xref: /openbmc/linux/fs/gfs2/trans.c (revision c1eba1b0)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
4  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
5  */
6 
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 
9 #include <linux/sched.h>
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/kallsyms.h>
15 #include <linux/gfs2_ondisk.h>
16 
17 #include "gfs2.h"
18 #include "incore.h"
19 #include "glock.h"
20 #include "inode.h"
21 #include "log.h"
22 #include "lops.h"
23 #include "meta_io.h"
24 #include "trans.h"
25 #include "util.h"
26 #include "trace_gfs2.h"
27 
28 static void gfs2_print_trans(struct gfs2_sbd *sdp, const struct gfs2_trans *tr)
29 {
30 	fs_warn(sdp, "Transaction created at: %pSR\n", (void *)tr->tr_ip);
31 	fs_warn(sdp, "blocks=%u revokes=%u reserved=%u touched=%u\n",
32 		tr->tr_blocks, tr->tr_revokes, tr->tr_reserved,
33 		test_bit(TR_TOUCHED, &tr->tr_flags));
34 	fs_warn(sdp, "Buf %u/%u Databuf %u/%u Revoke %u/%u\n",
35 		tr->tr_num_buf_new, tr->tr_num_buf_rm,
36 		tr->tr_num_databuf_new, tr->tr_num_databuf_rm,
37 		tr->tr_num_revoke, tr->tr_num_revoke_rm);
38 }
39 
40 int __gfs2_trans_begin(struct gfs2_trans *tr, struct gfs2_sbd *sdp,
41 		       unsigned int blocks, unsigned int revokes,
42 		       unsigned long ip)
43 {
44 	if (current->journal_info) {
45 		gfs2_print_trans(sdp, current->journal_info);
46 		BUG();
47 	}
48 	BUG_ON(blocks == 0 && revokes == 0);
49 
50 	if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
51 		return -EROFS;
52 
53 	tr->tr_ip = ip;
54 	tr->tr_blocks = blocks;
55 	tr->tr_revokes = revokes;
56 	tr->tr_reserved = 1;
57 	if (blocks)
58 		tr->tr_reserved += 6 + blocks;
59 	if (revokes)
60 		tr->tr_reserved += gfs2_struct2blk(sdp, revokes);
61 	INIT_LIST_HEAD(&tr->tr_databuf);
62 	INIT_LIST_HEAD(&tr->tr_buf);
63 	INIT_LIST_HEAD(&tr->tr_list);
64 	INIT_LIST_HEAD(&tr->tr_ail1_list);
65 	INIT_LIST_HEAD(&tr->tr_ail2_list);
66 
67 	if (gfs2_assert_warn(sdp, tr->tr_reserved <= sdp->sd_jdesc->jd_blocks))
68 		return -EINVAL;
69 
70 	sb_start_intwrite(sdp->sd_vfs);
71 
72 	gfs2_log_reserve(sdp, tr->tr_reserved);
73 
74 	down_read(&sdp->sd_log_flush_lock);
75 	if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
76 		gfs2_log_release(sdp, tr->tr_reserved);
77 		up_read(&sdp->sd_log_flush_lock);
78 		sb_end_intwrite(sdp->sd_vfs);
79 		wake_up(&sdp->sd_log_waitq);
80 		return -EROFS;
81 	}
82 
83 	current->journal_info = tr;
84 
85 	return 0;
86 }
87 
88 int gfs2_trans_begin(struct gfs2_sbd *sdp, unsigned int blocks,
89 		     unsigned int revokes)
90 {
91 	struct gfs2_trans *tr;
92 	int error;
93 
94 	tr = kmem_cache_zalloc(gfs2_trans_cachep, GFP_NOFS);
95 	if (!tr)
96 		return -ENOMEM;
97 	error = __gfs2_trans_begin(tr, sdp, blocks, revokes, _RET_IP_);
98 	if (error)
99 		kmem_cache_free(gfs2_trans_cachep, tr);
100 	return error;
101 }
102 
103 void gfs2_trans_end(struct gfs2_sbd *sdp)
104 {
105 	struct gfs2_trans *tr = current->journal_info;
106 	s64 nbuf;
107 
108 	current->journal_info = NULL;
109 
110 	if (!test_bit(TR_TOUCHED, &tr->tr_flags)) {
111 		gfs2_log_release(sdp, tr->tr_reserved);
112 		up_read(&sdp->sd_log_flush_lock);
113 		if (!test_bit(TR_ONSTACK, &tr->tr_flags))
114 			gfs2_trans_free(sdp, tr);
115 		sb_end_intwrite(sdp->sd_vfs);
116 		return;
117 	}
118 
119 	nbuf = tr->tr_num_buf_new + tr->tr_num_databuf_new;
120 	nbuf -= tr->tr_num_buf_rm;
121 	nbuf -= tr->tr_num_databuf_rm;
122 
123 	if (gfs2_assert_withdraw(sdp, nbuf <= tr->tr_blocks) ||
124 	    gfs2_assert_withdraw(sdp, tr->tr_num_revoke <= tr->tr_revokes))
125 		gfs2_print_trans(sdp, tr);
126 
127 	gfs2_log_commit(sdp, tr);
128 	if (!test_bit(TR_ONSTACK, &tr->tr_flags) &&
129 	    !test_bit(TR_ATTACHED, &tr->tr_flags))
130 		gfs2_trans_free(sdp, tr);
131 	up_read(&sdp->sd_log_flush_lock);
132 
133 	if (sdp->sd_vfs->s_flags & SB_SYNCHRONOUS)
134 		gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
135 			       GFS2_LFC_TRANS_END);
136 	sb_end_intwrite(sdp->sd_vfs);
137 }
138 
139 static struct gfs2_bufdata *gfs2_alloc_bufdata(struct gfs2_glock *gl,
140 					       struct buffer_head *bh)
141 {
142 	struct gfs2_bufdata *bd;
143 
144 	bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL);
145 	bd->bd_bh = bh;
146 	bd->bd_gl = gl;
147 	INIT_LIST_HEAD(&bd->bd_list);
148 	bh->b_private = bd;
149 	return bd;
150 }
151 
152 /**
153  * gfs2_trans_add_data - Add a databuf to the transaction.
154  * @gl: The inode glock associated with the buffer
155  * @bh: The buffer to add
156  *
157  * This is used in journaled data mode.
158  * We need to journal the data block in the same way as metadata in
159  * the functions above. The difference is that here we have a tag
160  * which is two __be64's being the block number (as per meta data)
161  * and a flag which says whether the data block needs escaping or
162  * not. This means we need a new log entry for each 251 or so data
163  * blocks, which isn't an enormous overhead but twice as much as
164  * for normal metadata blocks.
165  */
166 void gfs2_trans_add_data(struct gfs2_glock *gl, struct buffer_head *bh)
167 {
168 	struct gfs2_trans *tr = current->journal_info;
169 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
170 	struct gfs2_bufdata *bd;
171 
172 	lock_buffer(bh);
173 	if (buffer_pinned(bh)) {
174 		set_bit(TR_TOUCHED, &tr->tr_flags);
175 		goto out;
176 	}
177 	gfs2_log_lock(sdp);
178 	bd = bh->b_private;
179 	if (bd == NULL) {
180 		gfs2_log_unlock(sdp);
181 		unlock_buffer(bh);
182 		if (bh->b_private == NULL)
183 			bd = gfs2_alloc_bufdata(gl, bh);
184 		else
185 			bd = bh->b_private;
186 		lock_buffer(bh);
187 		gfs2_log_lock(sdp);
188 	}
189 	gfs2_assert(sdp, bd->bd_gl == gl);
190 	set_bit(TR_TOUCHED, &tr->tr_flags);
191 	if (list_empty(&bd->bd_list)) {
192 		set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
193 		set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
194 		gfs2_pin(sdp, bd->bd_bh);
195 		tr->tr_num_databuf_new++;
196 		list_add_tail(&bd->bd_list, &tr->tr_databuf);
197 	}
198 	gfs2_log_unlock(sdp);
199 out:
200 	unlock_buffer(bh);
201 }
202 
203 void gfs2_trans_add_meta(struct gfs2_glock *gl, struct buffer_head *bh)
204 {
205 
206 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
207 	struct gfs2_bufdata *bd;
208 	struct gfs2_meta_header *mh;
209 	struct gfs2_trans *tr = current->journal_info;
210 	enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
211 
212 	lock_buffer(bh);
213 	if (buffer_pinned(bh)) {
214 		set_bit(TR_TOUCHED, &tr->tr_flags);
215 		goto out;
216 	}
217 	gfs2_log_lock(sdp);
218 	bd = bh->b_private;
219 	if (bd == NULL) {
220 		gfs2_log_unlock(sdp);
221 		unlock_buffer(bh);
222 		lock_page(bh->b_page);
223 		if (bh->b_private == NULL)
224 			bd = gfs2_alloc_bufdata(gl, bh);
225 		else
226 			bd = bh->b_private;
227 		unlock_page(bh->b_page);
228 		lock_buffer(bh);
229 		gfs2_log_lock(sdp);
230 	}
231 	gfs2_assert(sdp, bd->bd_gl == gl);
232 	set_bit(TR_TOUCHED, &tr->tr_flags);
233 	if (!list_empty(&bd->bd_list))
234 		goto out_unlock;
235 	set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
236 	set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
237 	mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
238 	if (unlikely(mh->mh_magic != cpu_to_be32(GFS2_MAGIC))) {
239 		fs_err(sdp, "Attempting to add uninitialised block to "
240 		       "journal (inplace block=%lld)\n",
241 		       (unsigned long long)bd->bd_bh->b_blocknr);
242 		BUG();
243 	}
244 	if (unlikely(state == SFS_FROZEN)) {
245 		fs_info(sdp, "GFS2:adding buf while frozen\n");
246 		gfs2_assert_withdraw(sdp, 0);
247 	}
248 	if (unlikely(gfs2_withdrawn(sdp))) {
249 		fs_info(sdp, "GFS2:adding buf while withdrawn! 0x%llx\n",
250 			(unsigned long long)bd->bd_bh->b_blocknr);
251 	}
252 	gfs2_pin(sdp, bd->bd_bh);
253 	mh->__pad0 = cpu_to_be64(0);
254 	mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
255 	list_add(&bd->bd_list, &tr->tr_buf);
256 	tr->tr_num_buf_new++;
257 out_unlock:
258 	gfs2_log_unlock(sdp);
259 out:
260 	unlock_buffer(bh);
261 }
262 
263 void gfs2_trans_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
264 {
265 	struct gfs2_trans *tr = current->journal_info;
266 
267 	BUG_ON(!list_empty(&bd->bd_list));
268 	gfs2_add_revoke(sdp, bd);
269 	set_bit(TR_TOUCHED, &tr->tr_flags);
270 	tr->tr_num_revoke++;
271 }
272 
273 void gfs2_trans_remove_revoke(struct gfs2_sbd *sdp, u64 blkno, unsigned int len)
274 {
275 	struct gfs2_bufdata *bd, *tmp;
276 	struct gfs2_trans *tr = current->journal_info;
277 	unsigned int n = len;
278 
279 	gfs2_log_lock(sdp);
280 	list_for_each_entry_safe(bd, tmp, &sdp->sd_log_revokes, bd_list) {
281 		if ((bd->bd_blkno >= blkno) && (bd->bd_blkno < (blkno + len))) {
282 			list_del_init(&bd->bd_list);
283 			gfs2_assert_withdraw(sdp, sdp->sd_log_num_revoke);
284 			sdp->sd_log_num_revoke--;
285 			if (bd->bd_gl)
286 				gfs2_glock_remove_revoke(bd->bd_gl);
287 			kmem_cache_free(gfs2_bufdata_cachep, bd);
288 			tr->tr_num_revoke_rm++;
289 			if (--n == 0)
290 				break;
291 		}
292 	}
293 	gfs2_log_unlock(sdp);
294 }
295 
296 void gfs2_trans_free(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
297 {
298 	if (tr == NULL)
299 		return;
300 
301 	gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
302 	gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
303 	gfs2_assert_warn(sdp, list_empty(&tr->tr_databuf));
304 	gfs2_assert_warn(sdp, list_empty(&tr->tr_buf));
305 	kmem_cache_free(gfs2_trans_cachep, tr);
306 }
307