Kotresh HR [Mon, 24 Feb 2025 21:02:27 +0000 (02:32 +0530)]
multi-mds/link: Handle rollback for referent_inodes list
The referent inode added to the list on the primary/real
inode upon hardlink creation needs to rolled back properly
on failure. This patch takes care of the same.
Kotresh HR [Mon, 24 Feb 2025 10:32:36 +0000 (16:02 +0530)]
multi-mds/link: Reverse link primary inode to hardlink
Reverse link primary/real inode to the hardlink using
the referent inode. This patch takes care of link with
multiple active mds link. So the primary inode would
be auth on one mds and the new hardlink dentry being
created would be auth on different mds.
This is done as below.
1. Primary inode (CInode) maintains the referent_inodes list
2. Upon referent inode creation on auth mds of dentry being
created, the referent inode number is sent to auth mds of
the primary/real inode using MMDSPeerRequest::OP_LINKPREP.
The auth mds of primary/real inode adds it to the
referent_inodes list
Kotresh HR [Mon, 24 Feb 2025 11:07:05 +0000 (16:37 +0530)]
multi-mds/link: Send referent inode to dentry_replicas
On hardlink creation with multiple mdses, the
hardlink dentry being created could already exist.
In such cases, the change in inode is notified to
dentry replicas using MDCache::send_dentry_link.
If it's a referent remote, send the referent inode
to the dentry replicas and link it correctly.
Kotresh HR [Mon, 24 Feb 2025 08:12:44 +0000 (13:42 +0530)]
multi-mds/link: Create referent inode and store backtrace
On hardlink creation, create a referent inode (CInode)
and store backtrace for the hardlink. This patch
takes care of multiple active mds. So the primary inode
would be auth on one mds and the new hardlink dentry
being created would be auth on different mds.
Kotresh HR [Mon, 24 Feb 2025 05:54:19 +0000 (11:24 +0530)]
mds/link: Reverse link primary inode to hardlink
Reverse link primary/real inode to the hardlink using
the referent inode. This patch takes care of single
mds link. This is done as below.
1. Primary inode (CInode) maintains the referent_inodes list
2. Upon referent inode creation, the same is added the
referent_inodes list on the primary/real inode.
Kotresh HR [Fri, 21 Feb 2025 13:48:57 +0000 (19:18 +0530)]
mds/rejoin: Don't fetch the dir is already complete during rejoin
In MDCache::open_undef_inodes_dirfrags, it can so happen that
the same dir is added twice to fetch queue causing it to fetch
twice and hit the assert that dir is not complete for the
second fetch. So don't fetch if the dir is complete.
Kotresh HR [Fri, 21 Feb 2025 12:34:03 +0000 (18:04 +0530)]
mds/rejoin: Handle referent inode on MDSCache rejoin
The involves broadly following changes.
1. Add 'referent_ino' in the struct 'dn_strong' and required
encoding/decoding of the same. Noticed that the
MMDSCacheRejoin message actually isn't versioned yet.
There is a tracker [1] open for it. For now, CEPH_MDS_PROTOCOL
is bumped up as usual.
2. The following functions needs a change to construct the
in-memory referent inode from the inode number.
MDCache::rejoin_walk
- add_strong_dentry, pass referent inode number to build dn_strong
MDCache::handle_cache_rejoin_strong
- Construct referent inode from inode number if not found in memory and add_remote_dentry
MDCache::handle_cache_rejoin_ack
- Bad linkage check!, construct referent inode
MDCache::rejoin_send_acks
- add_strong_dentry, pass referent inode number to build dn_strong
MDCache::handle_cache_rejoin_weak
- add_strong_dentry, pass referent inode number as 0 to build dn_strong as it's weak rejoin
Kotresh HR [Wed, 19 Feb 2025 19:12:57 +0000 (00:42 +0530)]
mds/journal: Replay referent remote dentry
Add capability to EMetaBlob::replay to replay
the referent remote dentry journal. The replay
would take care of both remote dentry and
referent remote dentry.
Kotresh HR [Wed, 19 Feb 2025 18:48:12 +0000 (00:18 +0530)]
mds/journal: Journal referent remote dentry
Add machinery to journal the referent remote dentry.
The call to metablob's add_remote_dentry is just
adding remote dentry and not referent remote dentry
yet. This will be fixed as part of operations like
link, unlink, rename which makes use of referent
inodes.
Kotresh HR [Wed, 19 Feb 2025 16:44:30 +0000 (22:14 +0530)]
mds: Make referent inodes a optional feature
Hardlink referent inode plumbing work continues.
Add a fs option 'allow_referent_inodes'. Enabling the option,
allows the creation of referent inodes for the hardlinks
to store backtrace and viceversa. The option is disabled
by default so that it doesn't affect existing filesystems.
The option is introduced to handle upgrade as there is no
easy way to trace all the hardlinks of a file on existing
file systems to create referent inodes and update backtrace
online. So the feature is enabled only on new filesystems
Kotresh HR [Tue, 18 Feb 2025 18:49:14 +0000 (00:19 +0530)]
mds: Commit referent inode to disk
Hardlink referent inode plumbing work continues.
This patch adds the capability to commit the
referent inode created (would be part of hardlink
operation) to the disk.
The referent inode is marked with 'r'. If the
recovery tools recover the referent inode,
it's marked with 'R'
Kotresh HR [Tue, 4 Mar 2025 06:19:46 +0000 (11:49 +0530)]
mds: Store list of hardlinks on the inode of primary link
Hardlink referent inode plumbing continues. In order to
solve the snapshot hardlink lookup problem i.e., to be
able to identify all the hardlinks from a given file,
we need to maintain the list of referent inodes
associated with hardlink on the primary link. Now we
have two way link between primary link and corresponding
hardlink i.e., the hardlink points the primary link and
primary link points to all the hardlinks.
Kotresh HR [Tue, 4 Mar 2025 03:31:57 +0000 (09:01 +0530)]
mds: Store remote inode number in referent inode
Hardlink referent inode plumbing continues.
The remote inode information contains the following.
1. remote_ino - remote inode number
2. d_type
3. alternate_name
With the introduction of the referent inode as a full
blown CInode for the hardlink, these information needs
to be part of the inode. Hence add 'remote_ino' field
to the inode. The exisiting 'd_type' and 'alternate_name'
fields could be used.
Kotresh HR [Tue, 18 Feb 2025 11:17:53 +0000 (16:47 +0530)]
mds: Hardlink referent inode plumbing work
The linkage_t struct changes:
- Add referent_ino(inodeno_t) and referent_inode(CInode *) to the
linkage_t. These two fields becomes the core and identifies the
linkage as referent remote.
- Add 'is_referent_remote()' to check if it's referent remote
and modify 'is_remote()', 'is_null()' check accordingly.
- Add functions to access these fields.
Add the following new functions to link referent inode.
1. CDentry::push_projected_linkage - with referent_inode
2. CDir::link_null_referent_inode
3. CDir::link_referent_inode
Modify the following existing functions to link/unlink referent inode
1. CDentry::link_remote
2. CDentry::unlink_remote
3. CDentry::pop_projected_linakge
4. CDentry::CDentry
5. CDir::add_remote_dentry
6. CDir::link_inode_work
7. CDir::unlink_inode_work
* refs/pull/61321/head:
qa: update require-osd-release to tentacle
tools/monmaptool: bump new cluster version to X
doc/dev/release-checklists: remove ceph-container task
script/ceph-release-notes: add squid/tentacle
doc/dev/release-checklists:: mark task complete
doc/dev/release-checklist: add nightlies task
doc/dev/release-checklists: update ceph-build for tentacle
doc/dev/release-checklists: note redmine is done
qa: update to tentacle
doc/dev/release-checklist: question telemetry tentacle test
osd/OSDMap: update to tentacle
qa/workunits/cephtool/test: update to tentacle
mon/OSDMonitor: update to tentacle
common/options/global.yaml.in: update for tentacle
mon/MgrMonitor: update for tentacle
qa/standalone/mon/misc: update for tentacle
doc: update compatset for tentacle
doc: no deprecated features
include/ceph_features: add SERVER_TENTACLE feature bit
cephadm,ceph-volume: update to tentacle
doc/dev/release-checklist: add backport-create-issue
script: update backport-resolve-issue to tentacle
*: add constants and release names
ceph_release: update to tentacle
librbd: bump version
CMakeLists.txt: update VERSION
doc: remove obsolete checklist item
doc: reset for tentacle
Reviewed-by: Radoslaw Zarzynski <rzarzyns@redhat.com> Reviewed-by: Bill Scales <bill_scales@uk.ibm.com> Reviewed-by: Adam King <adking@redhat.com> Reviewed-by: Ilya Dryomov <idryomov@redhat.com>
* refs/pull/60746/head:
client: skip unexpected command replies
mgr: indicate map message is acked instead of unhandled
osdc/Objecter: convert to ms_dispatch2 for ack
client: indicate maps are acked not processed
msg: add alternate statuses for ms_dispatch2 handling
tools/cephfs_mirror: do not process maps with fast dispatch
doc: add docs for volumes interface for charmap
qa: add tests for subvolume charmap settings
pybind/mgr/volumes: wire up charmap for subvol/subvolgroup
pybind/mgr: send MDS commands through cephfs client
pybind/cephfs: wire up mds_command2
mgr: add module method to send notifications
libcephfs: add mds_command2 for asynchronous commands
mgr: excise CephFS client from mgr C++ base
mgr: use std namespace
doc: add docs for CephFS charmap config
qa: add charmap tests
qa: add helpful exceptions for attr changes
qa: ignore libicu leaks
client: add wrappings for charmap manipuluation of dentry names
client: add dir_result_t::dentry::print
win32: add libicu Windows build
CMakeLists: add boost::locale dependency for client
install-deps: unconditionally install boost libraries
test/libcephfs: update root operation return values
client: refactor all path traversals through path_walk
test/libcephfs: test parallel creates
test/libcephfs: add test for lookup failure after readdir
client: init dentry shared_gen with invalid value
client: add _lookup debugging
client: remove redundant check
client: dump InodeStat from mds
mds: encode optmetadata in InodeStat sent to clients
mds: check client features for charmap
mds: add client feature bit for charmap
mds: wire up vxattr for changing charmap
mds: inherit charmap on mkdir
mds,include: add charmap optmetadata
mds,include: add inode_t optional metadata
client: hide alternate_name from API
client: move alternate_name once
client: optimize alternate_name passing to helper
client: relocate definition
client: print dentry with alternate_name on dump
client: move inode dump to print method
mds: add debugging for encoding lease stat
mds: make encode_lease a proper method
mds: add fscrypt metadata for inode stat size
client: use DentryRef for ref counting in MetaRequest
client: add DentryRef
client: add helper for determining if a perm check is necessary
client: cache client_permissions config
client: add debugging for conf changes
client: sort configs
client/UserPerm: add print method
client: note mount parameters in debug log
client: print stat mode in octal
common: add missing op string
include/filepath: add empty path check
Matan Breizman [Sun, 2 Mar 2025 08:42:45 +0000 (08:42 +0000)]
cmake/modules/BuildISAL.cmake: set no-integrated-as on clang only
this option is only relevant to clang, gcc will fail with:
```
CMake Error at ceph/build/src/erasure-code/isa/isal_ext-prefix/src/isal_ext-stamp/isal_ext-configure-Debug-impl.cmake:19 (message):
Command failed (77):
J. Eric Ivancich [Fri, 28 Feb 2025 19:22:53 +0000 (14:22 -0500)]
doc/rgw: update dynamic resharding docs to reflect recent changes
The documentation on dynamic resharding is updated to include a) a
description of reducing the number of shards, b) related configuration
options, and c) the radosgw-admin sub-command to set a minimum number
of shards for a specific bucket.
Signed-off-by: J. Eric Ivancich <ivancich@redhat.com>
rgw: add radosgw-admin sub-command to set-min-shards for a bucket
There is now a mechansim to set the minimum number of shards when a
bucket is created, and dynamic resharding adheres to that
setting. This adds the ability to modify that minimum shard count that
exists within the bucket layout of the bucket instance
object. Example:
J. Eric Ivancich [Wed, 15 Jan 2025 16:26:59 +0000 (11:26 -0500)]
rgw: allow per-bucket minimum number of shards
Dynamic resharding can now reduce the number of shards. The code
currently has a hard-coded value of 11 as the minimum number of shards
dynamic resharding can reshard to. There may be cases where the user
wants to set an alternate minimum, such as when they have a sense of
how many objects the bucket will eventually hold.
This PR builds off of https://github.com/ceph/ceph/pull/61269 .
That PR allows the user to specify an initial number of shards during
bucket creation. This PR then takes that number to be the minimum and
saves it in the layout field of the bucket instance object
(RGWBucketInfo).
When dynamic resharding is triggered, it will use that stored value as
a minimum number of shards for resharing.
Signed-off-by: J. Eric Ivancich <ivancich@redhat.com>
as the system monotonic clock is used when the container is used
in Scrub implementation, and on some kernels there are rare cases
where the monotonic clock can go backwards, we need to tolerate
such events.
Patrick Donnelly [Fri, 28 Feb 2025 01:01:03 +0000 (20:01 -0500)]
Merge PR #61682 into main
* refs/pull/61682/head:
mds/snap: include cleanup
mds/MetricsHandler: include cleanup
mds/LogSegment: add missing includes
mds/JournalPointer: include cleanup
mds/cephfs_features: include cleanup
mds/MDSMap: include cleanup
mds/Migrator: include cleanup
mds/MDLog: include cleanup
mds/SnapServer: include cleanup
mds/MDSTableServer: include cleanup
mds/MetricAggregator: include cleanup
mds/SnapClient: include cleanup
mds/MDSPinger: include cleanup
mds/flock: include cleanup
mds/MDSAuthCaps: include cleanup
mds/QuiesceAgent: include cleanup
mds/QuiesceDb: include cleanup
mds/mdstypes: include cleanup
msg/Locker: include cleanup
mds/journal: add missing includes
mds/Server: include cleanup
mds/LogEvent: include cleanup
mds/Capability: include cleanup
mds/DamageTable: include cleanup
mds/StrayManager: include cleanup
mds/Mutation: include cleanup
mds/SnapRealm: include cleanup
mds/InoTable: add missing include
mds/ScatterLock: add missing include
mon/MDSMonitor: add missing include
mon/MDSDaemon: add missing include
Patrick Donnelly [Tue, 18 Feb 2025 17:41:52 +0000 (12:41 -0500)]
msg: add alternate statuses for ms_dispatch2 handling
Many dispatchers return false to allow other dispatchers also common messages
like MOSDMap or MFSMap. They implicitly depend on some dispatcher which is
always at the "tail" of the dispatcher queue to return "true" indicating the
msg was processed to avoid messages like:
but this cannot always happen when some libraries like the RadosClient used standalone.
So, add a variant for encapsulating other indications for how the message was
processed by dispatch2. For example, a message may be "acknowledged" but
explicitly allow other dispatchers to try processing the message.
Note: we're using a variant to avoid updating all of the ms_dispatch code to
use the sentinel classes.
Signed-off-by: Patrick Donnelly <pdonnell@ibm.com>
Patrick Donnelly [Thu, 30 Jan 2025 22:08:07 +0000 (17:08 -0500)]
mgr: excise CephFS client from mgr C++ base
Linking to the client causes two copies of the Client library to be linked in
the ceph-mgr when modules also dynamically link to libcephfs via the "cephfs"
python library. This creates problems with duplicate boost::locale.
Instead, modules should just use the "cephfs" library to send commands to the
MDS.
Signed-off-by: Patrick Donnelly <pdonnell@ibm.com>
Patrick Donnelly [Thu, 30 Jan 2025 22:05:25 +0000 (17:05 -0500)]
mgr: use std namespace
This C++ code was relying on the Client.h header to bring in these names from
the std:: namespace. A subsequent commit plans to remove that header so add
namespace qualifier now.
Signed-off-by: Patrick Donnelly <pdonnell@ibm.com>
If we add a new Boost component, the install-deps.sh script will not install it
because other Boost libraries are installed with the desired version. Just
unconditionally install. The package manager is smart enough to skip already
installed packages.
Signed-off-by: Patrick Donnelly <pdonnell@ibm.com>
Switch from EBUSY to reasonable and normal errors for operations on the root
directory. Use the Client::path_walk machinery changes from the previous commit
to catch most of them, otherwise return EINVAL (which is normal for other POSIX
fs).
Fixes: 6ed7f2364ae5507bab14c60b582929aa7b0ba400 Signed-off-by: Patrick Donnelly <pdonnell@ibm.com>
Patrick Donnelly [Thu, 14 Nov 2024 18:28:20 +0000 (13:28 -0500)]
client: refactor all path traversals through path_walk
This is groundwork to ensure that all code paths properly apply dentry name
transformations during traversal, specifically for casefolding.
Note some ancillary changes:
* readlink now correctly does lookup permission checks during path walk (previously a TODO).
* There has been some tweaking of the method signatures to accept an `const
InodeRef&`. This was convenient when passing walk_dentry_result::diri around
but the cascade in changes prompted me to leave it partially done to avoid
blowing up this changeset.
It's worth noting that this change is somewhat "half done". I had made an
effort to have all operations (whether fuse or libcephfs) call path_walk once.
This was done for Client::_mkdir and a few others. See for example:
- do_mkdirat -> _mkdir
- mksnap -> _mkdir
- ll_mkdir -> _mkdir
- mkdirs -> _mkdir (this had other significant simplification/cleanup);
however: note that we still have two calls to path_walk for any given
directory to be created. This is not a problem because:
Client::path_walk is now more tolerant of being called in a openat-style way.
The method now accepts an anchor directory and relative path (which may be a
single dentry name for ll_* calls). It is also tolerant of the directory inode
in fact referencing the target inode with relative path == "". That is a useful
property for many openat-style APIs (although largely unofficially). It also
means that if we resolve a path then we can pass the resolved Inode (directory)
/ string (Dentry name) pair to another method which may also call path_walk
with minimal replication of path walk work. This is done a few times in this
changeset.
Fixes: https://tracker.ceph.com/issues/66373 Fixes: https://tracker.ceph.com/issues/70100 Signed-off-by: Patrick Donnelly <pdonnell@ibm.com>
Patrick Donnelly [Wed, 19 Feb 2025 17:49:09 +0000 (12:49 -0500)]
client: init dentry shared_gen with invalid value
Directories and dentries are initialized with value 0 which makes detecting a
null (or placeholder) dentry created via Client::get_or_create difficult. We
already do checks to see if a dentry is invalid when the directory's shared_gen
changes so use an invalid value for these synthetic dentries until they can be
appropriately updated from Client::update_dentry_lease.
Signed-off-by: Patrick Donnelly <pdonnell@ibm.com>
projects / ceph.git / log