Skip to content

Commit 2ce439f

Browse files
robertmhaasrobertmhaas
committed
Recursively fsync() the data directory after a crash.
Otherwise, if there's another crash, some writes from after the first crash might make it to disk while writes from before the crash fail to make it to disk. This could lead to data corruption. Back-patch to all supported versions. Abhijit Menon-Sen, reviewed by Andres Freund and slightly revised by me.
1 parent ec3d976 commit 2ce439f

File tree

3 files changed

+159
-0
lines changed

3 files changed

+159
-0
lines changed

‎src/backend/access/transam/xlog.c

Lines changed: 42 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -845,6 +845,8 @@ static void WALInsertLockAcquireExclusive(void);
845845
static void WALInsertLockRelease(void);
846846
static void WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt);
847847

848+
static void fsync_pgdata(char *datadir);
849+
848850
/*
849851
* Insert an XLOG record represented by an already-constructed chain of data
850852
* chunks. This is a low-level routine; to construct the WAL record header
@@ -5910,6 +5912,18 @@ StartupXLOG(void)
59105912
(errmsg("database system was interrupted; last known up at %s",
59115913
str_time(ControlFile->time))));
59125914

5915+
/*
5916+
* If we previously crashed, there might be data which we had written,
5917+
* intending to fsync it, but which we had not actually fsync'd yet.
5918+
* Therefore, a power failure in the near future might cause earlier
5919+
* unflushed writes to be lost, even though more recent data written to
5920+
* disk from here on would be persisted. To avoid that, fsync the entire
5921+
* data directory.
5922+
*/
5923+
if (ControlFile->state != DB_SHUTDOWNED &&
5924+
ControlFile->state != DB_SHUTDOWNED_IN_RECOVERY)
5925+
fsync_pgdata(data_directory);
5926+
59135927
/* This is just to allow attaching to startup process with a debugger */
59145928
#ifdef XLOG_REPLAY_DELAY
59155929
if (ControlFile->state != DB_SHUTDOWNED)
@@ -11123,3 +11137,31 @@ SetWalWriterSleeping(bool sleeping)
1112311137
XLogCtl->WalWriterSleeping = sleeping;
1112411138
SpinLockRelease(&XLogCtl->info_lck);
1112511139
}
11140+
11141+
/*
11142+
* Issue fsync recursively on PGDATA and all its contents.
11143+
*/
11144+
static void
11145+
fsync_pgdata(char *datadir)
11146+
{
11147+
if (!enableFsync)
11148+
return;
11149+
11150+
/*
11151+
* If possible, hint to the kernel that we're soon going to fsync
11152+
* the data directory and its contents.
11153+
*/
11154+
#if defined(HAVE_SYNC_FILE_RANGE) || \
11155+
(defined(USE_POSIX_FADVISE) && defined(POSIX_FADV_DONTNEED))
11156+
walkdir(datadir, pre_sync_fname);
11157+
#endif
11158+
11159+
/*
11160+
* Now we do the fsync()s in the same order.
11161+
*
11162+
* It's important to fsync the destination directory itself as individual
11163+
* file fsyncs don't guarantee that the directory entry for the file is
11164+
* synced.
11165+
*/
11166+
walkdir(datadir, fsync_fname);
11167+
}

‎src/backend/storage/file/fd.c

Lines changed: 115 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2439,3 +2439,118 @@ looks_like_temp_rel_name(const char *name)
24392439
return false;
24402440
return true;
24412441
}
2442+
2443+
/*
2444+
* Hint to the OS that it should get ready to fsync() this file.
2445+
*
2446+
* Adapted from pre_sync_fname in initdb.c
2447+
*/
2448+
void
2449+
pre_sync_fname(char *fname, bool isdir)
2450+
{
2451+
int fd;
2452+
2453+
fd = open(fname, O_RDONLY | PG_BINARY);
2454+
2455+
/*
2456+
* Some OSs don't allow us to open directories at all (Windows returns
2457+
* EACCES)
2458+
*/
2459+
if (fd < 0 && isdir && (errno == EISDIR || errno == EACCES))
2460+
return;
2461+
2462+
if (fd < 0)
2463+
ereport(FATAL,
2464+
(errmsg("could not open file \"%s\" before fsync",
2465+
fname)));
2466+
2467+
pg_flush_data(fd, 0, 0);
2468+
2469+
close(fd);
2470+
}
2471+
2472+
/*
2473+
* walkdir: recursively walk a directory, applying the action to each
2474+
* regular file and directory (including the named directory itself)
2475+
* and following symbolic links.
2476+
*
2477+
* NB: There is another version of walkdir in initdb.c, but that version
2478+
* behaves differently with respect to symbolic links. Caveat emptor!
2479+
*/
2480+
void
2481+
walkdir(char *path, void (*action) (char *fname, bool isdir))
2482+
{
2483+
DIR *dir;
2484+
struct dirent *de;
2485+
2486+
dir = AllocateDir(path);
2487+
while ((de = ReadDir(dir, path)) != NULL)
2488+
{
2489+
char subpath[MAXPGPATH];
2490+
struct stat fst;
2491+
2492+
CHECK_FOR_INTERRUPTS();
2493+
2494+
if (strcmp(de->d_name, ".") == 0 ||
2495+
strcmp(de->d_name, "..") == 0)
2496+
continue;
2497+
2498+
snprintf(subpath, MAXPGPATH, "%s/%s", path, de->d_name);
2499+
2500+
if (lstat(subpath, &fst) < 0)
2501+
ereport(ERROR,
2502+
(errcode_for_file_access(),
2503+
errmsg("could not stat file \"%s\": %m", subpath)));
2504+
2505+
if (S_ISREG(fst.st_mode))
2506+
(*action) (subpath, false);
2507+
else if (S_ISDIR(fst.st_mode))
2508+
walkdir(subpath, action);
2509+
#ifndef WIN32
2510+
else if (S_ISLNK(fst.st_mode))
2511+
#else
2512+
else if (pg_win32_is_junction(subpath))
2513+
#endif
2514+
{
2515+
#if defined(HAVE_READLINK) || defined(WIN32)
2516+
char linkpath[MAXPGPATH];
2517+
int len;
2518+
struct stat lst;
2519+
2520+
len = readlink(subpath, linkpath, sizeof(linkpath)-1);
2521+
if (len < 0)
2522+
ereport(ERROR,
2523+
(errcode_for_file_access(),
2524+
errmsg("could not read symbolic link \"%s\": %m",
2525+
subpath)));
2526+
2527+
if (len >= sizeof(linkpath)-1)
2528+
ereport(ERROR,
2529+
(errmsg("symbolic link \"%s\" target is too long",
2530+
subpath)));
2531+
2532+
linkpath[len] = '\0';
2533+
2534+
if (lstat(linkpath, &lst) == 0)
2535+
{
2536+
if (S_ISREG(lst.st_mode))
2537+
(*action) (linkpath, false);
2538+
else if (S_ISDIR(lst.st_mode))
2539+
walkdir(subpath, action);
2540+
}
2541+
else if (errno != ENOENT)
2542+
ereport(ERROR,
2543+
(errcode_for_file_access(),
2544+
errmsg("could not stat file \"%s\": %m", linkpath)));
2545+
#else
2546+
ereport(WARNING,
2547+
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2548+
errmsg("this platform does not support symbolic links; ignoring \"%s\"",
2549+
subpath)));
2550+
#endif
2551+
}
2552+
}
2553+
FreeDir(dir);
2554+
2555+
(*action) (path, true);
2556+
}

‎src/include/storage/fd.h

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -114,6 +114,8 @@ extern int pg_fsync_writethrough(int fd);
114114
extern int pg_fdatasync(int fd);
115115
extern int pg_flush_data(int fd, off_t offset, off_t amount);
116116
extern void fsync_fname(char *fname, bool isdir);
117+
extern void pre_sync_fname(char *fname, bool isdir);
118+
extern void walkdir(char *path, void (*action) (char *fname, bool isdir));
117119

118120
/* Filename components for OpenTemporaryFile */
119121
#define PG_TEMP_FILES_DIR "pgsql_tmp"

0 commit comments

Comments
 (0)