xref: /openbmc/linux/fs/gfs2/aops.c (revision 22c6186e)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
4  *
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.
8  */
9 
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>
18 #include <linux/fs.h>
19 #include <linux/writeback.h>
20 #include <linux/swap.h>
21 #include <linux/gfs2_ondisk.h>
22 #include <linux/backing-dev.h>
23 #include <linux/uio.h>
24 #include <trace/events/writeback.h>
25 
26 #include "gfs2.h"
27 #include "incore.h"
28 #include "bmap.h"
29 #include "glock.h"
30 #include "inode.h"
31 #include "log.h"
32 #include "meta_io.h"
33 #include "quota.h"
34 #include "trans.h"
35 #include "rgrp.h"
36 #include "super.h"
37 #include "util.h"
38 #include "glops.h"
39 
40 
41 static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
42 				   unsigned int from, unsigned int to)
43 {
44 	struct buffer_head *head = page_buffers(page);
45 	unsigned int bsize = head->b_size;
46 	struct buffer_head *bh;
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 || start >= to)
53 			continue;
54 		if (gfs2_is_jdata(ip))
55 			set_buffer_uptodate(bh);
56 		gfs2_trans_add_data(ip->i_gl, bh);
57 	}
58 }
59 
60 /**
61  * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
62  * @inode: The inode
63  * @lblock: The block number to look up
64  * @bh_result: The buffer head to return the result in
65  * @create: Non-zero if we may add block to the file
66  *
67  * Returns: errno
68  */
69 
70 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
71 				  struct buffer_head *bh_result, int create)
72 {
73 	int error;
74 
75 	error = gfs2_block_map(inode, lblock, bh_result, 0);
76 	if (error)
77 		return error;
78 	if (!buffer_mapped(bh_result))
79 		return -EIO;
80 	return 0;
81 }
82 
83 static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
84 				 struct buffer_head *bh_result, int create)
85 {
86 	return gfs2_block_map(inode, lblock, bh_result, 0);
87 }
88 
89 /**
90  * gfs2_writepage_common - Common bits of writepage
91  * @page: The page to be written
92  * @wbc: The writeback control
93  *
94  * Returns: 1 if writepage is ok, otherwise an error code or zero if no error.
95  */
96 
97 static int gfs2_writepage_common(struct page *page,
98 				 struct writeback_control *wbc)
99 {
100 	struct inode *inode = page->mapping->host;
101 	struct gfs2_inode *ip = GFS2_I(inode);
102 	struct gfs2_sbd *sdp = GFS2_SB(inode);
103 	loff_t i_size = i_size_read(inode);
104 	pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
105 	unsigned offset;
106 
107 	if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
108 		goto out;
109 	if (current->journal_info)
110 		goto redirty;
111 	/* Is the page fully outside i_size? (truncate in progress) */
112 	offset = i_size & (PAGE_CACHE_SIZE-1);
113 	if (page->index > end_index || (page->index == end_index && !offset)) {
114 		page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
115 		goto out;
116 	}
117 	return 1;
118 redirty:
119 	redirty_page_for_writepage(wbc, page);
120 out:
121 	unlock_page(page);
122 	return 0;
123 }
124 
125 /**
126  * gfs2_writepage - Write page for writeback mappings
127  * @page: The page
128  * @wbc: The writeback control
129  *
130  */
131 
132 static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
133 {
134 	int ret;
135 
136 	ret = gfs2_writepage_common(page, wbc);
137 	if (ret <= 0)
138 		return ret;
139 
140 	return nobh_writepage(page, gfs2_get_block_noalloc, wbc);
141 }
142 
143 /**
144  * __gfs2_jdata_writepage - The core of jdata writepage
145  * @page: The page to write
146  * @wbc: The writeback control
147  *
148  * This is shared between writepage and writepages and implements the
149  * core of the writepage operation. If a transaction is required then
150  * PageChecked will have been set and the transaction will have
151  * already been started before this is called.
152  */
153 
154 static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
155 {
156 	struct inode *inode = page->mapping->host;
157 	struct gfs2_inode *ip = GFS2_I(inode);
158 	struct gfs2_sbd *sdp = GFS2_SB(inode);
159 
160 	if (PageChecked(page)) {
161 		ClearPageChecked(page);
162 		if (!page_has_buffers(page)) {
163 			create_empty_buffers(page, inode->i_sb->s_blocksize,
164 					     (1 << BH_Dirty)|(1 << BH_Uptodate));
165 		}
166 		gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
167 	}
168 	return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
169 }
170 
171 /**
172  * gfs2_jdata_writepage - Write complete page
173  * @page: Page to write
174  *
175  * Returns: errno
176  *
177  */
178 
179 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
180 {
181 	struct inode *inode = page->mapping->host;
182 	struct gfs2_sbd *sdp = GFS2_SB(inode);
183 	int ret;
184 	int done_trans = 0;
185 
186 	if (PageChecked(page)) {
187 		if (wbc->sync_mode != WB_SYNC_ALL)
188 			goto out_ignore;
189 		ret = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
190 		if (ret)
191 			goto out_ignore;
192 		done_trans = 1;
193 	}
194 	ret = gfs2_writepage_common(page, wbc);
195 	if (ret > 0)
196 		ret = __gfs2_jdata_writepage(page, wbc);
197 	if (done_trans)
198 		gfs2_trans_end(sdp);
199 	return ret;
200 
201 out_ignore:
202 	redirty_page_for_writepage(wbc, page);
203 	unlock_page(page);
204 	return 0;
205 }
206 
207 /**
208  * gfs2_writepages - Write a bunch of dirty pages back to disk
209  * @mapping: The mapping to write
210  * @wbc: Write-back control
211  *
212  * Used for both ordered and writeback modes.
213  */
214 static int gfs2_writepages(struct address_space *mapping,
215 			   struct writeback_control *wbc)
216 {
217 	return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
218 }
219 
220 /**
221  * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
222  * @mapping: The mapping
223  * @wbc: The writeback control
224  * @writepage: The writepage function to call for each page
225  * @pvec: The vector of pages
226  * @nr_pages: The number of pages to write
227  *
228  * Returns: non-zero if loop should terminate, zero otherwise
229  */
230 
231 static int gfs2_write_jdata_pagevec(struct address_space *mapping,
232 				    struct writeback_control *wbc,
233 				    struct pagevec *pvec,
234 				    int nr_pages, pgoff_t end,
235 				    pgoff_t *done_index)
236 {
237 	struct inode *inode = mapping->host;
238 	struct gfs2_sbd *sdp = GFS2_SB(inode);
239 	unsigned nrblocks = nr_pages * (PAGE_CACHE_SIZE/inode->i_sb->s_blocksize);
240 	int i;
241 	int ret;
242 
243 	ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
244 	if (ret < 0)
245 		return ret;
246 
247 	for(i = 0; i < nr_pages; i++) {
248 		struct page *page = pvec->pages[i];
249 
250 		/*
251 		 * At this point, the page may be truncated or
252 		 * invalidated (changing page->mapping to NULL), or
253 		 * even swizzled back from swapper_space to tmpfs file
254 		 * mapping. However, page->index will not change
255 		 * because we have a reference on the page.
256 		 */
257 		if (page->index > end) {
258 			/*
259 			 * can't be range_cyclic (1st pass) because
260 			 * end == -1 in that case.
261 			 */
262 			ret = 1;
263 			break;
264 		}
265 
266 		*done_index = page->index;
267 
268 		lock_page(page);
269 
270 		if (unlikely(page->mapping != mapping)) {
271 continue_unlock:
272 			unlock_page(page);
273 			continue;
274 		}
275 
276 		if (!PageDirty(page)) {
277 			/* someone wrote it for us */
278 			goto continue_unlock;
279 		}
280 
281 		if (PageWriteback(page)) {
282 			if (wbc->sync_mode != WB_SYNC_NONE)
283 				wait_on_page_writeback(page);
284 			else
285 				goto continue_unlock;
286 		}
287 
288 		BUG_ON(PageWriteback(page));
289 		if (!clear_page_dirty_for_io(page))
290 			goto continue_unlock;
291 
292 		trace_wbc_writepage(wbc, inode_to_bdi(inode));
293 
294 		ret = __gfs2_jdata_writepage(page, wbc);
295 		if (unlikely(ret)) {
296 			if (ret == AOP_WRITEPAGE_ACTIVATE) {
297 				unlock_page(page);
298 				ret = 0;
299 			} else {
300 
301 				/*
302 				 * done_index is set past this page,
303 				 * so media errors will not choke
304 				 * background writeout for the entire
305 				 * file. This has consequences for
306 				 * range_cyclic semantics (ie. it may
307 				 * not be suitable for data integrity
308 				 * writeout).
309 				 */
310 				*done_index = page->index + 1;
311 				ret = 1;
312 				break;
313 			}
314 		}
315 
316 		/*
317 		 * We stop writing back only if we are not doing
318 		 * integrity sync. In case of integrity sync we have to
319 		 * keep going until we have written all the pages
320 		 * we tagged for writeback prior to entering this loop.
321 		 */
322 		if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
323 			ret = 1;
324 			break;
325 		}
326 
327 	}
328 	gfs2_trans_end(sdp);
329 	return ret;
330 }
331 
332 /**
333  * gfs2_write_cache_jdata - Like write_cache_pages but different
334  * @mapping: The mapping to write
335  * @wbc: The writeback control
336  * @writepage: The writepage function to call
337  * @data: The data to pass to writepage
338  *
339  * The reason that we use our own function here is that we need to
340  * start transactions before we grab page locks. This allows us
341  * to get the ordering right.
342  */
343 
344 static int gfs2_write_cache_jdata(struct address_space *mapping,
345 				  struct writeback_control *wbc)
346 {
347 	int ret = 0;
348 	int done = 0;
349 	struct pagevec pvec;
350 	int nr_pages;
351 	pgoff_t uninitialized_var(writeback_index);
352 	pgoff_t index;
353 	pgoff_t end;
354 	pgoff_t done_index;
355 	int cycled;
356 	int range_whole = 0;
357 	int tag;
358 
359 	pagevec_init(&pvec, 0);
360 	if (wbc->range_cyclic) {
361 		writeback_index = mapping->writeback_index; /* prev offset */
362 		index = writeback_index;
363 		if (index == 0)
364 			cycled = 1;
365 		else
366 			cycled = 0;
367 		end = -1;
368 	} else {
369 		index = wbc->range_start >> PAGE_CACHE_SHIFT;
370 		end = wbc->range_end >> PAGE_CACHE_SHIFT;
371 		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
372 			range_whole = 1;
373 		cycled = 1; /* ignore range_cyclic tests */
374 	}
375 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
376 		tag = PAGECACHE_TAG_TOWRITE;
377 	else
378 		tag = PAGECACHE_TAG_DIRTY;
379 
380 retry:
381 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
382 		tag_pages_for_writeback(mapping, index, end);
383 	done_index = index;
384 	while (!done && (index <= end)) {
385 		nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
386 			      min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
387 		if (nr_pages == 0)
388 			break;
389 
390 		ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end, &done_index);
391 		if (ret)
392 			done = 1;
393 		if (ret > 0)
394 			ret = 0;
395 		pagevec_release(&pvec);
396 		cond_resched();
397 	}
398 
399 	if (!cycled && !done) {
400 		/*
401 		 * range_cyclic:
402 		 * We hit the last page and there is more work to be done: wrap
403 		 * back to the start of the file
404 		 */
405 		cycled = 1;
406 		index = 0;
407 		end = writeback_index - 1;
408 		goto retry;
409 	}
410 
411 	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
412 		mapping->writeback_index = done_index;
413 
414 	return ret;
415 }
416 
417 
418 /**
419  * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
420  * @mapping: The mapping to write
421  * @wbc: The writeback control
422  *
423  */
424 
425 static int gfs2_jdata_writepages(struct address_space *mapping,
426 				 struct writeback_control *wbc)
427 {
428 	struct gfs2_inode *ip = GFS2_I(mapping->host);
429 	struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
430 	int ret;
431 
432 	ret = gfs2_write_cache_jdata(mapping, wbc);
433 	if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
434 		gfs2_log_flush(sdp, ip->i_gl, NORMAL_FLUSH);
435 		ret = gfs2_write_cache_jdata(mapping, wbc);
436 	}
437 	return ret;
438 }
439 
440 /**
441  * stuffed_readpage - Fill in a Linux page with stuffed file data
442  * @ip: the inode
443  * @page: the page
444  *
445  * Returns: errno
446  */
447 
448 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
449 {
450 	struct buffer_head *dibh;
451 	u64 dsize = i_size_read(&ip->i_inode);
452 	void *kaddr;
453 	int error;
454 
455 	/*
456 	 * Due to the order of unstuffing files and ->fault(), we can be
457 	 * asked for a zero page in the case of a stuffed file being extended,
458 	 * so we need to supply one here. It doesn't happen often.
459 	 */
460 	if (unlikely(page->index)) {
461 		zero_user(page, 0, PAGE_CACHE_SIZE);
462 		SetPageUptodate(page);
463 		return 0;
464 	}
465 
466 	error = gfs2_meta_inode_buffer(ip, &dibh);
467 	if (error)
468 		return error;
469 
470 	kaddr = kmap_atomic(page);
471 	if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
472 		dsize = (dibh->b_size - sizeof(struct gfs2_dinode));
473 	memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
474 	memset(kaddr + dsize, 0, PAGE_CACHE_SIZE - dsize);
475 	kunmap_atomic(kaddr);
476 	flush_dcache_page(page);
477 	brelse(dibh);
478 	SetPageUptodate(page);
479 
480 	return 0;
481 }
482 
483 
484 /**
485  * __gfs2_readpage - readpage
486  * @file: The file to read a page for
487  * @page: The page to read
488  *
489  * This is the core of gfs2's readpage. Its used by the internal file
490  * reading code as in that case we already hold the glock. Also its
491  * called by gfs2_readpage() once the required lock has been granted.
492  *
493  */
494 
495 static int __gfs2_readpage(void *file, struct page *page)
496 {
497 	struct gfs2_inode *ip = GFS2_I(page->mapping->host);
498 	struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
499 	int error;
500 
501 	if (gfs2_is_stuffed(ip)) {
502 		error = stuffed_readpage(ip, page);
503 		unlock_page(page);
504 	} else {
505 		error = mpage_readpage(page, gfs2_block_map);
506 	}
507 
508 	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
509 		return -EIO;
510 
511 	return error;
512 }
513 
514 /**
515  * gfs2_readpage - read a page of a file
516  * @file: The file to read
517  * @page: The page of the file
518  *
519  * This deals with the locking required. We have to unlock and
520  * relock the page in order to get the locking in the right
521  * order.
522  */
523 
524 static int gfs2_readpage(struct file *file, struct page *page)
525 {
526 	struct address_space *mapping = page->mapping;
527 	struct gfs2_inode *ip = GFS2_I(mapping->host);
528 	struct gfs2_holder gh;
529 	int error;
530 
531 	unlock_page(page);
532 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
533 	error = gfs2_glock_nq(&gh);
534 	if (unlikely(error))
535 		goto out;
536 	error = AOP_TRUNCATED_PAGE;
537 	lock_page(page);
538 	if (page->mapping == mapping && !PageUptodate(page))
539 		error = __gfs2_readpage(file, page);
540 	else
541 		unlock_page(page);
542 	gfs2_glock_dq(&gh);
543 out:
544 	gfs2_holder_uninit(&gh);
545 	if (error && error != AOP_TRUNCATED_PAGE)
546 		lock_page(page);
547 	return error;
548 }
549 
550 /**
551  * gfs2_internal_read - read an internal file
552  * @ip: The gfs2 inode
553  * @buf: The buffer to fill
554  * @pos: The file position
555  * @size: The amount to read
556  *
557  */
558 
559 int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
560                        unsigned size)
561 {
562 	struct address_space *mapping = ip->i_inode.i_mapping;
563 	unsigned long index = *pos / PAGE_CACHE_SIZE;
564 	unsigned offset = *pos & (PAGE_CACHE_SIZE - 1);
565 	unsigned copied = 0;
566 	unsigned amt;
567 	struct page *page;
568 	void *p;
569 
570 	do {
571 		amt = size - copied;
572 		if (offset + size > PAGE_CACHE_SIZE)
573 			amt = PAGE_CACHE_SIZE - offset;
574 		page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
575 		if (IS_ERR(page))
576 			return PTR_ERR(page);
577 		p = kmap_atomic(page);
578 		memcpy(buf + copied, p + offset, amt);
579 		kunmap_atomic(p);
580 		page_cache_release(page);
581 		copied += amt;
582 		index++;
583 		offset = 0;
584 	} while(copied < size);
585 	(*pos) += size;
586 	return size;
587 }
588 
589 /**
590  * gfs2_readpages - Read a bunch of pages at once
591  *
592  * Some notes:
593  * 1. This is only for readahead, so we can simply ignore any things
594  *    which are slightly inconvenient (such as locking conflicts between
595  *    the page lock and the glock) and return having done no I/O. Its
596  *    obviously not something we'd want to do on too regular a basis.
597  *    Any I/O we ignore at this time will be done via readpage later.
598  * 2. We don't handle stuffed files here we let readpage do the honours.
599  * 3. mpage_readpages() does most of the heavy lifting in the common case.
600  * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
601  */
602 
603 static int gfs2_readpages(struct file *file, struct address_space *mapping,
604 			  struct list_head *pages, unsigned nr_pages)
605 {
606 	struct inode *inode = mapping->host;
607 	struct gfs2_inode *ip = GFS2_I(inode);
608 	struct gfs2_sbd *sdp = GFS2_SB(inode);
609 	struct gfs2_holder gh;
610 	int ret;
611 
612 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
613 	ret = gfs2_glock_nq(&gh);
614 	if (unlikely(ret))
615 		goto out_uninit;
616 	if (!gfs2_is_stuffed(ip))
617 		ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
618 	gfs2_glock_dq(&gh);
619 out_uninit:
620 	gfs2_holder_uninit(&gh);
621 	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
622 		ret = -EIO;
623 	return ret;
624 }
625 
626 /**
627  * gfs2_write_begin - Begin to write to a file
628  * @file: The file to write to
629  * @mapping: The mapping in which to write
630  * @pos: The file offset at which to start writing
631  * @len: Length of the write
632  * @flags: Various flags
633  * @pagep: Pointer to return the page
634  * @fsdata: Pointer to return fs data (unused by GFS2)
635  *
636  * Returns: errno
637  */
638 
639 static int gfs2_write_begin(struct file *file, struct address_space *mapping,
640 			    loff_t pos, unsigned len, unsigned flags,
641 			    struct page **pagep, void **fsdata)
642 {
643 	struct gfs2_inode *ip = GFS2_I(mapping->host);
644 	struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
645 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
646 	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
647 	unsigned requested = 0;
648 	int alloc_required;
649 	int error = 0;
650 	pgoff_t index = pos >> PAGE_CACHE_SHIFT;
651 	unsigned from = pos & (PAGE_CACHE_SIZE - 1);
652 	struct page *page;
653 
654 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
655 	error = gfs2_glock_nq(&ip->i_gh);
656 	if (unlikely(error))
657 		goto out_uninit;
658 	if (&ip->i_inode == sdp->sd_rindex) {
659 		error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
660 					   GL_NOCACHE, &m_ip->i_gh);
661 		if (unlikely(error)) {
662 			gfs2_glock_dq(&ip->i_gh);
663 			goto out_uninit;
664 		}
665 	}
666 
667 	alloc_required = gfs2_write_alloc_required(ip, pos, len);
668 
669 	if (alloc_required || gfs2_is_jdata(ip))
670 		gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
671 
672 	if (alloc_required) {
673 		struct gfs2_alloc_parms ap = { .aflags = 0, };
674 		error = gfs2_quota_lock_check(ip);
675 		if (error)
676 			goto out_unlock;
677 
678 		requested = data_blocks + ind_blocks;
679 		ap.target = requested;
680 		error = gfs2_inplace_reserve(ip, &ap);
681 		if (error)
682 			goto out_qunlock;
683 	}
684 
685 	rblocks = RES_DINODE + ind_blocks;
686 	if (gfs2_is_jdata(ip))
687 		rblocks += data_blocks ? data_blocks : 1;
688 	if (ind_blocks || data_blocks)
689 		rblocks += RES_STATFS + RES_QUOTA;
690 	if (&ip->i_inode == sdp->sd_rindex)
691 		rblocks += 2 * RES_STATFS;
692 	if (alloc_required)
693 		rblocks += gfs2_rg_blocks(ip, requested);
694 
695 	error = gfs2_trans_begin(sdp, rblocks,
696 				 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
697 	if (error)
698 		goto out_trans_fail;
699 
700 	error = -ENOMEM;
701 	flags |= AOP_FLAG_NOFS;
702 	page = grab_cache_page_write_begin(mapping, index, flags);
703 	*pagep = page;
704 	if (unlikely(!page))
705 		goto out_endtrans;
706 
707 	if (gfs2_is_stuffed(ip)) {
708 		error = 0;
709 		if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
710 			error = gfs2_unstuff_dinode(ip, page);
711 			if (error == 0)
712 				goto prepare_write;
713 		} else if (!PageUptodate(page)) {
714 			error = stuffed_readpage(ip, page);
715 		}
716 		goto out;
717 	}
718 
719 prepare_write:
720 	error = __block_write_begin(page, from, len, gfs2_block_map);
721 out:
722 	if (error == 0)
723 		return 0;
724 
725 	unlock_page(page);
726 	page_cache_release(page);
727 
728 	gfs2_trans_end(sdp);
729 	if (pos + len > ip->i_inode.i_size)
730 		gfs2_trim_blocks(&ip->i_inode);
731 	goto out_trans_fail;
732 
733 out_endtrans:
734 	gfs2_trans_end(sdp);
735 out_trans_fail:
736 	if (alloc_required) {
737 		gfs2_inplace_release(ip);
738 out_qunlock:
739 		gfs2_quota_unlock(ip);
740 	}
741 out_unlock:
742 	if (&ip->i_inode == sdp->sd_rindex) {
743 		gfs2_glock_dq(&m_ip->i_gh);
744 		gfs2_holder_uninit(&m_ip->i_gh);
745 	}
746 	gfs2_glock_dq(&ip->i_gh);
747 out_uninit:
748 	gfs2_holder_uninit(&ip->i_gh);
749 	return error;
750 }
751 
752 /**
753  * adjust_fs_space - Adjusts the free space available due to gfs2_grow
754  * @inode: the rindex inode
755  */
756 static void adjust_fs_space(struct inode *inode)
757 {
758 	struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
759 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
760 	struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
761 	struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
762 	struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
763 	struct buffer_head *m_bh, *l_bh;
764 	u64 fs_total, new_free;
765 
766 	/* Total up the file system space, according to the latest rindex. */
767 	fs_total = gfs2_ri_total(sdp);
768 	if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
769 		return;
770 
771 	spin_lock(&sdp->sd_statfs_spin);
772 	gfs2_statfs_change_in(m_sc, m_bh->b_data +
773 			      sizeof(struct gfs2_dinode));
774 	if (fs_total > (m_sc->sc_total + l_sc->sc_total))
775 		new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
776 	else
777 		new_free = 0;
778 	spin_unlock(&sdp->sd_statfs_spin);
779 	fs_warn(sdp, "File system extended by %llu blocks.\n",
780 		(unsigned long long)new_free);
781 	gfs2_statfs_change(sdp, new_free, new_free, 0);
782 
783 	if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0)
784 		goto out;
785 	update_statfs(sdp, m_bh, l_bh);
786 	brelse(l_bh);
787 out:
788 	brelse(m_bh);
789 }
790 
791 /**
792  * gfs2_stuffed_write_end - Write end for stuffed files
793  * @inode: The inode
794  * @dibh: The buffer_head containing the on-disk inode
795  * @pos: The file position
796  * @len: The length of the write
797  * @copied: How much was actually copied by the VFS
798  * @page: The page
799  *
800  * This copies the data from the page into the inode block after
801  * the inode data structure itself.
802  *
803  * Returns: errno
804  */
805 static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
806 				  loff_t pos, unsigned len, unsigned copied,
807 				  struct page *page)
808 {
809 	struct gfs2_inode *ip = GFS2_I(inode);
810 	struct gfs2_sbd *sdp = GFS2_SB(inode);
811 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
812 	u64 to = pos + copied;
813 	void *kaddr;
814 	unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
815 
816 	BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
817 	kaddr = kmap_atomic(page);
818 	memcpy(buf + pos, kaddr + pos, copied);
819 	memset(kaddr + pos + copied, 0, len - copied);
820 	flush_dcache_page(page);
821 	kunmap_atomic(kaddr);
822 
823 	if (!PageUptodate(page))
824 		SetPageUptodate(page);
825 	unlock_page(page);
826 	page_cache_release(page);
827 
828 	if (copied) {
829 		if (inode->i_size < to)
830 			i_size_write(inode, to);
831 		mark_inode_dirty(inode);
832 	}
833 
834 	if (inode == sdp->sd_rindex) {
835 		adjust_fs_space(inode);
836 		sdp->sd_rindex_uptodate = 0;
837 	}
838 
839 	brelse(dibh);
840 	gfs2_trans_end(sdp);
841 	if (inode == sdp->sd_rindex) {
842 		gfs2_glock_dq(&m_ip->i_gh);
843 		gfs2_holder_uninit(&m_ip->i_gh);
844 	}
845 	gfs2_glock_dq(&ip->i_gh);
846 	gfs2_holder_uninit(&ip->i_gh);
847 	return copied;
848 }
849 
850 /**
851  * gfs2_write_end
852  * @file: The file to write to
853  * @mapping: The address space to write to
854  * @pos: The file position
855  * @len: The length of the data
856  * @copied:
857  * @page: The page that has been written
858  * @fsdata: The fsdata (unused in GFS2)
859  *
860  * The main write_end function for GFS2. We have a separate one for
861  * stuffed files as they are slightly different, otherwise we just
862  * put our locking around the VFS provided functions.
863  *
864  * Returns: errno
865  */
866 
867 static int gfs2_write_end(struct file *file, struct address_space *mapping,
868 			  loff_t pos, unsigned len, unsigned copied,
869 			  struct page *page, void *fsdata)
870 {
871 	struct inode *inode = page->mapping->host;
872 	struct gfs2_inode *ip = GFS2_I(inode);
873 	struct gfs2_sbd *sdp = GFS2_SB(inode);
874 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
875 	struct buffer_head *dibh;
876 	unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
877 	unsigned int to = from + len;
878 	int ret;
879 	struct gfs2_trans *tr = current->journal_info;
880 	BUG_ON(!tr);
881 
882 	BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
883 
884 	ret = gfs2_meta_inode_buffer(ip, &dibh);
885 	if (unlikely(ret)) {
886 		unlock_page(page);
887 		page_cache_release(page);
888 		goto failed;
889 	}
890 
891 	if (gfs2_is_stuffed(ip))
892 		return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
893 
894 	if (!gfs2_is_writeback(ip))
895 		gfs2_page_add_databufs(ip, page, from, to);
896 
897 	ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
898 	if (tr->tr_num_buf_new)
899 		__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
900 	else
901 		gfs2_trans_add_meta(ip->i_gl, dibh);
902 
903 
904 	if (inode == sdp->sd_rindex) {
905 		adjust_fs_space(inode);
906 		sdp->sd_rindex_uptodate = 0;
907 	}
908 
909 	brelse(dibh);
910 failed:
911 	gfs2_trans_end(sdp);
912 	gfs2_inplace_release(ip);
913 	if (ip->i_res->rs_qa_qd_num)
914 		gfs2_quota_unlock(ip);
915 	if (inode == sdp->sd_rindex) {
916 		gfs2_glock_dq(&m_ip->i_gh);
917 		gfs2_holder_uninit(&m_ip->i_gh);
918 	}
919 	gfs2_glock_dq(&ip->i_gh);
920 	gfs2_holder_uninit(&ip->i_gh);
921 	return ret;
922 }
923 
924 /**
925  * gfs2_set_page_dirty - Page dirtying function
926  * @page: The page to dirty
927  *
928  * Returns: 1 if it dirtyed the page, or 0 otherwise
929  */
930 
931 static int gfs2_set_page_dirty(struct page *page)
932 {
933 	SetPageChecked(page);
934 	return __set_page_dirty_buffers(page);
935 }
936 
937 /**
938  * gfs2_bmap - Block map function
939  * @mapping: Address space info
940  * @lblock: The block to map
941  *
942  * Returns: The disk address for the block or 0 on hole or error
943  */
944 
945 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
946 {
947 	struct gfs2_inode *ip = GFS2_I(mapping->host);
948 	struct gfs2_holder i_gh;
949 	sector_t dblock = 0;
950 	int error;
951 
952 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
953 	if (error)
954 		return 0;
955 
956 	if (!gfs2_is_stuffed(ip))
957 		dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
958 
959 	gfs2_glock_dq_uninit(&i_gh);
960 
961 	return dblock;
962 }
963 
964 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
965 {
966 	struct gfs2_bufdata *bd;
967 
968 	lock_buffer(bh);
969 	gfs2_log_lock(sdp);
970 	clear_buffer_dirty(bh);
971 	bd = bh->b_private;
972 	if (bd) {
973 		if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
974 			list_del_init(&bd->bd_list);
975 		else
976 			gfs2_remove_from_journal(bh, current->journal_info, 0);
977 	}
978 	bh->b_bdev = NULL;
979 	clear_buffer_mapped(bh);
980 	clear_buffer_req(bh);
981 	clear_buffer_new(bh);
982 	gfs2_log_unlock(sdp);
983 	unlock_buffer(bh);
984 }
985 
986 static void gfs2_invalidatepage(struct page *page, unsigned int offset,
987 				unsigned int length)
988 {
989 	struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
990 	unsigned int stop = offset + length;
991 	int partial_page = (offset || length < PAGE_CACHE_SIZE);
992 	struct buffer_head *bh, *head;
993 	unsigned long pos = 0;
994 
995 	BUG_ON(!PageLocked(page));
996 	if (!partial_page)
997 		ClearPageChecked(page);
998 	if (!page_has_buffers(page))
999 		goto out;
1000 
1001 	bh = head = page_buffers(page);
1002 	do {
1003 		if (pos + bh->b_size > stop)
1004 			return;
1005 
1006 		if (offset <= pos)
1007 			gfs2_discard(sdp, bh);
1008 		pos += bh->b_size;
1009 		bh = bh->b_this_page;
1010 	} while (bh != head);
1011 out:
1012 	if (!partial_page)
1013 		try_to_release_page(page, 0);
1014 }
1015 
1016 /**
1017  * gfs2_ok_for_dio - check that dio is valid on this file
1018  * @ip: The inode
1019  * @offset: The offset at which we are reading or writing
1020  *
1021  * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
1022  *          1 (to accept the i/o request)
1023  */
1024 static int gfs2_ok_for_dio(struct gfs2_inode *ip, loff_t offset)
1025 {
1026 	/*
1027 	 * Should we return an error here? I can't see that O_DIRECT for
1028 	 * a stuffed file makes any sense. For now we'll silently fall
1029 	 * back to buffered I/O
1030 	 */
1031 	if (gfs2_is_stuffed(ip))
1032 		return 0;
1033 
1034 	if (offset >= i_size_read(&ip->i_inode))
1035 		return 0;
1036 	return 1;
1037 }
1038 
1039 
1040 
1041 static ssize_t gfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
1042 			      loff_t offset)
1043 {
1044 	struct file *file = iocb->ki_filp;
1045 	struct inode *inode = file->f_mapping->host;
1046 	struct address_space *mapping = inode->i_mapping;
1047 	struct gfs2_inode *ip = GFS2_I(inode);
1048 	struct gfs2_holder gh;
1049 	int rv;
1050 
1051 	/*
1052 	 * Deferred lock, even if its a write, since we do no allocation
1053 	 * on this path. All we need change is atime, and this lock mode
1054 	 * ensures that other nodes have flushed their buffered read caches
1055 	 * (i.e. their page cache entries for this inode). We do not,
1056 	 * unfortunately have the option of only flushing a range like
1057 	 * the VFS does.
1058 	 */
1059 	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
1060 	rv = gfs2_glock_nq(&gh);
1061 	if (rv)
1062 		return rv;
1063 	rv = gfs2_ok_for_dio(ip, offset);
1064 	if (rv != 1)
1065 		goto out; /* dio not valid, fall back to buffered i/o */
1066 
1067 	/*
1068 	 * Now since we are holding a deferred (CW) lock at this point, you
1069 	 * might be wondering why this is ever needed. There is a case however
1070 	 * where we've granted a deferred local lock against a cached exclusive
1071 	 * glock. That is ok provided all granted local locks are deferred, but
1072 	 * it also means that it is possible to encounter pages which are
1073 	 * cached and possibly also mapped. So here we check for that and sort
1074 	 * them out ahead of the dio. The glock state machine will take care of
1075 	 * everything else.
1076 	 *
1077 	 * If in fact the cached glock state (gl->gl_state) is deferred (CW) in
1078 	 * the first place, mapping->nr_pages will always be zero.
1079 	 */
1080 	if (mapping->nrpages) {
1081 		loff_t lstart = offset & (PAGE_CACHE_SIZE - 1);
1082 		loff_t len = iov_iter_count(iter);
1083 		loff_t end = PAGE_ALIGN(offset + len) - 1;
1084 
1085 		rv = 0;
1086 		if (len == 0)
1087 			goto out;
1088 		if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
1089 			unmap_shared_mapping_range(ip->i_inode.i_mapping, offset, len);
1090 		rv = filemap_write_and_wait_range(mapping, lstart, end);
1091 		if (rv)
1092 			goto out;
1093 		if (iov_iter_rw(iter) == WRITE)
1094 			truncate_inode_pages_range(mapping, lstart, end);
1095 	}
1096 
1097 	rv = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
1098 				  offset, gfs2_get_block_direct, NULL, NULL, 0);
1099 out:
1100 	gfs2_glock_dq(&gh);
1101 	gfs2_holder_uninit(&gh);
1102 	return rv;
1103 }
1104 
1105 /**
1106  * gfs2_releasepage - free the metadata associated with a page
1107  * @page: the page that's being released
1108  * @gfp_mask: passed from Linux VFS, ignored by us
1109  *
1110  * Call try_to_free_buffers() if the buffers in this page can be
1111  * released.
1112  *
1113  * Returns: 0
1114  */
1115 
1116 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1117 {
1118 	struct address_space *mapping = page->mapping;
1119 	struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
1120 	struct buffer_head *bh, *head;
1121 	struct gfs2_bufdata *bd;
1122 
1123 	if (!page_has_buffers(page))
1124 		return 0;
1125 
1126 	gfs2_log_lock(sdp);
1127 	spin_lock(&sdp->sd_ail_lock);
1128 	head = bh = page_buffers(page);
1129 	do {
1130 		if (atomic_read(&bh->b_count))
1131 			goto cannot_release;
1132 		bd = bh->b_private;
1133 		if (bd && bd->bd_tr)
1134 			goto cannot_release;
1135 		if (buffer_pinned(bh) || buffer_dirty(bh))
1136 			goto not_possible;
1137 		bh = bh->b_this_page;
1138 	} while(bh != head);
1139 	spin_unlock(&sdp->sd_ail_lock);
1140 
1141 	head = bh = page_buffers(page);
1142 	do {
1143 		bd = bh->b_private;
1144 		if (bd) {
1145 			gfs2_assert_warn(sdp, bd->bd_bh == bh);
1146 			if (!list_empty(&bd->bd_list))
1147 				list_del_init(&bd->bd_list);
1148 			bd->bd_bh = NULL;
1149 			bh->b_private = NULL;
1150 			kmem_cache_free(gfs2_bufdata_cachep, bd);
1151 		}
1152 
1153 		bh = bh->b_this_page;
1154 	} while (bh != head);
1155 	gfs2_log_unlock(sdp);
1156 
1157 	return try_to_free_buffers(page);
1158 
1159 not_possible: /* Should never happen */
1160 	WARN_ON(buffer_dirty(bh));
1161 	WARN_ON(buffer_pinned(bh));
1162 cannot_release:
1163 	spin_unlock(&sdp->sd_ail_lock);
1164 	gfs2_log_unlock(sdp);
1165 	return 0;
1166 }
1167 
1168 static const struct address_space_operations gfs2_writeback_aops = {
1169 	.writepage = gfs2_writepage,
1170 	.writepages = gfs2_writepages,
1171 	.readpage = gfs2_readpage,
1172 	.readpages = gfs2_readpages,
1173 	.write_begin = gfs2_write_begin,
1174 	.write_end = gfs2_write_end,
1175 	.bmap = gfs2_bmap,
1176 	.invalidatepage = gfs2_invalidatepage,
1177 	.releasepage = gfs2_releasepage,
1178 	.direct_IO = gfs2_direct_IO,
1179 	.migratepage = buffer_migrate_page,
1180 	.is_partially_uptodate = block_is_partially_uptodate,
1181 	.error_remove_page = generic_error_remove_page,
1182 };
1183 
1184 static const struct address_space_operations gfs2_ordered_aops = {
1185 	.writepage = gfs2_writepage,
1186 	.writepages = gfs2_writepages,
1187 	.readpage = gfs2_readpage,
1188 	.readpages = gfs2_readpages,
1189 	.write_begin = gfs2_write_begin,
1190 	.write_end = gfs2_write_end,
1191 	.set_page_dirty = gfs2_set_page_dirty,
1192 	.bmap = gfs2_bmap,
1193 	.invalidatepage = gfs2_invalidatepage,
1194 	.releasepage = gfs2_releasepage,
1195 	.direct_IO = gfs2_direct_IO,
1196 	.migratepage = buffer_migrate_page,
1197 	.is_partially_uptodate = block_is_partially_uptodate,
1198 	.error_remove_page = generic_error_remove_page,
1199 };
1200 
1201 static const struct address_space_operations gfs2_jdata_aops = {
1202 	.writepage = gfs2_jdata_writepage,
1203 	.writepages = gfs2_jdata_writepages,
1204 	.readpage = gfs2_readpage,
1205 	.readpages = gfs2_readpages,
1206 	.write_begin = gfs2_write_begin,
1207 	.write_end = gfs2_write_end,
1208 	.set_page_dirty = gfs2_set_page_dirty,
1209 	.bmap = gfs2_bmap,
1210 	.invalidatepage = gfs2_invalidatepage,
1211 	.releasepage = gfs2_releasepage,
1212 	.is_partially_uptodate = block_is_partially_uptodate,
1213 	.error_remove_page = generic_error_remove_page,
1214 };
1215 
1216 void gfs2_set_aops(struct inode *inode)
1217 {
1218 	struct gfs2_inode *ip = GFS2_I(inode);
1219 
1220 	if (gfs2_is_writeback(ip))
1221 		inode->i_mapping->a_ops = &gfs2_writeback_aops;
1222 	else if (gfs2_is_ordered(ip))
1223 		inode->i_mapping->a_ops = &gfs2_ordered_aops;
1224 	else if (gfs2_is_jdata(ip))
1225 		inode->i_mapping->a_ops = &gfs2_jdata_aops;
1226 	else
1227 		BUG();
1228 }
1229 
1230