + copied; 835 void *kaddr; 836 unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode); 837 838 BUG_ON(pos + copied > gfs2_max_stuffed_size(ip)); 839 840 kaddr = kmap_atomic(page); 841 memcpy(buf + pos, kaddr + pos, copied); 842 flush_dcache_page(page); 843 kunmap_atomic(kaddr); 844 845 WARN_ON(!PageUptodate(page)); 846 unlock_page(page); 847 put_page(page); 848 849 if (copied) { 850 if (inode->i_size < to) 851 i_size_write(inode, to); 852 mark_inode_dirty(inode); 853 } 854 return copied; 855 } 856 857 /** 858 * gfs2_write_end 859 * @file: The file to write to 860 * @mapping: The address space to write to 861 * @pos: The file position 862 * @len: The length of the data 863 * @copied: How much was actually copied by the VFS 864 * @page: The page that has been written 865 * @fsdata: The fsdata (unused in GFS2) 866 * 867 * The main write_end function for GFS2. We just put our locking around the VFS 868 * provided functions. 869 * 870 * Returns: copied bytes or errno 871 */ 872 873 static int gfs2_write_end(struct file *file, struct address_space *mapping, 874 loff_t pos, unsigned len, unsigned copied, 875 struct page *page, void *fsdata) 876 { 877 struct inode *inode = page->mapping->host; 878 struct gfs2_inode *ip = GFS2_I(inode); 879 struct gfs2_sbd *sdp = GFS2_SB(inode); 880 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); 881 struct buffer_head *dibh; 882 int ret; 883 struct gfs2_trans *tr = current->journal_info; 884 BUG_ON(!tr); 885 886 BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL); 887 888 ret = gfs2_meta_inode_buffer(ip, &dibh); 889 if (unlikely(ret)) 890 goto out; 891 892 if (gfs2_is_stuffed(ip)) { 893 ret = gfs2_stuffed_write_end(inode, dibh, pos, copied, page); 894 page = NULL; 895 goto out2; 896 } 897 898 if (gfs2_is_jdata(ip)) 899 gfs2_page_add_databufs(ip, page, pos & ~PAGE_MASK, len); 900 else 901 gfs2_ordered_add_inode(ip); 902 903 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); 904 page = NULL; 905 if (tr->tr_num_buf_new) 906 __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 907 else 908 gfs2_trans_add_meta(ip->i_gl, dibh); 909 910 out2: 911 if (inode == sdp->sd_rindex) { 912 adjust_fs_space(inode); 913 sdp->sd_rindex_uptodate = 0; 914 } 915 916 brelse(dibh); 917 out: 918 if (page) { 919 unlock_page(page); 920 put_page(page); 921 } 922 gfs2_trans_end(sdp); 923 gfs2_inplace_release(ip); 924 if (ip->i_qadata && ip->i_qadata->qa_qd_num) 925 gfs2_quota_unlock(ip); 926 if (inode == sdp->sd_rindex) { 927 gfs2_glock_dq(&m_ip->i_gh); 928 gfs2_holder_uninit(&m_ip->i_gh); 929 } 930 gfs2_glock_dq(&ip->i_gh); 931 gfs2_holder_uninit(&ip->i_gh); 932 return ret; 933 } 934 935 /** 936 * jdata_set_page_dirty - Page dirtying function 937 * @page: The page to dirty 938 * 939 * Returns: 1 if it dirtyed the page, or 0 otherwise 940 */ 941 942 static int jdata_set_page_dirty(struct page *page) 943 { 944 SetPageChecked(page); 945 return __set_page_dirty_buffers(page); 946 } 947 948 /** 949 * gfs2_bmap - Block map function 950 * @mapping: Address space info 951 * @lblock: The block to map 952 * 953 * Returns: The disk address for the block or 0 on hole or error 954 */ 955 956 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) 957 { 958 struct gfs2_inode *ip = GFS2_I(mapping->host); 959 struct gfs2_holder i_gh; 960 sector_t dblock = 0; 961 int error; 962 963 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); 964 if (error) 965 return 0; 966 967 if (!gfs2_is_stuffed(ip)) 968 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map); 969 970 gfs2_glock_dq_uninit(&i_gh); 971 972 return dblock; 973 } 974 975 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh) 976 { 977 struct gfs2_bufdata *bd; 978 979 lock_buffer(bh); 980 gfs2_log_lock(sdp); 981 clear_buffer_dirty(bh); 982 bd = bh->b_private; 983 if (bd) { 984 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh)) 985 list_del_init(&bd->bd_list); 986 else 987 gfs2_remove_from_journal(bh, REMOVE_JDATA); 988 } 989 bh->b_bdev = NULL; 990 clear_buffer_mapped(bh); 991 clear_buffer_req(bh); 992 clear_buffer_new(bh); 993 gfs2_log_unlock(sdp); 994 unlock_buffer(bh); 995 } 996 997 static void gfs2_invalidatepage(struct page *page, unsigned int offset, 998 unsigned int length) 999 { 1000 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); 1001 unsigned int stop = offset + length; 1002 int partial_page = (offset || length < PAGE_SIZE); 1003 struct buffer_head *bh, *head; 1004 unsigned long pos = 0; 1005 1006 BUG_ON(!PageLocked(page)); 1007 if (!partial_page) 1008 ClearPageChecked(page); 1009 if (!page_has_buffers(page)) 1010 goto out; 1011 1012 bh = head = page_buffers(page); 1013 do { 1014 if (pos + bh->b_size > stop) 1015 return; 1016 1017 if (offset <= pos) 1018 gfs2_discard(sdp, bh); 1019 pos += bh->b_size; 1020 bh = bh->b_this_page; 1021 } while (bh != head); 1022 out: 1023 if (!partial_page) 1024 try_to_release_page(page, 0); 1025 } 1026 1027 /** 1028 * gfs2_ok_for_dio - check that dio is valid on this file 1029 * @ip: The inode 1030 * @offset: The offset at which we are reading or writing 1031 * 1032 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o) 1033 * 1 (to accept the i/o request) 1034 */ 1035 static int gfs2_ok_for_dio(struct gfs2_inode *ip, loff_t offset) 1036 { 1037 /* 1038 * Should we return an error here? I can't see that O_DIRECT for 1039 * a stuffed file makes any sense. For now we'll silently fall 1040 * back to buffered I/O 1041 */ 1042 if (gfs2_is_stuffed(ip)) 1043 return 0; 1044 1045 if (offset >= i_size_read(&ip->i_inode)) 1046 return 0; 1047 return 1; 1048 } 1049 1050 1051 1052 static ssize_t gfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter) 1053 { 1054 struct file *file = iocb->ki_filp; 1055 struct inode *inode = file->f_mapping->host; 1056 struct address_space *mapping = inode->i_mapping; 1057 struct gfs2_inode *ip = GFS2_I(inode); 1058 loff_t offset = iocb->ki_pos; 1059 struct gfs2_holder gh; 1060 int rv; 1061 1062 /* 1063 * Deferred lock, even if its a write, since we do no allocation 1064 * on this path. All we need change is atime, and this lock mode 1065 * ensures that other nodes have flushed their buffered read caches 1066 * (i.e. their page cache entries for this inode). We do not, 1067 * unfortunately have the option of only flushing a range like 1068 * the VFS does. 1069 */ 1070 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh); 1071 rv = gfs2_glock_nq(&gh); 1072 if (rv) 1073 goto out_uninit; 1074 rv = gfs2_ok_for_dio(ip, offset); 1075 if (rv != 1) 1076 goto out; /* dio not valid, fall back to buffered i/o */ 1077 1078 /* 1079 * Now since we are holding a deferred (CW) lock at this point, you 1080 * might be wondering why this is ever needed. There is a case however 1081 * where we've granted a deferred local lock against a cached exclusive 1082 * glock. That is ok provided all granted local locks are deferred, but 1083 * it also means that it is possible to encounter pages which are 1084 * cached and possibly also mapped. So here we check for that and sort 1085 * them out ahead of the dio. The glock state machine will take care of 1086 * everything else. 1087 * 1088 * If in fact the cached glock state (gl->gl_state) is deferred (CW) in 1089 * the first place, mapping->nr_pages will always be zero. 1090 */ 1091 if (mapping->nrpages) { 1092 loff_t lstart = offset & ~(PAGE_SIZE - 1); 1093 loff_t len = iov_iter_count(iter); 1094 loff_t end = PAGE_ALIGN(offset + len) - 1; 1095 1096 rv = 0; 1097 if (len == 0) 1098 goto out; 1099 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags)) 1100 unmap_shared_mapping_range(ip->i_inode.i_mapping, offset, len); 1101 rv = filemap_write_and_wait_range(mapping, lstart, end); 1102 if (rv) 1103 goto out; 1104 if (iov_iter_rw(iter) == WRITE) 1105 truncate_inode_pages_range(mapping, lstart, end); 1106 } 1107 1108 rv = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter, 1109 gfs2_get_block_direct, NULL, NULL, 0); 1110 out: 1111 gfs2_glock_dq(&gh); 1112 out_uninit: 1113 gfs2_holder_uninit(&gh); 1114 return rv; 1115 } 1116 1117 /** 1118 * gfs2_releasepage - free the metadata associated with a page 1119 * @page: the page that's being released 1120 * @gfp_mask: passed from Linux VFS, ignored by us 1121 * 1122 * Call try_to_free_buffers() if the buffers in this page can be 1123 * released. 1124 * 1125 * Returns: 0 1126 */ 1127 1128 int gfs2_releasepage(struct page *page, gfp_t gfp_mask) 1129 { 1130 struct address_space *mapping = page->mapping; 1131 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); 1132 struct buffer_head *bh, *head; 1133 struct gfs2_bufdata *bd; 1134 1135 if (!page_has_buffers(page)) 1136 return 0; 1137 1138 /* 1139 * From xfs_vm_releasepage: mm accommodates an old ext3 case where 1140 * clean pages might not have had the dirty bit cleared. Thus, it can 1141 * send actual dirty pages to ->releasepage() via shrink_active_list(). 1142 * 1143 * As a workaround, we skip pages that contain dirty buffers below. 1144 * Once ->releasepage isn't called on dirty pages anymore, we can warn 1145 * on dirty buffers like we used to here again. 1146 */ 1147 1148 gfs2_log_lock(sdp); 1149 spin_lock(&sdp->sd_ail_lock); 1150 head = bh = page_buffers(page); 1151 do { 1152 if (atomic_read(&bh->b_count)) 1153 goto cannot_release; 1154 bd = bh->b_private; 1155 if (bd && bd->bd_tr) 1156 goto cannot_release; 1157 if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh))) 1158 goto cannot_release; 1159 bh = bh->b_this_page; 1160 } while(bh != head); 1161 spin_unlock(&sdp->sd_ail_lock); 1162 1163 head = bh = page_buffers(page); 1164 do { 1165 bd = bh->b_private; 1166 if (bd) { 1167 gfs2_assert_warn(sdp, bd->bd_bh == bh); 1168 if (!list_empty(&bd->bd_list)) 1169 list_del_init(&bd->bd_list); 1170 bd->bd_bh = NULL; 1171 bh->b_private = NULL; 1172 kmem_cache_free(gfs2_bufdata_cachep, bd); 1173 } 1174 1175 bh = bh->b_this_page; 1176 } while (bh != head); 1177 gfs2_log_unlock(sdp); 1178 1179 return try_to_free_buffers(page); 1180 1181 cannot_release: 1182 spin_unlock(&sdp->sd_ail_lock); 1183 gfs2_log_unlock(sdp); 1184 return 0; 1185 } 1186 1187 static const struct address_space_operations gfs2_writeback_aops = { 1188 .writepage = gfs2_writepage, 1189 .writepages = gfs2_writepages, 1190 .readpage = gfs2_readpage, 1191 .readpages = gfs2_readpages, 1192 .write_begin = gfs2_write_begin, 1193 .write_end = gfs2_write_end, 1194 .bmap = gfs2_bmap, 1195 .invalidatepage = gfs2_invalidatepage, 1196 .releasepage = gfs2_releasepage, 1197 .direct_IO = gfs2_direct_IO, 1198 .migratepage = buffer_migrate_page, 1199 .is_partially_uptodate = block_is_partially_uptodate, 1200 .error_remove_page = generic_error_remove_page, 1201 }; 1202 1203 static const struct address_space_operations gfs2_ordered_aops = { 1204 .writepage = gfs2_writepage, 1205 .writepages = gfs2_writepages, 1206 .readpage = gfs2_readpage, 1207 .readpages = gfs2_readpages, 1208 .write_begin = gfs2_write_begin, 1209 .write_end = gfs2_write_end, 1210 .set_page_dirty = __set_page_dirty_buffers, 1211 .bmap = gfs2_bmap, 1212 .invalidatepage = gfs2_invalidatepage, 1213 .releasepage = gfs2_releasepage, 1214 .direct_IO = gfs2_direct_IO, 1215 .migratepage = buffer_migrate_page, 1216 .is_partially_uptodate = block_is_partially_uptodate, 1217 .error_remove_page = generic_error_remove_page, 1218 }; 1219 1220 static const struct address_space_operations gfs2_jdata_aops = { 1221 .writepage = gfs2_jdata_writepage, 1222 .writepages = gfs2_jdata_writepages, 1223 .readpage = gfs2_readpage, 1224 .readpages = gfs2_readpages, 1225 .write_begin = gfs2_write_begin, 1226 .write_end = gfs2_write_end, 1227 .set_page_dirty = jdata_set_page_dirty, 1228 .bmap = gfs2_bmap, 1229 .invalidatepage = gfs2_invalidatepage, 1230 .releasepage = gfs2_releasepage, 1231 .is_partially_uptodate = block_is_partially_uptodate, 1232 .error_remove_page = generic_error_remove_page, 1233 }; 1234 1235 void gfs2_set_aops(struct inode *inode) 1236 { 1237 struct gfs2_inode *ip = GFS2_I(inode); 1238 1239 if (gfs2_is_writeback(ip)) 1240 inode->i_mapping->a_ops = &gfs2_writeback_aops; 1241 else if (gfs2_is_ordered(ip)) 1242 inode->i_mapping->a_ops = &gfs2_ordered_aops; 1243 else if (gfs2_is_jdata(ip)) 1244 inode->i_mapping->a_ops = &gfs2_jdata_aops; 1245 else 1246 BUG(); 1247 } 1248 1249