xref: /openbmc/linux/fs/gfs2/aops.c (revision 1d1997db)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
4  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
5  */
6 
7 #include <linux/sched.h>
8 #include <linux/slab.h>
9 #include <linux/spinlock.h>
10 #include <linux/completion.h>
11 #include <linux/buffer_head.h>
12 #include <linux/pagemap.h>
13 #include <linux/pagevec.h>
14 #include <linux/mpage.h>
15 #include <linux/fs.h>
16 #include <linux/writeback.h>
17 #include <linux/swap.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/backing-dev.h>
20 #include <linux/uio.h>
21 #include <trace/events/writeback.h>
22 #include <linux/sched/signal.h>
23 
24 #include "gfs2.h"
25 #include "incore.h"
26 #include "bmap.h"
27 #include "glock.h"
28 #include "inode.h"
29 #include "log.h"
30 #include "meta_io.h"
31 #include "quota.h"
32 #include "trans.h"
33 #include "rgrp.h"
34 #include "super.h"
35 #include "util.h"
36 #include "glops.h"
37 #include "aops.h"
38 
39 
40 void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
41 			    unsigned int from, unsigned int len)
42 {
43 	struct buffer_head *head = page_buffers(page);
44 	unsigned int bsize = head->b_size;
45 	struct buffer_head *bh;
46 	unsigned int to = from + len;
47 	unsigned int start, end;
48 
49 	for (bh = head, start = 0; bh != head || !start;
50 	     bh = bh->b_this_page, start = end) {
51 		end = start + bsize;
52 		if (end <= from)
53 			continue;
54 		if (start >= to)
55 			break;
56 		set_buffer_uptodate(bh);
57 		gfs2_trans_add_data(ip->i_gl, bh);
58 	}
59 }
60 
61 /**
62  * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
63  * @inode: The inode
64  * @lblock: The block number to look up
65  * @bh_result: The buffer head to return the result in
66  * @create: Non-zero if we may add block to the file
67  *
68  * Returns: errno
69  */
70 
71 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
72 				  struct buffer_head *bh_result, int create)
73 {
74 	int error;
75 
76 	error = gfs2_block_map(inode, lblock, bh_result, 0);
77 	if (error)
78 		return error;
79 	if (!buffer_mapped(bh_result))
80 		return -EIO;
81 	return 0;
82 }
83 
84 /**
85  * gfs2_writepage - Write page for writeback mappings
86  * @page: The page
87  * @wbc: The writeback control
88  */
89 static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
90 {
91 	struct inode *inode = page->mapping->host;
92 	struct gfs2_inode *ip = GFS2_I(inode);
93 	struct gfs2_sbd *sdp = GFS2_SB(inode);
94 	loff_t i_size = i_size_read(inode);
95 	pgoff_t end_index = i_size >> PAGE_SHIFT;
96 	unsigned offset;
97 
98 	if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
99 		goto out;
100 	if (current->journal_info)
101 		goto redirty;
102 	/* Is the page fully outside i_size? (truncate in progress) */
103 	offset = i_size & (PAGE_SIZE-1);
104 	if (page->index > end_index || (page->index == end_index && !offset)) {
105 		page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
106 		goto out;
107 	}
108 
109 	return nobh_writepage(page, gfs2_get_block_noalloc, wbc);
110 
111 redirty:
112 	redirty_page_for_writepage(wbc, page);
113 out:
114 	unlock_page(page);
115 	return 0;
116 }
117 
118 /* This is the same as calling block_write_full_page, but it also
119  * writes pages outside of i_size
120  */
121 static int gfs2_write_full_page(struct page *page, get_block_t *get_block,
122 				struct writeback_control *wbc)
123 {
124 	struct inode * const inode = page->mapping->host;
125 	loff_t i_size = i_size_read(inode);
126 	const pgoff_t end_index = i_size >> PAGE_SHIFT;
127 	unsigned offset;
128 
129 	/*
130 	 * The page straddles i_size.  It must be zeroed out on each and every
131 	 * writepage invocation because it may be mmapped.  "A file is mapped
132 	 * in multiples of the page size.  For a file that is not a multiple of
133 	 * the  page size, the remaining memory is zeroed when mapped, and
134 	 * writes to that region are not written out to the file."
135 	 */
136 	offset = i_size & (PAGE_SIZE - 1);
137 	if (page->index == end_index && offset)
138 		zero_user_segment(page, offset, PAGE_SIZE);
139 
140 	return __block_write_full_page(inode, page, get_block, wbc,
141 				       end_buffer_async_write);
142 }
143 
144 /**
145  * __gfs2_jdata_writepage - The core of jdata writepage
146  * @page: The page to write
147  * @wbc: The writeback control
148  *
149  * This is shared between writepage and writepages and implements the
150  * core of the writepage operation. If a transaction is required then
151  * PageChecked will have been set and the transaction will have
152  * already been started before this is called.
153  */
154 
155 static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
156 {
157 	struct inode *inode = page->mapping->host;
158 	struct gfs2_inode *ip = GFS2_I(inode);
159 	struct gfs2_sbd *sdp = GFS2_SB(inode);
160 
161 	if (PageChecked(page)) {
162 		ClearPageChecked(page);
163 		if (!page_has_buffers(page)) {
164 			create_empty_buffers(page, inode->i_sb->s_blocksize,
165 					     BIT(BH_Dirty)|BIT(BH_Uptodate));
166 		}
167 		gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize);
168 	}
169 	return gfs2_write_full_page(page, gfs2_get_block_noalloc, wbc);
170 }
171 
172 /**
173  * gfs2_jdata_writepage - Write complete page
174  * @page: Page to write
175  * @wbc: The writeback control
176  *
177  * Returns: errno
178  *
179  */
180 
181 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
182 {
183 	struct inode *inode = page->mapping->host;
184 	struct gfs2_inode *ip = GFS2_I(inode);
185 	struct gfs2_sbd *sdp = GFS2_SB(inode);
186 	int ret;
187 
188 	if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
189 		goto out;
190 	if (PageChecked(page) || current->journal_info)
191 		goto out_ignore;
192 	ret = __gfs2_jdata_writepage(page, wbc);
193 	return ret;
194 
195 out_ignore:
196 	redirty_page_for_writepage(wbc, page);
197 out:
198 	unlock_page(page);
199 	return 0;
200 }
201 
202 /**
203  * gfs2_writepages - Write a bunch of dirty pages back to disk
204  * @mapping: The mapping to write
205  * @wbc: Write-back control
206  *
207  * Used for both ordered and writeback modes.
208  */
209 static int gfs2_writepages(struct address_space *mapping,
210 			   struct writeback_control *wbc)
211 {
212 	struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
213 	int ret = mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
214 
215 	/*
216 	 * Even if we didn't write any pages here, we might still be holding
217 	 * dirty pages in the ail. We forcibly flush the ail because we don't
218 	 * want balance_dirty_pages() to loop indefinitely trying to write out
219 	 * pages held in the ail that it can't find.
220 	 */
221 	if (ret == 0)
222 		set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
223 
224 	return ret;
225 }
226 
227 /**
228  * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
229  * @mapping: The mapping
230  * @wbc: The writeback control
231  * @pvec: The vector of pages
232  * @nr_pages: The number of pages to write
233  * @done_index: Page index
234  *
235  * Returns: non-zero if loop should terminate, zero otherwise
236  */
237 
238 static int gfs2_write_jdata_pagevec(struct address_space *mapping,
239 				    struct writeback_control *wbc,
240 				    struct pagevec *pvec,
241 				    int nr_pages,
242 				    pgoff_t *done_index)
243 {
244 	struct inode *inode = mapping->host;
245 	struct gfs2_sbd *sdp = GFS2_SB(inode);
246 	unsigned nrblocks = nr_pages * (PAGE_SIZE >> inode->i_blkbits);
247 	int i;
248 	int ret;
249 
250 	ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
251 	if (ret < 0)
252 		return ret;
253 
254 	for(i = 0; i < nr_pages; i++) {
255 		struct page *page = pvec->pages[i];
256 
257 		*done_index = page->index;
258 
259 		lock_page(page);
260 
261 		if (unlikely(page->mapping != mapping)) {
262 continue_unlock:
263 			unlock_page(page);
264 			continue;
265 		}
266 
267 		if (!PageDirty(page)) {
268 			/* someone wrote it for us */
269 			goto continue_unlock;
270 		}
271 
272 		if (PageWriteback(page)) {
273 			if (wbc->sync_mode != WB_SYNC_NONE)
274 				wait_on_page_writeback(page);
275 			else
276 				goto continue_unlock;
277 		}
278 
279 		BUG_ON(PageWriteback(page));
280 		if (!clear_page_dirty_for_io(page))
281 			goto continue_unlock;
282 
283 		trace_wbc_writepage(wbc, inode_to_bdi(inode));
284 
285 		ret = __gfs2_jdata_writepage(page, wbc);
286 		if (unlikely(ret)) {
287 			if (ret == AOP_WRITEPAGE_ACTIVATE) {
288 				unlock_page(page);
289 				ret = 0;
290 			} else {
291 
292 				/*
293 				 * done_index is set past this page,
294 				 * so media errors will not choke
295 				 * background writeout for the entire
296 				 * file. This has consequences for
297 				 * range_cyclic semantics (ie. it may
298 				 * not be suitable for data integrity
299 				 * writeout).
300 				 */
301 				*done_index = page->index + 1;
302 				ret = 1;
303 				break;
304 			}
305 		}
306 
307 		/*
308 		 * We stop writing back only if we are not doing
309 		 * integrity sync. In case of integrity sync we have to
310 		 * keep going until we have written all the pages
311 		 * we tagged for writeback prior to entering this loop.
312 		 */
313 		if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
314 			ret = 1;
315 			break;
316 		}
317 
318 	}
319 	gfs2_trans_end(sdp);
320 	return ret;
321 }
322 
323 /**
324  * gfs2_write_cache_jdata - Like write_cache_pages but different
325  * @mapping: The mapping to write
326  * @wbc: The writeback control
327  *
328  * The reason that we use our own function here is that we need to
329  * start transactions before we grab page locks. This allows us
330  * to get the ordering right.
331  */
332 
333 static int gfs2_write_cache_jdata(struct address_space *mapping,
334 				  struct writeback_control *wbc)
335 {
336 	int ret = 0;
337 	int done = 0;
338 	struct pagevec pvec;
339 	int nr_pages;
340 	pgoff_t uninitialized_var(writeback_index);
341 	pgoff_t index;
342 	pgoff_t end;
343 	pgoff_t done_index;
344 	int cycled;
345 	int range_whole = 0;
346 	xa_mark_t tag;
347 
348 	pagevec_init(&pvec);
349 	if (wbc->range_cyclic) {
350 		writeback_index = mapping->writeback_index; /* prev offset */
351 		index = writeback_index;
352 		if (index == 0)
353 			cycled = 1;
354 		else
355 			cycled = 0;
356 		end = -1;
357 	} else {
358 		index = wbc->range_start >> PAGE_SHIFT;
359 		end = wbc->range_end >> PAGE_SHIFT;
360 		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
361 			range_whole = 1;
362 		cycled = 1; /* ignore range_cyclic tests */
363 	}
364 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
365 		tag = PAGECACHE_TAG_TOWRITE;
366 	else
367 		tag = PAGECACHE_TAG_DIRTY;
368 
369 retry:
370 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
371 		tag_pages_for_writeback(mapping, index, end);
372 	done_index = index;
373 	while (!done && (index <= end)) {
374 		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
375 				tag);
376 		if (nr_pages == 0)
377 			break;
378 
379 		ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, &done_index);
380 		if (ret)
381 			done = 1;
382 		if (ret > 0)
383 			ret = 0;
384 		pagevec_release(&pvec);
385 		cond_resched();
386 	}
387 
388 	if (!cycled && !done) {
389 		/*
390 		 * range_cyclic:
391 		 * We hit the last page and there is more work to be done: wrap
392 		 * back to the start of the file
393 		 */
394 		cycled = 1;
395 		index = 0;
396 		end = writeback_index - 1;
397 		goto retry;
398 	}
399 
400 	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
401 		mapping->writeback_index = done_index;
402 
403 	return ret;
404 }
405 
406 
407 /**
408  * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
409  * @mapping: The mapping to write
410  * @wbc: The writeback control
411  *
412  */
413 
414 static int gfs2_jdata_writepages(struct address_space *mapping,
415 				 struct writeback_control *wbc)
416 {
417 	struct gfs2_inode *ip = GFS2_I(mapping->host);
418 	struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
419 	int ret;
420 
421 	ret = gfs2_write_cache_jdata(mapping, wbc);
422 	if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
423 		gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
424 			       GFS2_LFC_JDATA_WPAGES);
425 		ret = gfs2_write_cache_jdata(mapping, wbc);
426 	}
427 	return ret;
428 }
429 
430 /**
431  * stuffed_readpage - Fill in a Linux page with stuffed file data
432  * @ip: the inode
433  * @page: the page
434  *
435  * Returns: errno
436  */
437 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
438 {
439 	struct buffer_head *dibh;
440 	u64 dsize = i_size_read(&ip->i_inode);
441 	void *kaddr;
442 	int error;
443 
444 	/*
445 	 * Due to the order of unstuffing files and ->fault(), we can be
446 	 * asked for a zero page in the case of a stuffed file being extended,
447 	 * so we need to supply one here. It doesn't happen often.
448 	 */
449 	if (unlikely(page->index)) {
450 		zero_user(page, 0, PAGE_SIZE);
451 		SetPageUptodate(page);
452 		return 0;
453 	}
454 
455 	error = gfs2_meta_inode_buffer(ip, &dibh);
456 	if (error)
457 		return error;
458 
459 	kaddr = kmap_atomic(page);
460 	if (dsize > gfs2_max_stuffed_size(ip))
461 		dsize = gfs2_max_stuffed_size(ip);
462 	memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
463 	memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
464 	kunmap_atomic(kaddr);
465 	flush_dcache_page(page);
466 	brelse(dibh);
467 	SetPageUptodate(page);
468 
469 	return 0;
470 }
471 
472 
473 /**
474  * __gfs2_readpage - readpage
475  * @file: The file to read a page for
476  * @page: The page to read
477  *
478  * This is the core of gfs2's readpage. It's used by the internal file
479  * reading code as in that case we already hold the glock. Also it's
480  * called by gfs2_readpage() once the required lock has been granted.
481  */
482 
483 static int __gfs2_readpage(void *file, struct page *page)
484 {
485 	struct gfs2_inode *ip = GFS2_I(page->mapping->host);
486 	struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
487 
488 	int error;
489 
490 	if (i_blocksize(page->mapping->host) == PAGE_SIZE &&
491 	    !page_has_buffers(page)) {
492 		error = iomap_readpage(page, &gfs2_iomap_ops);
493 	} else if (gfs2_is_stuffed(ip)) {
494 		error = stuffed_readpage(ip, page);
495 		unlock_page(page);
496 	} else {
497 		error = mpage_readpage(page, gfs2_block_map);
498 	}
499 
500 	if (unlikely(gfs2_withdrawn(sdp)))
501 		return -EIO;
502 
503 	return error;
504 }
505 
506 /**
507  * gfs2_readpage - read a page of a file
508  * @file: The file to read
509  * @page: The page of the file
510  *
511  * This deals with the locking required. We have to unlock and
512  * relock the page in order to get the locking in the right
513  * order.
514  */
515 
516 static int gfs2_readpage(struct file *file, struct page *page)
517 {
518 	struct address_space *mapping = page->mapping;
519 	struct gfs2_inode *ip = GFS2_I(mapping->host);
520 	struct gfs2_holder gh;
521 	int error;
522 
523 	unlock_page(page);
524 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
525 	error = gfs2_glock_nq(&gh);
526 	if (unlikely(error))
527 		goto out;
528 	error = AOP_TRUNCATED_PAGE;
529 	lock_page(page);
530 	if (page->mapping == mapping && !PageUptodate(page))
531 		error = __gfs2_readpage(file, page);
532 	else
533 		unlock_page(page);
534 	gfs2_glock_dq(&gh);
535 out:
536 	gfs2_holder_uninit(&gh);
537 	if (error && error != AOP_TRUNCATED_PAGE)
538 		lock_page(page);
539 	return error;
540 }
541 
542 /**
543  * gfs2_internal_read - read an internal file
544  * @ip: The gfs2 inode
545  * @buf: The buffer to fill
546  * @pos: The file position
547  * @size: The amount to read
548  *
549  */
550 
551 int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
552                        unsigned size)
553 {
554 	struct address_space *mapping = ip->i_inode.i_mapping;
555 	unsigned long index = *pos >> PAGE_SHIFT;
556 	unsigned offset = *pos & (PAGE_SIZE - 1);
557 	unsigned copied = 0;
558 	unsigned amt;
559 	struct page *page;
560 	void *p;
561 
562 	do {
563 		amt = size - copied;
564 		if (offset + size > PAGE_SIZE)
565 			amt = PAGE_SIZE - offset;
566 		page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
567 		if (IS_ERR(page))
568 			return PTR_ERR(page);
569 		p = kmap_atomic(page);
570 		memcpy(buf + copied, p + offset, amt);
571 		kunmap_atomic(p);
572 		put_page(page);
573 		copied += amt;
574 		index++;
575 		offset = 0;
576 	} while(copied < size);
577 	(*pos) += size;
578 	return size;
579 }
580 
581 /**
582  * gfs2_readpages - Read a bunch of pages at once
583  * @file: The file to read from
584  * @mapping: Address space info
585  * @pages: List of pages to read
586  * @nr_pages: Number of pages to read
587  *
588  * Some notes:
589  * 1. This is only for readahead, so we can simply ignore any things
590  *    which are slightly inconvenient (such as locking conflicts between
591  *    the page lock and the glock) and return having done no I/O. Its
592  *    obviously not something we'd want to do on too regular a basis.
593  *    Any I/O we ignore at this time will be done via readpage later.
594  * 2. We don't handle stuffed files here we let readpage do the honours.
595  * 3. mpage_readpages() does most of the heavy lifting in the common case.
596  * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
597  */
598 
599 static int gfs2_readpages(struct file *file, struct address_space *mapping,
600 			  struct list_head *pages, unsigned nr_pages)
601 {
602 	struct inode *inode = mapping->host;
603 	struct gfs2_inode *ip = GFS2_I(inode);
604 	struct gfs2_sbd *sdp = GFS2_SB(inode);
605 	struct gfs2_holder gh;
606 	int ret;
607 
608 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
609 	ret = gfs2_glock_nq(&gh);
610 	if (unlikely(ret))
611 		goto out_uninit;
612 	if (!gfs2_is_stuffed(ip))
613 		ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
614 	gfs2_glock_dq(&gh);
615 out_uninit:
616 	gfs2_holder_uninit(&gh);
617 	if (unlikely(gfs2_withdrawn(sdp)))
618 		ret = -EIO;
619 	return ret;
620 }
621 
622 /**
623  * adjust_fs_space - Adjusts the free space available due to gfs2_grow
624  * @inode: the rindex inode
625  */
626 void adjust_fs_space(struct inode *inode)
627 {
628 	struct gfs2_sbd *sdp = GFS2_SB(inode);
629 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
630 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
631 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
632 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
633 	struct buffer_head *m_bh, *l_bh;
634 	u64 fs_total, new_free;
635 
636 	if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0)
637 		return;
638 
639 	/* Total up the file system space, according to the latest rindex. */
640 	fs_total = gfs2_ri_total(sdp);
641 	if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
642 		goto out;
643 
644 	spin_lock(&sdp->sd_statfs_spin);
645 	gfs2_statfs_change_in(m_sc, m_bh->b_data +
646 			      sizeof(struct gfs2_dinode));
647 	if (fs_total > (m_sc->sc_total + l_sc->sc_total))
648 		new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
649 	else
650 		new_free = 0;
651 	spin_unlock(&sdp->sd_statfs_spin);
652 	fs_warn(sdp, "File system extended by %llu blocks.\n",
653 		(unsigned long long)new_free);
654 	gfs2_statfs_change(sdp, new_free, new_free, 0);
655 
656 	if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0)
657 		goto out2;
658 	update_statfs(sdp, m_bh, l_bh);
659 	brelse(l_bh);
660 out2:
661 	brelse(m_bh);
662 out:
663 	sdp->sd_rindex_uptodate = 0;
664 	gfs2_trans_end(sdp);
665 }
666 
667 /**
668  * jdata_set_page_dirty - Page dirtying function
669  * @page: The page to dirty
670  *
671  * Returns: 1 if it dirtyed the page, or 0 otherwise
672  */
673 
674 static int jdata_set_page_dirty(struct page *page)
675 {
676 	SetPageChecked(page);
677 	return __set_page_dirty_buffers(page);
678 }
679 
680 /**
681  * gfs2_bmap - Block map function
682  * @mapping: Address space info
683  * @lblock: The block to map
684  *
685  * Returns: The disk address for the block or 0 on hole or error
686  */
687 
688 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
689 {
690 	struct gfs2_inode *ip = GFS2_I(mapping->host);
691 	struct gfs2_holder i_gh;
692 	sector_t dblock = 0;
693 	int error;
694 
695 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
696 	if (error)
697 		return 0;
698 
699 	if (!gfs2_is_stuffed(ip))
700 		dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops);
701 
702 	gfs2_glock_dq_uninit(&i_gh);
703 
704 	return dblock;
705 }
706 
707 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
708 {
709 	struct gfs2_bufdata *bd;
710 
711 	lock_buffer(bh);
712 	gfs2_log_lock(sdp);
713 	clear_buffer_dirty(bh);
714 	bd = bh->b_private;
715 	if (bd) {
716 		if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
717 			list_del_init(&bd->bd_list);
718 		else
719 			gfs2_remove_from_journal(bh, REMOVE_JDATA);
720 	}
721 	bh->b_bdev = NULL;
722 	clear_buffer_mapped(bh);
723 	clear_buffer_req(bh);
724 	clear_buffer_new(bh);
725 	gfs2_log_unlock(sdp);
726 	unlock_buffer(bh);
727 }
728 
729 static void gfs2_invalidatepage(struct page *page, unsigned int offset,
730 				unsigned int length)
731 {
732 	struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
733 	unsigned int stop = offset + length;
734 	int partial_page = (offset || length < PAGE_SIZE);
735 	struct buffer_head *bh, *head;
736 	unsigned long pos = 0;
737 
738 	BUG_ON(!PageLocked(page));
739 	if (!partial_page)
740 		ClearPageChecked(page);
741 	if (!page_has_buffers(page))
742 		goto out;
743 
744 	bh = head = page_buffers(page);
745 	do {
746 		if (pos + bh->b_size > stop)
747 			return;
748 
749 		if (offset <= pos)
750 			gfs2_discard(sdp, bh);
751 		pos += bh->b_size;
752 		bh = bh->b_this_page;
753 	} while (bh != head);
754 out:
755 	if (!partial_page)
756 		try_to_release_page(page, 0);
757 }
758 
759 /**
760  * gfs2_releasepage - free the metadata associated with a page
761  * @page: the page that's being released
762  * @gfp_mask: passed from Linux VFS, ignored by us
763  *
764  * Calls try_to_free_buffers() to free the buffers and put the page if the
765  * buffers can be released.
766  *
767  * Returns: 1 if the page was put or else 0
768  */
769 
770 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
771 {
772 	struct address_space *mapping = page->mapping;
773 	struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
774 	struct buffer_head *bh, *head;
775 	struct gfs2_bufdata *bd;
776 
777 	if (!page_has_buffers(page))
778 		return 0;
779 
780 	/*
781 	 * From xfs_vm_releasepage: mm accommodates an old ext3 case where
782 	 * clean pages might not have had the dirty bit cleared.  Thus, it can
783 	 * send actual dirty pages to ->releasepage() via shrink_active_list().
784 	 *
785 	 * As a workaround, we skip pages that contain dirty buffers below.
786 	 * Once ->releasepage isn't called on dirty pages anymore, we can warn
787 	 * on dirty buffers like we used to here again.
788 	 */
789 
790 	gfs2_log_lock(sdp);
791 	spin_lock(&sdp->sd_ail_lock);
792 	head = bh = page_buffers(page);
793 	do {
794 		if (atomic_read(&bh->b_count))
795 			goto cannot_release;
796 		bd = bh->b_private;
797 		if (bd && bd->bd_tr)
798 			goto cannot_release;
799 		if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh)))
800 			goto cannot_release;
801 		bh = bh->b_this_page;
802 	} while(bh != head);
803 	spin_unlock(&sdp->sd_ail_lock);
804 
805 	head = bh = page_buffers(page);
806 	do {
807 		bd = bh->b_private;
808 		if (bd) {
809 			gfs2_assert_warn(sdp, bd->bd_bh == bh);
810 			if (!list_empty(&bd->bd_list))
811 				list_del_init(&bd->bd_list);
812 			bd->bd_bh = NULL;
813 			bh->b_private = NULL;
814 			kmem_cache_free(gfs2_bufdata_cachep, bd);
815 		}
816 
817 		bh = bh->b_this_page;
818 	} while (bh != head);
819 	gfs2_log_unlock(sdp);
820 
821 	return try_to_free_buffers(page);
822 
823 cannot_release:
824 	spin_unlock(&sdp->sd_ail_lock);
825 	gfs2_log_unlock(sdp);
826 	return 0;
827 }
828 
829 static const struct address_space_operations gfs2_aops = {
830 	.writepage = gfs2_writepage,
831 	.writepages = gfs2_writepages,
832 	.readpage = gfs2_readpage,
833 	.readpages = gfs2_readpages,
834 	.bmap = gfs2_bmap,
835 	.invalidatepage = gfs2_invalidatepage,
836 	.releasepage = gfs2_releasepage,
837 	.direct_IO = noop_direct_IO,
838 	.migratepage = buffer_migrate_page,
839 	.is_partially_uptodate = block_is_partially_uptodate,
840 	.error_remove_page = generic_error_remove_page,
841 };
842 
843 static const struct address_space_operations gfs2_jdata_aops = {
844 	.writepage = gfs2_jdata_writepage,
845 	.writepages = gfs2_jdata_writepages,
846 	.readpage = gfs2_readpage,
847 	.readpages = gfs2_readpages,
848 	.set_page_dirty = jdata_set_page_dirty,
849 	.bmap = gfs2_bmap,
850 	.invalidatepage = gfs2_invalidatepage,
851 	.releasepage = gfs2_releasepage,
852 	.is_partially_uptodate = block_is_partially_uptodate,
853 	.error_remove_page = generic_error_remove_page,
854 };
855 
856 void gfs2_set_aops(struct inode *inode)
857 {
858 	if (gfs2_is_jdata(GFS2_I(inode)))
859 		inode->i_mapping->a_ops = &gfs2_jdata_aops;
860 	else
861 		inode->i_mapping->a_ops = &gfs2_aops;
862 }
863