xref: /openbmc/linux/fs/ext4/ext4_jbd2.c (revision b7b3c35e)
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 (is_handle_aborted(handle))
166 		return -EROFS;
167 	if (jbd2_handle_buffer_credits(handle) >= check_cred &&
168 	    handle->h_revoke_credits >= revoke_cred)
169 		return 0;
170 	extend_cred = max(0, extend_cred - jbd2_handle_buffer_credits(handle));
171 	revoke_cred = max(0, revoke_cred - handle->h_revoke_credits);
172 	return ext4_journal_extend(handle, extend_cred, revoke_cred);
173 }
174 
175 static void ext4_journal_abort_handle(const char *caller, unsigned int line,
176 				      const char *err_fn,
177 				      struct buffer_head *bh,
178 				      handle_t *handle, int err)
179 {
180 	char nbuf[16];
181 	const char *errstr = ext4_decode_error(NULL, err, nbuf);
182 
183 	BUG_ON(!ext4_handle_valid(handle));
184 
185 	if (bh)
186 		BUFFER_TRACE(bh, "abort");
187 
188 	if (!handle->h_err)
189 		handle->h_err = err;
190 
191 	if (is_handle_aborted(handle))
192 		return;
193 
194 	printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
195 	       caller, line, errstr, err_fn);
196 
197 	jbd2_journal_abort_handle(handle);
198 }
199 
200 static void ext4_check_bdev_write_error(struct super_block *sb)
201 {
202 	struct address_space *mapping = sb->s_bdev->bd_inode->i_mapping;
203 	struct ext4_sb_info *sbi = EXT4_SB(sb);
204 	int err;
205 
206 	/*
207 	 * If the block device has write error flag, it may have failed to
208 	 * async write out metadata buffers in the background. In this case,
209 	 * we could read old data from disk and write it out again, which
210 	 * may lead to on-disk filesystem inconsistency.
211 	 */
212 	if (errseq_check(&mapping->wb_err, READ_ONCE(sbi->s_bdev_wb_err))) {
213 		spin_lock(&sbi->s_bdev_wb_lock);
214 		err = errseq_check_and_advance(&mapping->wb_err, &sbi->s_bdev_wb_err);
215 		spin_unlock(&sbi->s_bdev_wb_lock);
216 		if (err)
217 			ext4_error_err(sb, -err,
218 				       "Error while async write back metadata");
219 	}
220 }
221 
222 int __ext4_journal_get_write_access(const char *where, unsigned int line,
223 				    handle_t *handle, struct super_block *sb,
224 				    struct buffer_head *bh,
225 				    enum ext4_journal_trigger_type trigger_type)
226 {
227 	int err;
228 
229 	might_sleep();
230 
231 	if (bh->b_bdev->bd_super)
232 		ext4_check_bdev_write_error(bh->b_bdev->bd_super);
233 
234 	if (ext4_handle_valid(handle)) {
235 		err = jbd2_journal_get_write_access(handle, bh);
236 		if (err) {
237 			ext4_journal_abort_handle(where, line, __func__, bh,
238 						  handle, err);
239 			return err;
240 		}
241 	}
242 	if (trigger_type == EXT4_JTR_NONE || !ext4_has_metadata_csum(sb))
243 		return 0;
244 	BUG_ON(trigger_type >= EXT4_JOURNAL_TRIGGER_COUNT);
245 	jbd2_journal_set_triggers(bh,
246 		&EXT4_SB(sb)->s_journal_triggers[trigger_type].tr_triggers);
247 	return 0;
248 }
249 
250 /*
251  * The ext4 forget function must perform a revoke if we are freeing data
252  * which has been journaled.  Metadata (eg. indirect blocks) must be
253  * revoked in all cases.
254  *
255  * "bh" may be NULL: a metadata block may have been freed from memory
256  * but there may still be a record of it in the journal, and that record
257  * still needs to be revoked.
258  */
259 int __ext4_forget(const char *where, unsigned int line, handle_t *handle,
260 		  int is_metadata, struct inode *inode,
261 		  struct buffer_head *bh, ext4_fsblk_t blocknr)
262 {
263 	int err;
264 
265 	might_sleep();
266 
267 	trace_ext4_forget(inode, is_metadata, blocknr);
268 	BUFFER_TRACE(bh, "enter");
269 
270 	jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
271 		  "data mode %x\n",
272 		  bh, is_metadata, inode->i_mode,
273 		  test_opt(inode->i_sb, DATA_FLAGS));
274 
275 	/* In the no journal case, we can just do a bforget and return */
276 	if (!ext4_handle_valid(handle)) {
277 		bforget(bh);
278 		return 0;
279 	}
280 
281 	/* Never use the revoke function if we are doing full data
282 	 * journaling: there is no need to, and a V1 superblock won't
283 	 * support it.  Otherwise, only skip the revoke on un-journaled
284 	 * data blocks. */
285 
286 	if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
287 	    (!is_metadata && !ext4_should_journal_data(inode))) {
288 		if (bh) {
289 			BUFFER_TRACE(bh, "call jbd2_journal_forget");
290 			err = jbd2_journal_forget(handle, bh);
291 			if (err)
292 				ext4_journal_abort_handle(where, line, __func__,
293 							  bh, handle, err);
294 			return err;
295 		}
296 		return 0;
297 	}
298 
299 	/*
300 	 * data!=journal && (is_metadata || should_journal_data(inode))
301 	 */
302 	BUFFER_TRACE(bh, "call jbd2_journal_revoke");
303 	err = jbd2_journal_revoke(handle, blocknr, bh);
304 	if (err) {
305 		ext4_journal_abort_handle(where, line, __func__,
306 					  bh, handle, err);
307 		__ext4_error(inode->i_sb, where, line, true, -err, 0,
308 			     "error %d when attempting revoke", err);
309 	}
310 	BUFFER_TRACE(bh, "exit");
311 	return err;
312 }
313 
314 int __ext4_journal_get_create_access(const char *where, unsigned int line,
315 				handle_t *handle, struct super_block *sb,
316 				struct buffer_head *bh,
317 				enum ext4_journal_trigger_type trigger_type)
318 {
319 	int err;
320 
321 	if (!ext4_handle_valid(handle))
322 		return 0;
323 
324 	err = jbd2_journal_get_create_access(handle, bh);
325 	if (err) {
326 		ext4_journal_abort_handle(where, line, __func__, bh, handle,
327 					  err);
328 		return err;
329 	}
330 	if (trigger_type == EXT4_JTR_NONE || !ext4_has_metadata_csum(sb))
331 		return 0;
332 	BUG_ON(trigger_type >= EXT4_JOURNAL_TRIGGER_COUNT);
333 	jbd2_journal_set_triggers(bh,
334 		&EXT4_SB(sb)->s_journal_triggers[trigger_type].tr_triggers);
335 	return 0;
336 }
337 
338 int __ext4_handle_dirty_metadata(const char *where, unsigned int line,
339 				 handle_t *handle, struct inode *inode,
340 				 struct buffer_head *bh)
341 {
342 	int err = 0;
343 
344 	might_sleep();
345 
346 	set_buffer_meta(bh);
347 	set_buffer_prio(bh);
348 	set_buffer_uptodate(bh);
349 	if (ext4_handle_valid(handle)) {
350 		err = jbd2_journal_dirty_metadata(handle, bh);
351 		/* Errors can only happen due to aborted journal or a nasty bug */
352 		if (!is_handle_aborted(handle) && WARN_ON_ONCE(err)) {
353 			ext4_journal_abort_handle(where, line, __func__, bh,
354 						  handle, err);
355 			if (inode == NULL) {
356 				pr_err("EXT4: jbd2_journal_dirty_metadata "
357 				       "failed: handle type %u started at "
358 				       "line %u, credits %u/%u, errcode %d",
359 				       handle->h_type,
360 				       handle->h_line_no,
361 				       handle->h_requested_credits,
362 				       jbd2_handle_buffer_credits(handle), err);
363 				return err;
364 			}
365 			ext4_error_inode(inode, where, line,
366 					 bh->b_blocknr,
367 					 "journal_dirty_metadata failed: "
368 					 "handle type %u started at line %u, "
369 					 "credits %u/%u, errcode %d",
370 					 handle->h_type,
371 					 handle->h_line_no,
372 					 handle->h_requested_credits,
373 					 jbd2_handle_buffer_credits(handle),
374 					 err);
375 		}
376 	} else {
377 		if (inode)
378 			mark_buffer_dirty_inode(bh, inode);
379 		else
380 			mark_buffer_dirty(bh);
381 		if (inode && inode_needs_sync(inode)) {
382 			sync_dirty_buffer(bh);
383 			if (buffer_req(bh) && !buffer_uptodate(bh)) {
384 				ext4_error_inode_err(inode, where, line,
385 						     bh->b_blocknr, EIO,
386 					"IO error syncing itable block");
387 				err = -EIO;
388 			}
389 		}
390 	}
391 	return err;
392 }
393