Before this change cgofuse and libatexit would race to see who could
unmount the file system with unpredicatable results. On Linux it could
report an error or not, depending.
This change checks to see if umount is beng called from a signal and
if so leaves the unmounting to cgofuse/libfuse.
See #4804
Before this change if the user supplied `-o uid=XXX` then rclone would
write `-o uid=-1 -o uid=XXX` so duplicating the uid value.
After this change rclone doesn't write the default `-1` version.
This fix affects `uid` and `gid`.
See: https://forum.rclone.org/t/issue-with-rclone-mount-and-resilio-sync/14730/27
Before this change when reading directories we would use the directory
handle and the Readdir(-1) call on the directory handle. This worked
fine for the first read, but if the directory was read again on the
same handle Readdir(-1) returns nothing (as per its design).
It turns out that macOS leaves the directory handle open and just
re-reads the data from it, so this problem causes directories to start
out full then subsequently appear empty.
macOS/OSXFUSE is passing an offset of 0 to the Readdir call telling
rclone to seek in the directory, but we've told FUSE that we can't
seek by always returning ofst=0 in the fill function.
This fix works around the problem by reading the directory from the
path each time, ignoring the actual handle. This should be no less
efficient.
We will return an ESPIPE if offset is ever non 0.
There are possible corner cases reading deleted directories which this
ignores.
Before this change Windows would read a directory then immedately stat
every item in the directory.
After this change we return the stat information along with the
directory which stops so many callbacks.
The tests are now run for the mount commands and for the plain VFS.
This makes the tests much easier to debug when running with a VFS than
through a mount.
Before this change, files of unknown size (eg Google Docs) would
appear in file listings with 0 size and would only allow 0 bytes to be
read.
This change sets the direct_io flag in the FUSE return which bypasses
the cache for these files. This means that they can be read properly.
This is compatible with some, but not all applications.
OSX FUSE only supports 32 bit number of blocks which means that block
counts have been wrapping. This causes f_bavail to be 0 which in turn
causes problems with programs like borg backup.
Fixes#2356
By default the timeout is 60s which isn't long enough for long
transactions. The symptoms are rclone just quitting for no reason.
Supplying the --daemon-timeout flag fixes this causing the kernel to
wait longer for rclone.
Before this change we would unconditionally set the OSXFUSE options
noappledouble and noapplexattr.
However the noapplexattr options caused problems with copies in the
Finder.
Now the default for noapplexattr is false so we don't add the option
by default and the user can override the defaults using the
--noappledouble and --noapplexattr flags.
Before this change rclone would set the volume name from the
remote:path normally. However this has `:` and `/` in which make it
difficult to use in macOS.
Now rclone will remove the special characters and replace them with
spaces. It also allows the volume name to be set with the --volname
flag.
The written out list of tests was replaced with a nested test for
mount and cmount. The tests for each VFS cache mode were also replaced
with nested tests which makes the output and the code much cleaner.
Previously this was adding it in to all file opens which was causing
inefficiencies under Windows where it stats the file using
open/fstat/close.
This change will make stat operations run much quicker under Windows
as they won't have to open the underlying file
This problem was introduced in61b6159a05336bd7ba105766de2d2ff171f7fb81
where we added O_CREATE to all file opens and creates.
This flag allows the attribute caching in the kernel to be controlled.
The default is 0s - no caching - which is recommended for filesystems
which can change outside the control of the kernel.
Previously this was at the default meaning it was 60s for mount and 1s
for cmount. This showed strange effects when files changed on the
remote not via the kernel. For instance Caddy would serve corrupted
files for a while when serving from an rclone mount when a file
changed on the remote.
Before this change Open("name", os.O_RDONLY|os.O_TRUNC) would have
truncated the file. This is what Linux does, but is counterintuitive.
POSIX states this is undefined, so return an error in this case
instead. This preserves the invariant O_RDONLY => file is not
changed.
The purpose of this is to make it easier to maintain and eventually to
allow the rclone backends to be re-used in other projects without
having to use the rclone configuration system.
The new code layout is documented in CONTRIBUTING.
