xref: /linux/fs/bcachefs/error.c (revision 24168c5e6dfbdd5b414f048f47f75d64533296ca)
1 // SPDX-License-Identifier: GPL-2.0
2 #include "bcachefs.h"
3 #include "error.h"
4 #include "journal.h"
5 #include "recovery_passes.h"
6 #include "super.h"
7 #include "thread_with_file.h"
8 
9 #define FSCK_ERR_RATELIMIT_NR	10
10 
11 bool bch2_inconsistent_error(struct bch_fs *c)
12 {
13 	set_bit(BCH_FS_error, &c->flags);
14 
15 	switch (c->opts.errors) {
16 	case BCH_ON_ERROR_continue:
17 		return false;
18 	case BCH_ON_ERROR_ro:
19 		if (bch2_fs_emergency_read_only(c))
20 			bch_err(c, "inconsistency detected - emergency read only at journal seq %llu",
21 				journal_cur_seq(&c->journal));
22 		return true;
23 	case BCH_ON_ERROR_panic:
24 		panic(bch2_fmt(c, "panic after error"));
25 		return true;
26 	default:
27 		BUG();
28 	}
29 }
30 
31 int bch2_topology_error(struct bch_fs *c)
32 {
33 	set_bit(BCH_FS_topology_error, &c->flags);
34 	if (!test_bit(BCH_FS_fsck_running, &c->flags)) {
35 		bch2_inconsistent_error(c);
36 		return -BCH_ERR_btree_need_topology_repair;
37 	} else {
38 		return bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_topology) ?:
39 			-BCH_ERR_btree_node_read_validate_error;
40 	}
41 }
42 
43 void bch2_fatal_error(struct bch_fs *c)
44 {
45 	if (bch2_fs_emergency_read_only(c))
46 		bch_err(c, "fatal error - emergency read only");
47 }
48 
49 void bch2_io_error_work(struct work_struct *work)
50 {
51 	struct bch_dev *ca = container_of(work, struct bch_dev, io_error_work);
52 	struct bch_fs *c = ca->fs;
53 	bool dev;
54 
55 	down_write(&c->state_lock);
56 	dev = bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_ro,
57 				    BCH_FORCE_IF_DEGRADED);
58 	if (dev
59 	    ? __bch2_dev_set_state(c, ca, BCH_MEMBER_STATE_ro,
60 				  BCH_FORCE_IF_DEGRADED)
61 	    : bch2_fs_emergency_read_only(c))
62 		bch_err(ca,
63 			"too many IO errors, setting %s RO",
64 			dev ? "device" : "filesystem");
65 	up_write(&c->state_lock);
66 }
67 
68 void bch2_io_error(struct bch_dev *ca, enum bch_member_error_type type)
69 {
70 	atomic64_inc(&ca->errors[type]);
71 	//queue_work(system_long_wq, &ca->io_error_work);
72 }
73 
74 enum ask_yn {
75 	YN_NO,
76 	YN_YES,
77 	YN_ALLNO,
78 	YN_ALLYES,
79 };
80 
81 static enum ask_yn parse_yn_response(char *buf)
82 {
83 	buf = strim(buf);
84 
85 	if (strlen(buf) == 1)
86 		switch (buf[0]) {
87 		case 'n':
88 			return YN_NO;
89 		case 'y':
90 			return YN_YES;
91 		case 'N':
92 			return YN_ALLNO;
93 		case 'Y':
94 			return YN_ALLYES;
95 		}
96 	return -1;
97 }
98 
99 #ifdef __KERNEL__
100 static enum ask_yn bch2_fsck_ask_yn(struct bch_fs *c)
101 {
102 	struct stdio_redirect *stdio = c->stdio;
103 
104 	if (c->stdio_filter && c->stdio_filter != current)
105 		stdio = NULL;
106 
107 	if (!stdio)
108 		return YN_NO;
109 
110 	char buf[100];
111 	int ret;
112 
113 	do {
114 		bch2_print(c, " (y,n, or Y,N for all errors of this type) ");
115 
116 		int r = bch2_stdio_redirect_readline(stdio, buf, sizeof(buf) - 1);
117 		if (r < 0)
118 			return YN_NO;
119 		buf[r] = '\0';
120 	} while ((ret = parse_yn_response(buf)) < 0);
121 
122 	return ret;
123 }
124 #else
125 
126 #include "tools-util.h"
127 
128 static enum ask_yn bch2_fsck_ask_yn(struct bch_fs *c)
129 {
130 	char *buf = NULL;
131 	size_t buflen = 0;
132 	int ret;
133 
134 	do {
135 		fputs(" (y,n, or Y,N for all errors of this type) ", stdout);
136 		fflush(stdout);
137 
138 		if (getline(&buf, &buflen, stdin) < 0)
139 			die("error reading from standard input");
140 	} while ((ret = parse_yn_response(buf)) < 0);
141 
142 	free(buf);
143 	return ret;
144 }
145 
146 #endif
147 
148 static struct fsck_err_state *fsck_err_get(struct bch_fs *c, const char *fmt)
149 {
150 	struct fsck_err_state *s;
151 
152 	if (!test_bit(BCH_FS_fsck_running, &c->flags))
153 		return NULL;
154 
155 	list_for_each_entry(s, &c->fsck_error_msgs, list)
156 		if (s->fmt == fmt) {
157 			/*
158 			 * move it to the head of the list: repeated fsck errors
159 			 * are common
160 			 */
161 			list_move(&s->list, &c->fsck_error_msgs);
162 			return s;
163 		}
164 
165 	s = kzalloc(sizeof(*s), GFP_NOFS);
166 	if (!s) {
167 		if (!c->fsck_alloc_msgs_err)
168 			bch_err(c, "kmalloc err, cannot ratelimit fsck errs");
169 		c->fsck_alloc_msgs_err = true;
170 		return NULL;
171 	}
172 
173 	INIT_LIST_HEAD(&s->list);
174 	s->fmt = fmt;
175 	list_add(&s->list, &c->fsck_error_msgs);
176 	return s;
177 }
178 
179 /* s/fix?/fixing/ s/recreate?/recreating/ */
180 static void prt_actioning(struct printbuf *out, const char *action)
181 {
182 	unsigned len = strlen(action);
183 
184 	BUG_ON(action[len - 1] != '?');
185 	--len;
186 
187 	if (action[len - 1] == 'e')
188 		--len;
189 
190 	prt_bytes(out, action, len);
191 	prt_str(out, "ing");
192 }
193 
194 int bch2_fsck_err(struct bch_fs *c,
195 		  enum bch_fsck_flags flags,
196 		  enum bch_sb_error_id err,
197 		  const char *fmt, ...)
198 {
199 	struct fsck_err_state *s = NULL;
200 	va_list args;
201 	bool print = true, suppressing = false, inconsistent = false;
202 	struct printbuf buf = PRINTBUF, *out = &buf;
203 	int ret = -BCH_ERR_fsck_ignore;
204 	const char *action_orig = "fix?", *action = action_orig;
205 
206 	if ((flags & FSCK_CAN_FIX) &&
207 	    test_bit(err, c->sb.errors_silent))
208 		return -BCH_ERR_fsck_fix;
209 
210 	bch2_sb_error_count(c, err);
211 
212 	va_start(args, fmt);
213 	prt_vprintf(out, fmt, args);
214 	va_end(args);
215 
216 	/* Custom fix/continue/recreate/etc.? */
217 	if (out->buf[out->pos - 1] == '?') {
218 		const char *p = strrchr(out->buf, ',');
219 		if (p) {
220 			out->pos = p - out->buf;
221 			action = kstrdup(p + 2, GFP_KERNEL);
222 			if (!action) {
223 				ret = -ENOMEM;
224 				goto err;
225 			}
226 		}
227 	}
228 
229 	mutex_lock(&c->fsck_error_msgs_lock);
230 	s = fsck_err_get(c, fmt);
231 	if (s) {
232 		/*
233 		 * We may be called multiple times for the same error on
234 		 * transaction restart - this memoizes instead of asking the user
235 		 * multiple times for the same error:
236 		 */
237 		if (s->last_msg && !strcmp(buf.buf, s->last_msg)) {
238 			ret = s->ret;
239 			mutex_unlock(&c->fsck_error_msgs_lock);
240 			goto err;
241 		}
242 
243 		kfree(s->last_msg);
244 		s->last_msg = kstrdup(buf.buf, GFP_KERNEL);
245 		if (!s->last_msg) {
246 			mutex_unlock(&c->fsck_error_msgs_lock);
247 			ret = -ENOMEM;
248 			goto err;
249 		}
250 
251 		if (c->opts.ratelimit_errors &&
252 		    !(flags & FSCK_NO_RATELIMIT) &&
253 		    s->nr >= FSCK_ERR_RATELIMIT_NR) {
254 			if (s->nr == FSCK_ERR_RATELIMIT_NR)
255 				suppressing = true;
256 			else
257 				print = false;
258 		}
259 
260 		s->nr++;
261 	}
262 
263 #ifdef BCACHEFS_LOG_PREFIX
264 	if (!strncmp(fmt, "bcachefs:", 9))
265 		prt_printf(out, bch2_log_msg(c, ""));
266 #endif
267 
268 	if (!test_bit(BCH_FS_fsck_running, &c->flags)) {
269 		if (c->opts.errors != BCH_ON_ERROR_continue ||
270 		    !(flags & (FSCK_CAN_FIX|FSCK_CAN_IGNORE))) {
271 			prt_str(out, ", shutting down");
272 			inconsistent = true;
273 			ret = -BCH_ERR_fsck_errors_not_fixed;
274 		} else if (flags & FSCK_CAN_FIX) {
275 			prt_str(out, ", ");
276 			prt_actioning(out, action);
277 			ret = -BCH_ERR_fsck_fix;
278 		} else {
279 			prt_str(out, ", continuing");
280 			ret = -BCH_ERR_fsck_ignore;
281 		}
282 	} else if (c->opts.fix_errors == FSCK_FIX_exit) {
283 		prt_str(out, ", exiting");
284 		ret = -BCH_ERR_fsck_errors_not_fixed;
285 	} else if (flags & FSCK_CAN_FIX) {
286 		int fix = s && s->fix
287 			? s->fix
288 			: c->opts.fix_errors;
289 
290 		if (fix == FSCK_FIX_ask) {
291 			prt_str(out, ", ");
292 			prt_str(out, action);
293 
294 			if (bch2_fs_stdio_redirect(c))
295 				bch2_print(c, "%s", out->buf);
296 			else
297 				bch2_print_string_as_lines(KERN_ERR, out->buf);
298 			print = false;
299 
300 			int ask = bch2_fsck_ask_yn(c);
301 
302 			if (ask >= YN_ALLNO && s)
303 				s->fix = ask == YN_ALLNO
304 					? FSCK_FIX_no
305 					: FSCK_FIX_yes;
306 
307 			ret = ask & 1
308 				? -BCH_ERR_fsck_fix
309 				: -BCH_ERR_fsck_ignore;
310 		} else if (fix == FSCK_FIX_yes ||
311 			   (c->opts.nochanges &&
312 			    !(flags & FSCK_CAN_IGNORE))) {
313 			prt_str(out, ", ");
314 			prt_actioning(out, action);
315 			ret = -BCH_ERR_fsck_fix;
316 		} else {
317 			prt_str(out, ", not ");
318 			prt_actioning(out, action);
319 		}
320 	} else if (flags & FSCK_NEED_FSCK) {
321 		prt_str(out, " (run fsck to correct)");
322 	} else {
323 		prt_str(out, " (repair unimplemented)");
324 	}
325 
326 	if (ret == -BCH_ERR_fsck_ignore &&
327 	    (c->opts.fix_errors == FSCK_FIX_exit ||
328 	     !(flags & FSCK_CAN_IGNORE)))
329 		ret = -BCH_ERR_fsck_errors_not_fixed;
330 
331 	if (print) {
332 		if (bch2_fs_stdio_redirect(c))
333 			bch2_print(c, "%s\n", out->buf);
334 		else
335 			bch2_print_string_as_lines(KERN_ERR, out->buf);
336 	}
337 
338 	if (test_bit(BCH_FS_fsck_running, &c->flags) &&
339 	    (ret != -BCH_ERR_fsck_fix &&
340 	     ret != -BCH_ERR_fsck_ignore))
341 		bch_err(c, "Unable to continue, halting");
342 	else if (suppressing)
343 		bch_err(c, "Ratelimiting new instances of previous error");
344 
345 	if (s)
346 		s->ret = ret;
347 
348 	mutex_unlock(&c->fsck_error_msgs_lock);
349 
350 	if (inconsistent)
351 		bch2_inconsistent_error(c);
352 
353 	if (ret == -BCH_ERR_fsck_fix) {
354 		set_bit(BCH_FS_errors_fixed, &c->flags);
355 	} else {
356 		set_bit(BCH_FS_errors_not_fixed, &c->flags);
357 		set_bit(BCH_FS_error, &c->flags);
358 	}
359 err:
360 	if (action != action_orig)
361 		kfree(action);
362 	printbuf_exit(&buf);
363 	return ret;
364 }
365 
366 void bch2_flush_fsck_errs(struct bch_fs *c)
367 {
368 	struct fsck_err_state *s, *n;
369 
370 	mutex_lock(&c->fsck_error_msgs_lock);
371 
372 	list_for_each_entry_safe(s, n, &c->fsck_error_msgs, list) {
373 		if (s->ratelimited && s->last_msg)
374 			bch_err(c, "Saw %llu errors like:\n    %s", s->nr, s->last_msg);
375 
376 		list_del(&s->list);
377 		kfree(s->last_msg);
378 		kfree(s);
379 	}
380 
381 	mutex_unlock(&c->fsck_error_msgs_lock);
382 }
383