2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
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.
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>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/lm_interface.h>
29 #include "ops_address.h"
38 static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
39 unsigned int from, unsigned int to)
41 struct buffer_head *head = page_buffers(page);
42 unsigned int bsize = head->b_size;
43 struct buffer_head *bh;
44 unsigned int start, end;
46 for (bh = head, start = 0; bh != head || !start;
47 bh = bh->b_this_page, start = end) {
49 if (end <= from || start >= to)
51 gfs2_trans_add_bh(ip->i_gl, bh, 0);
56 * gfs2_get_block - Fills in a buffer head with details about a block
58 * @lblock: The block number to look up
59 * @bh_result: The buffer head to return the result in
60 * @create: Non-zero if we may add block to the file
65 int gfs2_get_block(struct inode *inode, sector_t lblock,
66 struct buffer_head *bh_result, int create)
68 return gfs2_block_map(inode, lblock, create, bh_result);
72 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
74 * @lblock: The block number to look up
75 * @bh_result: The buffer head to return the result in
76 * @create: Non-zero if we may add block to the file
81 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
82 struct buffer_head *bh_result, int create)
86 error = gfs2_block_map(inode, lblock, 0, bh_result);
89 if (bh_result->b_blocknr == 0)
94 static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
95 struct buffer_head *bh_result, int create)
97 return gfs2_block_map(inode, lblock, 0, bh_result);
101 * gfs2_writepage - Write complete page
102 * @page: Page to write
106 * Some of this is copied from block_write_full_page() although we still
107 * call it to do most of the work.
110 static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
112 struct inode *inode = page->mapping->host;
113 struct gfs2_inode *ip = GFS2_I(inode);
114 struct gfs2_sbd *sdp = GFS2_SB(inode);
115 loff_t i_size = i_size_read(inode);
116 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
121 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) {
125 if (current->journal_info)
128 /* Is the page fully outside i_size? (truncate in progress) */
129 offset = i_size & (PAGE_CACHE_SIZE-1);
130 if (page->index > end_index || (page->index == end_index && !offset)) {
131 page->mapping->a_ops->invalidatepage(page, 0);
133 return 0; /* don't care */
136 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) {
137 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
140 if (!page_has_buffers(page)) {
141 create_empty_buffers(page, inode->i_sb->s_blocksize,
142 (1 << BH_Dirty)|(1 << BH_Uptodate));
144 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
147 error = block_write_full_page(page, gfs2_get_block_noalloc, wbc);
150 gfs2_meta_cache_flush(ip);
154 redirty_page_for_writepage(wbc, page);
160 * stuffed_readpage - Fill in a Linux page with stuffed file data
167 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
169 struct buffer_head *dibh;
175 error = gfs2_meta_inode_buffer(ip, &dibh);
179 kaddr = kmap_atomic(page, KM_USER0);
180 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
182 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
183 kunmap_atomic(kaddr, KM_USER0);
187 SetPageUptodate(page);
194 * gfs2_readpage - readpage with locking
195 * @file: The file to read a page for. N.B. This may be NULL if we are
196 * reading an internal file.
197 * @page: The page to read
202 static int gfs2_readpage(struct file *file, struct page *page)
204 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
205 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
206 struct gfs2_file *gf = NULL;
207 struct gfs2_holder gh;
211 if (likely(file != &gfs2_internal_file_sentinel)) {
213 gf = file->private_data;
214 if (test_bit(GFF_EXLOCK, &gf->f_flags))
215 /* gfs2_sharewrite_nopage has grabbed the ip->i_gl already */
218 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh);
220 error = gfs2_glock_nq_atime(&gh);
226 if (gfs2_is_stuffed(ip)) {
227 error = stuffed_readpage(ip, page);
230 error = mpage_readpage(page, gfs2_get_block);
232 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
236 gfs2_glock_dq_m(1, &gh);
237 gfs2_holder_uninit(&gh);
242 if (error == GLR_TRYFAILED)
243 error = AOP_TRUNCATED_PAGE;
246 gfs2_holder_uninit(&gh);
251 * gfs2_readpages - Read a bunch of pages at once
254 * 1. This is only for readahead, so we can simply ignore any things
255 * which are slightly inconvenient (such as locking conflicts between
256 * the page lock and the glock) and return having done no I/O. Its
257 * obviously not something we'd want to do on too regular a basis.
258 * Any I/O we ignore at this time will be done via readpage later.
259 * 2. We don't handle stuffed files here we let readpage do the honours.
260 * 3. mpage_readpages() does most of the heavy lifting in the common case.
261 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
262 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
263 * well as read-ahead.
265 static int gfs2_readpages(struct file *file, struct address_space *mapping,
266 struct list_head *pages, unsigned nr_pages)
268 struct inode *inode = mapping->host;
269 struct gfs2_inode *ip = GFS2_I(inode);
270 struct gfs2_sbd *sdp = GFS2_SB(inode);
271 struct gfs2_holder gh;
275 if (likely(file != &gfs2_internal_file_sentinel)) {
277 struct gfs2_file *gf = file->private_data;
278 if (test_bit(GFF_EXLOCK, &gf->f_flags))
281 gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
282 LM_FLAG_TRY_1CB|GL_ATIME, &gh);
284 ret = gfs2_glock_nq_atime(&gh);
285 if (ret == GLR_TRYFAILED)
291 if (!gfs2_is_stuffed(ip))
292 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
295 gfs2_glock_dq_m(1, &gh);
296 gfs2_holder_uninit(&gh);
299 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
306 gfs2_holder_uninit(&gh);
311 * gfs2_prepare_write - Prepare to write a page to a file
312 * @file: The file to write to
313 * @page: The page which is to be prepared for writing
314 * @from: From (byte range within page)
315 * @to: To (byte range within page)
320 static int gfs2_prepare_write(struct file *file, struct page *page,
321 unsigned from, unsigned to)
323 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
324 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
325 unsigned int data_blocks, ind_blocks, rblocks;
328 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
329 loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
330 struct gfs2_alloc *al;
331 unsigned int write_len = to - from;
334 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|LM_FLAG_TRY_1CB, &ip->i_gh);
335 error = gfs2_glock_nq_atime(&ip->i_gh);
336 if (unlikely(error)) {
337 if (error == GLR_TRYFAILED)
338 error = AOP_TRUNCATED_PAGE;
342 gfs2_write_calc_reserv(ip, write_len, &data_blocks, &ind_blocks);
344 error = gfs2_write_alloc_required(ip, pos, write_len, &alloc_required);
349 ip->i_alloc.al_requested = 0;
350 if (alloc_required) {
351 al = gfs2_alloc_get(ip);
353 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
357 error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid);
361 al->al_requested = data_blocks + ind_blocks;
362 error = gfs2_inplace_reserve(ip);
367 rblocks = RES_DINODE + ind_blocks;
368 if (gfs2_is_jdata(ip))
369 rblocks += data_blocks ? data_blocks : 1;
370 if (ind_blocks || data_blocks)
371 rblocks += RES_STATFS + RES_QUOTA;
373 error = gfs2_trans_begin(sdp, rblocks, 0);
377 if (gfs2_is_stuffed(ip)) {
378 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
379 error = gfs2_unstuff_dinode(ip, page);
382 } else if (!PageUptodate(page))
383 error = stuffed_readpage(ip, page);
388 error = block_prepare_write(page, from, to, gfs2_get_block);
393 if (alloc_required) {
394 gfs2_inplace_release(ip);
396 gfs2_quota_unlock(ip);
401 gfs2_glock_dq_m(1, &ip->i_gh);
403 gfs2_holder_uninit(&ip->i_gh);
410 * gfs2_commit_write - Commit write to a file
411 * @file: The file to write to
412 * @page: The page containing the data
413 * @from: From (byte range within page)
414 * @to: To (byte range within page)
419 static int gfs2_commit_write(struct file *file, struct page *page,
420 unsigned from, unsigned to)
422 struct inode *inode = page->mapping->host;
423 struct gfs2_inode *ip = GFS2_I(inode);
424 struct gfs2_sbd *sdp = GFS2_SB(inode);
425 int error = -EOPNOTSUPP;
426 struct buffer_head *dibh;
427 struct gfs2_alloc *al = &ip->i_alloc;
428 struct gfs2_dinode *di;
430 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
433 error = gfs2_meta_inode_buffer(ip, &dibh);
437 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
438 di = (struct gfs2_dinode *)dibh->b_data;
440 if (gfs2_is_stuffed(ip)) {
444 file_size = ((u64)page->index << PAGE_CACHE_SHIFT) + to;
446 kaddr = kmap_atomic(page, KM_USER0);
447 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
448 kaddr + from, to - from);
449 kunmap_atomic(kaddr, KM_USER0);
451 SetPageUptodate(page);
453 if (inode->i_size < file_size) {
454 i_size_write(inode, file_size);
455 mark_inode_dirty(inode);
458 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
460 gfs2_page_add_databufs(ip, page, from, to);
461 error = generic_commit_write(file, page, from, to);
466 if (ip->i_di.di_size < inode->i_size) {
467 ip->i_di.di_size = inode->i_size;
468 di->di_size = cpu_to_be64(inode->i_size);
473 if (al->al_requested) {
474 gfs2_inplace_release(ip);
475 gfs2_quota_unlock(ip);
478 gfs2_glock_dq_m(1, &ip->i_gh);
479 gfs2_holder_uninit(&ip->i_gh);
486 if (al->al_requested) {
487 gfs2_inplace_release(ip);
488 gfs2_quota_unlock(ip);
491 gfs2_glock_dq_m(1, &ip->i_gh);
492 gfs2_holder_uninit(&ip->i_gh);
494 ClearPageUptodate(page);
499 * gfs2_bmap - Block map function
500 * @mapping: Address space info
501 * @lblock: The block to map
503 * Returns: The disk address for the block or 0 on hole or error
506 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
508 struct gfs2_inode *ip = GFS2_I(mapping->host);
509 struct gfs2_holder i_gh;
513 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
517 if (!gfs2_is_stuffed(ip))
518 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
520 gfs2_glock_dq_uninit(&i_gh);
525 static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
527 struct gfs2_bufdata *bd;
533 bh->b_private = NULL;
535 gfs2_log_unlock(sdp);
538 clear_buffer_dirty(bh);
540 clear_buffer_mapped(bh);
541 clear_buffer_req(bh);
542 clear_buffer_new(bh);
543 clear_buffer_delay(bh);
547 static void gfs2_invalidatepage(struct page *page, unsigned long offset)
549 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
550 struct buffer_head *head, *bh, *next;
551 unsigned int curr_off = 0;
553 BUG_ON(!PageLocked(page));
554 if (!page_has_buffers(page))
557 bh = head = page_buffers(page);
559 unsigned int next_off = curr_off + bh->b_size;
560 next = bh->b_this_page;
562 if (offset <= curr_off)
563 discard_buffer(sdp, bh);
567 } while (bh != head);
570 try_to_release_page(page, 0);
576 * gfs2_ok_for_dio - check that dio is valid on this file
579 * @offset: The offset at which we are reading or writing
581 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
582 * 1 (to accept the i/o request)
584 static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
587 * Should we return an error here? I can't see that O_DIRECT for
588 * a journaled file makes any sense. For now we'll silently fall
589 * back to buffered I/O, likewise we do the same for stuffed
590 * files since they are (a) small and (b) unaligned.
592 if (gfs2_is_jdata(ip))
595 if (gfs2_is_stuffed(ip))
598 if (offset > i_size_read(&ip->i_inode))
605 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
606 const struct iovec *iov, loff_t offset,
607 unsigned long nr_segs)
609 struct file *file = iocb->ki_filp;
610 struct inode *inode = file->f_mapping->host;
611 struct gfs2_inode *ip = GFS2_I(inode);
612 struct gfs2_holder gh;
616 * Deferred lock, even if its a write, since we do no allocation
617 * on this path. All we need change is atime, and this lock mode
618 * ensures that other nodes have flushed their buffered read caches
619 * (i.e. their page cache entries for this inode). We do not,
620 * unfortunately have the option of only flushing a range like
623 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh);
624 rv = gfs2_glock_nq_atime(&gh);
627 rv = gfs2_ok_for_dio(ip, rw, offset);
629 goto out; /* dio not valid, fall back to buffered i/o */
631 rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
632 iov, offset, nr_segs,
633 gfs2_get_block_direct, NULL);
635 gfs2_glock_dq_m(1, &gh);
636 gfs2_holder_uninit(&gh);
641 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
642 * @bh: the buffer we're stuck on
646 static void stuck_releasepage(struct buffer_head *bh)
648 struct inode *inode = bh->b_page->mapping->host;
649 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
650 struct gfs2_bufdata *bd = bh->b_private;
651 struct gfs2_glock *gl;
652 static unsigned limit = 0;
658 fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode);
659 fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
660 (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count));
661 fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
662 fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL");
669 fs_warn(sdp, "gl = (%u, %llu)\n",
670 gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number);
672 fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
673 (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
674 (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
676 if (gl->gl_ops == &gfs2_inode_glops) {
677 struct gfs2_inode *ip = gl->gl_object;
683 fs_warn(sdp, "ip = %llu %llu\n",
684 (unsigned long long)ip->i_num.no_formal_ino,
685 (unsigned long long)ip->i_num.no_addr);
687 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++)
688 fs_warn(sdp, "ip->i_cache[%u] = %s\n",
689 x, (ip->i_cache[x]) ? "!NULL" : "NULL");
694 * gfs2_releasepage - free the metadata associated with a page
695 * @page: the page that's being released
696 * @gfp_mask: passed from Linux VFS, ignored by us
698 * Call try_to_free_buffers() if the buffers in this page can be
704 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
706 struct inode *aspace = page->mapping->host;
707 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
708 struct buffer_head *bh, *head;
709 struct gfs2_bufdata *bd;
710 unsigned long t = jiffies + gfs2_tune_get(sdp, gt_stall_secs) * HZ;
712 if (!page_has_buffers(page))
715 head = bh = page_buffers(page);
717 while (atomic_read(&bh->b_count)) {
718 if (!atomic_read(&aspace->i_writecount))
721 if (!(gfp_mask & __GFP_WAIT))
724 if (time_after_eq(jiffies, t)) {
725 stuck_releasepage(bh);
726 /* should we withdraw here? */
733 gfs2_assert_warn(sdp, !buffer_pinned(bh));
734 gfs2_assert_warn(sdp, !buffer_dirty(bh));
739 gfs2_assert_warn(sdp, bd->bd_bh == bh);
740 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
741 gfs2_assert_warn(sdp, !bd->bd_ail);
743 if (!list_empty(&bd->bd_le.le_list))
745 bh->b_private = NULL;
747 gfs2_log_unlock(sdp);
749 kmem_cache_free(gfs2_bufdata_cachep, bd);
751 bh = bh->b_this_page;
752 } while (bh != head);
755 return try_to_free_buffers(page);
758 const struct address_space_operations gfs2_file_aops = {
759 .writepage = gfs2_writepage,
760 .readpage = gfs2_readpage,
761 .readpages = gfs2_readpages,
762 .sync_page = block_sync_page,
763 .prepare_write = gfs2_prepare_write,
764 .commit_write = gfs2_commit_write,
766 .invalidatepage = gfs2_invalidatepage,
767 .releasepage = gfs2_releasepage,
768 .direct_IO = gfs2_direct_IO,