xref: /openbmc/linux/fs/gfs2/meta_io.c (revision 35267cea)
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/mm.h>
13 #include <linux/pagemap.h>
14 #include <linux/writeback.h>
15 #include <linux/swap.h>
16 #include <linux/delay.h>
17 #include <linux/bio.h>
18 #include <linux/gfs2_ondisk.h>
19 
20 #include "gfs2.h"
21 #include "incore.h"
22 #include "glock.h"
23 #include "glops.h"
24 #include "inode.h"
25 #include "log.h"
26 #include "lops.h"
27 #include "meta_io.h"
28 #include "rgrp.h"
29 #include "trans.h"
30 #include "util.h"
31 #include "trace_gfs2.h"
32 
33 static int gfs2_aspace_writepage(struct page *page, struct writeback_control *wbc)
34 {
35 	struct buffer_head *bh, *head;
36 	int nr_underway = 0;
37 	int write_flags = REQ_META | REQ_PRIO | wbc_to_write_flags(wbc);
38 
39 	BUG_ON(!PageLocked(page));
40 	BUG_ON(!page_has_buffers(page));
41 
42 	head = page_buffers(page);
43 	bh = head;
44 
45 	do {
46 		if (!buffer_mapped(bh))
47 			continue;
48 		/*
49 		 * If it's a fully non-blocking write attempt and we cannot
50 		 * lock the buffer then redirty the page.  Note that this can
51 		 * potentially cause a busy-wait loop from flusher thread and kswapd
52 		 * activity, but those code paths have their own higher-level
53 		 * throttling.
54 		 */
55 		if (wbc->sync_mode != WB_SYNC_NONE) {
56 			lock_buffer(bh);
57 		} else if (!trylock_buffer(bh)) {
58 			redirty_page_for_writepage(wbc, page);
59 			continue;
60 		}
61 		if (test_clear_buffer_dirty(bh)) {
62 			mark_buffer_async_write(bh);
63 		} else {
64 			unlock_buffer(bh);
65 		}
66 	} while ((bh = bh->b_this_page) != head);
67 
68 	/*
69 	 * The page and its buffers are protected by PageWriteback(), so we can
70 	 * drop the bh refcounts early.
71 	 */
72 	BUG_ON(PageWriteback(page));
73 	set_page_writeback(page);
74 
75 	do {
76 		struct buffer_head *next = bh->b_this_page;
77 		if (buffer_async_write(bh)) {
78 			submit_bh(REQ_OP_WRITE, write_flags, bh);
79 			nr_underway++;
80 		}
81 		bh = next;
82 	} while (bh != head);
83 	unlock_page(page);
84 
85 	if (nr_underway == 0)
86 		end_page_writeback(page);
87 
88 	return 0;
89 }
90 
91 const struct address_space_operations gfs2_meta_aops = {
92 	.set_page_dirty	= __set_page_dirty_buffers,
93 	.writepage = gfs2_aspace_writepage,
94 	.releasepage = gfs2_releasepage,
95 };
96 
97 const struct address_space_operations gfs2_rgrp_aops = {
98 	.set_page_dirty	= __set_page_dirty_buffers,
99 	.writepage = gfs2_aspace_writepage,
100 	.releasepage = gfs2_releasepage,
101 };
102 
103 /**
104  * gfs2_getbuf - Get a buffer with a given address space
105  * @gl: the glock
106  * @blkno: the block number (filesystem scope)
107  * @create: 1 if the buffer should be created
108  *
109  * Returns: the buffer
110  */
111 
112 struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create)
113 {
114 	struct address_space *mapping = gfs2_glock2aspace(gl);
115 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
116 	struct page *page;
117 	struct buffer_head *bh;
118 	unsigned int shift;
119 	unsigned long index;
120 	unsigned int bufnum;
121 
122 	if (mapping == NULL)
123 		mapping = &sdp->sd_aspace;
124 
125 	shift = PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift;
126 	index = blkno >> shift;             /* convert block to page */
127 	bufnum = blkno - (index << shift);  /* block buf index within page */
128 
129 	if (create) {
130 		for (;;) {
131 			page = grab_cache_page(mapping, index);
132 			if (page)
133 				break;
134 			yield();
135 		}
136 		if (!page_has_buffers(page))
137 			create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0);
138 	} else {
139 		page = find_get_page_flags(mapping, index,
140 						FGP_LOCK|FGP_ACCESSED);
141 		if (!page)
142 			return NULL;
143 		if (!page_has_buffers(page)) {
144 			bh = NULL;
145 			goto out_unlock;
146 		}
147 	}
148 
149 	/* Locate header for our buffer within our page */
150 	for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
151 		/* Do nothing */;
152 	get_bh(bh);
153 
154 	if (!buffer_mapped(bh))
155 		map_bh(bh, sdp->sd_vfs, blkno);
156 
157 out_unlock:
158 	unlock_page(page);
159 	put_page(page);
160 
161 	return bh;
162 }
163 
164 static void meta_prep_new(struct buffer_head *bh)
165 {
166 	struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
167 
168 	lock_buffer(bh);
169 	clear_buffer_dirty(bh);
170 	set_buffer_uptodate(bh);
171 	unlock_buffer(bh);
172 
173 	mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
174 }
175 
176 /**
177  * gfs2_meta_new - Get a block
178  * @gl: The glock associated with this block
179  * @blkno: The block number
180  *
181  * Returns: The buffer
182  */
183 
184 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno)
185 {
186 	struct buffer_head *bh;
187 	bh = gfs2_getbuf(gl, blkno, CREATE);
188 	meta_prep_new(bh);
189 	return bh;
190 }
191 
192 static void gfs2_meta_read_endio(struct bio *bio)
193 {
194 	struct bio_vec *bvec;
195 	struct bvec_iter_all iter_all;
196 
197 	bio_for_each_segment_all(bvec, bio, iter_all) {
198 		struct page *page = bvec->bv_page;
199 		struct buffer_head *bh = page_buffers(page);
200 		unsigned int len = bvec->bv_len;
201 
202 		while (bh_offset(bh) < bvec->bv_offset)
203 			bh = bh->b_this_page;
204 		do {
205 			struct buffer_head *next = bh->b_this_page;
206 			len -= bh->b_size;
207 			bh->b_end_io(bh, !bio->bi_status);
208 			bh = next;
209 		} while (bh && len);
210 	}
211 	bio_put(bio);
212 }
213 
214 /*
215  * Submit several consecutive buffer head I/O requests as a single bio I/O
216  * request.  (See submit_bh_wbc.)
217  */
218 static void gfs2_submit_bhs(int op, int op_flags, struct buffer_head *bhs[],
219 			    int num)
220 {
221 	while (num > 0) {
222 		struct buffer_head *bh = *bhs;
223 		struct bio *bio;
224 
225 		bio = bio_alloc(GFP_NOIO, num);
226 		bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
227 		bio_set_dev(bio, bh->b_bdev);
228 		while (num > 0) {
229 			bh = *bhs;
230 			if (!bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh))) {
231 				BUG_ON(bio->bi_iter.bi_size == 0);
232 				break;
233 			}
234 			bhs++;
235 			num--;
236 		}
237 		bio->bi_end_io = gfs2_meta_read_endio;
238 		bio_set_op_attrs(bio, op, op_flags);
239 		submit_bio(bio);
240 	}
241 }
242 
243 /**
244  * gfs2_meta_read - Read a block from disk
245  * @gl: The glock covering the block
246  * @blkno: The block number
247  * @flags: flags
248  * @rahead: Do read-ahead
249  * @bhp: the place where the buffer is returned (NULL on failure)
250  *
251  * Returns: errno
252  */
253 
254 int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
255 		   int rahead, struct buffer_head **bhp)
256 {
257 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
258 	struct buffer_head *bh, *bhs[2];
259 	int num = 0;
260 
261 	if (unlikely(gfs2_withdrawn(sdp)) &&
262 	    (!sdp->sd_jdesc || gl != sdp->sd_jinode_gl)) {
263 		*bhp = NULL;
264 		return -EIO;
265 	}
266 
267 	*bhp = bh = gfs2_getbuf(gl, blkno, CREATE);
268 
269 	lock_buffer(bh);
270 	if (buffer_uptodate(bh)) {
271 		unlock_buffer(bh);
272 		flags &= ~DIO_WAIT;
273 	} else {
274 		bh->b_end_io = end_buffer_read_sync;
275 		get_bh(bh);
276 		bhs[num++] = bh;
277 	}
278 
279 	if (rahead) {
280 		bh = gfs2_getbuf(gl, blkno + 1, CREATE);
281 
282 		lock_buffer(bh);
283 		if (buffer_uptodate(bh)) {
284 			unlock_buffer(bh);
285 			brelse(bh);
286 		} else {
287 			bh->b_end_io = end_buffer_read_sync;
288 			bhs[num++] = bh;
289 		}
290 	}
291 
292 	gfs2_submit_bhs(REQ_OP_READ, REQ_META | REQ_PRIO, bhs, num);
293 	if (!(flags & DIO_WAIT))
294 		return 0;
295 
296 	bh = *bhp;
297 	wait_on_buffer(bh);
298 	if (unlikely(!buffer_uptodate(bh))) {
299 		struct gfs2_trans *tr = current->journal_info;
300 		if (tr && test_bit(TR_TOUCHED, &tr->tr_flags))
301 			gfs2_io_error_bh_wd(sdp, bh);
302 		brelse(bh);
303 		*bhp = NULL;
304 		return -EIO;
305 	}
306 
307 	return 0;
308 }
309 
310 /**
311  * gfs2_meta_wait - Reread a block from disk
312  * @sdp: the filesystem
313  * @bh: The block to wait for
314  *
315  * Returns: errno
316  */
317 
318 int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh)
319 {
320 	if (unlikely(gfs2_withdrawn(sdp)))
321 		return -EIO;
322 
323 	wait_on_buffer(bh);
324 
325 	if (!buffer_uptodate(bh)) {
326 		struct gfs2_trans *tr = current->journal_info;
327 		if (tr && test_bit(TR_TOUCHED, &tr->tr_flags))
328 			gfs2_io_error_bh_wd(sdp, bh);
329 		return -EIO;
330 	}
331 	if (unlikely(gfs2_withdrawn(sdp)))
332 		return -EIO;
333 
334 	return 0;
335 }
336 
337 void gfs2_remove_from_journal(struct buffer_head *bh, int meta)
338 {
339 	struct address_space *mapping = bh->b_page->mapping;
340 	struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
341 	struct gfs2_bufdata *bd = bh->b_private;
342 	struct gfs2_trans *tr = current->journal_info;
343 	int was_pinned = 0;
344 
345 	if (test_clear_buffer_pinned(bh)) {
346 		trace_gfs2_pin(bd, 0);
347 		atomic_dec(&sdp->sd_log_pinned);
348 		list_del_init(&bd->bd_list);
349 		if (meta == REMOVE_META)
350 			tr->tr_num_buf_rm++;
351 		else
352 			tr->tr_num_databuf_rm++;
353 		set_bit(TR_TOUCHED, &tr->tr_flags);
354 		was_pinned = 1;
355 		brelse(bh);
356 	}
357 	if (bd) {
358 		if (bd->bd_tr) {
359 			gfs2_trans_add_revoke(sdp, bd);
360 		} else if (was_pinned) {
361 			bh->b_private = NULL;
362 			kmem_cache_free(gfs2_bufdata_cachep, bd);
363 		} else if (!list_empty(&bd->bd_ail_st_list) &&
364 					!list_empty(&bd->bd_ail_gl_list)) {
365 			gfs2_remove_from_ail(bd);
366 		}
367 	}
368 	clear_buffer_dirty(bh);
369 	clear_buffer_uptodate(bh);
370 }
371 
372 /**
373  * gfs2_ail1_wipe - remove deleted/freed buffers from the ail1 list
374  * @sdp: superblock
375  * @bstart: starting block address of buffers to remove
376  * @blen: length of buffers to be removed
377  *
378  * This function is called from gfs2_journal wipe, whose job is to remove
379  * buffers, corresponding to deleted blocks, from the journal. If we find any
380  * bufdata elements on the system ail1 list, they haven't been written to
381  * the journal yet. So we remove them.
382  */
383 static void gfs2_ail1_wipe(struct gfs2_sbd *sdp, u64 bstart, u32 blen)
384 {
385 	struct gfs2_trans *tr, *s;
386 	struct gfs2_bufdata *bd, *bs;
387 	struct buffer_head *bh;
388 	u64 end = bstart + blen;
389 
390 	gfs2_log_lock(sdp);
391 	spin_lock(&sdp->sd_ail_lock);
392 	list_for_each_entry_safe(tr, s, &sdp->sd_ail1_list, tr_list) {
393 		list_for_each_entry_safe(bd, bs, &tr->tr_ail1_list,
394 					 bd_ail_st_list) {
395 			bh = bd->bd_bh;
396 			if (bh->b_blocknr < bstart || bh->b_blocknr >= end)
397 				continue;
398 
399 			gfs2_remove_from_journal(bh, REMOVE_JDATA);
400 		}
401 	}
402 	spin_unlock(&sdp->sd_ail_lock);
403 	gfs2_log_unlock(sdp);
404 }
405 
406 static struct buffer_head *gfs2_getjdatabuf(struct gfs2_inode *ip, u64 blkno)
407 {
408 	struct address_space *mapping = ip->i_inode.i_mapping;
409 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
410 	struct page *page;
411 	struct buffer_head *bh;
412 	unsigned int shift = PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift;
413 	unsigned long index = blkno >> shift; /* convert block to page */
414 	unsigned int bufnum = blkno - (index << shift);
415 
416 	page = find_get_page_flags(mapping, index, FGP_LOCK|FGP_ACCESSED);
417 	if (!page)
418 		return NULL;
419 	if (!page_has_buffers(page)) {
420 		unlock_page(page);
421 		put_page(page);
422 		return NULL;
423 	}
424 	/* Locate header for our buffer within our page */
425 	for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
426 		/* Do nothing */;
427 	get_bh(bh);
428 	unlock_page(page);
429 	put_page(page);
430 	return bh;
431 }
432 
433 /**
434  * gfs2_journal_wipe - make inode's buffers so they aren't dirty/pinned anymore
435  * @ip: the inode who owns the buffers
436  * @bstart: the first buffer in the run
437  * @blen: the number of buffers in the run
438  *
439  */
440 
441 void gfs2_journal_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen)
442 {
443 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
444 	struct buffer_head *bh;
445 	int ty;
446 
447 	gfs2_ail1_wipe(sdp, bstart, blen);
448 	while (blen) {
449 		ty = REMOVE_META;
450 		bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE);
451 		if (!bh && gfs2_is_jdata(ip)) {
452 			bh = gfs2_getjdatabuf(ip, bstart);
453 			ty = REMOVE_JDATA;
454 		}
455 		if (bh) {
456 			lock_buffer(bh);
457 			gfs2_log_lock(sdp);
458 			spin_lock(&sdp->sd_ail_lock);
459 			gfs2_remove_from_journal(bh, ty);
460 			spin_unlock(&sdp->sd_ail_lock);
461 			gfs2_log_unlock(sdp);
462 			unlock_buffer(bh);
463 			brelse(bh);
464 		}
465 
466 		bstart++;
467 		blen--;
468 	}
469 }
470 
471 /**
472  * gfs2_meta_buffer - Get a metadata buffer
473  * @ip: The GFS2 inode
474  * @mtype: The block type (GFS2_METATYPE_*)
475  * @num: The block number (device relative) of the buffer
476  * @bhp: the buffer is returned here
477  *
478  * Returns: errno
479  */
480 
481 int gfs2_meta_buffer(struct gfs2_inode *ip, u32 mtype, u64 num,
482 		     struct buffer_head **bhp)
483 {
484 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
485 	struct gfs2_glock *gl = ip->i_gl;
486 	struct buffer_head *bh;
487 	int ret = 0;
488 	int rahead = 0;
489 
490 	if (num == ip->i_no_addr)
491 		rahead = ip->i_rahead;
492 
493 	ret = gfs2_meta_read(gl, num, DIO_WAIT, rahead, &bh);
494 	if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) {
495 		brelse(bh);
496 		ret = -EIO;
497 	} else {
498 		*bhp = bh;
499 	}
500 	return ret;
501 }
502 
503 /**
504  * gfs2_meta_ra - start readahead on an extent of a file
505  * @gl: the glock the blocks belong to
506  * @dblock: the starting disk block
507  * @extlen: the number of blocks in the extent
508  *
509  * returns: the first buffer in the extent
510  */
511 
512 struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen)
513 {
514 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
515 	struct buffer_head *first_bh, *bh;
516 	u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
517 			  sdp->sd_sb.sb_bsize_shift;
518 
519 	BUG_ON(!extlen);
520 
521 	if (max_ra < 1)
522 		max_ra = 1;
523 	if (extlen > max_ra)
524 		extlen = max_ra;
525 
526 	first_bh = gfs2_getbuf(gl, dblock, CREATE);
527 
528 	if (buffer_uptodate(first_bh))
529 		goto out;
530 	if (!buffer_locked(first_bh))
531 		ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &first_bh);
532 
533 	dblock++;
534 	extlen--;
535 
536 	while (extlen) {
537 		bh = gfs2_getbuf(gl, dblock, CREATE);
538 
539 		if (!buffer_uptodate(bh) && !buffer_locked(bh))
540 			ll_rw_block(REQ_OP_READ,
541 				    REQ_RAHEAD | REQ_META | REQ_PRIO,
542 				    1, &bh);
543 		brelse(bh);
544 		dblock++;
545 		extlen--;
546 		if (!buffer_locked(first_bh) && buffer_uptodate(first_bh))
547 			goto out;
548 	}
549 
550 	wait_on_buffer(first_bh);
551 out:
552 	return first_bh;
553 }
554 
555