xref: /linux/fs/ext4/ext4_jbd2.c (revision c84d3e3130dfe1058cb27dc78e7ad8bd36f0545a)
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 
ext4_inode_journal_mode(struct inode * inode)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 */
ext4_get_nojournal(void)33 static handle_t *ext4_get_nojournal(void)
34 {
35 	handle_t *handle = current->journal_info;
36 
37 	BUG_ON(handle && !handle->h_invalid);
38 
39 	if (!handle) {
40 		handle = jbd2_alloc_handle(GFP_NOFS);
41 		if (!handle)
42 			return ERR_PTR(-ENOMEM);
43 		handle->h_invalid = 1;
44 		/*
45 		 * This is done by start_this_handle() if journalling
46 		 * is enabled.
47 		 */
48 		handle->saved_alloc_context = memalloc_nofs_save();
49 		current->journal_info = handle;
50 	}
51 	handle->h_ref++;
52 	return handle;
53 }
54 
55 
56 /* Decrement the non-pointer handle value */
ext4_put_nojournal(handle_t * handle)57 static void ext4_put_nojournal(handle_t *handle)
58 {
59 	BUG_ON(handle->h_ref == 0);
60 
61 	handle->h_ref--;
62 	if (handle->h_ref == 0) {
63 		memalloc_nofs_restore(handle->saved_alloc_context);
64 		jbd2_free_handle(handle);
65 		current->journal_info = NULL;
66 	}
67 }
68 
69 /*
70  * Wrappers for jbd2_journal_start/end.
71  */
ext4_journal_check_start(struct super_block * sb)72 static int ext4_journal_check_start(struct super_block *sb)
73 {
74 	int ret;
75 	journal_t *journal;
76 
77 	might_sleep();
78 
79 	ret = ext4_emergency_state(sb);
80 	if (unlikely(ret))
81 		return ret;
82 
83 	if (WARN_ON_ONCE(sb_rdonly(sb)))
84 		return -EROFS;
85 
86 	WARN_ON(sb->s_writers.frozen == SB_FREEZE_COMPLETE);
87 	journal = EXT4_SB(sb)->s_journal;
88 	/*
89 	 * Special case here: if the journal has aborted behind our
90 	 * backs (eg. EIO in the commit thread), then we still need to
91 	 * take the FS itself readonly cleanly.
92 	 */
93 	if (journal && is_journal_aborted(journal)) {
94 		ext4_abort(sb, -journal->j_errno, "Detected aborted journal");
95 		return -EROFS;
96 	}
97 	return 0;
98 }
99 
__ext4_journal_start_sb(struct inode * inode,struct super_block * sb,unsigned int line,int type,int blocks,int rsv_blocks,int revoke_creds)100 handle_t *__ext4_journal_start_sb(struct inode *inode,
101 				  struct super_block *sb, unsigned int line,
102 				  int type, int blocks, int rsv_blocks,
103 				  int revoke_creds)
104 {
105 	journal_t *journal;
106 	int err;
107 	if (inode)
108 		trace_ext4_journal_start_inode(inode, blocks, rsv_blocks,
109 					revoke_creds, type,
110 					_RET_IP_);
111 	else
112 		trace_ext4_journal_start_sb(sb, blocks, rsv_blocks,
113 					revoke_creds, type,
114 					_RET_IP_);
115 	err = ext4_journal_check_start(sb);
116 	if (err < 0)
117 		return ERR_PTR(err);
118 
119 	journal = EXT4_SB(sb)->s_journal;
120 	if (!journal || (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))
121 		return ext4_get_nojournal();
122 	return jbd2__journal_start(journal, blocks, rsv_blocks, revoke_creds,
123 				   GFP_NOFS, type, line);
124 }
125 
__ext4_journal_stop(const char * where,unsigned int line,handle_t * handle)126 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
127 {
128 	struct super_block *sb;
129 	int err;
130 	int rc;
131 
132 	if (!ext4_handle_valid(handle)) {
133 		ext4_put_nojournal(handle);
134 		return 0;
135 	}
136 
137 	err = handle->h_err;
138 	if (!handle->h_transaction) {
139 		rc = jbd2_journal_stop(handle);
140 		return err ? err : rc;
141 	}
142 
143 	sb = handle->h_transaction->t_journal->j_private;
144 	rc = jbd2_journal_stop(handle);
145 
146 	if (!err)
147 		err = rc;
148 	if (err)
149 		__ext4_std_error(sb, where, line, err);
150 	return err;
151 }
152 
__ext4_journal_start_reserved(handle_t * handle,unsigned int line,int type)153 handle_t *__ext4_journal_start_reserved(handle_t *handle, unsigned int line,
154 					int type)
155 {
156 	struct super_block *sb;
157 	int err;
158 
159 	if (!ext4_handle_valid(handle))
160 		return ext4_get_nojournal();
161 
162 	sb = handle->h_journal->j_private;
163 	trace_ext4_journal_start_reserved(sb,
164 				jbd2_handle_buffer_credits(handle), _RET_IP_);
165 	err = ext4_journal_check_start(sb);
166 	if (err < 0) {
167 		jbd2_journal_free_reserved(handle);
168 		return ERR_PTR(err);
169 	}
170 
171 	err = jbd2_journal_start_reserved(handle, type, line);
172 	if (err < 0)
173 		return ERR_PTR(err);
174 	return handle;
175 }
176 
__ext4_journal_ensure_credits(handle_t * handle,int check_cred,int extend_cred,int revoke_cred)177 int __ext4_journal_ensure_credits(handle_t *handle, int check_cred,
178 				  int extend_cred, int revoke_cred)
179 {
180 	if (!ext4_handle_valid(handle))
181 		return 0;
182 	if (is_handle_aborted(handle))
183 		return -EROFS;
184 	if (jbd2_handle_buffer_credits(handle) >= check_cred &&
185 	    handle->h_revoke_credits >= revoke_cred)
186 		return 0;
187 	extend_cred = max(0, extend_cred - jbd2_handle_buffer_credits(handle));
188 	revoke_cred = max(0, revoke_cred - handle->h_revoke_credits);
189 	return ext4_journal_extend(handle, extend_cred, revoke_cred);
190 }
191 
ext4_journal_abort_handle(const char * caller,unsigned int line,const char * err_fn,struct buffer_head * bh,handle_t * handle,int err)192 static void ext4_journal_abort_handle(const char *caller, unsigned int line,
193 				      const char *err_fn,
194 				      struct buffer_head *bh,
195 				      handle_t *handle, int err)
196 {
197 	char nbuf[16];
198 	const char *errstr = ext4_decode_error(NULL, err, nbuf);
199 
200 	BUG_ON(!ext4_handle_valid(handle));
201 
202 	if (bh)
203 		BUFFER_TRACE(bh, "abort");
204 
205 	if (!handle->h_err)
206 		handle->h_err = err;
207 
208 	if (is_handle_aborted(handle))
209 		return;
210 
211 	printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
212 	       caller, line, errstr, err_fn);
213 
214 	jbd2_journal_abort_handle(handle);
215 }
216 
ext4_check_bdev_write_error(struct super_block * sb)217 static void ext4_check_bdev_write_error(struct super_block *sb)
218 {
219 	struct address_space *mapping = sb->s_bdev->bd_mapping;
220 	struct ext4_sb_info *sbi = EXT4_SB(sb);
221 	int err;
222 
223 	/*
224 	 * If the block device has write error flag, it may have failed to
225 	 * async write out metadata buffers in the background. In this case,
226 	 * we could read old data from disk and write it out again, which
227 	 * may lead to on-disk filesystem inconsistency.
228 	 */
229 	if (errseq_check(&mapping->wb_err, READ_ONCE(sbi->s_bdev_wb_err))) {
230 		spin_lock(&sbi->s_bdev_wb_lock);
231 		err = errseq_check_and_advance(&mapping->wb_err, &sbi->s_bdev_wb_err);
232 		spin_unlock(&sbi->s_bdev_wb_lock);
233 		if (err)
234 			ext4_error_err(sb, -err,
235 				       "Error while async write back metadata");
236 	}
237 }
238 
__ext4_journal_get_write_access(const char * where,unsigned int line,handle_t * handle,struct super_block * sb,struct buffer_head * bh,enum ext4_journal_trigger_type trigger_type)239 int __ext4_journal_get_write_access(const char *where, unsigned int line,
240 				    handle_t *handle, struct super_block *sb,
241 				    struct buffer_head *bh,
242 				    enum ext4_journal_trigger_type trigger_type)
243 {
244 	int err;
245 
246 	might_sleep();
247 
248 	if (ext4_handle_valid(handle)) {
249 		err = jbd2_journal_get_write_access(handle, bh);
250 		if (err) {
251 			ext4_journal_abort_handle(where, line, __func__, bh,
252 						  handle, err);
253 			return err;
254 		}
255 	} else
256 		ext4_check_bdev_write_error(sb);
257 	if (trigger_type == EXT4_JTR_NONE ||
258 	    !ext4_has_feature_metadata_csum(sb))
259 		return 0;
260 	BUG_ON(trigger_type >= EXT4_JOURNAL_TRIGGER_COUNT);
261 	jbd2_journal_set_triggers(bh,
262 		&EXT4_SB(sb)->s_journal_triggers[trigger_type].tr_triggers);
263 	return 0;
264 }
265 
266 /*
267  * The ext4 forget function must perform a revoke if we are freeing data
268  * which has been journaled.  Metadata (eg. indirect blocks) must be
269  * revoked in all cases.
270  *
271  * "bh" may be NULL: a metadata block may have been freed from memory
272  * but there may still be a record of it in the journal, and that record
273  * still needs to be revoked.
274  */
__ext4_forget(const char * where,unsigned int line,handle_t * handle,int is_metadata,struct inode * inode,struct buffer_head * bh,ext4_fsblk_t blocknr)275 int __ext4_forget(const char *where, unsigned int line, handle_t *handle,
276 		  int is_metadata, struct inode *inode,
277 		  struct buffer_head *bh, ext4_fsblk_t blocknr)
278 {
279 	int err;
280 
281 	might_sleep();
282 
283 	trace_ext4_forget(inode, is_metadata, blocknr);
284 	BUFFER_TRACE(bh, "enter");
285 
286 	ext4_debug("forgetting bh %p: is_metadata=%d, mode %o, data mode %x\n",
287 		  bh, is_metadata, inode->i_mode,
288 		  test_opt(inode->i_sb, DATA_FLAGS));
289 
290 	/*
291 	 * In the no journal case, we should wait for the ongoing buffer
292 	 * to complete and do a forget.
293 	 */
294 	if (!ext4_handle_valid(handle)) {
295 		if (bh) {
296 			clear_buffer_dirty(bh);
297 			wait_on_buffer(bh);
298 			__bforget(bh);
299 		}
300 		return 0;
301 	}
302 
303 	/* Never use the revoke function if we are doing full data
304 	 * journaling: there is no need to, and a V1 superblock won't
305 	 * support it.  Otherwise, only skip the revoke on un-journaled
306 	 * data blocks. */
307 
308 	if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
309 	    (!is_metadata && !ext4_should_journal_data(inode))) {
310 		if (bh) {
311 			BUFFER_TRACE(bh, "call jbd2_journal_forget");
312 			err = jbd2_journal_forget(handle, bh);
313 			if (err)
314 				ext4_journal_abort_handle(where, line, __func__,
315 							  bh, handle, err);
316 			return err;
317 		}
318 		return 0;
319 	}
320 
321 	/*
322 	 * data!=journal && (is_metadata || should_journal_data(inode))
323 	 */
324 	BUFFER_TRACE(bh, "call jbd2_journal_revoke");
325 	err = jbd2_journal_revoke(handle, blocknr, bh);
326 	if (err) {
327 		ext4_journal_abort_handle(where, line, __func__,
328 					  bh, handle, err);
329 		__ext4_error(inode->i_sb, where, line, true, -err, 0,
330 			     "error %d when attempting revoke", err);
331 	}
332 	BUFFER_TRACE(bh, "exit");
333 	return err;
334 }
335 
__ext4_journal_get_create_access(const char * where,unsigned int line,handle_t * handle,struct super_block * sb,struct buffer_head * bh,enum ext4_journal_trigger_type trigger_type)336 int __ext4_journal_get_create_access(const char *where, unsigned int line,
337 				handle_t *handle, struct super_block *sb,
338 				struct buffer_head *bh,
339 				enum ext4_journal_trigger_type trigger_type)
340 {
341 	int err;
342 
343 	if (!ext4_handle_valid(handle))
344 		return 0;
345 
346 	err = jbd2_journal_get_create_access(handle, bh);
347 	if (err) {
348 		ext4_journal_abort_handle(where, line, __func__, bh, handle,
349 					  err);
350 		return err;
351 	}
352 	if (trigger_type == EXT4_JTR_NONE ||
353 	    !ext4_has_feature_metadata_csum(sb))
354 		return 0;
355 	BUG_ON(trigger_type >= EXT4_JOURNAL_TRIGGER_COUNT);
356 	jbd2_journal_set_triggers(bh,
357 		&EXT4_SB(sb)->s_journal_triggers[trigger_type].tr_triggers);
358 	return 0;
359 }
360 
ext4_inode_attach_mmb(struct inode * inode)361 static void ext4_inode_attach_mmb(struct inode *inode)
362 {
363 	struct mapping_metadata_bhs *mmb;
364 
365 	/*
366 	 * It's difficult to handle failure when marking buffer dirty without
367 	 * leaving filesystem corrupted
368 	 */
369 	mmb = kmalloc_obj(*mmb, GFP_NOFS | __GFP_NOFAIL | __GFP_ACCOUNT);
370 	mmb_init(mmb, &inode->i_data);
371 	/* Someone swapped another mmb before us? */
372 	if (cmpxchg(&EXT4_I(inode)->i_metadata_bhs, NULL, mmb))
373 		kfree(mmb);
374 }
375 
__ext4_handle_dirty_metadata(const char * where,unsigned int line,handle_t * handle,struct inode * inode,struct buffer_head * bh)376 int __ext4_handle_dirty_metadata(const char *where, unsigned int line,
377 				 handle_t *handle, struct inode *inode,
378 				 struct buffer_head *bh)
379 {
380 	int err = 0;
381 
382 	might_sleep();
383 
384 	set_buffer_meta(bh);
385 	set_buffer_prio(bh);
386 	set_buffer_uptodate(bh);
387 	if (ext4_handle_valid(handle)) {
388 		err = jbd2_journal_dirty_metadata(handle, bh);
389 		/* Errors can only happen due to aborted journal or a nasty bug */
390 		if (!is_handle_aborted(handle) && WARN_ON_ONCE(err)) {
391 			ext4_journal_abort_handle(where, line, __func__, bh,
392 						  handle, err);
393 			if (inode == NULL) {
394 				pr_err("EXT4: jbd2_journal_dirty_metadata "
395 				       "failed: handle type %u started at "
396 				       "line %u, credits %u/%u, errcode %d",
397 				       handle->h_type,
398 				       handle->h_line_no,
399 				       handle->h_requested_credits,
400 				       jbd2_handle_buffer_credits(handle), err);
401 				return err;
402 			}
403 			ext4_error_inode(inode, where, line,
404 					 bh->b_blocknr,
405 					 "journal_dirty_metadata failed: "
406 					 "handle type %u started at line %u, "
407 					 "credits %u/%u, errcode %d",
408 					 handle->h_type,
409 					 handle->h_line_no,
410 					 handle->h_requested_credits,
411 					 jbd2_handle_buffer_credits(handle),
412 					 err);
413 		}
414 	} else {
415 		if (inode) {
416 			if (!ext4_i_metadata_bhs(inode))
417 				ext4_inode_attach_mmb(inode);
418 			mmb_mark_buffer_dirty(bh, ext4_i_metadata_bhs(inode));
419 		} else {
420 			mark_buffer_dirty(bh);
421 		}
422 		if (inode && inode_needs_sync(inode)) {
423 			sync_dirty_buffer(bh);
424 			if (buffer_req(bh) && !buffer_uptodate(bh)) {
425 				ext4_error_inode_err(inode, where, line,
426 						     bh->b_blocknr, EIO,
427 					"IO error syncing itable block");
428 				err = -EIO;
429 			}
430 		}
431 	}
432 	return err;
433 }
434