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 v.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);
122 * gfs2_writepage - Write complete page
123 * @page: Page to write
127 * Some of this is copied from block_write_full_page() although we still
128 * call it to do most of the work.
131 static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
133 struct inode *inode = page->mapping->host;
134 struct gfs2_inode *ip = GFS2_I(inode);
135 struct gfs2_sbd *sdp = GFS2_SB(inode);
136 loff_t i_size = i_size_read(inode);
137 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
142 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) {
146 if (current->journal_info)
149 /* Is the page fully outside i_size? (truncate in progress) */
150 offset = i_size & (PAGE_CACHE_SIZE-1);
151 if (page->index > end_index || (page->index == end_index && !offset)) {
152 page->mapping->a_ops->invalidatepage(page, 0);
154 return 0; /* don't care */
157 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) {
158 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
161 if (!page_has_buffers(page)) {
162 create_empty_buffers(page, inode->i_sb->s_blocksize,
163 (1 << BH_Dirty)|(1 << BH_Uptodate));
165 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
168 error = block_write_full_page(page, get_block_noalloc, wbc);
171 gfs2_meta_cache_flush(ip);
175 redirty_page_for_writepage(wbc, page);
180 static int zero_readpage(struct page *page)
184 kaddr = kmap_atomic(page, KM_USER0);
185 memset(kaddr, 0, PAGE_CACHE_SIZE);
186 kunmap_atomic(page, KM_USER0);
188 SetPageUptodate(page);
194 * stuffed_readpage - Fill in a Linux page with stuffed file data
201 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
203 struct buffer_head *dibh;
207 /* Only the first page of a stuffed file might contain data */
208 if (unlikely(page->index))
209 return zero_readpage(page);
211 error = gfs2_meta_inode_buffer(ip, &dibh);
215 kaddr = kmap_atomic(page, KM_USER0);
216 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
218 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
219 kunmap_atomic(page, KM_USER0);
223 SetPageUptodate(page);
230 * gfs2_readpage - readpage with locking
231 * @file: The file to read a page for. N.B. This may be NULL if we are
232 * reading an internal file.
233 * @page: The page to read
238 static int gfs2_readpage(struct file *file, struct page *page)
240 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
241 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
242 struct gfs2_holder gh;
246 if (likely(file != &gfs2_internal_file_sentinal)) {
248 struct gfs2_file *gf = file->private_data;
249 if (test_bit(GFF_EXLOCK, &gf->f_flags))
252 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|GL_AOP, &gh);
254 error = gfs2_glock_nq_m_atime(1, &gh);
260 if (gfs2_is_stuffed(ip)) {
261 error = stuffed_readpage(ip, page);
264 error = mpage_readpage(page, gfs2_get_block);
266 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
269 if (file != &gfs2_internal_file_sentinal) {
270 gfs2_glock_dq_m(1, &gh);
271 gfs2_holder_uninit(&gh);
278 gfs2_holder_uninit(&gh);
283 * gfs2_readpages - Read a bunch of pages at once
286 * 1. This is only for readahead, so we can simply ignore any things
287 * which are slightly inconvenient (such as locking conflicts between
288 * the page lock and the glock) and return having done no I/O. Its
289 * obviously not something we'd want to do on too regular a basis.
290 * Any I/O we ignore at this time will be done via readpage later.
291 * 2. We have to handle stuffed files here too.
292 * 3. mpage_readpages() does most of the heavy lifting in the common case.
293 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
294 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
295 * well as read-ahead.
297 static int gfs2_readpages(struct file *file, struct address_space *mapping,
298 struct list_head *pages, unsigned nr_pages)
300 struct inode *inode = mapping->host;
301 struct gfs2_inode *ip = GFS2_I(inode);
302 struct gfs2_sbd *sdp = GFS2_SB(inode);
303 struct gfs2_holder gh;
308 if (likely(file != &gfs2_internal_file_sentinal)) {
310 struct gfs2_file *gf = file->private_data;
311 if (test_bit(GFF_EXLOCK, &gf->f_flags))
314 gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
315 LM_FLAG_TRY_1CB|GL_ATIME|GL_AOP, &gh);
317 ret = gfs2_glock_nq_m_atime(1, &gh);
318 if (ret == GLR_TRYFAILED)
324 if (gfs2_is_stuffed(ip)) {
325 struct pagevec lru_pvec;
326 pagevec_init(&lru_pvec, 0);
327 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
328 struct page *page = list_entry(pages->prev, struct page, lru);
329 prefetchw(&page->flags);
330 list_del(&page->lru);
331 if (!add_to_page_cache(page, mapping,
332 page->index, GFP_KERNEL)) {
333 ret = stuffed_readpage(ip, page);
335 if (!pagevec_add(&lru_pvec, page))
336 __pagevec_lru_add(&lru_pvec);
338 page_cache_release(page);
341 pagevec_lru_add(&lru_pvec);
344 /* What we really want to do .... */
345 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
349 gfs2_glock_dq_m(1, &gh);
350 gfs2_holder_uninit(&gh);
353 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
359 /* unlock all pages, we can't do any I/O right now */
360 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
361 struct page *page = list_entry(pages->prev, struct page, lru);
362 list_del(&page->lru);
364 page_cache_release(page);
367 gfs2_holder_uninit(&gh);
372 * gfs2_prepare_write - Prepare to write a page to a file
373 * @file: The file to write to
374 * @page: The page which is to be prepared for writing
375 * @from: From (byte range within page)
376 * @to: To (byte range within page)
381 static int gfs2_prepare_write(struct file *file, struct page *page,
382 unsigned from, unsigned to)
384 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
385 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
386 unsigned int data_blocks, ind_blocks, rblocks;
389 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
390 loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
391 struct gfs2_alloc *al;
393 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|GL_AOP, &ip->i_gh);
394 error = gfs2_glock_nq_m_atime(1, &ip->i_gh);
398 gfs2_write_calc_reserv(ip, to - from, &data_blocks, &ind_blocks);
400 error = gfs2_write_alloc_required(ip, pos, from - to, &alloc_required);
405 if (alloc_required) {
406 al = gfs2_alloc_get(ip);
408 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
412 error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid);
416 al->al_requested = data_blocks + ind_blocks;
417 error = gfs2_inplace_reserve(ip);
422 rblocks = RES_DINODE + ind_blocks;
423 if (gfs2_is_jdata(ip))
424 rblocks += data_blocks ? data_blocks : 1;
425 if (ind_blocks || data_blocks)
426 rblocks += RES_STATFS + RES_QUOTA;
428 error = gfs2_trans_begin(sdp, rblocks, 0);
432 if (gfs2_is_stuffed(ip)) {
433 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
434 error = gfs2_unstuff_dinode(ip, page);
437 } else if (!PageUptodate(page))
438 error = stuffed_readpage(ip, page);
443 error = block_prepare_write(page, from, to, gfs2_get_block);
448 if (alloc_required) {
449 gfs2_inplace_release(ip);
451 gfs2_quota_unlock(ip);
456 gfs2_glock_dq_m(1, &ip->i_gh);
458 gfs2_holder_uninit(&ip->i_gh);
465 * gfs2_commit_write - Commit write to a file
466 * @file: The file to write to
467 * @page: The page containing the data
468 * @from: From (byte range within page)
469 * @to: To (byte range within page)
474 static int gfs2_commit_write(struct file *file, struct page *page,
475 unsigned from, unsigned to)
477 struct inode *inode = page->mapping->host;
478 struct gfs2_inode *ip = GFS2_I(inode);
479 struct gfs2_sbd *sdp = GFS2_SB(inode);
480 int error = -EOPNOTSUPP;
481 struct buffer_head *dibh;
482 struct gfs2_alloc *al = &ip->i_alloc;;
484 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
487 error = gfs2_meta_inode_buffer(ip, &dibh);
491 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
493 if (gfs2_is_stuffed(ip)) {
497 file_size = ((uint64_t)page->index << PAGE_CACHE_SHIFT) + to;
499 kaddr = kmap_atomic(page, KM_USER0);
500 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
501 (char *)kaddr + from, to - from);
502 kunmap_atomic(page, KM_USER0);
504 SetPageUptodate(page);
506 if (inode->i_size < file_size)
507 i_size_write(inode, file_size);
509 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
511 gfs2_page_add_databufs(ip, page, from, to);
512 error = generic_commit_write(file, page, from, to);
517 if (ip->i_di.di_size < inode->i_size)
518 ip->i_di.di_size = inode->i_size;
520 gfs2_dinode_out(&ip->i_di, dibh->b_data);
523 if (al->al_requested) {
524 gfs2_inplace_release(ip);
525 gfs2_quota_unlock(ip);
528 gfs2_glock_dq_m(1, &ip->i_gh);
529 gfs2_holder_uninit(&ip->i_gh);
536 if (al->al_requested) {
537 gfs2_inplace_release(ip);
538 gfs2_quota_unlock(ip);
541 gfs2_glock_dq_m(1, &ip->i_gh);
542 gfs2_holder_uninit(&ip->i_gh);
544 ClearPageUptodate(page);
549 * gfs2_bmap - Block map function
550 * @mapping: Address space info
551 * @lblock: The block to map
553 * Returns: The disk address for the block or 0 on hole or error
556 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
558 struct gfs2_inode *ip = GFS2_I(mapping->host);
559 struct gfs2_holder i_gh;
563 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
567 if (!gfs2_is_stuffed(ip))
568 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
570 gfs2_glock_dq_uninit(&i_gh);
575 static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
577 struct gfs2_bufdata *bd;
583 bh->b_private = NULL;
585 gfs2_log_unlock(sdp);
588 clear_buffer_dirty(bh);
590 clear_buffer_mapped(bh);
591 clear_buffer_req(bh);
592 clear_buffer_new(bh);
593 clear_buffer_delay(bh);
597 static void gfs2_invalidatepage(struct page *page, unsigned long offset)
599 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
600 struct buffer_head *head, *bh, *next;
601 unsigned int curr_off = 0;
603 BUG_ON(!PageLocked(page));
604 if (!page_has_buffers(page))
607 bh = head = page_buffers(page);
609 unsigned int next_off = curr_off + bh->b_size;
610 next = bh->b_this_page;
612 if (offset <= curr_off)
613 discard_buffer(sdp, bh);
617 } while (bh != head);
620 try_to_release_page(page, 0);
625 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
626 const struct iovec *iov, loff_t offset,
627 unsigned long nr_segs)
629 struct file *file = iocb->ki_filp;
630 struct inode *inode = file->f_mapping->host;
631 struct gfs2_inode *ip = GFS2_I(inode);
632 struct gfs2_holder gh;
636 mutex_lock(&inode->i_mutex);
638 * Shared lock, even if its a write, since we do no allocation
639 * on this path. All we need change is atime.
641 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
642 rv = gfs2_glock_nq_m_atime(1, &gh);
646 if (offset > i_size_read(inode))
650 * Should we return an error here? I can't see that O_DIRECT for
651 * a journaled file makes any sense. For now we'll silently fall
652 * back to buffered I/O, likewise we do the same for stuffed
653 * files since they are (a) small and (b) unaligned.
655 if (gfs2_is_jdata(ip))
658 if (gfs2_is_stuffed(ip))
661 rv = blockdev_direct_IO_own_locking(rw, iocb, inode,
663 iov, offset, nr_segs,
664 gfs2_get_block, NULL);
666 gfs2_glock_dq_m(1, &gh);
667 gfs2_holder_uninit(&gh);
669 mutex_unlock(&inode->i_mutex);
675 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
676 * @bh: the buffer we're stuck on
680 static void stuck_releasepage(struct buffer_head *bh)
682 struct inode *inode = bh->b_page->mapping->host;
683 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
684 struct gfs2_bufdata *bd = bh->b_private;
685 struct gfs2_glock *gl;
686 static unsigned limit = 0;
691 fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode);
692 fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
693 (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count));
694 fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
695 fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL");
702 fs_warn(sdp, "gl = (%u, %llu)\n",
703 gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number);
705 fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
706 (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
707 (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
709 if (gl->gl_ops == &gfs2_inode_glops) {
710 struct gfs2_inode *ip = gl->gl_object;
716 fs_warn(sdp, "ip = %llu %llu\n",
717 (unsigned long long)ip->i_num.no_formal_ino,
718 (unsigned long long)ip->i_num.no_addr);
720 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++)
721 fs_warn(sdp, "ip->i_cache[%u] = %s\n",
722 x, (ip->i_cache[x]) ? "!NULL" : "NULL");
727 * gfs2_aspace_releasepage - free the metadata associated with a page
728 * @page: the page that's being released
729 * @gfp_mask: passed from Linux VFS, ignored by us
731 * Call try_to_free_buffers() if the buffers in this page can be
737 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
739 struct inode *aspace = page->mapping->host;
740 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
741 struct buffer_head *bh, *head;
742 struct gfs2_bufdata *bd;
743 unsigned long t = jiffies + gfs2_tune_get(sdp, gt_stall_secs) * HZ;
745 if (!page_has_buffers(page))
748 head = bh = page_buffers(page);
750 while (atomic_read(&bh->b_count)) {
751 if (!atomic_read(&aspace->i_writecount))
754 if (time_after_eq(jiffies, t)) {
755 stuck_releasepage(bh);
756 /* should we withdraw here? */
763 gfs2_assert_warn(sdp, !buffer_pinned(bh));
767 gfs2_assert_warn(sdp, bd->bd_bh == bh);
768 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
769 gfs2_assert_warn(sdp, list_empty(&bd->bd_le.le_list));
770 gfs2_assert_warn(sdp, !bd->bd_ail);
771 kmem_cache_free(gfs2_bufdata_cachep, bd);
772 bh->b_private = NULL;
775 bh = bh->b_this_page;
776 } while (bh != head);
779 return try_to_free_buffers(page);
782 const struct address_space_operations gfs2_file_aops = {
783 .writepage = gfs2_writepage,
784 .readpage = gfs2_readpage,
785 .readpages = gfs2_readpages,
786 .sync_page = block_sync_page,
787 .prepare_write = gfs2_prepare_write,
788 .commit_write = gfs2_commit_write,
790 .invalidatepage = gfs2_invalidatepage,
791 .releasepage = gfs2_releasepage,
792 .direct_IO = gfs2_direct_IO,