ext3 bug in ftruncate() with O_SYNC?
Hi. We're doing some checking on Linux file systems and found what appears to be a bug in the Linux 2.6.11 implementation of
ext3: when ftruncate shrinks a file, using a file descriptor opened with O_SYNC, the file size is not updated synchronously. I've appended a test program that illustrates the problem. After this program runs, the file system shows a file with length 1031, not 4 as would be expected if the ftruncate completed synchronously. (If I insert an fsync before closing the file, the file length is correct.)
Does this look like a bug to you guys?
Thanks,
Ben.
----------------------------------------------------------------------
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <stdarg.h>
#include <assert.h>
#define CHECK(ret) if(ret < 0) {perror(0); assert(0);}
int systemf(const char *fmt, ...)
{
static char cmd[1024];
va_list ap;
va_start(ap, fmt);
vsprintf(cmd, fmt, ap);
va_end(ap);
fprintf(stderr, "running cmd \"%s\"\n", cmd);
return system(cmd);
}main(int argc, char *argv[])
{
int ret, fd;
systemf("umount /dev/sbd0");
systemf("sbin/mkfs.ext3 -F -j -b 1024 /dev/sbd0 2048");
systemf("sbin/e2fsck.shared -y /dev/sbd0");
systemf("mount -t ext3 /dev/sbd0 /mnt/sbd0 -o commit=65535");
systemf("umount /dev/sbd0");
systemf("mount -t ext3 /dev/sbd0 /mnt/sbd0 -o commit=65535");
fd = open("/mnt/sbd0/0001", O_CREAT | O_RDWR | O_SYNC, 0700);
CHECK(fd);
ret = write(fd, "foobar", 6);
CHECK(ret);
ret = ftruncate(fd, 1031);
CHECK(ret);
ret = pwrite(fd, "bazzle", 6, 1031 - 6);
CHECK(ret);
ret = ftruncate(fd, 4);
CHECK(ret);
ret = close(fd);
CHECK(ret);#if 0
{
#include "../sbd/sbd.h"
int sbd_fd = open("/dev/sbd0", O_RDONLY);
ret = ioctl(sbd_fd, SBD_COPY_DISK, 1);
CHECK(ret);
close(sbd_fd);
}
#else
systemf("reboot -f -n");
#endif
return 0;
}Signature · 3 lines
Ben Pfaff email: [email protected] web: http://benpfaff.org
Quoted 11 lines · Ben Pfaff
Hi. We're doing some checking on Linux file systems and found what appears to be a bug in the Linux 2.6.11 implementation of
ext3: when ftruncate shrinks a file, using a file descriptor opened withO_SYNC, the file size is not updated synchronously. I've appended a test program that illustrates the problem. After this program runs, the file system shows a file with length 1031, not 4 as would be expected if the ftruncate completed synchronously. (If I insert an fsync before closing the file, the file length is correct.)
Quoted 2 lines
Does this look like a bug to you guys?
The spec says "Write I/O operations on the file descriptor shall complete as defined by synchronized I/O file integrity completion".
Is ftruncate a "write I/O operation"? No. Should ftruncate() honour O_SYNC? I'd say so, yes. After all, an extending ftruncate is a sort-of-write.
Unfortunately Linux doesn't pass the file* down to the filesytem's ->truncate() handler, so the fs doesn't know that the ftruncated fd was opened O_SYNC. I don't think _any_ Linux filesystems get this right.
Quoted 75 lines
Ben.
----------------------------------------------------------------------
#include <
sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <stdarg.h>
#include <assert.h>#define CHECK(ret) if(ret < 0) {perror(0); assert(0);}
int
systemf(const char *fmt, ...){ static char cmd[1024]; va_list ap; va_start(ap, fmt); vsprintf(cmd, fmt, ap); va_end(ap); fprintf(stderr, "running cmd \"%s\"\n", cmd); return system(cmd); }
main(int argc, char *argv[]){ int ret, fd; systemf("umount /dev/sbd0"); systemf("sbin/mkfs.ext3 -F -j -b 1024 /dev/sbd0 2048"); systemf("sbin/e2fsck.shared -y /dev/sbd0"); systemf("mount -t ext3 /dev/sbd0 /mnt/sbd0 -o commit=65535"); systemf("umount /dev/sbd0"); systemf("mount -t ext3 /dev/sbd0 /mnt/sbd0 -o commit=65535"); fd = open("/mnt/sbd0/0001", O_CREAT | O_RDWR | O_SYNC, 0700); CHECK(fd); ret = write(fd, "foobar", 6); CHECK(ret); ret = ftruncate(fd, 1031); CHECK(ret); ret = pwrite(fd, "bazzle", 6, 1031 - 6); CHECK(ret); ret = ftruncate(fd, 4); CHECK(ret); ret = close(fd); CHECK(ret);#if 0 { #include "../sbd/sbd.h" int sbd_fd = open("/dev/sbd0", O_RDONLY); ret = ioctl(sbd_fd, SBD_COPY_DISK, 1); CHECK(ret); close(sbd_fd); } #else systemf("reboot -f -n"); #endif return 0; }Signature · 8 lines
Ben Pfaff email: [email protected] web: http://benpfaff.org - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Hi,
Quoted 4 lines · Andrew Morton
The spec says "Write I/O operations on the file descriptor shall complete as defined by synchronized I/O file integrity completion".
Is ftruncate a "write I/O operation"? No.
SUS seems to be pretty clear on this. The syscall descriptions for write(2) and pwrite(2) explicitly describe O_SYNC as requiring synchronized I/O file integrity completion. ftruncate() has no such requirement.
It would certainly be a reasonable thing to do, but I don't think it strictly counts as a bug that we're not honouring O_SYNC here.
--Stephen