X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Ferror.h;h=fec17d1353d18042ca77132106d2e2318de2be01;hb=f0334bbc1032b3be485767ec804ec868a8b1e7f0;hp=750c676a65d60c647455ef0753ce888c4d23086b;hpb=819f2dde79241915a6edda2c20bb4ca5d4017030;p=bcachefs-tools-debian

diff --git a/libbcachefs/error.h b/libbcachefs/error.h
index 750c676..fec17d1 100644
--- a/libbcachefs/error.h
+++ b/libbcachefs/error.h
@@ -1,7 +1,10 @@
-#ifndef _BCACHE_ERROR_H
-#define _BCACHE_ERROR_H
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_ERROR_H
+#define _BCACHEFS_ERROR_H
 
+#include <linux/list.h>
 #include <linux/printk.h>
+#include "sb-errors.h"
 
 struct bch_dev;
 struct bch_fs;
@@ -14,26 +17,6 @@ struct work_struct;
 
 /* Error messages: */
 
-/*
- * Very fatal logic/inconsistency errors: these indicate that we've majorly
- * screwed up at runtime, i.e. it's not likely that it was just caused by the
- * data on disk being inconsistent. These BUG():
- *
- * XXX: audit and convert to inconsistent() checks
- */
-
-#define bch2_fs_bug(c, ...)						\
-do {									\
-	bch_err(c, __VA_ARGS__);					\
-	BUG();								\
-} while (0)
-
-#define bch2_fs_bug_on(cond, c, ...)					\
-do {									\
-	if (cond)							\
-		bch2_fs_bug(c, __VA_ARGS__);				\
-} while (0)
-
 /*
  * Inconsistency errors: The on disk data is inconsistent. If these occur during
  * initial recovery, they don't indicate a bug in the running code - we walk all
@@ -45,17 +28,19 @@ do {									\
  * BCH_ON_ERROR_CONTINUE mode
  */
 
-void bch2_inconsistent_error(struct bch_fs *);
+bool bch2_inconsistent_error(struct bch_fs *);
+
+void bch2_topology_error(struct bch_fs *);
 
 #define bch2_fs_inconsistent(c, ...)					\
-do {									\
+({									\
 	bch_err(c, __VA_ARGS__);					\
 	bch2_inconsistent_error(c);					\
-} while (0)
+})
 
 #define bch2_fs_inconsistent_on(cond, c, ...)				\
 ({									\
-	int _ret = !!(cond);						\
+	bool _ret = unlikely(!!(cond));					\
 									\
 	if (_ret)							\
 		bch2_fs_inconsistent(c, __VA_ARGS__);			\
@@ -75,7 +60,7 @@ do {									\
 
 #define bch2_dev_inconsistent_on(cond, ca, ...)				\
 ({									\
-	int _ret = !!(cond);						\
+	bool _ret = unlikely(!!(cond));					\
 									\
 	if (_ret)							\
 		bch2_dev_inconsistent(ca, __VA_ARGS__);			\
@@ -83,79 +68,121 @@ do {									\
 })
 
 /*
- * Fsck errors: inconsistency errors we detect at mount time, and should ideally
- * be able to repair:
+ * When a transaction update discovers or is causing a fs inconsistency, it's
+ * helpful to also dump the pending updates:
  */
+#define bch2_trans_inconsistent(trans, ...)				\
+({									\
+	bch_err(trans->c, __VA_ARGS__);					\
+	bch2_dump_trans_updates(trans);					\
+	bch2_inconsistent_error(trans->c);				\
+})
 
-enum {
-	BCH_FSCK_OK			= 0,
-	BCH_FSCK_ERRORS_NOT_FIXED	= 1,
-	BCH_FSCK_REPAIR_UNIMPLEMENTED	= 2,
-	BCH_FSCK_REPAIR_IMPOSSIBLE	= 3,
-	BCH_FSCK_UNKNOWN_VERSION	= 4,
-};
-
-enum fsck_err_opts {
-	FSCK_ERR_NO,
-	FSCK_ERR_YES,
-	FSCK_ERR_ASK,
-};
+#define bch2_trans_inconsistent_on(cond, trans, ...)			\
+({									\
+	bool _ret = unlikely(!!(cond));					\
+									\
+	if (_ret)							\
+		bch2_trans_inconsistent(trans, __VA_ARGS__);		\
+	_ret;								\
+})
 
-enum fsck_err_ret {
-	FSCK_ERR_IGNORE	= 0,
-	FSCK_ERR_FIX	= 1,
-	FSCK_ERR_EXIT	= 2,
-};
+/*
+ * Fsck errors: inconsistency errors we detect at mount time, and should ideally
+ * be able to repair:
+ */
 
 struct fsck_err_state {
 	struct list_head	list;
 	const char		*fmt;
 	u64			nr;
-	char			buf[512];
+	bool			ratelimited;
+	int			ret;
+	int			fix;
+	char			*last_msg;
+};
+
+enum bch_fsck_flags {
+	FSCK_CAN_FIX		= 1 << 0,
+	FSCK_CAN_IGNORE		= 1 << 1,
+	FSCK_NEED_FSCK		= 1 << 2,
+	FSCK_NO_RATELIMIT	= 1 << 3,
 };
 
-#define FSCK_CAN_FIX		(1 << 0)
-#define FSCK_CAN_IGNORE		(1 << 1)
-#define FSCK_NEED_FSCK		(1 << 2)
+#define fsck_err_count(_c, _err)	bch2_sb_err_count(_c, BCH_FSCK_ERR_##_err)
 
-enum fsck_err_ret bch2_fsck_err(struct bch_fs *,
-				unsigned, const char *, ...);
+__printf(4, 5) __cold
+int bch2_fsck_err(struct bch_fs *,
+		  enum bch_fsck_flags,
+		  enum bch_sb_error_id,
+		  const char *, ...);
 void bch2_flush_fsck_errs(struct bch_fs *);
 
-#define __fsck_err(c, _flags, msg, ...)					\
+#define __fsck_err(c, _flags, _err_type, ...)				\
 ({									\
-	int _fix = bch2_fsck_err(c, _flags, msg, ##__VA_ARGS__);\
+	int _ret = bch2_fsck_err(c, _flags, BCH_FSCK_ERR_##_err_type,	\
+				 __VA_ARGS__);				\
 									\
-	if (_fix == FSCK_ERR_EXIT) {					\
-		bch_err(c, "Unable to continue, halting");		\
-		ret = BCH_FSCK_ERRORS_NOT_FIXED;			\
+	if (_ret != -BCH_ERR_fsck_fix &&				\
+	    _ret != -BCH_ERR_fsck_ignore) {				\
+		ret = _ret;						\
 		goto fsck_err;						\
 	}								\
 									\
-	_fix;								\
+	_ret == -BCH_ERR_fsck_fix;					\
 })
 
 /* These macros return true if error should be fixed: */
 
 /* XXX: mark in superblock that filesystem contains errors, if we ignore: */
 
-#define __fsck_err_on(cond, c, _flags, ...)				\
-	((cond) ? __fsck_err(c, _flags,	##__VA_ARGS__) : false)
+#define __fsck_err_on(cond, c, _flags, _err_type, ...)			\
+	(unlikely(cond) ? __fsck_err(c, _flags, _err_type, __VA_ARGS__) : false)
 
-#define unfixable_fsck_err_on(cond, c, ...)				\
-	__fsck_err_on(cond, c, FSCK_CAN_IGNORE, ##__VA_ARGS__)
+#define need_fsck_err_on(cond, c, _err_type, ...)				\
+	__fsck_err_on(cond, c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, _err_type, __VA_ARGS__)
 
-#define need_fsck_err_on(cond, c, ...)					\
-	__fsck_err_on(cond, c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, ##__VA_ARGS__)
+#define need_fsck_err(c, _err_type, ...)				\
+	__fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, _err_type, __VA_ARGS__)
 
-#define mustfix_fsck_err(c, ...)					\
-	__fsck_err(c, FSCK_CAN_FIX, ##__VA_ARGS__)
+#define mustfix_fsck_err(c, _err_type, ...)				\
+	__fsck_err(c, FSCK_CAN_FIX, _err_type, __VA_ARGS__)
 
-#define mustfix_fsck_err_on(cond, c, ...)				\
-	__fsck_err_on(cond, c, FSCK_CAN_FIX, ##__VA_ARGS__)
+#define mustfix_fsck_err_on(cond, c, _err_type, ...)			\
+	__fsck_err_on(cond, c, FSCK_CAN_FIX, _err_type, __VA_ARGS__)
 
-#define fsck_err_on(cond, c, ...)					\
-	__fsck_err_on(cond, c, FSCK_CAN_FIX|FSCK_CAN_IGNORE, ##__VA_ARGS__)
+#define fsck_err(c, _err_type, ...)					\
+	__fsck_err(c, FSCK_CAN_FIX|FSCK_CAN_IGNORE, _err_type, __VA_ARGS__)
+
+#define fsck_err_on(cond, c, _err_type, ...)				\
+	__fsck_err_on(cond, c, FSCK_CAN_FIX|FSCK_CAN_IGNORE, _err_type, __VA_ARGS__)
+
+__printf(4, 0)
+static inline void bch2_bkey_fsck_err(struct bch_fs *c,
+				     struct printbuf *err_msg,
+				     enum bch_sb_error_id err_type,
+				     const char *fmt, ...)
+{
+	va_list args;
+
+	va_start(args, fmt);
+	prt_vprintf(err_msg, fmt, args);
+	va_end(args);
+}
+
+#define bkey_fsck_err(c, _err_msg, _err_type, ...)			\
+do {									\
+	prt_printf(_err_msg, __VA_ARGS__);				\
+	bch2_sb_error_count(c, BCH_FSCK_ERR_##_err_type);		\
+	ret = -BCH_ERR_invalid_bkey;					\
+	goto fsck_err;							\
+} while (0)
+
+#define bkey_fsck_err_on(cond, ...)					\
+do {									\
+	if (unlikely(cond))						\
+		bkey_fsck_err(__VA_ARGS__);				\
+} while (0)
 
 /*
  * Fatal errors: these don't indicate a bug, but we can't continue running in RW
@@ -172,83 +199,44 @@ do {									\
 
 #define bch2_fs_fatal_err_on(cond, c, ...)				\
 ({									\
-	int _ret = !!(cond);						\
+	bool _ret = unlikely(!!(cond));					\
 									\
 	if (_ret)							\
 		bch2_fs_fatal_error(c, __VA_ARGS__);			\
 	_ret;								\
 })
 
-#define bch2_dev_fatal_error(ca, ...)					\
-do {									\
-	bch_err(ca, __VA_ARGS__);					\
-	bch2_fatal_error(c);						\
-} while (0)
-
-#define bch2_dev_fatal_io_error(ca, fmt, ...)				\
-do {									\
-	printk_ratelimited(KERN_ERR bch2_fmt((ca)->fs,			\
-		"fatal IO error on %s for " fmt),			\
-		(ca)->name, ##__VA_ARGS__);				\
-	bch2_fatal_error((ca)->fs);					\
-} while (0)
-
-#define bch2_dev_fatal_io_err_on(cond, ca, ...)				\
-({									\
-	int _ret = !!(cond);						\
-									\
-	if (_ret)							\
-		bch2_dev_fatal_io_error(ca, __VA_ARGS__);		\
-	_ret;								\
-})
-
 /*
- * Nonfatal IO errors: either recoverable metadata IO (because we have
- * replicas), or data IO - we need to log it and print out a message, but we
- * don't (necessarily) want to shut down the fs:
+ * IO errors: either recoverable metadata IO (because we have replicas), or data
+ * IO - we need to log it and print out a message, but we don't (necessarily)
+ * want to shut down the fs:
  */
 
-void bch2_nonfatal_io_error_work(struct work_struct *);
+void bch2_io_error_work(struct work_struct *);
 
 /* Does the error handling without logging a message */
-void bch2_nonfatal_io_error(struct bch_dev *);
+void bch2_io_error(struct bch_dev *, enum bch_member_error_type);
 
-#if 0
-#define bch2_fs_nonfatal_io_error(c, ...)				\
-do {									\
-	bch_err(c, __VA_ARGS__);					\
-	bch2_nonfatal_io_error(c);					\
-} while (0)
-#endif
-
-/* Logs message and handles the error: */
-#define bch2_dev_nonfatal_io_error(ca, fmt, ...)				\
-do {									\
-	printk_ratelimited(KERN_ERR bch2_fmt((ca)->fs,			\
-		"IO error on %s for " fmt),				\
-		(ca)->name, ##__VA_ARGS__);				\
-	bch2_nonfatal_io_error(ca);					\
-} while (0)
-
-#define bch2_dev_nonfatal_io_err_on(cond, ca, ...)			\
+#define bch2_dev_io_err_on(cond, ca, _type, ...)			\
 ({									\
 	bool _ret = (cond);						\
 									\
-	if (_ret)							\
-		bch2_dev_nonfatal_io_error(ca, __VA_ARGS__);		\
+	if (_ret) {							\
+		bch_err_dev_ratelimited(ca, __VA_ARGS__);		\
+		bch2_io_error(ca, _type);				\
+	}								\
 	_ret;								\
 })
 
-/* kill? */
-
-#define __bcache_io_error(c, fmt, ...)					\
-	printk_ratelimited(KERN_ERR bch2_fmt(c,				\
-			"IO error: " fmt), ##__VA_ARGS__)
-
-#define bcache_io_error(c, bio, fmt, ...)				\
-do {									\
-	__bcache_io_error(c, fmt, ##__VA_ARGS__);			\
-	(bio)->bi_error = -EIO;						\
-} while (0)
+#define bch2_dev_inum_io_err_on(cond, ca, _type, ...)			\
+({									\
+	bool _ret = (cond);						\
+									\
+	if (_ret) {							\
+		bch_err_inum_offset_ratelimited(ca, __VA_ARGS__);	\
+		bch2_io_error(ca, _type);				\
+	}								\
+	_ret;								\
+})
 
-#endif /* _BCACHE_ERROR_H */
+#endif /* _BCACHEFS_ERROR_H */