xref: /linux/fs/bcachefs/error.c (revision fa8a4d3659d0c1ad73d5f59b2e0a6d408de5b317)
1 // SPDX-License-Identifier: GPL-2.0
2 #include "bcachefs.h"
3 #include "btree_iter.h"
4 #include "error.h"
5 #include "journal.h"
6 #include "recovery_passes.h"
7 #include "super.h"
8 #include "thread_with_file.h"
9 
10 #define FSCK_ERR_RATELIMIT_NR	10
11 
12 bool bch2_inconsistent_error(struct bch_fs *c)
13 {
14 	set_bit(BCH_FS_error, &c->flags);
15 
16 	switch (c->opts.errors) {
17 	case BCH_ON_ERROR_continue:
18 		return false;
19 	case BCH_ON_ERROR_fix_safe:
20 	case BCH_ON_ERROR_ro:
21 		if (bch2_fs_emergency_read_only(c))
22 			bch_err(c, "inconsistency detected - emergency read only at journal seq %llu",
23 				journal_cur_seq(&c->journal));
24 		return true;
25 	case BCH_ON_ERROR_panic:
26 		panic(bch2_fmt(c, "panic after error"));
27 		return true;
28 	default:
29 		BUG();
30 	}
31 }
32 
33 int bch2_topology_error(struct bch_fs *c)
34 {
35 	set_bit(BCH_FS_topology_error, &c->flags);
36 	if (!test_bit(BCH_FS_fsck_running, &c->flags)) {
37 		bch2_inconsistent_error(c);
38 		return -BCH_ERR_btree_need_topology_repair;
39 	} else {
40 		return bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_topology) ?:
41 			-BCH_ERR_btree_node_read_validate_error;
42 	}
43 }
44 
45 void bch2_fatal_error(struct bch_fs *c)
46 {
47 	if (bch2_fs_emergency_read_only(c))
48 		bch_err(c, "fatal error - emergency read only");
49 }
50 
51 void bch2_io_error_work(struct work_struct *work)
52 {
53 	struct bch_dev *ca = container_of(work, struct bch_dev, io_error_work);
54 	struct bch_fs *c = ca->fs;
55 	bool dev;
56 
57 	down_write(&c->state_lock);
58 	dev = bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_ro,
59 				    BCH_FORCE_IF_DEGRADED);
60 	if (dev
61 	    ? __bch2_dev_set_state(c, ca, BCH_MEMBER_STATE_ro,
62 				  BCH_FORCE_IF_DEGRADED)
63 	    : bch2_fs_emergency_read_only(c))
64 		bch_err(ca,
65 			"too many IO errors, setting %s RO",
66 			dev ? "device" : "filesystem");
67 	up_write(&c->state_lock);
68 }
69 
70 void bch2_io_error(struct bch_dev *ca, enum bch_member_error_type type)
71 {
72 	atomic64_inc(&ca->errors[type]);
73 	//queue_work(system_long_wq, &ca->io_error_work);
74 }
75 
76 enum ask_yn {
77 	YN_NO,
78 	YN_YES,
79 	YN_ALLNO,
80 	YN_ALLYES,
81 };
82 
83 static enum ask_yn parse_yn_response(char *buf)
84 {
85 	buf = strim(buf);
86 
87 	if (strlen(buf) == 1)
88 		switch (buf[0]) {
89 		case 'n':
90 			return YN_NO;
91 		case 'y':
92 			return YN_YES;
93 		case 'N':
94 			return YN_ALLNO;
95 		case 'Y':
96 			return YN_ALLYES;
97 		}
98 	return -1;
99 }
100 
101 #ifdef __KERNEL__
102 static enum ask_yn bch2_fsck_ask_yn(struct bch_fs *c, struct btree_trans *trans)
103 {
104 	struct stdio_redirect *stdio = c->stdio;
105 
106 	if (c->stdio_filter && c->stdio_filter != current)
107 		stdio = NULL;
108 
109 	if (!stdio)
110 		return YN_NO;
111 
112 	if (trans)
113 		bch2_trans_unlock(trans);
114 
115 	unsigned long unlock_long_at = trans ? jiffies + HZ * 2 : 0;
116 	darray_char line = {};
117 	int ret;
118 
119 	do {
120 		unsigned long t;
121 		bch2_print(c, " (y,n, or Y,N for all errors of this type) ");
122 rewait:
123 		t = unlock_long_at
124 			? max_t(long, unlock_long_at - jiffies, 0)
125 			: MAX_SCHEDULE_TIMEOUT;
126 
127 		int r = bch2_stdio_redirect_readline_timeout(stdio, &line, t);
128 		if (r == -ETIME) {
129 			bch2_trans_unlock_long(trans);
130 			unlock_long_at = 0;
131 			goto rewait;
132 		}
133 
134 		if (r < 0) {
135 			ret = YN_NO;
136 			break;
137 		}
138 
139 		darray_last(line) = '\0';
140 	} while ((ret = parse_yn_response(line.data)) < 0);
141 
142 	darray_exit(&line);
143 	return ret;
144 }
145 #else
146 
147 #include "tools-util.h"
148 
149 static enum ask_yn bch2_fsck_ask_yn(struct bch_fs *c, struct btree_trans *trans)
150 {
151 	char *buf = NULL;
152 	size_t buflen = 0;
153 	int ret;
154 
155 	do {
156 		fputs(" (y,n, or Y,N for all errors of this type) ", stdout);
157 		fflush(stdout);
158 
159 		if (getline(&buf, &buflen, stdin) < 0)
160 			die("error reading from standard input");
161 	} while ((ret = parse_yn_response(buf)) < 0);
162 
163 	free(buf);
164 	return ret;
165 }
166 
167 #endif
168 
169 static struct fsck_err_state *fsck_err_get(struct bch_fs *c, const char *fmt)
170 {
171 	struct fsck_err_state *s;
172 
173 	if (!test_bit(BCH_FS_fsck_running, &c->flags))
174 		return NULL;
175 
176 	list_for_each_entry(s, &c->fsck_error_msgs, list)
177 		if (s->fmt == fmt) {
178 			/*
179 			 * move it to the head of the list: repeated fsck errors
180 			 * are common
181 			 */
182 			list_move(&s->list, &c->fsck_error_msgs);
183 			return s;
184 		}
185 
186 	s = kzalloc(sizeof(*s), GFP_NOFS);
187 	if (!s) {
188 		if (!c->fsck_alloc_msgs_err)
189 			bch_err(c, "kmalloc err, cannot ratelimit fsck errs");
190 		c->fsck_alloc_msgs_err = true;
191 		return NULL;
192 	}
193 
194 	INIT_LIST_HEAD(&s->list);
195 	s->fmt = fmt;
196 	list_add(&s->list, &c->fsck_error_msgs);
197 	return s;
198 }
199 
200 /* s/fix?/fixing/ s/recreate?/recreating/ */
201 static void prt_actioning(struct printbuf *out, const char *action)
202 {
203 	unsigned len = strlen(action);
204 
205 	BUG_ON(action[len - 1] != '?');
206 	--len;
207 
208 	if (action[len - 1] == 'e')
209 		--len;
210 
211 	prt_bytes(out, action, len);
212 	prt_str(out, "ing");
213 }
214 
215 static const u8 fsck_flags_extra[] = {
216 #define x(t, n, flags)		[BCH_FSCK_ERR_##t] = flags,
217 	BCH_SB_ERRS()
218 #undef x
219 };
220 
221 int __bch2_fsck_err(struct bch_fs *c,
222 		  struct btree_trans *trans,
223 		  enum bch_fsck_flags flags,
224 		  enum bch_sb_error_id err,
225 		  const char *fmt, ...)
226 {
227 	struct fsck_err_state *s = NULL;
228 	va_list args;
229 	bool print = true, suppressing = false, inconsistent = false;
230 	struct printbuf buf = PRINTBUF, *out = &buf;
231 	int ret = -BCH_ERR_fsck_ignore;
232 	const char *action_orig = "fix?", *action = action_orig;
233 
234 	might_sleep();
235 
236 	if (!WARN_ON(err >= ARRAY_SIZE(fsck_flags_extra)))
237 		flags |= fsck_flags_extra[err];
238 
239 	if (!c)
240 		c = trans->c;
241 
242 	WARN_ON(!trans && bch2_current_has_btree_trans(c));
243 
244 	if ((flags & FSCK_CAN_FIX) &&
245 	    test_bit(err, c->sb.errors_silent))
246 		return -BCH_ERR_fsck_fix;
247 
248 	bch2_sb_error_count(c, err);
249 
250 	va_start(args, fmt);
251 	prt_vprintf(out, fmt, args);
252 	va_end(args);
253 
254 	/* Custom fix/continue/recreate/etc.? */
255 	if (out->buf[out->pos - 1] == '?') {
256 		const char *p = strrchr(out->buf, ',');
257 		if (p) {
258 			out->pos = p - out->buf;
259 			action = kstrdup(p + 2, GFP_KERNEL);
260 			if (!action) {
261 				ret = -ENOMEM;
262 				goto err;
263 			}
264 		}
265 	}
266 
267 	mutex_lock(&c->fsck_error_msgs_lock);
268 	s = fsck_err_get(c, fmt);
269 	if (s) {
270 		/*
271 		 * We may be called multiple times for the same error on
272 		 * transaction restart - this memoizes instead of asking the user
273 		 * multiple times for the same error:
274 		 */
275 		if (s->last_msg && !strcmp(buf.buf, s->last_msg)) {
276 			ret = s->ret;
277 			mutex_unlock(&c->fsck_error_msgs_lock);
278 			goto err;
279 		}
280 
281 		kfree(s->last_msg);
282 		s->last_msg = kstrdup(buf.buf, GFP_KERNEL);
283 		if (!s->last_msg) {
284 			mutex_unlock(&c->fsck_error_msgs_lock);
285 			ret = -ENOMEM;
286 			goto err;
287 		}
288 
289 		if (c->opts.ratelimit_errors &&
290 		    !(flags & FSCK_NO_RATELIMIT) &&
291 		    s->nr >= FSCK_ERR_RATELIMIT_NR) {
292 			if (s->nr == FSCK_ERR_RATELIMIT_NR)
293 				suppressing = true;
294 			else
295 				print = false;
296 		}
297 
298 		s->nr++;
299 	}
300 
301 #ifdef BCACHEFS_LOG_PREFIX
302 	if (!strncmp(fmt, "bcachefs:", 9))
303 		prt_printf(out, bch2_log_msg(c, ""));
304 #endif
305 
306 	if ((flags & FSCK_CAN_FIX) &&
307 	    (flags & FSCK_AUTOFIX) &&
308 	    (c->opts.errors == BCH_ON_ERROR_continue ||
309 	     c->opts.errors == BCH_ON_ERROR_fix_safe)) {
310 		prt_str(out, ", ");
311 		prt_actioning(out, action);
312 		ret = -BCH_ERR_fsck_fix;
313 	} else if (!test_bit(BCH_FS_fsck_running, &c->flags)) {
314 		if (c->opts.errors != BCH_ON_ERROR_continue ||
315 		    !(flags & (FSCK_CAN_FIX|FSCK_CAN_IGNORE))) {
316 			prt_str(out, ", shutting down");
317 			inconsistent = true;
318 			ret = -BCH_ERR_fsck_errors_not_fixed;
319 		} else if (flags & FSCK_CAN_FIX) {
320 			prt_str(out, ", ");
321 			prt_actioning(out, action);
322 			ret = -BCH_ERR_fsck_fix;
323 		} else {
324 			prt_str(out, ", continuing");
325 			ret = -BCH_ERR_fsck_ignore;
326 		}
327 	} else if (c->opts.fix_errors == FSCK_FIX_exit) {
328 		prt_str(out, ", exiting");
329 		ret = -BCH_ERR_fsck_errors_not_fixed;
330 	} else if (flags & FSCK_CAN_FIX) {
331 		int fix = s && s->fix
332 			? s->fix
333 			: c->opts.fix_errors;
334 
335 		if (fix == FSCK_FIX_ask) {
336 			prt_str(out, ", ");
337 			prt_str(out, action);
338 
339 			if (bch2_fs_stdio_redirect(c))
340 				bch2_print(c, "%s", out->buf);
341 			else
342 				bch2_print_string_as_lines(KERN_ERR, out->buf);
343 			print = false;
344 
345 			int ask = bch2_fsck_ask_yn(c, trans);
346 
347 			if (trans) {
348 				ret = bch2_trans_relock(trans);
349 				if (ret) {
350 					mutex_unlock(&c->fsck_error_msgs_lock);
351 					goto err;
352 				}
353 			}
354 
355 			if (ask >= YN_ALLNO && s)
356 				s->fix = ask == YN_ALLNO
357 					? FSCK_FIX_no
358 					: FSCK_FIX_yes;
359 
360 			ret = ask & 1
361 				? -BCH_ERR_fsck_fix
362 				: -BCH_ERR_fsck_ignore;
363 		} else if (fix == FSCK_FIX_yes ||
364 			   (c->opts.nochanges &&
365 			    !(flags & FSCK_CAN_IGNORE))) {
366 			prt_str(out, ", ");
367 			prt_actioning(out, action);
368 			ret = -BCH_ERR_fsck_fix;
369 		} else {
370 			prt_str(out, ", not ");
371 			prt_actioning(out, action);
372 		}
373 	} else if (flags & FSCK_NEED_FSCK) {
374 		prt_str(out, " (run fsck to correct)");
375 	} else {
376 		prt_str(out, " (repair unimplemented)");
377 	}
378 
379 	if (ret == -BCH_ERR_fsck_ignore &&
380 	    (c->opts.fix_errors == FSCK_FIX_exit ||
381 	     !(flags & FSCK_CAN_IGNORE)))
382 		ret = -BCH_ERR_fsck_errors_not_fixed;
383 
384 	if (print) {
385 		if (bch2_fs_stdio_redirect(c))
386 			bch2_print(c, "%s\n", out->buf);
387 		else
388 			bch2_print_string_as_lines(KERN_ERR, out->buf);
389 	}
390 
391 	if (test_bit(BCH_FS_fsck_running, &c->flags) &&
392 	    (ret != -BCH_ERR_fsck_fix &&
393 	     ret != -BCH_ERR_fsck_ignore))
394 		bch_err(c, "Unable to continue, halting");
395 	else if (suppressing)
396 		bch_err(c, "Ratelimiting new instances of previous error");
397 
398 	if (s)
399 		s->ret = ret;
400 
401 	mutex_unlock(&c->fsck_error_msgs_lock);
402 
403 	if (inconsistent)
404 		bch2_inconsistent_error(c);
405 
406 	if (ret == -BCH_ERR_fsck_fix) {
407 		set_bit(BCH_FS_errors_fixed, &c->flags);
408 	} else {
409 		set_bit(BCH_FS_errors_not_fixed, &c->flags);
410 		set_bit(BCH_FS_error, &c->flags);
411 	}
412 err:
413 	if (action != action_orig)
414 		kfree(action);
415 	printbuf_exit(&buf);
416 	return ret;
417 }
418 
419 int __bch2_bkey_fsck_err(struct bch_fs *c,
420 			 struct bkey_s_c k,
421 			 enum bch_fsck_flags flags,
422 			 enum bch_sb_error_id err,
423 			 const char *fmt, ...)
424 {
425 	struct printbuf buf = PRINTBUF;
426 	va_list args;
427 
428 	prt_str(&buf, "invalid bkey ");
429 	bch2_bkey_val_to_text(&buf, c, k);
430 	prt_str(&buf, "\n  ");
431 	va_start(args, fmt);
432 	prt_vprintf(&buf, fmt, args);
433 	va_end(args);
434 	prt_str(&buf, ": delete?");
435 
436 	int ret = __bch2_fsck_err(c, NULL, flags, err, "%s", buf.buf);
437 	printbuf_exit(&buf);
438 	return ret;
439 }
440 
441 void bch2_flush_fsck_errs(struct bch_fs *c)
442 {
443 	struct fsck_err_state *s, *n;
444 
445 	mutex_lock(&c->fsck_error_msgs_lock);
446 
447 	list_for_each_entry_safe(s, n, &c->fsck_error_msgs, list) {
448 		if (s->ratelimited && s->last_msg)
449 			bch_err(c, "Saw %llu errors like:\n    %s", s->nr, s->last_msg);
450 
451 		list_del(&s->list);
452 		kfree(s->last_msg);
453 		kfree(s);
454 	}
455 
456 	mutex_unlock(&c->fsck_error_msgs_lock);
457 }
458