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)
68 return gfs2_block_map(inode, lblock, create, bh_result, 32);
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, 1);
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, 512);
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);
159 static int zero_readpage(struct page *page)
163 kaddr = kmap_atomic(page, KM_USER0);
164 memset(kaddr, 0, PAGE_CACHE_SIZE);
165 kunmap_atomic(page, KM_USER0);
167 SetPageUptodate(page);
173 * stuffed_readpage - Fill in a Linux page with stuffed file data
180 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
182 struct buffer_head *dibh;
186 /* Only the first page of a stuffed file might contain data */
187 if (unlikely(page->index))
188 return zero_readpage(page);
190 error = gfs2_meta_inode_buffer(ip, &dibh);
194 kaddr = kmap_atomic(page, KM_USER0);
195 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
197 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
198 kunmap_atomic(page, KM_USER0);
202 SetPageUptodate(page);
209 * gfs2_readpage - readpage with locking
210 * @file: The file to read a page for. N.B. This may be NULL if we are
211 * reading an internal file.
212 * @page: The page to read
217 static int gfs2_readpage(struct file *file, struct page *page)
219 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
220 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
221 struct gfs2_holder gh;
225 if (likely(file != &gfs2_internal_file_sentinel)) {
227 struct gfs2_file *gf = file->private_data;
228 if (test_bit(GFF_EXLOCK, &gf->f_flags))
231 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|GL_AOP, &gh);
233 error = gfs2_glock_nq_m_atime(1, &gh);
239 if (gfs2_is_stuffed(ip)) {
240 error = stuffed_readpage(ip, page);
243 error = mpage_readpage(page, gfs2_get_block);
245 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
248 if (file != &gfs2_internal_file_sentinel) {
249 gfs2_glock_dq_m(1, &gh);
250 gfs2_holder_uninit(&gh);
257 gfs2_holder_uninit(&gh);
262 * gfs2_readpages - Read a bunch of pages at once
265 * 1. This is only for readahead, so we can simply ignore any things
266 * which are slightly inconvenient (such as locking conflicts between
267 * the page lock and the glock) and return having done no I/O. Its
268 * obviously not something we'd want to do on too regular a basis.
269 * Any I/O we ignore at this time will be done via readpage later.
270 * 2. We have to handle stuffed files here too.
271 * 3. mpage_readpages() does most of the heavy lifting in the common case.
272 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
273 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
274 * well as read-ahead.
276 static int gfs2_readpages(struct file *file, struct address_space *mapping,
277 struct list_head *pages, unsigned nr_pages)
279 struct inode *inode = mapping->host;
280 struct gfs2_inode *ip = GFS2_I(inode);
281 struct gfs2_sbd *sdp = GFS2_SB(inode);
282 struct gfs2_holder gh;
287 if (likely(file != &gfs2_internal_file_sentinel)) {
289 struct gfs2_file *gf = file->private_data;
290 if (test_bit(GFF_EXLOCK, &gf->f_flags))
293 gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
294 LM_FLAG_TRY_1CB|GL_ATIME|GL_AOP, &gh);
296 ret = gfs2_glock_nq_m_atime(1, &gh);
297 if (ret == GLR_TRYFAILED)
303 if (gfs2_is_stuffed(ip)) {
304 struct pagevec lru_pvec;
305 pagevec_init(&lru_pvec, 0);
306 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
307 struct page *page = list_entry(pages->prev, struct page, lru);
308 prefetchw(&page->flags);
309 list_del(&page->lru);
310 if (!add_to_page_cache(page, mapping,
311 page->index, GFP_KERNEL)) {
312 ret = stuffed_readpage(ip, page);
314 if (!pagevec_add(&lru_pvec, page))
315 __pagevec_lru_add(&lru_pvec);
317 page_cache_release(page);
320 pagevec_lru_add(&lru_pvec);
323 /* What we really want to do .... */
324 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
328 gfs2_glock_dq_m(1, &gh);
329 gfs2_holder_uninit(&gh);
332 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
338 /* unlock all pages, we can't do any I/O right now */
339 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
340 struct page *page = list_entry(pages->prev, struct page, lru);
341 list_del(&page->lru);
343 page_cache_release(page);
346 gfs2_holder_uninit(&gh);
351 * gfs2_prepare_write - Prepare to write a page to a file
352 * @file: The file to write to
353 * @page: The page which is to be prepared for writing
354 * @from: From (byte range within page)
355 * @to: To (byte range within page)
360 static int gfs2_prepare_write(struct file *file, struct page *page,
361 unsigned from, unsigned to)
363 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
364 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
365 unsigned int data_blocks, ind_blocks, rblocks;
368 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
369 loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
370 struct gfs2_alloc *al;
372 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|GL_AOP, &ip->i_gh);
373 error = gfs2_glock_nq_m_atime(1, &ip->i_gh);
377 gfs2_write_calc_reserv(ip, to - from, &data_blocks, &ind_blocks);
379 error = gfs2_write_alloc_required(ip, pos, from - to, &alloc_required);
384 if (alloc_required) {
385 al = gfs2_alloc_get(ip);
387 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
391 error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid);
395 al->al_requested = data_blocks + ind_blocks;
396 error = gfs2_inplace_reserve(ip);
401 rblocks = RES_DINODE + ind_blocks;
402 if (gfs2_is_jdata(ip))
403 rblocks += data_blocks ? data_blocks : 1;
404 if (ind_blocks || data_blocks)
405 rblocks += RES_STATFS + RES_QUOTA;
407 error = gfs2_trans_begin(sdp, rblocks, 0);
411 if (gfs2_is_stuffed(ip)) {
412 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
413 error = gfs2_unstuff_dinode(ip, page);
416 } else if (!PageUptodate(page))
417 error = stuffed_readpage(ip, page);
422 error = block_prepare_write(page, from, to, gfs2_get_block);
427 if (alloc_required) {
428 gfs2_inplace_release(ip);
430 gfs2_quota_unlock(ip);
435 gfs2_glock_dq_m(1, &ip->i_gh);
437 gfs2_holder_uninit(&ip->i_gh);
444 * gfs2_commit_write - Commit write to a file
445 * @file: The file to write to
446 * @page: The page containing the data
447 * @from: From (byte range within page)
448 * @to: To (byte range within page)
453 static int gfs2_commit_write(struct file *file, struct page *page,
454 unsigned from, unsigned to)
456 struct inode *inode = page->mapping->host;
457 struct gfs2_inode *ip = GFS2_I(inode);
458 struct gfs2_sbd *sdp = GFS2_SB(inode);
459 int error = -EOPNOTSUPP;
460 struct buffer_head *dibh;
461 struct gfs2_alloc *al = &ip->i_alloc;;
463 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
466 error = gfs2_meta_inode_buffer(ip, &dibh);
470 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
472 if (gfs2_is_stuffed(ip)) {
476 file_size = ((u64)page->index << PAGE_CACHE_SHIFT) + to;
478 kaddr = kmap_atomic(page, KM_USER0);
479 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
480 kaddr + from, to - from);
481 kunmap_atomic(page, KM_USER0);
483 SetPageUptodate(page);
485 if (inode->i_size < file_size)
486 i_size_write(inode, file_size);
488 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
490 gfs2_page_add_databufs(ip, page, from, to);
491 error = generic_commit_write(file, page, from, to);
496 if (ip->i_di.di_size < inode->i_size)
497 ip->i_di.di_size = inode->i_size;
499 gfs2_dinode_out(&ip->i_di, dibh->b_data);
502 if (al->al_requested) {
503 gfs2_inplace_release(ip);
504 gfs2_quota_unlock(ip);
507 gfs2_glock_dq_m(1, &ip->i_gh);
508 gfs2_holder_uninit(&ip->i_gh);
515 if (al->al_requested) {
516 gfs2_inplace_release(ip);
517 gfs2_quota_unlock(ip);
520 gfs2_glock_dq_m(1, &ip->i_gh);
521 gfs2_holder_uninit(&ip->i_gh);
523 ClearPageUptodate(page);
528 * gfs2_bmap - Block map function
529 * @mapping: Address space info
530 * @lblock: The block to map
532 * Returns: The disk address for the block or 0 on hole or error
535 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
537 struct gfs2_inode *ip = GFS2_I(mapping->host);
538 struct gfs2_holder i_gh;
542 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
546 if (!gfs2_is_stuffed(ip))
547 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
549 gfs2_glock_dq_uninit(&i_gh);
554 static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
556 struct gfs2_bufdata *bd;
562 bh->b_private = NULL;
564 gfs2_log_unlock(sdp);
567 clear_buffer_dirty(bh);
569 clear_buffer_mapped(bh);
570 clear_buffer_req(bh);
571 clear_buffer_new(bh);
572 clear_buffer_delay(bh);
576 static void gfs2_invalidatepage(struct page *page, unsigned long offset)
578 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
579 struct buffer_head *head, *bh, *next;
580 unsigned int curr_off = 0;
582 BUG_ON(!PageLocked(page));
583 if (!page_has_buffers(page))
586 bh = head = page_buffers(page);
588 unsigned int next_off = curr_off + bh->b_size;
589 next = bh->b_this_page;
591 if (offset <= curr_off)
592 discard_buffer(sdp, bh);
596 } while (bh != head);
599 try_to_release_page(page, 0);
604 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
605 const struct iovec *iov, loff_t offset,
606 unsigned long nr_segs)
608 struct file *file = iocb->ki_filp;
609 struct inode *inode = file->f_mapping->host;
610 struct gfs2_inode *ip = GFS2_I(inode);
611 struct gfs2_holder gh;
615 mutex_lock(&inode->i_mutex);
617 * Shared lock, even if its a write, since we do no allocation
618 * on this path. All we need change is atime.
620 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
621 rv = gfs2_glock_nq_m_atime(1, &gh);
625 if (offset > i_size_read(inode))
629 * Should we return an error here? I can't see that O_DIRECT for
630 * a journaled file makes any sense. For now we'll silently fall
631 * back to buffered I/O, likewise we do the same for stuffed
632 * files since they are (a) small and (b) unaligned.
634 if (gfs2_is_jdata(ip))
637 if (gfs2_is_stuffed(ip))
640 rv = blockdev_direct_IO_own_locking(rw, iocb, inode,
642 iov, offset, nr_segs,
643 gfs2_get_block_direct, NULL);
645 gfs2_glock_dq_m(1, &gh);
646 gfs2_holder_uninit(&gh);
648 mutex_unlock(&inode->i_mutex);
654 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
655 * @bh: the buffer we're stuck on
659 static void stuck_releasepage(struct buffer_head *bh)
661 struct inode *inode = bh->b_page->mapping->host;
662 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
663 struct gfs2_bufdata *bd = bh->b_private;
664 struct gfs2_glock *gl;
665 static unsigned limit = 0;
671 fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode);
672 fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
673 (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count));
674 fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
675 fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL");
682 fs_warn(sdp, "gl = (%u, %llu)\n",
683 gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number);
685 fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
686 (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
687 (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
689 if (gl->gl_ops == &gfs2_inode_glops) {
690 struct gfs2_inode *ip = gl->gl_object;
696 fs_warn(sdp, "ip = %llu %llu\n",
697 (unsigned long long)ip->i_num.no_formal_ino,
698 (unsigned long long)ip->i_num.no_addr);
700 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++)
701 fs_warn(sdp, "ip->i_cache[%u] = %s\n",
702 x, (ip->i_cache[x]) ? "!NULL" : "NULL");
707 * gfs2_releasepage - free the metadata associated with a page
708 * @page: the page that's being released
709 * @gfp_mask: passed from Linux VFS, ignored by us
711 * Call try_to_free_buffers() if the buffers in this page can be
717 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
719 struct inode *aspace = page->mapping->host;
720 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
721 struct buffer_head *bh, *head;
722 struct gfs2_bufdata *bd;
723 unsigned long t = jiffies + gfs2_tune_get(sdp, gt_stall_secs) * HZ;
725 if (!page_has_buffers(page))
728 head = bh = page_buffers(page);
730 while (atomic_read(&bh->b_count)) {
731 if (!atomic_read(&aspace->i_writecount))
734 if (time_after_eq(jiffies, t)) {
735 stuck_releasepage(bh);
736 /* should we withdraw here? */
743 gfs2_assert_warn(sdp, !buffer_pinned(bh));
744 gfs2_assert_warn(sdp, !buffer_dirty(bh));
749 gfs2_assert_warn(sdp, bd->bd_bh == bh);
750 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
751 gfs2_assert_warn(sdp, !bd->bd_ail);
753 if (!list_empty(&bd->bd_le.le_list))
755 bh->b_private = NULL;
757 gfs2_log_unlock(sdp);
759 kmem_cache_free(gfs2_bufdata_cachep, bd);
761 bh = bh->b_this_page;
762 } while (bh != head);
765 return try_to_free_buffers(page);
768 const struct address_space_operations gfs2_file_aops = {
769 .writepage = gfs2_writepage,
770 .readpage = gfs2_readpage,
771 .readpages = gfs2_readpages,
772 .sync_page = block_sync_page,
773 .prepare_write = gfs2_prepare_write,
774 .commit_write = gfs2_commit_write,
776 .invalidatepage = gfs2_invalidatepage,
777 .releasepage = gfs2_releasepage,
778 .direct_IO = gfs2_direct_IO,