On Fedora, /tmp is tmpfs, which behaves differently than ext4
(inode numbers are never reused, for example).
Use /var/tmp, which is ext4 on Fedora, to get a more realistic
test environment.
This also allows us to drop the xattr workaround.
When running
$ go test ./tests/matrix/
in isolation, it failed like this:
fd leak? before, after:
[0r=/dev/null 3w=/dev/null 5r=/proc/8078/fd (hidden:4)]
[0r=/dev/null 3w=/dev/null 5w=/tmp/go-build366655199/b001/testlog.txt 7r=/proc/8078/fd (hidden:4)]
Filter by prefix to get rid of this spurious test failure.
When we run tests as root, they will leave root-owned files
in testParentDir, which causes trouble when we run tests as
a normal user later on. Split by UID.
CheckTrailingGarbage was called even when "-passfile" was
used, which is stupid, and causes false positives:
https://github.com/rfjakob/gocryptfs/issues/391
(false error "Received trailing garbage after the password"
when using -passfile in .bash_profile)
Instead of trying to improve the logic to handle that case
and make everything even more complicated, delete the function.
It is unclear if actually helps in some cases, and it definitely
harms as shown by the above bug report.
This adds support for gitignore-like wildcards and exclude patters in
reverse mode. It (somewhat) fixes#273: no regexp support, but the
syntax should be powerful enough to satisfy most needs.
Also, since adding a lot of --exclude options can be tedious, it adds
the --exclude-from option to read patterns from a file (or files).
Ironically, the test for DirIV races had a data race itself
by writing to a bool without taking a lock.
Found by running "./test.bash -race":
WARNING: DATA RACE
Write at 0x00c00001dea5 by goroutine 22:
github.com/rfjakob/gocryptfs/tests/defaults.TestDirIVRace.func1()
/home/jakob/go/src/github.com/rfjakob/gocryptfs/tests/defaults/diriv_test.go:39 +0x38
github.com/rfjakob/gocryptfs/tests/defaults.TestDirIVRace()
/home/jakob/go/src/github.com/rfjakob/gocryptfs/tests/defaults/diriv_test.go:73 +0x65c
testing.tRunner()
/usr/local/go/src/testing/testing.go:865 +0x163
Previous read at 0x00c00001dea5 by goroutine 23:
github.com/rfjakob/gocryptfs/tests/defaults.TestDirIVRace.func2()
/home/jakob/go/src/github.com/rfjakob/gocryptfs/tests/defaults/diriv_test.go:51 +0x8b
Goroutine 22 (running) created at:
testing.(*T).Run()
/usr/local/go/src/testing/testing.go:916 +0x699
testing.runTests.func1()
/usr/local/go/src/testing/testing.go:1157 +0xa8
testing.tRunner()
/usr/local/go/src/testing/testing.go:865 +0x163
testing.runTests()
/usr/local/go/src/testing/testing.go:1155 +0x523
testing.(*M).Run()
/usr/local/go/src/testing/testing.go:1072 +0x2eb
github.com/rfjakob/gocryptfs/tests/defaults.TestMain()
/home/jakob/go/src/github.com/rfjakob/gocryptfs/tests/defaults/main_test.go:21 +0xe1
main.main()
_testmain.go:76 +0x222
Goroutine 23 (running) created at:
github.com/rfjakob/gocryptfs/tests/defaults.TestDirIVRace()
/home/jakob/go/src/github.com/rfjakob/gocryptfs/tests/defaults/diriv_test.go:43 +0x48d
testing.tRunner()
/usr/local/go/src/testing/testing.go:865 +0x163
==================
--- FAIL: TestDirIVRace (0.00s)
testing.go:809: race detected during execution of test
FAIL
This should get rid of
Openat: O_NOFOLLOW missing: flags = 0x0
Fchmodat: adding missing AT_SYMLINK_NOFOLLOW flag
sys_common_test.go:203: chmod on symlink should have failed, but did not. New mode=0333
UnmountErr: "[...]/057376762.mnt" was not found in MountInfo, cannot check for FD leak
and add some context to
--- FAIL: TestUtimesNano (0.00s)
matrix_test.go:628: no such file or directory
See https://github.com/rfjakob/gocryptfs/pull/343#issuecomment-453888006
for full test output
Setting/removing extended attributes on directories was partially fixed with
commit eff35e60b6. However, on most file systems
it is also possible to do these operations without read access (see tests).
Since we cannot open a write-access fd to a directory, we have to use the
/proc/self/fd trick (already used for ListXAttr) for the other operations aswell.
For simplicity, let's separate the Linux and Darwin code again (basically revert
commit f320b76fd1), and always use the
/proc/self/fd trick on Linux. On Darwin we use the best-effort approach with
openBackingFile() as a fallback.
More discussion about the available options is available in
https://github.com/rfjakob/gocryptfs/issues/308.
Give the gocryptfs process one extra millisecond to close
files. Allows us to drop several other sleeps.
UnmountErr now really returns an error when it detects an fd leak
instead of just printing a message.
An Open() a fifo blocks until it is opened for writing.
This meant that xattr operations on FIFOs would block.
Pass O_NONBLOCK to fix that, and add a test.
Retry with length 1000 if length 4000 fails, which
should work on all filesystems.
Failure was:
--- FAIL: TestTooLongSymlink (0.00s)
correctness_test.go:198: symlink xxx[...]xxxx /tmp/xfs.mnt/gocryptfs-test-parent/549823072/365091391/TooLongSymlink: file name too long
https://github.com/rfjakob/gocryptfs/issues/267
Instead of reporting the consequence:
matrix_test.go:906: modeHave 0664 != modeWant 0777
Report it if chmod itself fails, and also report the old file mode:
matrix_test.go:901: chmod 000 -> 777 failed: bad file descriptor
The cipherdir path is used as the fsname, as displayed
in "df -T". Now, having a comma in fsname triggers a sanity check
in go-fuse, aborting the mount with:
/bin/fusermount: mount failed: Invalid argument
fuse.NewServer failed: fusermount exited with code 256
Sanitize fsname by replacing any commas with underscores.
https://github.com/rfjakob/gocryptfs/issues/262
Rename openBackingPath to openBackingDir and use OpenDirNofollow
to be safe against symlink races. Note that openBackingDir is
not used in several important code paths like Create().
But it is used in Unlink, and the performance impact in the RM benchmark
to be acceptable:
Before
$ ./benchmark.bash
Testing gocryptfs at /tmp/benchmark.bash.bYO: gocryptfs v1.6-12-g930c37e-dirty; go-fuse v20170619-49-gb11e293; 2018-09-08 go1.10.3
WRITE: 262144000 bytes (262 MB, 250 MiB) copied, 1.07979 s, 243 MB/s
READ: 262144000 bytes (262 MB, 250 MiB) copied, 0.882413 s, 297 MB/s
UNTAR: 16.703
MD5: 7.606
LS: 1.349
RM: 3.237
After
$ ./benchmark.bash
Testing gocryptfs at /tmp/benchmark.bash.jK3: gocryptfs v1.6-13-g84d6faf-dirty; go-fuse v20170619-49-gb11e293; 2018-09-08 go1.10.3
WRITE: 262144000 bytes (262 MB, 250 MiB) copied, 1.06261 s, 247 MB/s
READ: 262144000 bytes (262 MB, 250 MiB) copied, 0.947228 s, 277 MB/s
UNTAR: 17.197
MD5: 7.540
LS: 1.364
RM: 3.410
If we encounter a 128KB block of zeros, try to skip to the next
data section by calling File.SeekData().
This fixes xfstests generic/285, which creates a 17TB sparse file,
and runs fsck afterwards. Without this optimization, fsck would
take ages.
TrezorPayload stores 32 random bytes used for unlocking
the master key using a Trezor security module. The randomness makes sure
that a unique unlock value is used for each gocryptfs filesystem.
When mounted via /etc/fstab like this,
/a /b fuse.gocryptfs default 0 0
we always get extra options passed. As reported by @mahkoh
at https://github.com/rfjakob/gocryptfs/pull/233 :
mount passes `-o noexec` if `-o user` is set and `-o exec` is not set.
If both `-o user` and `-o exec` are set, it passes `-o exec`.
Make these options work, and in addtion, also make -suid and -rw
work the same way.
Reported-by: @mahkoh
mv is unhappy when we return EPERM when it tries to set
system.posix_acl_access:
mv: preserving permissions for ‘b/x’: Operation not permitted
Now we return EOPNOTSUPP like tmpfs does and mv seems happy.
