xref: /openbmc/linux/fs/gfs2/meta_io.c (revision 367b8112)
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/mm.h>
16 #include <linux/pagemap.h>
17 #include <linux/writeback.h>
18 #include <linux/swap.h>
19 #include <linux/delay.h>
20 #include <linux/bio.h>
21 #include <linux/gfs2_ondisk.h>
22 #include <linux/lm_interface.h>
23 
24 #include "gfs2.h"
25 #include "incore.h"
26 #include "glock.h"
27 #include "glops.h"
28 #include "inode.h"
29 #include "log.h"
30 #include "lops.h"
31 #include "meta_io.h"
32 #include "rgrp.h"
33 #include "trans.h"
34 #include "util.h"
35 #include "ops_address.h"
36 
37 static int aspace_get_block(struct inode *inode, sector_t lblock,
38 			    struct buffer_head *bh_result, int create)
39 {
40 	gfs2_assert_warn(inode->i_sb->s_fs_info, 0);
41 	return -EOPNOTSUPP;
42 }
43 
44 static int gfs2_aspace_writepage(struct page *page,
45 				 struct writeback_control *wbc)
46 {
47 	return block_write_full_page(page, aspace_get_block, wbc);
48 }
49 
50 static const struct address_space_operations aspace_aops = {
51 	.writepage = gfs2_aspace_writepage,
52 	.releasepage = gfs2_releasepage,
53 	.sync_page = block_sync_page,
54 };
55 
56 /**
57  * gfs2_aspace_get - Create and initialize a struct inode structure
58  * @sdp: the filesystem the aspace is in
59  *
60  * Right now a struct inode is just a struct inode.  Maybe Linux
61  * will supply a more lightweight address space construct (that works)
62  * in the future.
63  *
64  * Make sure pages/buffers in this aspace aren't in high memory.
65  *
66  * Returns: the aspace
67  */
68 
69 struct inode *gfs2_aspace_get(struct gfs2_sbd *sdp)
70 {
71 	struct inode *aspace;
72 	struct gfs2_inode *ip;
73 
74 	aspace = new_inode(sdp->sd_vfs);
75 	if (aspace) {
76 		mapping_set_gfp_mask(aspace->i_mapping, GFP_NOFS);
77 		aspace->i_mapping->a_ops = &aspace_aops;
78 		aspace->i_size = ~0ULL;
79 		ip = GFS2_I(aspace);
80 		clear_bit(GIF_USER, &ip->i_flags);
81 		insert_inode_hash(aspace);
82 	}
83 	return aspace;
84 }
85 
86 void gfs2_aspace_put(struct inode *aspace)
87 {
88 	remove_inode_hash(aspace);
89 	iput(aspace);
90 }
91 
92 /**
93  * gfs2_meta_inval - Invalidate all buffers associated with a glock
94  * @gl: the glock
95  *
96  */
97 
98 void gfs2_meta_inval(struct gfs2_glock *gl)
99 {
100 	struct gfs2_sbd *sdp = gl->gl_sbd;
101 	struct inode *aspace = gl->gl_aspace;
102 	struct address_space *mapping = gl->gl_aspace->i_mapping;
103 
104 	gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
105 
106 	atomic_inc(&aspace->i_writecount);
107 	truncate_inode_pages(mapping, 0);
108 	atomic_dec(&aspace->i_writecount);
109 
110 	gfs2_assert_withdraw(sdp, !mapping->nrpages);
111 }
112 
113 /**
114  * gfs2_meta_sync - Sync all buffers associated with a glock
115  * @gl: The glock
116  *
117  */
118 
119 void gfs2_meta_sync(struct gfs2_glock *gl)
120 {
121 	struct address_space *mapping = gl->gl_aspace->i_mapping;
122 	int error;
123 
124 	filemap_fdatawrite(mapping);
125 	error = filemap_fdatawait(mapping);
126 
127 	if (error)
128 		gfs2_io_error(gl->gl_sbd);
129 }
130 
131 /**
132  * gfs2_getbuf - Get a buffer with a given address space
133  * @gl: the glock
134  * @blkno: the block number (filesystem scope)
135  * @create: 1 if the buffer should be created
136  *
137  * Returns: the buffer
138  */
139 
140 struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create)
141 {
142 	struct address_space *mapping = gl->gl_aspace->i_mapping;
143 	struct gfs2_sbd *sdp = gl->gl_sbd;
144 	struct page *page;
145 	struct buffer_head *bh;
146 	unsigned int shift;
147 	unsigned long index;
148 	unsigned int bufnum;
149 
150 	shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift;
151 	index = blkno >> shift;             /* convert block to page */
152 	bufnum = blkno - (index << shift);  /* block buf index within page */
153 
154 	if (create) {
155 		for (;;) {
156 			page = grab_cache_page(mapping, index);
157 			if (page)
158 				break;
159 			yield();
160 		}
161 	} else {
162 		page = find_lock_page(mapping, index);
163 		if (!page)
164 			return NULL;
165 	}
166 
167 	if (!page_has_buffers(page))
168 		create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0);
169 
170 	/* Locate header for our buffer within our page */
171 	for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
172 		/* Do nothing */;
173 	get_bh(bh);
174 
175 	if (!buffer_mapped(bh))
176 		map_bh(bh, sdp->sd_vfs, blkno);
177 
178 	unlock_page(page);
179 	mark_page_accessed(page);
180 	page_cache_release(page);
181 
182 	return bh;
183 }
184 
185 static void meta_prep_new(struct buffer_head *bh)
186 {
187 	struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
188 
189 	lock_buffer(bh);
190 	clear_buffer_dirty(bh);
191 	set_buffer_uptodate(bh);
192 	unlock_buffer(bh);
193 
194 	mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
195 }
196 
197 /**
198  * gfs2_meta_new - Get a block
199  * @gl: The glock associated with this block
200  * @blkno: The block number
201  *
202  * Returns: The buffer
203  */
204 
205 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno)
206 {
207 	struct buffer_head *bh;
208 	bh = gfs2_getbuf(gl, blkno, CREATE);
209 	meta_prep_new(bh);
210 	return bh;
211 }
212 
213 /**
214  * gfs2_meta_read - Read a block from disk
215  * @gl: The glock covering the block
216  * @blkno: The block number
217  * @flags: flags
218  * @bhp: the place where the buffer is returned (NULL on failure)
219  *
220  * Returns: errno
221  */
222 
223 int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
224 		   struct buffer_head **bhp)
225 {
226 	*bhp = gfs2_getbuf(gl, blkno, CREATE);
227 	if (!buffer_uptodate(*bhp)) {
228 		ll_rw_block(READ_META, 1, bhp);
229 		if (flags & DIO_WAIT) {
230 			int error = gfs2_meta_wait(gl->gl_sbd, *bhp);
231 			if (error) {
232 				brelse(*bhp);
233 				return error;
234 			}
235 		}
236 	}
237 
238 	return 0;
239 }
240 
241 /**
242  * gfs2_meta_wait - Reread a block from disk
243  * @sdp: the filesystem
244  * @bh: The block to wait for
245  *
246  * Returns: errno
247  */
248 
249 int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh)
250 {
251 	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
252 		return -EIO;
253 
254 	wait_on_buffer(bh);
255 
256 	if (!buffer_uptodate(bh)) {
257 		struct gfs2_trans *tr = current->journal_info;
258 		if (tr && tr->tr_touched)
259 			gfs2_io_error_bh(sdp, bh);
260 		return -EIO;
261 	}
262 	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
263 		return -EIO;
264 
265 	return 0;
266 }
267 
268 /**
269  * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer
270  * @gl: the glock the buffer belongs to
271  * @bh: The buffer to be attached to
272  * @meta: Flag to indicate whether its metadata or not
273  */
274 
275 void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh,
276 			 int meta)
277 {
278 	struct gfs2_bufdata *bd;
279 
280 	if (meta)
281 		lock_page(bh->b_page);
282 
283 	if (bh->b_private) {
284 		if (meta)
285 			unlock_page(bh->b_page);
286 		return;
287 	}
288 
289 	bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL);
290 	bd->bd_bh = bh;
291 	bd->bd_gl = gl;
292 
293 	INIT_LIST_HEAD(&bd->bd_list_tr);
294 	if (meta)
295 		lops_init_le(&bd->bd_le, &gfs2_buf_lops);
296 	else
297 		lops_init_le(&bd->bd_le, &gfs2_databuf_lops);
298 	bh->b_private = bd;
299 
300 	if (meta)
301 		unlock_page(bh->b_page);
302 }
303 
304 void gfs2_remove_from_journal(struct buffer_head *bh, struct gfs2_trans *tr, int meta)
305 {
306 	struct gfs2_sbd *sdp = GFS2_SB(bh->b_page->mapping->host);
307 	struct gfs2_bufdata *bd = bh->b_private;
308 	if (test_clear_buffer_pinned(bh)) {
309 		list_del_init(&bd->bd_le.le_list);
310 		if (meta) {
311 			gfs2_assert_warn(sdp, sdp->sd_log_num_buf);
312 			sdp->sd_log_num_buf--;
313 			tr->tr_num_buf_rm++;
314 		} else {
315 			gfs2_assert_warn(sdp, sdp->sd_log_num_databuf);
316 			sdp->sd_log_num_databuf--;
317 			tr->tr_num_databuf_rm++;
318 		}
319 		tr->tr_touched = 1;
320 		brelse(bh);
321 	}
322 	if (bd) {
323 		if (bd->bd_ail) {
324 			gfs2_remove_from_ail(bd);
325 			bh->b_private = NULL;
326 			bd->bd_bh = NULL;
327 			bd->bd_blkno = bh->b_blocknr;
328 			gfs2_trans_add_revoke(sdp, bd);
329 		}
330 	}
331 	clear_buffer_dirty(bh);
332 	clear_buffer_uptodate(bh);
333 }
334 
335 /**
336  * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore
337  * @ip: the inode who owns the buffers
338  * @bstart: the first buffer in the run
339  * @blen: the number of buffers in the run
340  *
341  */
342 
343 void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen)
344 {
345 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
346 	struct buffer_head *bh;
347 
348 	while (blen) {
349 		bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE);
350 		if (bh) {
351 			lock_buffer(bh);
352 			gfs2_log_lock(sdp);
353 			gfs2_remove_from_journal(bh, current->journal_info, 1);
354 			gfs2_log_unlock(sdp);
355 			unlock_buffer(bh);
356 			brelse(bh);
357 		}
358 
359 		bstart++;
360 		blen--;
361 	}
362 }
363 
364 /**
365  * gfs2_meta_indirect_buffer - Get a metadata buffer
366  * @ip: The GFS2 inode
367  * @height: The level of this buf in the metadata (indir addr) tree (if any)
368  * @num: The block number (device relative) of the buffer
369  * @new: Non-zero if we may create a new buffer
370  * @bhp: the buffer is returned here
371  *
372  * Returns: errno
373  */
374 
375 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num,
376 			      int new, struct buffer_head **bhp)
377 {
378 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
379 	struct gfs2_glock *gl = ip->i_gl;
380 	struct buffer_head *bh;
381 	int ret = 0;
382 
383 	if (new) {
384 		BUG_ON(height == 0);
385 		bh = gfs2_meta_new(gl, num);
386 		gfs2_trans_add_bh(ip->i_gl, bh, 1);
387 		gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN);
388 		gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
389 	} else {
390 		u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI;
391 		ret = gfs2_meta_read(gl, num, DIO_WAIT, &bh);
392 		if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) {
393 			brelse(bh);
394 			ret = -EIO;
395 		}
396 	}
397 	*bhp = bh;
398 	return ret;
399 }
400 
401 /**
402  * gfs2_meta_ra - start readahead on an extent of a file
403  * @gl: the glock the blocks belong to
404  * @dblock: the starting disk block
405  * @extlen: the number of blocks in the extent
406  *
407  * returns: the first buffer in the extent
408  */
409 
410 struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen)
411 {
412 	struct gfs2_sbd *sdp = gl->gl_sbd;
413 	struct buffer_head *first_bh, *bh;
414 	u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
415 			  sdp->sd_sb.sb_bsize_shift;
416 
417 	BUG_ON(!extlen);
418 
419 	if (max_ra < 1)
420 		max_ra = 1;
421 	if (extlen > max_ra)
422 		extlen = max_ra;
423 
424 	first_bh = gfs2_getbuf(gl, dblock, CREATE);
425 
426 	if (buffer_uptodate(first_bh))
427 		goto out;
428 	if (!buffer_locked(first_bh))
429 		ll_rw_block(READ_META, 1, &first_bh);
430 
431 	dblock++;
432 	extlen--;
433 
434 	while (extlen) {
435 		bh = gfs2_getbuf(gl, dblock, CREATE);
436 
437 		if (!buffer_uptodate(bh) && !buffer_locked(bh))
438 			ll_rw_block(READA, 1, &bh);
439 		brelse(bh);
440 		dblock++;
441 		extlen--;
442 		if (!buffer_locked(first_bh) && buffer_uptodate(first_bh))
443 			goto out;
444 	}
445 
446 	wait_on_buffer(first_bh);
447 out:
448 	return first_bh;
449 }
450 
451