Truncate can lead to data loss

[benrubson]

Hi,

I found a bug in truncate function, which leads to data loss.

You can easily reproduce it using the following tiny C++ example :

#include <fcntl.h>
#include <unistd.h>
#include <stdio.h>
#include <string.h>
#include <cerrno>

int main() {

int error=0;

int fd = open("trunc.tmp", O_CREAT|O_TRUNC|O_WRONLY);
if(fd<0) {
	printf("open WR failed: %s\n",strerror(errno));
	error++;
}
if(pwrite(fd, "1\n2\n3\n4\n5\n", 10, 0)<=0) {
	printf("write failed: %s\n",strerror(errno));
	error++;
}
close(fd);

fd = open("trunc.tmp", O_RDONLY);
if(fd<0) {
	printf("open RD failed: %s\n",strerror(errno));
	error++;
}
if(ftruncate(fd, 6)>=0) {
	printf("ftruncate on the read-only fd succeeded, but should have failed !\n");
	error++;
}
if(truncate("trunc.tmp", 8)<0) {
	printf("truncate on the read-only opened path failed: %s\n",strerror(errno));
	error++;
}
close(fd);
if(truncate("trunc.tmp", 6)<0) {
	printf("truncate on the closed path failed: %s\n",strerror(errno));
	error++;
}

printf("Errors : %d\n",error);
return 0;
}

This will produce the following and corrupt trunc.tmp :

truncate on the read-only opened path failed: Input/output error
truncate on the closed path failed: Input/output error
Errors : 2

Encfs will log the following :

VERBOSE op: truncate : /tmp/encfs_test/ciphertext/qQD8mBQ1SItyykZ7MCh8xLle [encfs.cpp:150]
VERBOSE streamRead(data, 10, IV) [CipherFileIO.cpp:340]
VERBOSE Called streamWrite [CipherFileIO.cpp:407]
ERROR Assert failed: canWrite [RawFileIO.cpp:216]
ERROR withFileNode: error caught in truncate: canWrite [encfs.cpp:176]
VERBOSE created FileNode for /tmp/encfs_test/ciphertext/qQD8mBQ1SItyykZ7MCh8xLle [DirNode.cpp:675]
VERBOSE op: truncate : /tmp/encfs_test/ciphertext/qQD8mBQ1SItyykZ7MCh8xLle [encfs.cpp:150]
ERROR Assert failed: fd >= 0 [RawFileIO.cpp:202]
ERROR withFileNode: error caught in truncate: fd >= 0 [encfs.cpp:176]

In other words :

• ftruncate should only truncate write-opened fds.
• truncate should always truncate.

Ben

[rfjakob]

Where is the data loss?

[benrubson]

In trunc.tmp, it will get garbage instead of expected result :

1
2
3

[benrubson]

To reproduce the bug you need a configuration file created in reverse mode, so with :

        <uniqueIV>0</uniqueIV>
        <chainedNameIV>0</chainedNameIV>

[benrubson]

OK, issue solved in #380 👍

[benrubson]

Committed !