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>
22 #include "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)
73 error = gfs2_block_map(inode, lblock, &new, &dblock, &boundary);
80 map_bh(bh_result, inode->i_sb, dblock);
82 set_buffer_new(bh_result);
84 set_buffer_boundary(bh_result);
90 * get_block_noalloc - Fills in a buffer head with details about a block
92 * @lblock: The block number to look up
93 * @bh_result: The buffer head to return the result in
94 * @create: Non-zero if we may add block to the file
99 static int get_block_noalloc(struct inode *inode, sector_t lblock,
100 struct buffer_head *bh_result, int create)
107 error = gfs2_block_map(inode, lblock, &new, &dblock, &boundary);
112 map_bh(bh_result, inode->i_sb, dblock);
113 else if (gfs2_assert_withdraw(GFS2_SB(inode), !create))
116 set_buffer_boundary(bh_result);
121 static int get_block_direct(struct inode *inode, sector_t lblock,
122 struct buffer_head *bh_result, int create)
128 error = gfs2_block_map(inode, lblock, &new, &dblock, &boundary);
133 map_bh(bh_result, inode->i_sb, dblock);
135 set_buffer_boundary(bh_result);
141 * gfs2_writepage - Write complete page
142 * @page: Page to write
146 * Some of this is copied from block_write_full_page() although we still
147 * call it to do most of the work.
150 static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
152 struct inode *inode = page->mapping->host;
153 struct gfs2_inode *ip = GFS2_I(inode);
154 struct gfs2_sbd *sdp = GFS2_SB(inode);
155 loff_t i_size = i_size_read(inode);
156 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
161 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) {
165 if (current->journal_info)
168 /* Is the page fully outside i_size? (truncate in progress) */
169 offset = i_size & (PAGE_CACHE_SIZE-1);
170 if (page->index > end_index || (page->index == end_index && !offset)) {
171 page->mapping->a_ops->invalidatepage(page, 0);
173 return 0; /* don't care */
176 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) {
177 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
180 if (!page_has_buffers(page)) {
181 create_empty_buffers(page, inode->i_sb->s_blocksize,
182 (1 << BH_Dirty)|(1 << BH_Uptodate));
184 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
187 error = block_write_full_page(page, get_block_noalloc, wbc);
190 gfs2_meta_cache_flush(ip);
194 redirty_page_for_writepage(wbc, page);
199 static int zero_readpage(struct page *page)
203 kaddr = kmap_atomic(page, KM_USER0);
204 memset(kaddr, 0, PAGE_CACHE_SIZE);
205 kunmap_atomic(page, KM_USER0);
207 SetPageUptodate(page);
213 * stuffed_readpage - Fill in a Linux page with stuffed file data
220 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
222 struct buffer_head *dibh;
226 /* Only the first page of a stuffed file might contain data */
227 if (unlikely(page->index))
228 return zero_readpage(page);
230 error = gfs2_meta_inode_buffer(ip, &dibh);
234 kaddr = kmap_atomic(page, KM_USER0);
235 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
237 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
238 kunmap_atomic(page, KM_USER0);
242 SetPageUptodate(page);
249 * gfs2_readpage - readpage with locking
250 * @file: The file to read a page for. N.B. This may be NULL if we are
251 * reading an internal file.
252 * @page: The page to read
257 static int gfs2_readpage(struct file *file, struct page *page)
259 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
260 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
261 struct gfs2_holder gh;
265 if (likely(file != &gfs2_internal_file_sentinel)) {
267 struct gfs2_file *gf = file->private_data;
268 if (test_bit(GFF_EXLOCK, &gf->f_flags))
271 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|GL_AOP, &gh);
273 error = gfs2_glock_nq_m_atime(1, &gh);
279 if (gfs2_is_stuffed(ip)) {
280 error = stuffed_readpage(ip, page);
283 error = mpage_readpage(page, gfs2_get_block);
285 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
288 if (file != &gfs2_internal_file_sentinel) {
289 gfs2_glock_dq_m(1, &gh);
290 gfs2_holder_uninit(&gh);
297 gfs2_holder_uninit(&gh);
302 * gfs2_readpages - Read a bunch of pages at once
305 * 1. This is only for readahead, so we can simply ignore any things
306 * which are slightly inconvenient (such as locking conflicts between
307 * the page lock and the glock) and return having done no I/O. Its
308 * obviously not something we'd want to do on too regular a basis.
309 * Any I/O we ignore at this time will be done via readpage later.
310 * 2. We have to handle stuffed files here too.
311 * 3. mpage_readpages() does most of the heavy lifting in the common case.
312 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
313 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
314 * well as read-ahead.
316 static int gfs2_readpages(struct file *file, struct address_space *mapping,
317 struct list_head *pages, unsigned nr_pages)
319 struct inode *inode = mapping->host;
320 struct gfs2_inode *ip = GFS2_I(inode);
321 struct gfs2_sbd *sdp = GFS2_SB(inode);
322 struct gfs2_holder gh;
327 if (likely(file != &gfs2_internal_file_sentinel)) {
329 struct gfs2_file *gf = file->private_data;
330 if (test_bit(GFF_EXLOCK, &gf->f_flags))
333 gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
334 LM_FLAG_TRY_1CB|GL_ATIME|GL_AOP, &gh);
336 ret = gfs2_glock_nq_m_atime(1, &gh);
337 if (ret == GLR_TRYFAILED)
343 if (gfs2_is_stuffed(ip)) {
344 struct pagevec lru_pvec;
345 pagevec_init(&lru_pvec, 0);
346 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
347 struct page *page = list_entry(pages->prev, struct page, lru);
348 prefetchw(&page->flags);
349 list_del(&page->lru);
350 if (!add_to_page_cache(page, mapping,
351 page->index, GFP_KERNEL)) {
352 ret = stuffed_readpage(ip, page);
354 if (!pagevec_add(&lru_pvec, page))
355 __pagevec_lru_add(&lru_pvec);
357 page_cache_release(page);
360 pagevec_lru_add(&lru_pvec);
363 /* What we really want to do .... */
364 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
368 gfs2_glock_dq_m(1, &gh);
369 gfs2_holder_uninit(&gh);
372 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
378 /* unlock all pages, we can't do any I/O right now */
379 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
380 struct page *page = list_entry(pages->prev, struct page, lru);
381 list_del(&page->lru);
383 page_cache_release(page);
386 gfs2_holder_uninit(&gh);
391 * gfs2_prepare_write - Prepare to write a page to a file
392 * @file: The file to write to
393 * @page: The page which is to be prepared for writing
394 * @from: From (byte range within page)
395 * @to: To (byte range within page)
400 static int gfs2_prepare_write(struct file *file, struct page *page,
401 unsigned from, unsigned to)
403 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
404 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
405 unsigned int data_blocks, ind_blocks, rblocks;
408 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
409 loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
410 struct gfs2_alloc *al;
412 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|GL_AOP, &ip->i_gh);
413 error = gfs2_glock_nq_m_atime(1, &ip->i_gh);
417 gfs2_write_calc_reserv(ip, to - from, &data_blocks, &ind_blocks);
419 error = gfs2_write_alloc_required(ip, pos, from - to, &alloc_required);
424 if (alloc_required) {
425 al = gfs2_alloc_get(ip);
427 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
431 error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid);
435 al->al_requested = data_blocks + ind_blocks;
436 error = gfs2_inplace_reserve(ip);
441 rblocks = RES_DINODE + ind_blocks;
442 if (gfs2_is_jdata(ip))
443 rblocks += data_blocks ? data_blocks : 1;
444 if (ind_blocks || data_blocks)
445 rblocks += RES_STATFS + RES_QUOTA;
447 error = gfs2_trans_begin(sdp, rblocks, 0);
451 if (gfs2_is_stuffed(ip)) {
452 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
453 error = gfs2_unstuff_dinode(ip, page);
456 } else if (!PageUptodate(page))
457 error = stuffed_readpage(ip, page);
462 error = block_prepare_write(page, from, to, gfs2_get_block);
467 if (alloc_required) {
468 gfs2_inplace_release(ip);
470 gfs2_quota_unlock(ip);
475 gfs2_glock_dq_m(1, &ip->i_gh);
477 gfs2_holder_uninit(&ip->i_gh);
484 * gfs2_commit_write - Commit write to a file
485 * @file: The file to write to
486 * @page: The page containing the data
487 * @from: From (byte range within page)
488 * @to: To (byte range within page)
493 static int gfs2_commit_write(struct file *file, struct page *page,
494 unsigned from, unsigned to)
496 struct inode *inode = page->mapping->host;
497 struct gfs2_inode *ip = GFS2_I(inode);
498 struct gfs2_sbd *sdp = GFS2_SB(inode);
499 int error = -EOPNOTSUPP;
500 struct buffer_head *dibh;
501 struct gfs2_alloc *al = &ip->i_alloc;;
503 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
506 error = gfs2_meta_inode_buffer(ip, &dibh);
510 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
512 if (gfs2_is_stuffed(ip)) {
516 file_size = ((u64)page->index << PAGE_CACHE_SHIFT) + to;
518 kaddr = kmap_atomic(page, KM_USER0);
519 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
520 kaddr + from, to - from);
521 kunmap_atomic(page, KM_USER0);
523 SetPageUptodate(page);
525 if (inode->i_size < file_size)
526 i_size_write(inode, file_size);
528 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
530 gfs2_page_add_databufs(ip, page, from, to);
531 error = generic_commit_write(file, page, from, to);
536 if (ip->i_di.di_size < inode->i_size)
537 ip->i_di.di_size = inode->i_size;
539 gfs2_dinode_out(&ip->i_di, dibh->b_data);
542 if (al->al_requested) {
543 gfs2_inplace_release(ip);
544 gfs2_quota_unlock(ip);
547 gfs2_glock_dq_m(1, &ip->i_gh);
548 gfs2_holder_uninit(&ip->i_gh);
555 if (al->al_requested) {
556 gfs2_inplace_release(ip);
557 gfs2_quota_unlock(ip);
560 gfs2_glock_dq_m(1, &ip->i_gh);
561 gfs2_holder_uninit(&ip->i_gh);
563 ClearPageUptodate(page);
568 * gfs2_bmap - Block map function
569 * @mapping: Address space info
570 * @lblock: The block to map
572 * Returns: The disk address for the block or 0 on hole or error
575 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
577 struct gfs2_inode *ip = GFS2_I(mapping->host);
578 struct gfs2_holder i_gh;
582 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
586 if (!gfs2_is_stuffed(ip))
587 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
589 gfs2_glock_dq_uninit(&i_gh);
594 static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
596 struct gfs2_bufdata *bd;
602 bh->b_private = NULL;
604 gfs2_log_unlock(sdp);
607 clear_buffer_dirty(bh);
609 clear_buffer_mapped(bh);
610 clear_buffer_req(bh);
611 clear_buffer_new(bh);
612 clear_buffer_delay(bh);
616 static void gfs2_invalidatepage(struct page *page, unsigned long offset)
618 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
619 struct buffer_head *head, *bh, *next;
620 unsigned int curr_off = 0;
622 BUG_ON(!PageLocked(page));
623 if (!page_has_buffers(page))
626 bh = head = page_buffers(page);
628 unsigned int next_off = curr_off + bh->b_size;
629 next = bh->b_this_page;
631 if (offset <= curr_off)
632 discard_buffer(sdp, bh);
636 } while (bh != head);
639 try_to_release_page(page, 0);
644 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
645 const struct iovec *iov, loff_t offset,
646 unsigned long nr_segs)
648 struct file *file = iocb->ki_filp;
649 struct inode *inode = file->f_mapping->host;
650 struct gfs2_inode *ip = GFS2_I(inode);
651 struct gfs2_holder gh;
655 mutex_lock(&inode->i_mutex);
657 * Shared lock, even if its a write, since we do no allocation
658 * on this path. All we need change is atime.
660 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
661 rv = gfs2_glock_nq_m_atime(1, &gh);
665 if (offset > i_size_read(inode))
669 * Should we return an error here? I can't see that O_DIRECT for
670 * a journaled file makes any sense. For now we'll silently fall
671 * back to buffered I/O, likewise we do the same for stuffed
672 * files since they are (a) small and (b) unaligned.
674 if (gfs2_is_jdata(ip))
677 if (gfs2_is_stuffed(ip))
680 rv = blockdev_direct_IO_own_locking(rw, iocb, inode,
682 iov, offset, nr_segs,
683 get_block_direct, NULL);
685 gfs2_glock_dq_m(1, &gh);
686 gfs2_holder_uninit(&gh);
688 mutex_unlock(&inode->i_mutex);
694 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
695 * @bh: the buffer we're stuck on
699 static void stuck_releasepage(struct buffer_head *bh)
701 struct inode *inode = bh->b_page->mapping->host;
702 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
703 struct gfs2_bufdata *bd = bh->b_private;
704 struct gfs2_glock *gl;
705 static unsigned limit = 0;
711 fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode);
712 fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
713 (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count));
714 fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
715 fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL");
722 fs_warn(sdp, "gl = (%u, %llu)\n",
723 gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number);
725 fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
726 (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
727 (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
729 if (gl->gl_ops == &gfs2_inode_glops) {
730 struct gfs2_inode *ip = gl->gl_object;
736 fs_warn(sdp, "ip = %llu %llu\n",
737 (unsigned long long)ip->i_num.no_formal_ino,
738 (unsigned long long)ip->i_num.no_addr);
740 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++)
741 fs_warn(sdp, "ip->i_cache[%u] = %s\n",
742 x, (ip->i_cache[x]) ? "!NULL" : "NULL");
747 * gfs2_releasepage - free the metadata associated with a page
748 * @page: the page that's being released
749 * @gfp_mask: passed from Linux VFS, ignored by us
751 * Call try_to_free_buffers() if the buffers in this page can be
757 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
759 struct inode *aspace = page->mapping->host;
760 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
761 struct buffer_head *bh, *head;
762 struct gfs2_bufdata *bd;
763 unsigned long t = jiffies + gfs2_tune_get(sdp, gt_stall_secs) * HZ;
765 if (!page_has_buffers(page))
768 head = bh = page_buffers(page);
770 while (atomic_read(&bh->b_count)) {
771 if (!atomic_read(&aspace->i_writecount))
774 if (time_after_eq(jiffies, t)) {
775 stuck_releasepage(bh);
776 /* should we withdraw here? */
783 gfs2_assert_warn(sdp, !buffer_pinned(bh));
784 gfs2_assert_warn(sdp, !buffer_dirty(bh));
789 gfs2_assert_warn(sdp, bd->bd_bh == bh);
790 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
791 gfs2_assert_warn(sdp, !bd->bd_ail);
793 if (!list_empty(&bd->bd_le.le_list))
795 bh->b_private = NULL;
797 gfs2_log_unlock(sdp);
799 kmem_cache_free(gfs2_bufdata_cachep, bd);
801 bh = bh->b_this_page;
802 } while (bh != head);
805 return try_to_free_buffers(page);
808 const struct address_space_operations gfs2_file_aops = {
809 .writepage = gfs2_writepage,
810 .readpage = gfs2_readpage,
811 .readpages = gfs2_readpages,
812 .sync_page = block_sync_page,
813 .prepare_write = gfs2_prepare_write,
814 .commit_write = gfs2_commit_write,
816 .invalidatepage = gfs2_invalidatepage,
817 .releasepage = gfs2_releasepage,
818 .direct_IO = gfs2_direct_IO,