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(page->mapping->host);
135 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
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 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
164 error = block_write_full_page(page, get_block_noalloc, wbc);
167 gfs2_meta_cache_flush(ip);
171 redirty_page_for_writepage(wbc, page);
176 static int zero_readpage(struct page *page)
180 kaddr = kmap_atomic(page, KM_USER0);
181 memset(kaddr, 0, PAGE_CACHE_SIZE);
182 kunmap_atomic(page, KM_USER0);
184 SetPageUptodate(page);
190 * stuffed_readpage - Fill in a Linux page with stuffed file data
197 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
199 struct buffer_head *dibh;
203 /* Only the first page of a stuffed file might contain data */
204 if (unlikely(page->index))
205 return zero_readpage(page);
207 error = gfs2_meta_inode_buffer(ip, &dibh);
211 kaddr = kmap_atomic(page, KM_USER0);
212 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
214 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
215 kunmap_atomic(page, KM_USER0);
219 SetPageUptodate(page);
226 * gfs2_readpage - readpage with locking
227 * @file: The file to read a page for. N.B. This may be NULL if we are
228 * reading an internal file.
229 * @page: The page to read
234 static int gfs2_readpage(struct file *file, struct page *page)
236 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
237 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
238 struct gfs2_holder gh;
242 if (likely(file != &gfs2_internal_file_sentinal)) {
244 struct gfs2_file *gf = file->private_data;
245 if (test_bit(GFF_EXLOCK, &gf->f_flags))
248 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|GL_AOP, &gh);
250 error = gfs2_glock_nq_m_atime(1, &gh);
256 if (gfs2_is_stuffed(ip)) {
257 error = stuffed_readpage(ip, page);
260 error = mpage_readpage(page, gfs2_get_block);
262 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
265 if (file != &gfs2_internal_file_sentinal) {
266 gfs2_glock_dq_m(1, &gh);
267 gfs2_holder_uninit(&gh);
274 gfs2_holder_uninit(&gh);
279 * gfs2_readpages - Read a bunch of pages at once
282 * 1. This is only for readahead, so we can simply ignore any things
283 * which are slightly inconvenient (such as locking conflicts between
284 * the page lock and the glock) and return having done no I/O. Its
285 * obviously not something we'd want to do on too regular a basis.
286 * Any I/O we ignore at this time will be done via readpage later.
287 * 2. We have to handle stuffed files here too.
288 * 3. mpage_readpages() does most of the heavy lifting in the common case.
289 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
290 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
291 * well as read-ahead.
293 static int gfs2_readpages(struct file *file, struct address_space *mapping,
294 struct list_head *pages, unsigned nr_pages)
296 struct inode *inode = mapping->host;
297 struct gfs2_inode *ip = GFS2_I(inode);
298 struct gfs2_sbd *sdp = GFS2_SB(inode);
299 struct gfs2_holder gh;
304 if (likely(file != &gfs2_internal_file_sentinal)) {
306 struct gfs2_file *gf = file->private_data;
307 if (test_bit(GFF_EXLOCK, &gf->f_flags))
310 gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
311 LM_FLAG_TRY_1CB|GL_ATIME|GL_AOP, &gh);
313 ret = gfs2_glock_nq_m_atime(1, &gh);
314 if (ret == GLR_TRYFAILED)
320 if (gfs2_is_stuffed(ip)) {
321 struct pagevec lru_pvec;
322 pagevec_init(&lru_pvec, 0);
323 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
324 struct page *page = list_entry(pages->prev, struct page, lru);
325 prefetchw(&page->flags);
326 list_del(&page->lru);
327 if (!add_to_page_cache(page, mapping,
328 page->index, GFP_KERNEL)) {
329 ret = stuffed_readpage(ip, page);
331 if (!pagevec_add(&lru_pvec, page))
332 __pagevec_lru_add(&lru_pvec);
334 page_cache_release(page);
337 pagevec_lru_add(&lru_pvec);
340 /* What we really want to do .... */
341 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
345 gfs2_glock_dq_m(1, &gh);
346 gfs2_holder_uninit(&gh);
349 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
355 /* unlock all pages, we can't do any I/O right now */
356 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
357 struct page *page = list_entry(pages->prev, struct page, lru);
358 list_del(&page->lru);
360 page_cache_release(page);
363 gfs2_holder_uninit(&gh);
368 * gfs2_prepare_write - Prepare to write a page to a file
369 * @file: The file to write to
370 * @page: The page which is to be prepared for writing
371 * @from: From (byte range within page)
372 * @to: To (byte range within page)
377 static int gfs2_prepare_write(struct file *file, struct page *page,
378 unsigned from, unsigned to)
380 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
381 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
382 unsigned int data_blocks, ind_blocks, rblocks;
385 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
386 loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
387 struct gfs2_alloc *al;
389 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|GL_AOP, &ip->i_gh);
390 error = gfs2_glock_nq_m_atime(1, &ip->i_gh);
394 gfs2_write_calc_reserv(ip, to - from, &data_blocks, &ind_blocks);
396 error = gfs2_write_alloc_required(ip, pos, from - to, &alloc_required);
401 if (alloc_required) {
402 al = gfs2_alloc_get(ip);
404 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
408 error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid);
412 al->al_requested = data_blocks + ind_blocks;
413 error = gfs2_inplace_reserve(ip);
418 rblocks = RES_DINODE + ind_blocks;
419 if (gfs2_is_jdata(ip))
420 rblocks += data_blocks ? data_blocks : 1;
421 if (ind_blocks || data_blocks)
422 rblocks += RES_STATFS + RES_QUOTA;
424 error = gfs2_trans_begin(sdp, rblocks, 0);
428 if (gfs2_is_stuffed(ip)) {
429 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
430 error = gfs2_unstuff_dinode(ip, page);
433 } else if (!PageUptodate(page))
434 error = stuffed_readpage(ip, page);
439 error = block_prepare_write(page, from, to, gfs2_get_block);
444 if (alloc_required) {
445 gfs2_inplace_release(ip);
447 gfs2_quota_unlock(ip);
452 gfs2_glock_dq_m(1, &ip->i_gh);
454 gfs2_holder_uninit(&ip->i_gh);
461 * gfs2_commit_write - Commit write to a file
462 * @file: The file to write to
463 * @page: The page containing the data
464 * @from: From (byte range within page)
465 * @to: To (byte range within page)
470 static int gfs2_commit_write(struct file *file, struct page *page,
471 unsigned from, unsigned to)
473 struct inode *inode = page->mapping->host;
474 struct gfs2_inode *ip = GFS2_I(inode);
475 struct gfs2_sbd *sdp = GFS2_SB(inode);
476 int error = -EOPNOTSUPP;
477 struct buffer_head *dibh;
478 struct gfs2_alloc *al = &ip->i_alloc;;
480 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
483 error = gfs2_meta_inode_buffer(ip, &dibh);
487 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
489 if (gfs2_is_stuffed(ip)) {
493 file_size = ((uint64_t)page->index << PAGE_CACHE_SHIFT) + to;
495 kaddr = kmap_atomic(page, KM_USER0);
496 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
497 (char *)kaddr + from, to - from);
498 kunmap_atomic(page, KM_USER0);
500 SetPageUptodate(page);
502 if (inode->i_size < file_size)
503 i_size_write(inode, file_size);
505 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
507 gfs2_page_add_databufs(ip, page, from, to);
508 error = generic_commit_write(file, page, from, to);
513 if (ip->i_di.di_size < inode->i_size)
514 ip->i_di.di_size = inode->i_size;
516 gfs2_dinode_out(&ip->i_di, dibh->b_data);
519 if (al->al_requested) {
520 gfs2_inplace_release(ip);
521 gfs2_quota_unlock(ip);
524 gfs2_glock_dq_m(1, &ip->i_gh);
525 gfs2_holder_uninit(&ip->i_gh);
532 if (al->al_requested) {
533 gfs2_inplace_release(ip);
534 gfs2_quota_unlock(ip);
537 gfs2_glock_dq_m(1, &ip->i_gh);
538 gfs2_holder_uninit(&ip->i_gh);
540 ClearPageUptodate(page);
545 * gfs2_bmap - Block map function
546 * @mapping: Address space info
547 * @lblock: The block to map
549 * Returns: The disk address for the block or 0 on hole or error
552 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
554 struct gfs2_inode *ip = GFS2_I(mapping->host);
555 struct gfs2_holder i_gh;
559 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
563 if (!gfs2_is_stuffed(ip))
564 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
566 gfs2_glock_dq_uninit(&i_gh);
571 static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
573 struct gfs2_bufdata *bd;
579 bh->b_private = NULL;
580 gfs2_log_unlock(sdp);
583 gfs2_log_unlock(sdp);
586 clear_buffer_dirty(bh);
588 clear_buffer_mapped(bh);
589 clear_buffer_req(bh);
590 clear_buffer_new(bh);
591 clear_buffer_delay(bh);
595 static void gfs2_invalidatepage(struct page *page, unsigned long offset)
597 struct gfs2_sbd *sdp = page->mapping->host->i_sb->s_fs_info;
598 struct buffer_head *head, *bh, *next;
599 unsigned int curr_off = 0;
601 BUG_ON(!PageLocked(page));
602 if (!page_has_buffers(page))
605 bh = head = page_buffers(page);
607 unsigned int next_off = curr_off + bh->b_size;
608 next = bh->b_this_page;
610 if (offset <= curr_off)
611 discard_buffer(sdp, bh);
615 } while (bh != head);
618 try_to_release_page(page, 0);
623 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
624 const struct iovec *iov, loff_t offset,
625 unsigned long nr_segs)
627 struct file *file = iocb->ki_filp;
628 struct inode *inode = file->f_mapping->host;
629 struct gfs2_inode *ip = GFS2_I(inode);
630 struct gfs2_holder gh;
634 mutex_lock(&inode->i_mutex);
636 * Shared lock, even if its a write, since we do no allocation
637 * on this path. All we need change is atime.
639 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
640 rv = gfs2_glock_nq_m_atime(1, &gh);
644 if (offset > i_size_read(inode))
648 * Should we return an error here? I can't see that O_DIRECT for
649 * a journaled file makes any sense. For now we'll silently fall
650 * back to buffered I/O, likewise we do the same for stuffed
651 * files since they are (a) small and (b) unaligned.
653 if (gfs2_is_jdata(ip))
656 if (gfs2_is_stuffed(ip))
659 rv = blockdev_direct_IO_own_locking(rw, iocb, inode,
661 iov, offset, nr_segs,
662 gfs2_get_block, NULL);
664 gfs2_glock_dq_m(1, &gh);
665 gfs2_holder_uninit(&gh);
667 mutex_unlock(&inode->i_mutex);
673 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
674 * @bh: the buffer we're stuck on
678 static void stuck_releasepage(struct buffer_head *bh)
680 struct inode *inode = bh->b_page->mapping->host;
681 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
682 struct gfs2_bufdata *bd = bh->b_private;
683 struct gfs2_glock *gl;
685 fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode);
686 fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
687 (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count));
688 fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
689 fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL");
696 fs_warn(sdp, "gl = (%u, %llu)\n",
697 gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number);
699 fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
700 (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
701 (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
703 if (gl->gl_ops == &gfs2_inode_glops) {
704 struct gfs2_inode *ip = gl->gl_object;
710 fs_warn(sdp, "ip = %llu %llu\n",
711 (unsigned long long)ip->i_num.no_formal_ino,
712 (unsigned long long)ip->i_num.no_addr);
714 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++)
715 fs_warn(sdp, "ip->i_cache[%u] = %s\n",
716 x, (ip->i_cache[x]) ? "!NULL" : "NULL");
721 * gfs2_aspace_releasepage - free the metadata associated with a page
722 * @page: the page that's being released
723 * @gfp_mask: passed from Linux VFS, ignored by us
725 * Call try_to_free_buffers() if the buffers in this page can be
731 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
733 struct inode *aspace = page->mapping->host;
734 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
735 struct buffer_head *bh, *head;
736 struct gfs2_bufdata *bd;
739 if (!page_has_buffers(page))
742 head = bh = page_buffers(page);
746 while (atomic_read(&bh->b_count)) {
747 if (atomic_read(&aspace->i_writecount)) {
748 if (time_after_eq(jiffies, t +
749 gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
750 stuck_releasepage(bh);
761 gfs2_assert_warn(sdp, !buffer_pinned(bh));
765 gfs2_assert_warn(sdp, bd->bd_bh == bh);
766 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
767 gfs2_assert_warn(sdp, list_empty(&bd->bd_le.le_list));
768 gfs2_assert_warn(sdp, !bd->bd_ail);
769 kmem_cache_free(gfs2_bufdata_cachep, bd);
770 bh->b_private = NULL;
773 bh = bh->b_this_page;
778 return try_to_free_buffers(page);
781 const struct address_space_operations gfs2_file_aops = {
782 .writepage = gfs2_writepage,
783 .readpage = gfs2_readpage,
784 .readpages = gfs2_readpages,
785 .sync_page = block_sync_page,
786 .prepare_write = gfs2_prepare_write,
787 .commit_write = gfs2_commit_write,
789 .invalidatepage = gfs2_invalidatepage,
790 .releasepage = gfs2_releasepage,
791 .direct_IO = gfs2_direct_IO,