xref: /linux/fs/gfs2/trans.c (revision c4ee0af3fa0dc65f690fc908f02b8355f9576ea0)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2006 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/kallsyms.h>
16 #include <linux/gfs2_ondisk.h>
17 
18 #include "gfs2.h"
19 #include "incore.h"
20 #include "glock.h"
21 #include "inode.h"
22 #include "log.h"
23 #include "lops.h"
24 #include "meta_io.h"
25 #include "trans.h"
26 #include "util.h"
27 #include "trace_gfs2.h"
28 
29 int gfs2_trans_begin(struct gfs2_sbd *sdp, unsigned int blocks,
30 		     unsigned int revokes)
31 {
32 	struct gfs2_trans *tr;
33 	int error;
34 
35 	BUG_ON(current->journal_info);
36 	BUG_ON(blocks == 0 && revokes == 0);
37 
38 	if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
39 		return -EROFS;
40 
41 	tr = kzalloc(sizeof(struct gfs2_trans), GFP_NOFS);
42 	if (!tr)
43 		return -ENOMEM;
44 
45 	tr->tr_ip = (unsigned long)__builtin_return_address(0);
46 	tr->tr_blocks = blocks;
47 	tr->tr_revokes = revokes;
48 	tr->tr_reserved = 1;
49 	if (blocks)
50 		tr->tr_reserved += 6 + blocks;
51 	if (revokes)
52 		tr->tr_reserved += gfs2_struct2blk(sdp, revokes,
53 						   sizeof(u64));
54 	sb_start_intwrite(sdp->sd_vfs);
55 	gfs2_holder_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &tr->tr_t_gh);
56 
57 	error = gfs2_glock_nq(&tr->tr_t_gh);
58 	if (error)
59 		goto fail_holder_uninit;
60 
61 	error = gfs2_log_reserve(sdp, tr->tr_reserved);
62 	if (error)
63 		goto fail_gunlock;
64 
65 	current->journal_info = tr;
66 
67 	return 0;
68 
69 fail_gunlock:
70 	gfs2_glock_dq(&tr->tr_t_gh);
71 
72 fail_holder_uninit:
73 	sb_end_intwrite(sdp->sd_vfs);
74 	gfs2_holder_uninit(&tr->tr_t_gh);
75 	kfree(tr);
76 
77 	return error;
78 }
79 
80 /**
81  * gfs2_log_release - Release a given number of log blocks
82  * @sdp: The GFS2 superblock
83  * @blks: The number of blocks
84  *
85  */
86 
87 static void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
88 {
89 
90 	atomic_add(blks, &sdp->sd_log_blks_free);
91 	trace_gfs2_log_blocks(sdp, blks);
92 	gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
93 				  sdp->sd_jdesc->jd_blocks);
94 	up_read(&sdp->sd_log_flush_lock);
95 }
96 
97 static void gfs2_print_trans(const struct gfs2_trans *tr)
98 {
99 	printk(KERN_WARNING "GFS2: Transaction created at: %pSR\n",
100 	       (void *)tr->tr_ip);
101 	printk(KERN_WARNING "GFS2: blocks=%u revokes=%u reserved=%u touched=%d\n",
102 	       tr->tr_blocks, tr->tr_revokes, tr->tr_reserved, tr->tr_touched);
103 	printk(KERN_WARNING "GFS2: Buf %u/%u Databuf %u/%u Revoke %u/%u\n",
104 	       tr->tr_num_buf_new, tr->tr_num_buf_rm,
105 	       tr->tr_num_databuf_new, tr->tr_num_databuf_rm,
106 	       tr->tr_num_revoke, tr->tr_num_revoke_rm);
107 }
108 
109 void gfs2_trans_end(struct gfs2_sbd *sdp)
110 {
111 	struct gfs2_trans *tr = current->journal_info;
112 	s64 nbuf;
113 	BUG_ON(!tr);
114 	current->journal_info = NULL;
115 
116 	if (!tr->tr_touched) {
117 		gfs2_log_release(sdp, tr->tr_reserved);
118 		if (tr->tr_t_gh.gh_gl) {
119 			gfs2_glock_dq(&tr->tr_t_gh);
120 			gfs2_holder_uninit(&tr->tr_t_gh);
121 			kfree(tr);
122 		}
123 		sb_end_intwrite(sdp->sd_vfs);
124 		return;
125 	}
126 
127 	nbuf = tr->tr_num_buf_new + tr->tr_num_databuf_new;
128 	nbuf -= tr->tr_num_buf_rm;
129 	nbuf -= tr->tr_num_databuf_rm;
130 
131 	if (gfs2_assert_withdraw(sdp, (nbuf <= tr->tr_blocks) &&
132 				       (tr->tr_num_revoke <= tr->tr_revokes)))
133 		gfs2_print_trans(tr);
134 
135 	gfs2_log_commit(sdp, tr);
136 	if (tr->tr_t_gh.gh_gl) {
137 		gfs2_glock_dq(&tr->tr_t_gh);
138 		gfs2_holder_uninit(&tr->tr_t_gh);
139 		if (!tr->tr_attached)
140 			kfree(tr);
141 	}
142 	up_read(&sdp->sd_log_flush_lock);
143 
144 	if (sdp->sd_vfs->s_flags & MS_SYNCHRONOUS)
145 		gfs2_log_flush(sdp, NULL);
146 	sb_end_intwrite(sdp->sd_vfs);
147 }
148 
149 static struct gfs2_bufdata *gfs2_alloc_bufdata(struct gfs2_glock *gl,
150 					       struct buffer_head *bh,
151 					       const struct gfs2_log_operations *lops)
152 {
153 	struct gfs2_bufdata *bd;
154 
155 	bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL);
156 	bd->bd_bh = bh;
157 	bd->bd_gl = gl;
158 	bd->bd_ops = lops;
159 	INIT_LIST_HEAD(&bd->bd_list);
160 	bh->b_private = bd;
161 	return bd;
162 }
163 
164 /**
165  * gfs2_trans_add_data - Add a databuf to the transaction.
166  * @gl: The inode glock associated with the buffer
167  * @bh: The buffer to add
168  *
169  * This is used in two distinct cases:
170  * i) In ordered write mode
171  *    We put the data buffer on a list so that we can ensure that its
172  *    synced to disk at the right time
173  * ii) In journaled data mode
174  *    We need to journal the data block in the same way as metadata in
175  *    the functions above. The difference is that here we have a tag
176  *    which is two __be64's being the block number (as per meta data)
177  *    and a flag which says whether the data block needs escaping or
178  *    not. This means we need a new log entry for each 251 or so data
179  *    blocks, which isn't an enormous overhead but twice as much as
180  *    for normal metadata blocks.
181  */
182 void gfs2_trans_add_data(struct gfs2_glock *gl, struct buffer_head *bh)
183 {
184 	struct gfs2_trans *tr = current->journal_info;
185 	struct gfs2_sbd *sdp = gl->gl_sbd;
186 	struct address_space *mapping = bh->b_page->mapping;
187 	struct gfs2_inode *ip = GFS2_I(mapping->host);
188 	struct gfs2_bufdata *bd;
189 
190 	if (!gfs2_is_jdata(ip)) {
191 		gfs2_ordered_add_inode(ip);
192 		return;
193 	}
194 
195 	lock_buffer(bh);
196 	gfs2_log_lock(sdp);
197 	bd = bh->b_private;
198 	if (bd == NULL) {
199 		gfs2_log_unlock(sdp);
200 		unlock_buffer(bh);
201 		if (bh->b_private == NULL)
202 			bd = gfs2_alloc_bufdata(gl, bh, &gfs2_databuf_lops);
203 		lock_buffer(bh);
204 		gfs2_log_lock(sdp);
205 	}
206 	gfs2_assert(sdp, bd->bd_gl == gl);
207 	tr->tr_touched = 1;
208 	if (list_empty(&bd->bd_list)) {
209 		set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
210 		set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
211 		gfs2_pin(sdp, bd->bd_bh);
212 		tr->tr_num_databuf_new++;
213 		sdp->sd_log_num_databuf++;
214 		list_add_tail(&bd->bd_list, &sdp->sd_log_le_databuf);
215 	}
216 	gfs2_log_unlock(sdp);
217 	unlock_buffer(bh);
218 }
219 
220 static void meta_lo_add(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
221 {
222 	struct gfs2_meta_header *mh;
223 	struct gfs2_trans *tr;
224 
225 	tr = current->journal_info;
226 	tr->tr_touched = 1;
227 	if (!list_empty(&bd->bd_list))
228 		return;
229 	set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
230 	set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
231 	mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
232 	if (unlikely(mh->mh_magic != cpu_to_be32(GFS2_MAGIC))) {
233 		printk(KERN_ERR
234 		       "Attempting to add uninitialised block to journal (inplace block=%lld)\n",
235 		       (unsigned long long)bd->bd_bh->b_blocknr);
236 		BUG();
237 	}
238 	gfs2_pin(sdp, bd->bd_bh);
239 	mh->__pad0 = cpu_to_be64(0);
240 	mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
241 	sdp->sd_log_num_buf++;
242 	list_add(&bd->bd_list, &sdp->sd_log_le_buf);
243 	tr->tr_num_buf_new++;
244 }
245 
246 void gfs2_trans_add_meta(struct gfs2_glock *gl, struct buffer_head *bh)
247 {
248 
249 	struct gfs2_sbd *sdp = gl->gl_sbd;
250 	struct gfs2_bufdata *bd;
251 
252 	lock_buffer(bh);
253 	gfs2_log_lock(sdp);
254 	bd = bh->b_private;
255 	if (bd == NULL) {
256 		gfs2_log_unlock(sdp);
257 		unlock_buffer(bh);
258 		lock_page(bh->b_page);
259 		if (bh->b_private == NULL)
260 			bd = gfs2_alloc_bufdata(gl, bh, &gfs2_buf_lops);
261 		unlock_page(bh->b_page);
262 		lock_buffer(bh);
263 		gfs2_log_lock(sdp);
264 	}
265 	gfs2_assert(sdp, bd->bd_gl == gl);
266 	meta_lo_add(sdp, bd);
267 	gfs2_log_unlock(sdp);
268 	unlock_buffer(bh);
269 }
270 
271 void gfs2_trans_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
272 {
273 	struct gfs2_trans *tr = current->journal_info;
274 
275 	BUG_ON(!list_empty(&bd->bd_list));
276 	gfs2_add_revoke(sdp, bd);
277 	tr->tr_touched = 1;
278 	tr->tr_num_revoke++;
279 }
280 
281 void gfs2_trans_add_unrevoke(struct gfs2_sbd *sdp, u64 blkno, unsigned int len)
282 {
283 	struct gfs2_bufdata *bd, *tmp;
284 	struct gfs2_trans *tr = current->journal_info;
285 	unsigned int n = len;
286 
287 	gfs2_log_lock(sdp);
288 	list_for_each_entry_safe(bd, tmp, &sdp->sd_log_le_revoke, bd_list) {
289 		if ((bd->bd_blkno >= blkno) && (bd->bd_blkno < (blkno + len))) {
290 			list_del_init(&bd->bd_list);
291 			gfs2_assert_withdraw(sdp, sdp->sd_log_num_revoke);
292 			sdp->sd_log_num_revoke--;
293 			kmem_cache_free(gfs2_bufdata_cachep, bd);
294 			tr->tr_num_revoke_rm++;
295 			if (--n == 0)
296 				break;
297 		}
298 	}
299 	gfs2_log_unlock(sdp);
300 }
301 
302