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 
gfs2_page_add_databufs(struct gfs2_inode * ip,struct page * page,unsigned int from,unsigned int len)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 
gfs2_get_block_noalloc(struct inode * inode,sector_t lblock,struct buffer_head * bh_result,int create)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 -ENODATA;
81 	return 0;
82 }
83 
84 /**
85  * gfs2_writepage - Write page for writeback mappings
86  * @page: The page
87  * @wbc: The writeback control
88  */
gfs2_writepage(struct page * page,struct writeback_control * wbc)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 	struct iomap_writepage_ctx wpc = { };
95 
96 	if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
97 		goto out;
98 	if (current->journal_info)
99 		goto redirty;
100 	return iomap_writepage(page, wbc, &wpc, &gfs2_writeback_ops);
101 
102 redirty:
103 	redirty_page_for_writepage(wbc, page);
104 out:
105 	unlock_page(page);
106 	return 0;
107 }
108 
109 /**
110  * gfs2_write_jdata_page - gfs2 jdata-specific version of block_write_full_page
111  * @page: The page to write
112  * @wbc: The writeback control
113  *
114  * This is the same as calling block_write_full_page, but it also
115  * writes pages outside of i_size
116  */
gfs2_write_jdata_page(struct page * page,struct writeback_control * wbc)117 static int gfs2_write_jdata_page(struct page *page,
118 				 struct writeback_control *wbc)
119 {
120 	struct inode * const inode = page->mapping->host;
121 	loff_t i_size = i_size_read(inode);
122 	const pgoff_t end_index = i_size >> PAGE_SHIFT;
123 	unsigned offset;
124 
125 	/*
126 	 * The page straddles i_size.  It must be zeroed out on each and every
127 	 * writepage invocation because it may be mmapped.  "A file is mapped
128 	 * in multiples of the page size.  For a file that is not a multiple of
129 	 * the  page size, the remaining memory is zeroed when mapped, and
130 	 * writes to that region are not written out to the file."
131 	 */
132 	offset = i_size & (PAGE_SIZE - 1);
133 	if (page->index == end_index && offset)
134 		zero_user_segment(page, offset, PAGE_SIZE);
135 
136 	return __block_write_full_page(inode, page, gfs2_get_block_noalloc, wbc,
137 				       end_buffer_async_write);
138 }
139 
140 /**
141  * __gfs2_jdata_writepage - The core of jdata writepage
142  * @page: The page to write
143  * @wbc: The writeback control
144  *
145  * This is shared between writepage and writepages and implements the
146  * core of the writepage operation. If a transaction is required then
147  * PageChecked will have been set and the transaction will have
148  * already been started before this is called.
149  */
150 
__gfs2_jdata_writepage(struct page * page,struct writeback_control * wbc)151 static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
152 {
153 	struct inode *inode = page->mapping->host;
154 	struct gfs2_inode *ip = GFS2_I(inode);
155 	struct gfs2_sbd *sdp = GFS2_SB(inode);
156 
157 	if (PageChecked(page)) {
158 		ClearPageChecked(page);
159 		if (!page_has_buffers(page)) {
160 			create_empty_buffers(page, inode->i_sb->s_blocksize,
161 					     BIT(BH_Dirty)|BIT(BH_Uptodate));
162 		}
163 		gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize);
164 	}
165 	return gfs2_write_jdata_page(page, wbc);
166 }
167 
168 /**
169  * gfs2_jdata_writepage - Write complete page
170  * @page: Page to write
171  * @wbc: The writeback control
172  *
173  * Returns: errno
174  *
175  */
176 
gfs2_jdata_writepage(struct page * page,struct writeback_control * wbc)177 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
178 {
179 	struct inode *inode = page->mapping->host;
180 	struct gfs2_inode *ip = GFS2_I(inode);
181 	struct gfs2_sbd *sdp = GFS2_SB(inode);
182 
183 	if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
184 		goto out;
185 	if (PageChecked(page) || current->journal_info)
186 		goto out_ignore;
187 	return __gfs2_jdata_writepage(page, wbc);
188 
189 out_ignore:
190 	redirty_page_for_writepage(wbc, page);
191 out:
192 	unlock_page(page);
193 	return 0;
194 }
195 
196 /**
197  * gfs2_writepages - Write a bunch of dirty pages back to disk
198  * @mapping: The mapping to write
199  * @wbc: Write-back control
200  *
201  * Used for both ordered and writeback modes.
202  */
gfs2_writepages(struct address_space * mapping,struct writeback_control * wbc)203 static int gfs2_writepages(struct address_space *mapping,
204 			   struct writeback_control *wbc)
205 {
206 	struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
207 	struct iomap_writepage_ctx wpc = { };
208 	int ret;
209 
210 	/*
211 	 * Even if we didn't write any pages here, we might still be holding
212 	 * dirty pages in the ail. We forcibly flush the ail because we don't
213 	 * want balance_dirty_pages() to loop indefinitely trying to write out
214 	 * pages held in the ail that it can't find.
215 	 */
216 	ret = iomap_writepages(mapping, wbc, &wpc, &gfs2_writeback_ops);
217 	if (ret == 0)
218 		set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
219 	return ret;
220 }
221 
222 /**
223  * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
224  * @mapping: The mapping
225  * @wbc: The writeback control
226  * @pvec: The vector of pages
227  * @nr_pages: The number of pages to write
228  * @done_index: Page index
229  *
230  * Returns: non-zero if loop should terminate, zero otherwise
231  */
232 
gfs2_write_jdata_pagevec(struct address_space * mapping,struct writeback_control * wbc,struct pagevec * pvec,int nr_pages,pgoff_t * done_index)233 static int gfs2_write_jdata_pagevec(struct address_space *mapping,
234 				    struct writeback_control *wbc,
235 				    struct pagevec *pvec,
236 				    int nr_pages,
237 				    pgoff_t *done_index)
238 {
239 	struct inode *inode = mapping->host;
240 	struct gfs2_sbd *sdp = GFS2_SB(inode);
241 	unsigned nrblocks = nr_pages * (PAGE_SIZE >> inode->i_blkbits);
242 	int i;
243 	int ret;
244 
245 	ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
246 	if (ret < 0)
247 		return ret;
248 
249 	for(i = 0; i < nr_pages; i++) {
250 		struct page *page = pvec->pages[i];
251 
252 		*done_index = page->index;
253 
254 		lock_page(page);
255 
256 		if (unlikely(page->mapping != mapping)) {
257 continue_unlock:
258 			unlock_page(page);
259 			continue;
260 		}
261 
262 		if (!PageDirty(page)) {
263 			/* someone wrote it for us */
264 			goto continue_unlock;
265 		}
266 
267 		if (PageWriteback(page)) {
268 			if (wbc->sync_mode != WB_SYNC_NONE)
269 				wait_on_page_writeback(page);
270 			else
271 				goto continue_unlock;
272 		}
273 
274 		BUG_ON(PageWriteback(page));
275 		if (!clear_page_dirty_for_io(page))
276 			goto continue_unlock;
277 
278 		trace_wbc_writepage(wbc, inode_to_bdi(inode));
279 
280 		ret = __gfs2_jdata_writepage(page, wbc);
281 		if (unlikely(ret)) {
282 			if (ret == AOP_WRITEPAGE_ACTIVATE) {
283 				unlock_page(page);
284 				ret = 0;
285 			} else {
286 
287 				/*
288 				 * done_index is set past this page,
289 				 * so media errors will not choke
290 				 * background writeout for the entire
291 				 * file. This has consequences for
292 				 * range_cyclic semantics (ie. it may
293 				 * not be suitable for data integrity
294 				 * writeout).
295 				 */
296 				*done_index = page->index + 1;
297 				ret = 1;
298 				break;
299 			}
300 		}
301 
302 		/*
303 		 * We stop writing back only if we are not doing
304 		 * integrity sync. In case of integrity sync we have to
305 		 * keep going until we have written all the pages
306 		 * we tagged for writeback prior to entering this loop.
307 		 */
308 		if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
309 			ret = 1;
310 			break;
311 		}
312 
313 	}
314 	gfs2_trans_end(sdp);
315 	return ret;
316 }
317 
318 /**
319  * gfs2_write_cache_jdata - Like write_cache_pages but different
320  * @mapping: The mapping to write
321  * @wbc: The writeback control
322  *
323  * The reason that we use our own function here is that we need to
324  * start transactions before we grab page locks. This allows us
325  * to get the ordering right.
326  */
327 
gfs2_write_cache_jdata(struct address_space * mapping,struct writeback_control * wbc)328 static int gfs2_write_cache_jdata(struct address_space *mapping,
329 				  struct writeback_control *wbc)
330 {
331 	int ret = 0;
332 	int done = 0;
333 	struct pagevec pvec;
334 	int nr_pages;
335 	pgoff_t writeback_index;
336 	pgoff_t index;
337 	pgoff_t end;
338 	pgoff_t done_index;
339 	int cycled;
340 	int range_whole = 0;
341 	xa_mark_t tag;
342 
343 	pagevec_init(&pvec);
344 	if (wbc->range_cyclic) {
345 		writeback_index = mapping->writeback_index; /* prev offset */
346 		index = writeback_index;
347 		if (index == 0)
348 			cycled = 1;
349 		else
350 			cycled = 0;
351 		end = -1;
352 	} else {
353 		index = wbc->range_start >> PAGE_SHIFT;
354 		end = wbc->range_end >> PAGE_SHIFT;
355 		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
356 			range_whole = 1;
357 		cycled = 1; /* ignore range_cyclic tests */
358 	}
359 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
360 		tag = PAGECACHE_TAG_TOWRITE;
361 	else
362 		tag = PAGECACHE_TAG_DIRTY;
363 
364 retry:
365 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
366 		tag_pages_for_writeback(mapping, index, end);
367 	done_index = index;
368 	while (!done && (index <= end)) {
369 		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
370 				tag);
371 		if (nr_pages == 0)
372 			break;
373 
374 		ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, &done_index);
375 		if (ret)
376 			done = 1;
377 		if (ret > 0)
378 			ret = 0;
379 		pagevec_release(&pvec);
380 		cond_resched();
381 	}
382 
383 	if (!cycled && !done) {
384 		/*
385 		 * range_cyclic:
386 		 * We hit the last page and there is more work to be done: wrap
387 		 * back to the start of the file
388 		 */
389 		cycled = 1;
390 		index = 0;
391 		end = writeback_index - 1;
392 		goto retry;
393 	}
394 
395 	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
396 		mapping->writeback_index = done_index;
397 
398 	return ret;
399 }
400 
401 
402 /**
403  * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
404  * @mapping: The mapping to write
405  * @wbc: The writeback control
406  *
407  */
408 
gfs2_jdata_writepages(struct address_space * mapping,struct writeback_control * wbc)409 static int gfs2_jdata_writepages(struct address_space *mapping,
410 				 struct writeback_control *wbc)
411 {
412 	struct gfs2_inode *ip = GFS2_I(mapping->host);
413 	struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
414 	int ret;
415 
416 	ret = gfs2_write_cache_jdata(mapping, wbc);
417 	if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
418 		gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
419 			       GFS2_LFC_JDATA_WPAGES);
420 		ret = gfs2_write_cache_jdata(mapping, wbc);
421 	}
422 	return ret;
423 }
424 
425 /**
426  * stuffed_readpage - Fill in a Linux page with stuffed file data
427  * @ip: the inode
428  * @page: the page
429  *
430  * Returns: errno
431  */
stuffed_readpage(struct gfs2_inode * ip,struct page * page)432 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
433 {
434 	struct buffer_head *dibh;
435 	u64 dsize = i_size_read(&ip->i_inode);
436 	void *kaddr;
437 	int error;
438 
439 	/*
440 	 * Due to the order of unstuffing files and ->fault(), we can be
441 	 * asked for a zero page in the case of a stuffed file being extended,
442 	 * so we need to supply one here. It doesn't happen often.
443 	 */
444 	if (unlikely(page->index)) {
445 		zero_user(page, 0, PAGE_SIZE);
446 		SetPageUptodate(page);
447 		return 0;
448 	}
449 
450 	error = gfs2_meta_inode_buffer(ip, &dibh);
451 	if (error)
452 		return error;
453 
454 	kaddr = kmap_atomic(page);
455 	if (dsize > gfs2_max_stuffed_size(ip))
456 		dsize = gfs2_max_stuffed_size(ip);
457 	memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
458 	memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
459 	kunmap_atomic(kaddr);
460 	flush_dcache_page(page);
461 	brelse(dibh);
462 	SetPageUptodate(page);
463 
464 	return 0;
465 }
466 
467 /**
468  * gfs2_read_folio - read a folio from a file
469  * @file: The file to read
470  * @folio: The folio in the file
471  */
gfs2_read_folio(struct file * file,struct folio * folio)472 static int gfs2_read_folio(struct file *file, struct folio *folio)
473 {
474 	struct inode *inode = folio->mapping->host;
475 	struct gfs2_inode *ip = GFS2_I(inode);
476 	struct gfs2_sbd *sdp = GFS2_SB(inode);
477 	int error;
478 
479 	if (!gfs2_is_jdata(ip) ||
480 	    (i_blocksize(inode) == PAGE_SIZE && !folio_buffers(folio))) {
481 		error = iomap_read_folio(folio, &gfs2_iomap_ops);
482 	} else if (gfs2_is_stuffed(ip)) {
483 		error = stuffed_readpage(ip, &folio->page);
484 		folio_unlock(folio);
485 	} else {
486 		error = mpage_read_folio(folio, gfs2_block_map);
487 	}
488 
489 	if (unlikely(gfs2_withdrawn(sdp)))
490 		return -EIO;
491 
492 	return error;
493 }
494 
495 /**
496  * gfs2_internal_read - read an internal file
497  * @ip: The gfs2 inode
498  * @buf: The buffer to fill
499  * @pos: The file position
500  * @size: The amount to read
501  *
502  */
503 
gfs2_internal_read(struct gfs2_inode * ip,char * buf,loff_t * pos,unsigned size)504 int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
505                        unsigned size)
506 {
507 	struct address_space *mapping = ip->i_inode.i_mapping;
508 	unsigned long index = *pos >> PAGE_SHIFT;
509 	unsigned offset = *pos & (PAGE_SIZE - 1);
510 	unsigned copied = 0;
511 	unsigned amt;
512 	struct page *page;
513 	void *p;
514 
515 	do {
516 		amt = size - copied;
517 		if (offset + size > PAGE_SIZE)
518 			amt = PAGE_SIZE - offset;
519 		page = read_cache_page(mapping, index, gfs2_read_folio, NULL);
520 		if (IS_ERR(page))
521 			return PTR_ERR(page);
522 		p = kmap_atomic(page);
523 		memcpy(buf + copied, p + offset, amt);
524 		kunmap_atomic(p);
525 		put_page(page);
526 		copied += amt;
527 		index++;
528 		offset = 0;
529 	} while(copied < size);
530 	(*pos) += size;
531 	return size;
532 }
533 
534 /**
535  * gfs2_readahead - Read a bunch of pages at once
536  * @rac: Read-ahead control structure
537  *
538  * Some notes:
539  * 1. This is only for readahead, so we can simply ignore any things
540  *    which are slightly inconvenient (such as locking conflicts between
541  *    the page lock and the glock) and return having done no I/O. Its
542  *    obviously not something we'd want to do on too regular a basis.
543  *    Any I/O we ignore at this time will be done via readpage later.
544  * 2. We don't handle stuffed files here we let readpage do the honours.
545  * 3. mpage_readahead() does most of the heavy lifting in the common case.
546  * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
547  */
548 
gfs2_readahead(struct readahead_control * rac)549 static void gfs2_readahead(struct readahead_control *rac)
550 {
551 	struct inode *inode = rac->mapping->host;
552 	struct gfs2_inode *ip = GFS2_I(inode);
553 
554 	if (gfs2_is_stuffed(ip))
555 		;
556 	else if (gfs2_is_jdata(ip))
557 		mpage_readahead(rac, gfs2_block_map);
558 	else
559 		iomap_readahead(rac, &gfs2_iomap_ops);
560 }
561 
562 /**
563  * adjust_fs_space - Adjusts the free space available due to gfs2_grow
564  * @inode: the rindex inode
565  */
adjust_fs_space(struct inode * inode)566 void adjust_fs_space(struct inode *inode)
567 {
568 	struct gfs2_sbd *sdp = GFS2_SB(inode);
569 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
570 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
571 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
572 	struct buffer_head *m_bh;
573 	u64 fs_total, new_free;
574 
575 	if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0)
576 		return;
577 
578 	/* Total up the file system space, according to the latest rindex. */
579 	fs_total = gfs2_ri_total(sdp);
580 	if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
581 		goto out;
582 
583 	spin_lock(&sdp->sd_statfs_spin);
584 	gfs2_statfs_change_in(m_sc, m_bh->b_data +
585 			      sizeof(struct gfs2_dinode));
586 	if (fs_total > (m_sc->sc_total + l_sc->sc_total))
587 		new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
588 	else
589 		new_free = 0;
590 	spin_unlock(&sdp->sd_statfs_spin);
591 	fs_warn(sdp, "File system extended by %llu blocks.\n",
592 		(unsigned long long)new_free);
593 	gfs2_statfs_change(sdp, new_free, new_free, 0);
594 
595 	update_statfs(sdp, m_bh);
596 	brelse(m_bh);
597 out:
598 	sdp->sd_rindex_uptodate = 0;
599 	gfs2_trans_end(sdp);
600 }
601 
jdata_dirty_folio(struct address_space * mapping,struct folio * folio)602 static bool jdata_dirty_folio(struct address_space *mapping,
603 		struct folio *folio)
604 {
605 	if (current->journal_info)
606 		folio_set_checked(folio);
607 	return block_dirty_folio(mapping, folio);
608 }
609 
610 /**
611  * gfs2_bmap - Block map function
612  * @mapping: Address space info
613  * @lblock: The block to map
614  *
615  * Returns: The disk address for the block or 0 on hole or error
616  */
617 
gfs2_bmap(struct address_space * mapping,sector_t lblock)618 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
619 {
620 	struct gfs2_inode *ip = GFS2_I(mapping->host);
621 	struct gfs2_holder i_gh;
622 	sector_t dblock = 0;
623 	int error;
624 
625 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
626 	if (error)
627 		return 0;
628 
629 	if (!gfs2_is_stuffed(ip))
630 		dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops);
631 
632 	gfs2_glock_dq_uninit(&i_gh);
633 
634 	return dblock;
635 }
636 
gfs2_discard(struct gfs2_sbd * sdp,struct buffer_head * bh)637 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
638 {
639 	struct gfs2_bufdata *bd;
640 
641 	lock_buffer(bh);
642 	gfs2_log_lock(sdp);
643 	clear_buffer_dirty(bh);
644 	bd = bh->b_private;
645 	if (bd) {
646 		if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
647 			list_del_init(&bd->bd_list);
648 		else {
649 			spin_lock(&sdp->sd_ail_lock);
650 			gfs2_remove_from_journal(bh, REMOVE_JDATA);
651 			spin_unlock(&sdp->sd_ail_lock);
652 		}
653 	}
654 	bh->b_bdev = NULL;
655 	clear_buffer_mapped(bh);
656 	clear_buffer_req(bh);
657 	clear_buffer_new(bh);
658 	gfs2_log_unlock(sdp);
659 	unlock_buffer(bh);
660 }
661 
gfs2_invalidate_folio(struct folio * folio,size_t offset,size_t length)662 static void gfs2_invalidate_folio(struct folio *folio, size_t offset,
663 				size_t length)
664 {
665 	struct gfs2_sbd *sdp = GFS2_SB(folio->mapping->host);
666 	size_t stop = offset + length;
667 	int partial_page = (offset || length < folio_size(folio));
668 	struct buffer_head *bh, *head;
669 	unsigned long pos = 0;
670 
671 	BUG_ON(!folio_test_locked(folio));
672 	if (!partial_page)
673 		folio_clear_checked(folio);
674 	head = folio_buffers(folio);
675 	if (!head)
676 		goto out;
677 
678 	bh = head;
679 	do {
680 		if (pos + bh->b_size > stop)
681 			return;
682 
683 		if (offset <= pos)
684 			gfs2_discard(sdp, bh);
685 		pos += bh->b_size;
686 		bh = bh->b_this_page;
687 	} while (bh != head);
688 out:
689 	if (!partial_page)
690 		filemap_release_folio(folio, 0);
691 }
692 
693 /**
694  * gfs2_release_folio - free the metadata associated with a folio
695  * @folio: the folio that's being released
696  * @gfp_mask: passed from Linux VFS, ignored by us
697  *
698  * Calls try_to_free_buffers() to free the buffers and put the folio if the
699  * buffers can be released.
700  *
701  * Returns: true if the folio was put or else false
702  */
703 
gfs2_release_folio(struct folio * folio,gfp_t gfp_mask)704 bool gfs2_release_folio(struct folio *folio, gfp_t gfp_mask)
705 {
706 	struct address_space *mapping = folio->mapping;
707 	struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
708 	struct buffer_head *bh, *head;
709 	struct gfs2_bufdata *bd;
710 
711 	head = folio_buffers(folio);
712 	if (!head)
713 		return false;
714 
715 	/*
716 	 * mm accommodates an old ext3 case where clean folios might
717 	 * not have had the dirty bit cleared.	Thus, it can send actual
718 	 * dirty folios to ->release_folio() via shrink_active_list().
719 	 *
720 	 * As a workaround, we skip folios that contain dirty buffers
721 	 * below.  Once ->release_folio isn't called on dirty folios
722 	 * anymore, we can warn on dirty buffers like we used to here
723 	 * again.
724 	 */
725 
726 	gfs2_log_lock(sdp);
727 	bh = head;
728 	do {
729 		if (atomic_read(&bh->b_count))
730 			goto cannot_release;
731 		bd = bh->b_private;
732 		if (bd && bd->bd_tr)
733 			goto cannot_release;
734 		if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh)))
735 			goto cannot_release;
736 		bh = bh->b_this_page;
737 	} while (bh != head);
738 
739 	bh = head;
740 	do {
741 		bd = bh->b_private;
742 		if (bd) {
743 			gfs2_assert_warn(sdp, bd->bd_bh == bh);
744 			bd->bd_bh = NULL;
745 			bh->b_private = NULL;
746 			/*
747 			 * The bd may still be queued as a revoke, in which
748 			 * case we must not dequeue nor free it.
749 			 */
750 			if (!bd->bd_blkno && !list_empty(&bd->bd_list))
751 				list_del_init(&bd->bd_list);
752 			if (list_empty(&bd->bd_list))
753 				kmem_cache_free(gfs2_bufdata_cachep, bd);
754 		}
755 
756 		bh = bh->b_this_page;
757 	} while (bh != head);
758 	gfs2_log_unlock(sdp);
759 
760 	return try_to_free_buffers(folio);
761 
762 cannot_release:
763 	gfs2_log_unlock(sdp);
764 	return false;
765 }
766 
767 static const struct address_space_operations gfs2_aops = {
768 	.writepage = gfs2_writepage,
769 	.writepages = gfs2_writepages,
770 	.read_folio = gfs2_read_folio,
771 	.readahead = gfs2_readahead,
772 	.dirty_folio = filemap_dirty_folio,
773 	.release_folio = iomap_release_folio,
774 	.invalidate_folio = iomap_invalidate_folio,
775 	.bmap = gfs2_bmap,
776 	.direct_IO = noop_direct_IO,
777 	.migratepage = iomap_migrate_page,
778 	.is_partially_uptodate = iomap_is_partially_uptodate,
779 	.error_remove_page = generic_error_remove_page,
780 };
781 
782 static const struct address_space_operations gfs2_jdata_aops = {
783 	.writepage = gfs2_jdata_writepage,
784 	.writepages = gfs2_jdata_writepages,
785 	.read_folio = gfs2_read_folio,
786 	.readahead = gfs2_readahead,
787 	.dirty_folio = jdata_dirty_folio,
788 	.bmap = gfs2_bmap,
789 	.invalidate_folio = gfs2_invalidate_folio,
790 	.release_folio = gfs2_release_folio,
791 	.is_partially_uptodate = block_is_partially_uptodate,
792 	.error_remove_page = generic_error_remove_page,
793 };
794 
gfs2_set_aops(struct inode * inode)795 void gfs2_set_aops(struct inode *inode)
796 {
797 	if (gfs2_is_jdata(GFS2_I(inode)))
798 		inode->i_mapping->a_ops = &gfs2_jdata_aops;
799 	else
800 		inode->i_mapping->a_ops = &gfs2_aops;
801 }
802