because buffer::raw is not a standard layout type. so `offsetof()`
does not always work. let's use a homebrew macro to calculate the
address of `bptr_storage`.
this change silences the warning like:
src/common/buffer.cc:2156:15: warning: ‘offsetof’ within non-standard-layout type ‘ceph::buffer::v15_2_0::raw’ is conditionally-supported [-Winvalid-offsetof]
2156 | offsetof(buffer::raw, bptr_storage));
Signed-off-by: Kefu Chai <kchai@redhat.com>
RGWStore::create_bucket() only returns EEXIST in case of conflicts that
should be returned to the client as errors. the code that applies
changes to metadata should instead use the s->bucket_exists flag
Signed-off-by: Casey Bodley <cbodley@redhat.com>
cmake: add cls_fifo as a dependency of osd
Reviewed-by: Adam C. Emerson <aemerson@redhat.com>
Reviewed-by: Soumya Koduri <skoduri@redhat.com>
Reviewed-by: Ali Maredia <amaredia@redhat.com>
Modify filesystem.Filesystem.delete_all_filesystems() method to make it
more succinct, move it to class MDSCluster instead and update every call
to it accordingly.
Signed-off-by: Rishabh Dave <ridave@redhat.com>
RGWOp_Bucket_Remove::execute() was storing failures from
bucket->remove_bucket() in op_ret, but left http_ret=0 so we responded
to the client with '200 OK'
to avoid bugs like this, remove the extra http_ret variable and only use
the op_ret from RGWOp
Signed-off-by: Casey Bodley <cbodley@redhat.com>
Previously it still could race when unmap_device returned success
because the device was not found in `rbd-nbd list-mapped` (the nbd
device was removed) but the test failed because the process was still
found in the ps table.
Fixes: https://tracker.ceph.com/issues/47394
Signed-off-by: Mykola Golub <mgolub@suse.com>
This commit exposes HACKING.rst through the
docs website. Formerly, HACKING.rst was just
a file that existed at /src/pybind/mgr/dashboard/
HACKING.rst. This commit also updates README.rst
to point to the new location of HACKING.rst.
Fixes: https://tracker.ceph.com/issues/47230
Signed-off-by: Zac Dover <zac.dover@gmail.com>
* cmake: bump Seastar_API_LEVEL to "5"
* errorator.h: add invoke() methods to futurize variants, as in seastar
API level 5, the future's store type is now either monostate or
a single T, instead of tuple. so it'd be simpler to integrate
with seastar with invoke() semantic without the boxing-unboxing
dance.
* errorator.h: drop apply() methods, as they are not used anymore.
* errorator.h: rename and revise tuple2future. as the store type
is not tuple now, the trait's name is changed accordingly to
match seastar::internal::future_stored_type
Signed-off-by: Kefu Chai <kchai@redhat.com>
crimson uses seastar logging facility for logging. and the latter uses
fmt::format(). currently, we collect the log message into a string and
pass it to log(fmt,...) as "fmt", but seastar/core/sstring.hh defines
the operator<<(ostream&, const vector<T>&) which is a better match than
our the operator<<(ostream&, const vector<T, Allocator>&). and seastar's
operator<<(ostream&, const vector<T>&) uses "{" and "}" to mark the
begin and end of a vector when printing it. and "{}" is also used by
libfmt to enclose its replacement fields. see
https://fmt.dev/latest/syntax.html. so when a vector is printed using
logging facility in crimson, libfmt chokes when trying to parse it as a
format string. so we have some options, like:
- disable seastar's operator<< implementation
- escape the "{}" when writing the vector to the output stream
- print the message as the args, and use "{}" as the fmt.
the 3rd one is the most straightforward, and probably more performant.
Signed-off-by: Kefu Chai <kchai@redhat.com>
Modify cephfs.filesystem.Filesystem.recreate() method to delete only the
FS represented by the object instead of deleting the every FS on the
Ceph cluster.
Signed-off-by: Rishabh Dave <ridave@redhat.com>
And reset_obj_attrs parameter to it so that the caller of the method can
choose to destroy the Ceph FS represented by the object without
disturbing the object attributes.
Signed-off-by: Rishabh Dave <ridave@redhat.com>
This commit adds a new argument check_status to mount methods of
KernelMount, FuseMount, LocalKernelMount and LocalFuseMount. When value
of this argument is False, these methods would catch the
CommandFailedError exception and would return a tuple consisting of the
exception itself, and stdout and stderr of the mount command. This
allows reusing these mount methods while running negative tests for
commands.
The name "check_status" is selected so since teuthology's run() and
vstart_runner's run() use a variable with same name for the very same
purpose.
Signed-off-by: Rishabh Dave <ridave@redhat.com>
This commit introduces following two set of changes -
First, make client keyring path, mountpoint on host FS and CephFS and
CephFS's name attributes of the object representing the mount
and update all the mount object creation calls accordingly. Also,
rewrite all the mount object creation to use keyword arguments instead
of positional arguments to avoid mistakes, especially since a new
argument was added in this commit.
Second, add remount method to mount.py so that it's possible to unmount
safely, modify the attributes of the object representing the mount and
mount again based on new state of the object *in a single call*. The
method is placed in mount.py to avoid duplication.
This change has two leads to two more changes: upgrading interface of
mount() and mount_wait() and upgrading testsuites to adapt to these
change.
Signed-off-by: Rishabh Dave <ridave@redhat.com>
So far running "fs authorize" subcommand sets "allow r" as MON cap;
change that to "allow <perms> fsname=<fsname>".
Signed-off-by: Rishabh Dave <ridave@redhat.com>
Allow passing path along with fs names and capspec while creating an MDS
cap. The new syntax looks as follows -
allow rw fsname=<fsname> path=<path>
To provide caps for multiple file systems, pass the same phrase multiple
times separated by commas -
allow rw fsname=<fsname1> path=<path1>, allow rw fsname=<fsname2>
path=<path2>, ...
This commit also makes sure that the old syntax 'allow rw path=<path>'
is supported for backwards compatibility. The old syntax would imply
'allow rw fsname=* path=<path>' and would grant read-write permission for
all FSs containing the path <path>.
Signed-off-by: Rishabh Dave <ridave@redhat.com>
Add new auth caps to restrict access to clients based on fsnames. To
specify this, for example:
mds 'allow rw fsname=cephfs1'
This will restrict client access to fs name "cephfs1" only. Messages to
active MDS assigned to any other FSMap will be dropped. Standby MDS not
associated with an FSMap will accept messages from clients. To allow
multiple file systems, create MDS cap as follows -
mds 'allow rw fsname=cephfs1, allow rw fsname=cephfs2'
Fixes: http://tracker.ceph.com/issues/15070
Signed-off-by: Douglas Fuller <dfuller@redhat.com>
Signed-off-by: Rishabh Dave <ridave@redhat.com>
Modify MMDSMap to include the name of the file system from the FSMap
when bringing up a new MDS. Store the name in the MDSRank structure for
use in security validation.
Fixes: http://tracker.ceph.com/issues/15070
Signed-off-by: Douglas Fuller <dfuller@redhat.com>
Signed-off-by: Rishabh Dave <ridave@redhat.com>
Add a 'fsname' clause to mon auth caps to restrict a client's view
of the FSMap. Example:
mon 'allow rw fsname=cephfs2'
This would restrict the client's view of the FSMap to the MDSMap for
cephfs2. Any MDS allocated to a different filesystem will be invisible.
Global standby daemons are always visible. To allow multiple CephFSs,
add multiple caps:
mon 'allow rw fsname=cephfs2, allow rw fsname=cephfs2'
Fixes: http://tracker.ceph.com/issues/15070
Signed-off-by: Douglas Fuller <dfuller@redhat.com>
Signed-off-by: Rishabh Dave <ridave@redhat.com>
* use "code-block:: yaml" for qa suite samples
* use "prompt:: bash $" for shell commands using a non-root user
Signed-off-by: Kefu Chai <kchai@redhat.com>
