xref: /openbmc/linux/fs/ext4/ext4_jbd2.c (revision 188c299e)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Interface between ext4 and JBD
4  */
5 
6 #include "ext4_jbd2.h"
7 
8 #include <trace/events/ext4.h>
9 
10 int ext4_inode_journal_mode(struct inode *inode)
11 {
12 	if (EXT4_JOURNAL(inode) == NULL)
13 		return EXT4_INODE_WRITEBACK_DATA_MODE;	/* writeback */
14 	/* We do not support data journalling with delayed allocation */
15 	if (!S_ISREG(inode->i_mode) ||
16 	    ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE) ||
17 	    test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
18 	    (ext4_test_inode_flag(inode, EXT4_INODE_JOURNAL_DATA) &&
19 	    !test_opt(inode->i_sb, DELALLOC))) {
20 		/* We do not support data journalling for encrypted data */
21 		if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode))
22 			return EXT4_INODE_ORDERED_DATA_MODE;  /* ordered */
23 		return EXT4_INODE_JOURNAL_DATA_MODE;	/* journal data */
24 	}
25 	if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
26 		return EXT4_INODE_ORDERED_DATA_MODE;	/* ordered */
27 	if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
28 		return EXT4_INODE_WRITEBACK_DATA_MODE;	/* writeback */
29 	BUG();
30 }
31 
32 /* Just increment the non-pointer handle value */
33 static handle_t *ext4_get_nojournal(void)
34 {
35 	handle_t *handle = current->journal_info;
36 	unsigned long ref_cnt = (unsigned long)handle;
37 
38 	BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
39 
40 	ref_cnt++;
41 	handle = (handle_t *)ref_cnt;
42 
43 	current->journal_info = handle;
44 	return handle;
45 }
46 
47 
48 /* Decrement the non-pointer handle value */
49 static void ext4_put_nojournal(handle_t *handle)
50 {
51 	unsigned long ref_cnt = (unsigned long)handle;
52 
53 	BUG_ON(ref_cnt == 0);
54 
55 	ref_cnt--;
56 	handle = (handle_t *)ref_cnt;
57 
58 	current->journal_info = handle;
59 }
60 
61 /*
62  * Wrappers for jbd2_journal_start/end.
63  */
64 static int ext4_journal_check_start(struct super_block *sb)
65 {
66 	journal_t *journal;
67 
68 	might_sleep();
69 
70 	if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
71 		return -EIO;
72 
73 	if (sb_rdonly(sb))
74 		return -EROFS;
75 	WARN_ON(sb->s_writers.frozen == SB_FREEZE_COMPLETE);
76 	journal = EXT4_SB(sb)->s_journal;
77 	/*
78 	 * Special case here: if the journal has aborted behind our
79 	 * backs (eg. EIO in the commit thread), then we still need to
80 	 * take the FS itself readonly cleanly.
81 	 */
82 	if (journal && is_journal_aborted(journal)) {
83 		ext4_abort(sb, -journal->j_errno, "Detected aborted journal");
84 		return -EROFS;
85 	}
86 	return 0;
87 }
88 
89 handle_t *__ext4_journal_start_sb(struct super_block *sb, unsigned int line,
90 				  int type, int blocks, int rsv_blocks,
91 				  int revoke_creds)
92 {
93 	journal_t *journal;
94 	int err;
95 
96 	trace_ext4_journal_start(sb, blocks, rsv_blocks, revoke_creds,
97 				 _RET_IP_);
98 	err = ext4_journal_check_start(sb);
99 	if (err < 0)
100 		return ERR_PTR(err);
101 
102 	journal = EXT4_SB(sb)->s_journal;
103 	if (!journal || (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))
104 		return ext4_get_nojournal();
105 	return jbd2__journal_start(journal, blocks, rsv_blocks, revoke_creds,
106 				   GFP_NOFS, type, line);
107 }
108 
109 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
110 {
111 	struct super_block *sb;
112 	int err;
113 	int rc;
114 
115 	if (!ext4_handle_valid(handle)) {
116 		ext4_put_nojournal(handle);
117 		return 0;
118 	}
119 
120 	err = handle->h_err;
121 	if (!handle->h_transaction) {
122 		rc = jbd2_journal_stop(handle);
123 		return err ? err : rc;
124 	}
125 
126 	sb = handle->h_transaction->t_journal->j_private;
127 	rc = jbd2_journal_stop(handle);
128 
129 	if (!err)
130 		err = rc;
131 	if (err)
132 		__ext4_std_error(sb, where, line, err);
133 	return err;
134 }
135 
136 handle_t *__ext4_journal_start_reserved(handle_t *handle, unsigned int line,
137 					int type)
138 {
139 	struct super_block *sb;
140 	int err;
141 
142 	if (!ext4_handle_valid(handle))
143 		return ext4_get_nojournal();
144 
145 	sb = handle->h_journal->j_private;
146 	trace_ext4_journal_start_reserved(sb,
147 				jbd2_handle_buffer_credits(handle), _RET_IP_);
148 	err = ext4_journal_check_start(sb);
149 	if (err < 0) {
150 		jbd2_journal_free_reserved(handle);
151 		return ERR_PTR(err);
152 	}
153 
154 	err = jbd2_journal_start_reserved(handle, type, line);
155 	if (err < 0)
156 		return ERR_PTR(err);
157 	return handle;
158 }
159 
160 int __ext4_journal_ensure_credits(handle_t *handle, int check_cred,
161 				  int extend_cred, int revoke_cred)
162 {
163 	if (!ext4_handle_valid(handle))
164 		return 0;
165 	if (jbd2_handle_buffer_credits(handle) >= check_cred &&
166 	    handle->h_revoke_credits >= revoke_cred)
167 		return 0;
168 	extend_cred = max(0, extend_cred - jbd2_handle_buffer_credits(handle));
169 	revoke_cred = max(0, revoke_cred - handle->h_revoke_credits);
170 	return ext4_journal_extend(handle, extend_cred, revoke_cred);
171 }
172 
173 static void ext4_journal_abort_handle(const char *caller, unsigned int line,
174 				      const char *err_fn,
175 				      struct buffer_head *bh,
176 				      handle_t *handle, int err)
177 {
178 	char nbuf[16];
179 	const char *errstr = ext4_decode_error(NULL, err, nbuf);
180 
181 	BUG_ON(!ext4_handle_valid(handle));
182 
183 	if (bh)
184 		BUFFER_TRACE(bh, "abort");
185 
186 	if (!handle->h_err)
187 		handle->h_err = err;
188 
189 	if (is_handle_aborted(handle))
190 		return;
191 
192 	printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
193 	       caller, line, errstr, err_fn);
194 
195 	jbd2_journal_abort_handle(handle);
196 }
197 
198 static void ext4_check_bdev_write_error(struct super_block *sb)
199 {
200 	struct address_space *mapping = sb->s_bdev->bd_inode->i_mapping;
201 	struct ext4_sb_info *sbi = EXT4_SB(sb);
202 	int err;
203 
204 	/*
205 	 * If the block device has write error flag, it may have failed to
206 	 * async write out metadata buffers in the background. In this case,
207 	 * we could read old data from disk and write it out again, which
208 	 * may lead to on-disk filesystem inconsistency.
209 	 */
210 	if (errseq_check(&mapping->wb_err, READ_ONCE(sbi->s_bdev_wb_err))) {
211 		spin_lock(&sbi->s_bdev_wb_lock);
212 		err = errseq_check_and_advance(&mapping->wb_err, &sbi->s_bdev_wb_err);
213 		spin_unlock(&sbi->s_bdev_wb_lock);
214 		if (err)
215 			ext4_error_err(sb, -err,
216 				       "Error while async write back metadata");
217 	}
218 }
219 
220 int __ext4_journal_get_write_access(const char *where, unsigned int line,
221 				    handle_t *handle, struct super_block *sb,
222 				    struct buffer_head *bh,
223 				    enum ext4_journal_trigger_type trigger_type)
224 {
225 	int err;
226 
227 	might_sleep();
228 
229 	if (bh->b_bdev->bd_super)
230 		ext4_check_bdev_write_error(bh->b_bdev->bd_super);
231 
232 	if (ext4_handle_valid(handle)) {
233 		err = jbd2_journal_get_write_access(handle, bh);
234 		if (err) {
235 			ext4_journal_abort_handle(where, line, __func__, bh,
236 						  handle, err);
237 			return err;
238 		}
239 	}
240 	if (trigger_type == EXT4_JTR_NONE || !ext4_has_metadata_csum(sb))
241 		return 0;
242 	BUG_ON(trigger_type >= EXT4_JOURNAL_TRIGGER_COUNT);
243 	jbd2_journal_set_triggers(bh,
244 		&EXT4_SB(sb)->s_journal_triggers[trigger_type].tr_triggers);
245 	return 0;
246 }
247 
248 /*
249  * The ext4 forget function must perform a revoke if we are freeing data
250  * which has been journaled.  Metadata (eg. indirect blocks) must be
251  * revoked in all cases.
252  *
253  * "bh" may be NULL: a metadata block may have been freed from memory
254  * but there may still be a record of it in the journal, and that record
255  * still needs to be revoked.
256  */
257 int __ext4_forget(const char *where, unsigned int line, handle_t *handle,
258 		  int is_metadata, struct inode *inode,
259 		  struct buffer_head *bh, ext4_fsblk_t blocknr)
260 {
261 	int err;
262 
263 	might_sleep();
264 
265 	trace_ext4_forget(inode, is_metadata, blocknr);
266 	BUFFER_TRACE(bh, "enter");
267 
268 	jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
269 		  "data mode %x\n",
270 		  bh, is_metadata, inode->i_mode,
271 		  test_opt(inode->i_sb, DATA_FLAGS));
272 
273 	/* In the no journal case, we can just do a bforget and return */
274 	if (!ext4_handle_valid(handle)) {
275 		bforget(bh);
276 		return 0;
277 	}
278 
279 	/* Never use the revoke function if we are doing full data
280 	 * journaling: there is no need to, and a V1 superblock won't
281 	 * support it.  Otherwise, only skip the revoke on un-journaled
282 	 * data blocks. */
283 
284 	if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
285 	    (!is_metadata && !ext4_should_journal_data(inode))) {
286 		if (bh) {
287 			BUFFER_TRACE(bh, "call jbd2_journal_forget");
288 			err = jbd2_journal_forget(handle, bh);
289 			if (err)
290 				ext4_journal_abort_handle(where, line, __func__,
291 							  bh, handle, err);
292 			return err;
293 		}
294 		return 0;
295 	}
296 
297 	/*
298 	 * data!=journal && (is_metadata || should_journal_data(inode))
299 	 */
300 	BUFFER_TRACE(bh, "call jbd2_journal_revoke");
301 	err = jbd2_journal_revoke(handle, blocknr, bh);
302 	if (err) {
303 		ext4_journal_abort_handle(where, line, __func__,
304 					  bh, handle, err);
305 		__ext4_error(inode->i_sb, where, line, true, -err, 0,
306 			     "error %d when attempting revoke", err);
307 	}
308 	BUFFER_TRACE(bh, "exit");
309 	return err;
310 }
311 
312 int __ext4_journal_get_create_access(const char *where, unsigned int line,
313 				handle_t *handle, struct super_block *sb,
314 				struct buffer_head *bh,
315 				enum ext4_journal_trigger_type trigger_type)
316 {
317 	int err;
318 
319 	if (!ext4_handle_valid(handle))
320 		return 0;
321 
322 	err = jbd2_journal_get_create_access(handle, bh);
323 	if (err) {
324 		ext4_journal_abort_handle(where, line, __func__, bh, handle,
325 					  err);
326 		return err;
327 	}
328 	if (trigger_type == EXT4_JTR_NONE || !ext4_has_metadata_csum(sb))
329 		return 0;
330 	BUG_ON(trigger_type >= EXT4_JOURNAL_TRIGGER_COUNT);
331 	jbd2_journal_set_triggers(bh,
332 		&EXT4_SB(sb)->s_journal_triggers[trigger_type].tr_triggers);
333 	return 0;
334 }
335 
336 int __ext4_handle_dirty_metadata(const char *where, unsigned int line,
337 				 handle_t *handle, struct inode *inode,
338 				 struct buffer_head *bh)
339 {
340 	int err = 0;
341 
342 	might_sleep();
343 
344 	set_buffer_meta(bh);
345 	set_buffer_prio(bh);
346 	set_buffer_uptodate(bh);
347 	if (ext4_handle_valid(handle)) {
348 		err = jbd2_journal_dirty_metadata(handle, bh);
349 		/* Errors can only happen due to aborted journal or a nasty bug */
350 		if (!is_handle_aborted(handle) && WARN_ON_ONCE(err)) {
351 			ext4_journal_abort_handle(where, line, __func__, bh,
352 						  handle, err);
353 			if (inode == NULL) {
354 				pr_err("EXT4: jbd2_journal_dirty_metadata "
355 				       "failed: handle type %u started at "
356 				       "line %u, credits %u/%u, errcode %d",
357 				       handle->h_type,
358 				       handle->h_line_no,
359 				       handle->h_requested_credits,
360 				       jbd2_handle_buffer_credits(handle), err);
361 				return err;
362 			}
363 			ext4_error_inode(inode, where, line,
364 					 bh->b_blocknr,
365 					 "journal_dirty_metadata failed: "
366 					 "handle type %u started at line %u, "
367 					 "credits %u/%u, errcode %d",
368 					 handle->h_type,
369 					 handle->h_line_no,
370 					 handle->h_requested_credits,
371 					 jbd2_handle_buffer_credits(handle),
372 					 err);
373 		}
374 	} else {
375 		if (inode)
376 			mark_buffer_dirty_inode(bh, inode);
377 		else
378 			mark_buffer_dirty(bh);
379 		if (inode && inode_needs_sync(inode)) {
380 			sync_dirty_buffer(bh);
381 			if (buffer_req(bh) && !buffer_uptodate(bh)) {
382 				ext4_error_inode_err(inode, where, line,
383 						     bh->b_blocknr, EIO,
384 					"IO error syncing itable block");
385 				err = -EIO;
386 			}
387 		}
388 	}
389 	return err;
390 }
391