#include <common/DecayCounter.h>
+#include "include/common_fwd.h"
+
#include "messages/MClientReconnect.h"
#include "messages/MClientReply.h"
#include "messages/MClientRequest.h"
#include "messages/MClientReclaimReply.h"
#include "messages/MLock.h"
+#include "CInode.h"
#include "MDSRank.h"
#include "Mutation.h"
#include "MDSContext.h"
class OSDMap;
-class PerfCounters;
class LogEvent;
class EMetaBlob;
class EUpdate;
class MDLog;
struct SnapInfo;
+class MetricsHandler;
enum {
l_mdss_first = 1000,
l_mdss_dispatch_client_request,
- l_mdss_dispatch_slave_request,
+ l_mdss_dispatch_peer_request,
l_mdss_handle_client_request,
l_mdss_handle_client_session,
- l_mdss_handle_slave_request,
+ l_mdss_handle_peer_request,
l_mdss_req_create_latency,
l_mdss_req_getattr_latency,
l_mdss_req_getfilelock_latency,
l_mdss_req_symlink_latency,
l_mdss_req_unlink_latency,
l_mdss_cap_revoke_eviction,
+ l_mdss_cap_acquisition_throttle,
+ l_mdss_req_getvxattr_latency,
l_mdss_last,
};
using clock = ceph::coarse_mono_clock;
using time = ceph::coarse_mono_time;
-private:
- MDSRank *mds;
- MDCache *mdcache;
- MDLog *mdlog;
- PerfCounters *logger;
-
- // OSDMap full status, used to generate ENOSPC on some operations
- bool is_full;
-
- // State for while in reconnect
- MDSContext *reconnect_done;
- int failed_reconnects;
- bool reconnect_evicting; // true if I am waiting for evictions to complete
- // before proceeding to reconnect_gather_finish
- time reconnect_start = clock::zero();
- time reconnect_last_seen = clock::zero();
- set<client_t> client_reconnect_gather; // clients i need a reconnect msg from.
-
- feature_bitset_t supported_features;
- feature_bitset_t required_client_features;
-
- bool replay_unsafe_with_closed_session = false;
- double cap_revoke_eviction_timeout = 0;
-
- friend class MDSContinuation;
- friend class ServerContext;
- friend class ServerLogContext;
-
-public:
- bool terminating_sessions;
+ enum class RecallFlags : uint64_t {
+ NONE = 0,
+ STEADY = (1<<0),
+ ENFORCE_MAX = (1<<1),
+ TRIM = (1<<2),
+ ENFORCE_LIVENESS = (1<<3),
+ };
- explicit Server(MDSRank *m);
+ explicit Server(MDSRank *m, MetricsHandler *metrics_handler);
~Server() {
g_ceph_context->get_perfcounters_collection()->remove(logger);
delete logger;
void create_logger();
// message handler
- void dispatch(const Message::const_ref &m);
+ void dispatch(const cref_t<Message> &m);
void handle_osd_map();
return last_recall_state;
}
- void handle_client_session(const MClientSession::const_ref &m);
- void _session_logged(Session *session, uint64_t state_seq,
- bool open, version_t pv, interval_set<inodeno_t>& inos,version_t piv);
- version_t prepare_force_open_sessions(map<client_t,entity_inst_t> &cm,
- map<client_t,client_metadata_t>& cmm,
- map<client_t,pair<Session*,uint64_t> >& smap);
- void finish_force_open_sessions(const map<client_t,pair<Session*,uint64_t> >& smap,
+ void handle_client_session(const cref_t<MClientSession> &m);
+ void _session_logged(Session *session, uint64_t state_seq, bool open, version_t pv,
+ const interval_set<inodeno_t>& inos_to_free, version_t piv,
+ const interval_set<inodeno_t>& inos_to_purge, LogSegment *ls);
+ version_t prepare_force_open_sessions(std::map<client_t,entity_inst_t> &cm,
+ std::map<client_t,client_metadata_t>& cmm,
+ std::map<client_t,std::pair<Session*,uint64_t> >& smap);
+ void finish_force_open_sessions(const std::map<client_t,std::pair<Session*,uint64_t> >& smap,
bool dec_import=true);
- void flush_client_sessions(set<client_t>& client_set, MDSGatherBuilder& gather);
+ void flush_client_sessions(std::set<client_t>& client_set, MDSGatherBuilder& gather);
void finish_flush_session(Session *session, version_t seq);
void terminate_sessions();
void find_idle_sessions();
+
void kill_session(Session *session, Context *on_safe);
- size_t apply_blacklist(const std::set<entity_addr_t> &blacklist);
+ size_t apply_blocklist();
void journal_close_session(Session *session, int state, Context *on_safe);
- set<client_t> client_reclaim_gather;
size_t get_num_pending_reclaim() const { return client_reclaim_gather.size(); }
Session *find_session_by_uuid(std::string_view uuid);
- void reclaim_session(Session *session, const MClientReclaim::const_ref &m);
- void finish_reclaim_session(Session *session, const MClientReclaimReply::ref &reply=nullptr);
- void handle_client_reclaim(const MClientReclaim::const_ref &m);
+ void reclaim_session(Session *session, const cref_t<MClientReclaim> &m);
+ void finish_reclaim_session(Session *session, const ref_t<MClientReclaimReply> &reply=nullptr);
+ void handle_client_reclaim(const cref_t<MClientReclaim> &m);
void reconnect_clients(MDSContext *reconnect_done_);
- void handle_client_reconnect(const MClientReconnect::const_ref &m);
+ void handle_client_reconnect(const cref_t<MClientReconnect> &m);
void infer_supported_features(Session *session, client_metadata_t& client_metadata);
void update_required_client_features();
void reconnect_tick();
void recover_filelocks(CInode *in, bufferlist locks, int64_t client);
- enum class RecallFlags : uint64_t {
- NONE = 0,
- STEADY = (1<<0),
- ENFORCE_MAX = (1<<1),
- TRIM = (1<<2),
- ENFORCE_LIVENESS = (1<<3),
- };
std::pair<bool, uint64_t> recall_client_state(MDSGatherBuilder* gather, RecallFlags=RecallFlags::NONE);
void force_clients_readonly();
// -- requests --
- void handle_client_request(const MClientRequest::const_ref &m);
+ void handle_client_request(const cref_t<MClientRequest> &m);
+ void handle_client_reply(const cref_t<MClientReply> &m);
void journal_and_reply(MDRequestRef& mdr, CInode *tracei, CDentry *tracedn,
LogEvent *le, MDSLogContextBase *fin);
void submit_mdlog_entry(LogEvent *le, MDSLogContextBase *fin,
MDRequestRef& mdr, std::string_view event);
void dispatch_client_request(MDRequestRef& mdr);
- void perf_gather_op_latency(const MClientRequest::const_ref &req, utime_t lat);
+ void perf_gather_op_latency(const cref_t<MClientRequest> &req, utime_t lat);
void early_reply(MDRequestRef& mdr, CInode *tracei, CDentry *tracedn);
void respond_to_request(MDRequestRef& mdr, int r = 0);
- void set_trace_dist(Session *session, const MClientReply::ref &reply, CInode *in, CDentry *dn,
- snapid_t snapid,
- int num_dentries_wanted,
+ void set_trace_dist(const ref_t<MClientReply> &reply, CInode *in, CDentry *dn,
MDRequestRef& mdr);
-
- void handle_slave_request(const MMDSSlaveRequest::const_ref &m);
- void handle_slave_request_reply(const MMDSSlaveRequest::const_ref &m);
- void dispatch_slave_request(MDRequestRef& mdr);
- void handle_slave_auth_pin(MDRequestRef& mdr);
- void handle_slave_auth_pin_ack(MDRequestRef& mdr, const MMDSSlaveRequest::const_ref &ack);
+ void handle_peer_request(const cref_t<MMDSPeerRequest> &m);
+ void handle_peer_request_reply(const cref_t<MMDSPeerRequest> &m);
+ void dispatch_peer_request(MDRequestRef& mdr);
+ void handle_peer_auth_pin(MDRequestRef& mdr);
+ void handle_peer_auth_pin_ack(MDRequestRef& mdr, const cref_t<MMDSPeerRequest> &ack);
// some helpers
bool check_fragment_space(MDRequestRef& mdr, CDir *in);
+ bool check_dir_max_entries(MDRequestRef& mdr, CDir *in);
bool check_access(MDRequestRef& mdr, CInode *in, unsigned mask);
bool _check_access(Session *session, CInode *in, unsigned mask, int caller_uid, int caller_gid, int setattr_uid, int setattr_gid);
- CDir *validate_dentry_dir(MDRequestRef& mdr, CInode *diri, std::string_view dname);
- CDir *traverse_to_auth_dir(MDRequestRef& mdr, vector<CDentry*> &trace, filepath refpath);
- CDentry *prepare_null_dentry(MDRequestRef& mdr, CDir *dir, std::string_view dname, bool okexist=false);
CDentry *prepare_stray_dentry(MDRequestRef& mdr, CInode *in);
CInode* prepare_new_inode(MDRequestRef& mdr, CDir *dir, inodeno_t useino, unsigned mode,
- file_layout_t *layout=NULL);
+ const file_layout_t *layout=nullptr);
void journal_allocated_inos(MDRequestRef& mdr, EMetaBlob *blob);
void apply_allocated_inos(MDRequestRef& mdr, Session *session);
- CInode* rdlock_path_pin_ref(MDRequestRef& mdr, int n, MutationImpl::LockOpVec& lov,
- bool want_auth, bool no_want_auth=false,
- file_layout_t **layout=nullptr,
- bool no_lookup=false);
- CDentry* rdlock_path_xlock_dentry(MDRequestRef& mdr, int n,
- MutationImpl::LockOpVec& lov,
- bool okexist, bool mustexist, bool alwaysxlock,
- file_layout_t **layout=nullptr);
+ void _try_open_ino(MDRequestRef& mdr, int r, inodeno_t ino);
+ CInode* rdlock_path_pin_ref(MDRequestRef& mdr, bool want_auth,
+ bool no_want_auth=false);
+ CDentry* rdlock_path_xlock_dentry(MDRequestRef& mdr, bool create,
+ bool okexist=false, bool authexist=false,
+ bool want_layout=false);
+ std::pair<CDentry*, CDentry*>
+ rdlock_two_paths_xlock_destdn(MDRequestRef& mdr, bool xlock_srcdn);
CDir* try_open_auth_dirfrag(CInode *diri, frag_t fg, MDRequestRef& mdr);
-
// requests on existing inodes.
void handle_client_getattr(MDRequestRef& mdr, bool is_lookup);
void handle_client_lookup_ino(MDRequestRef& mdr,
void handle_client_file_setlock(MDRequestRef& mdr);
void handle_client_file_readlock(MDRequestRef& mdr);
+ bool xlock_policylock(MDRequestRef& mdr, CInode *in,
+ bool want_layout=false, bool xlock_snaplock=false);
+ CInode* try_get_auth_inode(MDRequestRef& mdr, inodeno_t ino);
void handle_client_setattr(MDRequestRef& mdr);
void handle_client_setlayout(MDRequestRef& mdr);
void handle_client_setdirlayout(MDRequestRef& mdr);
- int parse_quota_vxattr(string name, string value, quota_info_t *quota);
+ int parse_quota_vxattr(std::string name, std::string value, quota_info_t *quota);
void create_quota_realm(CInode *in);
- int parse_layout_vxattr(string name, string value, const OSDMap& osdmap,
+ int parse_layout_vxattr_json(std::string name, std::string value,
+ const OSDMap& osdmap, file_layout_t *layout);
+ int parse_layout_vxattr_string(std::string name, std::string value, const OSDMap& osdmap,
+ file_layout_t *layout);
+ int parse_layout_vxattr(std::string name, std::string value, const OSDMap& osdmap,
file_layout_t *layout, bool validate=true);
int check_layout_vxattr(MDRequestRef& mdr,
- string name,
- string value,
+ std::string name,
+ std::string value,
file_layout_t *layout);
- void handle_set_vxattr(MDRequestRef& mdr, CInode *cur,
- file_layout_t *dir_layout,
- MutationImpl::LockOpVec& lov);
- void handle_remove_vxattr(MDRequestRef& mdr, CInode *cur,
- file_layout_t *dir_layout,
- MutationImpl::LockOpVec& lov);
+ void handle_set_vxattr(MDRequestRef& mdr, CInode *cur);
+ void handle_remove_vxattr(MDRequestRef& mdr, CInode *cur);
+ void handle_client_getvxattr(MDRequestRef& mdr);
void handle_client_setxattr(MDRequestRef& mdr);
void handle_client_removexattr(MDRequestRef& mdr);
void handle_client_fsync(MDRequestRef& mdr);
+ bool is_unlink_pending(CDentry *dn);
+ void wait_for_pending_unlink(CDentry *dn, MDRequestRef& mdr);
+
+ bool is_reintegrate_pending(CDentry *dn);
+ void wait_for_pending_reintegrate(CDentry *dn, MDRequestRef& mdr);
+
// open
void handle_client_open(MDRequestRef& mdr);
void handle_client_openc(MDRequestRef& mdr); // O_CREAT variant.
// link
void handle_client_link(MDRequestRef& mdr);
- void _link_local(MDRequestRef& mdr, CDentry *dn, CInode *targeti);
+ void _link_local(MDRequestRef& mdr, CDentry *dn, CInode *targeti, SnapRealm *target_realm);
void _link_local_finish(MDRequestRef& mdr, CDentry *dn, CInode *targeti,
version_t, version_t, bool);
void _link_remote_finish(MDRequestRef& mdr, bool inc, CDentry *dn, CInode *targeti,
version_t);
- void handle_slave_link_prep(MDRequestRef& mdr);
- void _logged_slave_link(MDRequestRef& mdr, CInode *targeti, bool adjust_realm);
- void _commit_slave_link(MDRequestRef& mdr, int r, CInode *targeti);
- void _committed_slave(MDRequestRef& mdr); // use for rename, too
- void handle_slave_link_prep_ack(MDRequestRef& mdr, const MMDSSlaveRequest::const_ref &m);
- void do_link_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef& mdr);
+ void handle_peer_link_prep(MDRequestRef& mdr);
+ void _logged_peer_link(MDRequestRef& mdr, CInode *targeti, bool adjust_realm);
+ void _commit_peer_link(MDRequestRef& mdr, int r, CInode *targeti);
+ void _committed_peer(MDRequestRef& mdr); // use for rename, too
+ void handle_peer_link_prep_ack(MDRequestRef& mdr, const cref_t<MMDSPeerRequest> &m);
+ void do_link_rollback(bufferlist &rbl, mds_rank_t leader, MDRequestRef& mdr);
void _link_rollback_finish(MutationRef& mut, MDRequestRef& mdr,
- map<client_t,MClientSnap::ref>& split);
+ std::map<client_t,ref_t<MClientSnap>>& split);
// unlink
void handle_client_unlink(MDRequestRef& mdr);
void _unlink_local_finish(MDRequestRef& mdr,
CDentry *dn, CDentry *straydn,
version_t);
- bool _rmdir_prepare_witness(MDRequestRef& mdr, mds_rank_t who, vector<CDentry*>& trace, CDentry *straydn);
- void handle_slave_rmdir_prep(MDRequestRef& mdr);
- void _logged_slave_rmdir(MDRequestRef& mdr, CDentry *srcdn, CDentry *straydn);
- void _commit_slave_rmdir(MDRequestRef& mdr, int r, CDentry *straydn);
- void handle_slave_rmdir_prep_ack(MDRequestRef& mdr, const MMDSSlaveRequest::const_ref &ack);
- void do_rmdir_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef& mdr);
+ bool _rmdir_prepare_witness(MDRequestRef& mdr, mds_rank_t who, std::vector<CDentry*>& trace, CDentry *straydn);
+ void handle_peer_rmdir_prep(MDRequestRef& mdr);
+ void _logged_peer_rmdir(MDRequestRef& mdr, CDentry *srcdn, CDentry *straydn);
+ void _commit_peer_rmdir(MDRequestRef& mdr, int r, CDentry *straydn);
+ void handle_peer_rmdir_prep_ack(MDRequestRef& mdr, const cref_t<MMDSPeerRequest> &ack);
+ void do_rmdir_rollback(bufferlist &rbl, mds_rank_t leader, MDRequestRef& mdr);
void _rmdir_rollback_finish(MDRequestRef& mdr, metareqid_t reqid, CDentry *dn, CDentry *straydn);
// rename
void handle_client_renamesnap(MDRequestRef& mdr);
void _renamesnap_finish(MDRequestRef& mdr, CInode *diri, snapid_t snapid);
-
// helpers
- bool _rename_prepare_witness(MDRequestRef& mdr, mds_rank_t who, set<mds_rank_t> &witnesse,
- vector<CDentry*>& srctrace, vector<CDentry*>& dsttrace, CDentry *straydn);
+ bool _rename_prepare_witness(MDRequestRef& mdr, mds_rank_t who, std::set<mds_rank_t> &witnesse,
+ std::vector<CDentry*>& srctrace, std::vector<CDentry*>& dsttrace, CDentry *straydn);
version_t _rename_prepare_import(MDRequestRef& mdr, CDentry *srcdn, bufferlist *client_map_bl);
bool _need_force_journal(CInode *diri, bool empty);
void _rename_prepare(MDRequestRef& mdr,
EMetaBlob *metablob, bufferlist *client_map_bl,
- CDentry *srcdn, CDentry *destdn, CDentry *straydn);
+ CDentry *srcdn, CDentry *destdn, std::string_view alternate_name,
+ CDentry *straydn);
/* set not_journaling=true if you're going to discard the results --
* this bypasses the asserts to make sure we're journaling the right
* things on the right nodes */
void _rename_apply(MDRequestRef& mdr, CDentry *srcdn, CDentry *destdn, CDentry *straydn);
// slaving
- void handle_slave_rename_prep(MDRequestRef& mdr);
- void handle_slave_rename_prep_ack(MDRequestRef& mdr, const MMDSSlaveRequest::const_ref &m);
- void handle_slave_rename_notify_ack(MDRequestRef& mdr, const MMDSSlaveRequest::const_ref &m);
- void _slave_rename_sessions_flushed(MDRequestRef& mdr);
- void _logged_slave_rename(MDRequestRef& mdr, CDentry *srcdn, CDentry *destdn, CDentry *straydn);
- void _commit_slave_rename(MDRequestRef& mdr, int r, CDentry *srcdn, CDentry *destdn, CDentry *straydn);
- void do_rename_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef& mdr, bool finish_mdr=false);
+ void handle_peer_rename_prep(MDRequestRef& mdr);
+ void handle_peer_rename_prep_ack(MDRequestRef& mdr, const cref_t<MMDSPeerRequest> &m);
+ void handle_peer_rename_notify_ack(MDRequestRef& mdr, const cref_t<MMDSPeerRequest> &m);
+ void _peer_rename_sessions_flushed(MDRequestRef& mdr);
+ void _logged_peer_rename(MDRequestRef& mdr, CDentry *srcdn, CDentry *destdn, CDentry *straydn);
+ void _commit_peer_rename(MDRequestRef& mdr, int r, CDentry *srcdn, CDentry *destdn, CDentry *straydn);
+ void do_rename_rollback(bufferlist &rbl, mds_rank_t leader, MDRequestRef& mdr, bool finish_mdr=false);
void _rename_rollback_finish(MutationRef& mut, MDRequestRef& mdr, CDentry *srcdn, version_t srcdnpv,
- CDentry *destdn, CDentry *staydn, map<client_t,MClientSnap::ref> splits[2],
+ CDentry *destdn, CDentry *staydn, std::map<client_t,ref_t<MClientSnap>> splits[2],
bool finish_mdr);
void evict_cap_revoke_non_responders();
void handle_conf_change(const std::set<std::string>& changed);
+ bool terminating_sessions = false;
+
+ std::set<client_t> client_reclaim_gather;
+
+ const bufferlist& get_snap_trace(Session *session, SnapRealm *realm) const;
+ const bufferlist& get_snap_trace(client_t client, SnapRealm *realm) const;
+
private:
- void reply_client_request(MDRequestRef& mdr, const MClientReply::ref &reply);
- void flush_session(Session *session, MDSGatherBuilder *gather);
+ friend class MDSContinuation;
+ friend class ServerContext;
+ friend class ServerLogContext;
+ friend class Batch_Getattr_Lookup;
+
+ // placeholder for validation handler to store xattr specific
+ // data
+ struct XattrInfo {
+ virtual ~XattrInfo() {
+ }
+ };
+
+ struct MirrorXattrInfo : XattrInfo {
+ std::string cluster_id;
+ std::string fs_id;
+
+ static const std::string MIRROR_INFO_REGEX;
+ static const std::string CLUSTER_ID;
+ static const std::string FS_ID;
+
+ MirrorXattrInfo(std::string_view cluster_id,
+ std::string_view fs_id)
+ : cluster_id(cluster_id),
+ fs_id(fs_id) {
+ }
+ };
+
+ struct XattrOp {
+ int op;
+ std::string xattr_name;
+ const bufferlist &xattr_value;
+ int flags = 0;
+
+ std::unique_ptr<XattrInfo> xinfo;
+
+ XattrOp(int op, std::string_view xattr_name, const bufferlist &xattr_value, int flags)
+ : op(op),
+ xattr_name(xattr_name),
+ xattr_value(xattr_value),
+ flags (flags) {
+ }
+ };
+
+ struct XattrHandler {
+ const std::string xattr_name;
+ const std::string description;
+
+ // basic checks are to be done in this handler. return -errno to
+ // reject xattr request (set or remove), zero to proceed. handlers
+ // may parse xattr value for verification if needed and have an
+ // option to store custom data in XattrOp::xinfo.
+ int (Server::*validate)(CInode *cur, const InodeStoreBase::xattr_map_const_ptr xattrs,
+ XattrOp *xattr_op);
+
+ // set xattr for an inode in xattr_map
+ void (Server::*setxattr)(CInode *cur, InodeStoreBase::xattr_map_ptr xattrs,
+ const XattrOp &xattr_op);
+
+ // remove xattr for an inode from xattr_map
+ void (Server::*removexattr)(CInode *cur, InodeStoreBase::xattr_map_ptr xattrs,
+ const XattrOp &xattr_op);
+ };
+
+ inline static const std::string DEFAULT_HANDLER = "<default>";
+ static const XattrHandler xattr_handlers[];
+
+ const XattrHandler* get_xattr_or_default_handler(std::string_view xattr_name);
+
+ // generic variant to set/remove xattr in/from xattr_map
+ int xattr_validate(CInode *cur, const InodeStoreBase::xattr_map_const_ptr xattrs,
+ const std::string &xattr_name, int op, int flags);
+ void xattr_set(InodeStoreBase::xattr_map_ptr xattrs, const std::string &xattr_name,
+ const bufferlist &xattr_value);
+ void xattr_rm(InodeStoreBase::xattr_map_ptr xattrs, const std::string &xattr_name);
+
+ // default xattr handlers
+ int default_xattr_validate(CInode *cur, const InodeStoreBase::xattr_map_const_ptr xattrs,
+ XattrOp *xattr_op);
+ void default_setxattr_handler(CInode *cur, InodeStoreBase::xattr_map_ptr xattrs,
+ const XattrOp &xattr_op);
+ void default_removexattr_handler(CInode *cur, InodeStoreBase::xattr_map_ptr xattrs,
+ const XattrOp &xattr_op);
+
+ // mirror info xattr handler
+ int parse_mirror_info_xattr(const std::string &name, const std::string &value,
+ std::string &cluster_id, std::string &fs_id);
+ int mirror_info_xattr_validate(CInode *cur, const InodeStoreBase::xattr_map_const_ptr xattrs,
+ XattrOp *xattr_op);
+ void mirror_info_setxattr_handler(CInode *cur, InodeStoreBase::xattr_map_ptr xattrs,
+ const XattrOp &xattr_op);
+ void mirror_info_removexattr_handler(CInode *cur, InodeStoreBase::xattr_map_ptr xattrs,
+ const XattrOp &xattr_op);
+
+ static bool is_ceph_vxattr(std::string_view xattr_name) {
+ return xattr_name.rfind("ceph.dir.layout", 0) == 0 ||
+ xattr_name.rfind("ceph.file.layout", 0) == 0 ||
+ xattr_name.rfind("ceph.quota", 0) == 0 ||
+ xattr_name == "ceph.dir.subvolume" ||
+ xattr_name == "ceph.dir.pin" ||
+ xattr_name == "ceph.dir.pin.random" ||
+ xattr_name == "ceph.dir.pin.distributed";
+ }
+
+ static bool is_ceph_dir_vxattr(std::string_view xattr_name) {
+ return (xattr_name == "ceph.dir.layout" ||
+ xattr_name == "ceph.dir.layout.json" ||
+ xattr_name == "ceph.dir.layout.object_size" ||
+ xattr_name == "ceph.dir.layout.stripe_unit" ||
+ xattr_name == "ceph.dir.layout.stripe_count" ||
+ xattr_name == "ceph.dir.layout.pool" ||
+ xattr_name == "ceph.dir.layout.pool_name" ||
+ xattr_name == "ceph.dir.layout.pool_id" ||
+ xattr_name == "ceph.dir.layout.pool_namespace" ||
+ xattr_name == "ceph.dir.pin" ||
+ xattr_name == "ceph.dir.pin.random" ||
+ xattr_name == "ceph.dir.pin.distributed");
+ }
+
+ static bool is_ceph_file_vxattr(std::string_view xattr_name) {
+ return (xattr_name == "ceph.file.layout" ||
+ xattr_name == "ceph.file.layout.json" ||
+ xattr_name == "ceph.file.layout.object_size" ||
+ xattr_name == "ceph.file.layout.stripe_unit" ||
+ xattr_name == "ceph.file.layout.stripe_count" ||
+ xattr_name == "ceph.file.layout.pool" ||
+ xattr_name == "ceph.file.layout.pool_name" ||
+ xattr_name == "ceph.file.layout.pool_id" ||
+ xattr_name == "ceph.file.layout.pool_namespace");
+ }
+
+ static bool is_allowed_ceph_xattr(std::string_view xattr_name) {
+ // not a ceph xattr -- allow!
+ if (xattr_name.rfind("ceph.", 0) != 0) {
+ return true;
+ }
+
+ return xattr_name == "ceph.mirror.info" ||
+ xattr_name == "ceph.mirror.dirty_snap_id";
+ }
+
+ void reply_client_request(MDRequestRef& mdr, const ref_t<MClientReply> &reply);
+ void flush_session(Session *session, MDSGatherBuilder& gather);
+
+ MDSRank *mds;
+ MDCache *mdcache;
+ MDLog *mdlog;
+ PerfCounters *logger = nullptr;
+
+ // OSDMap full status, used to generate CEPHFS_ENOSPC on some operations
+ bool is_full = false;
+
+ // State for while in reconnect
+ MDSContext *reconnect_done = nullptr;
+ int failed_reconnects = 0;
+ bool reconnect_evicting = false; // true if I am waiting for evictions to complete
+ // before proceeding to reconnect_gather_finish
+ time reconnect_start = clock::zero();
+ time reconnect_last_seen = clock::zero();
+ std::set<client_t> client_reconnect_gather; // clients i need a reconnect msg from.
+ std::set<client_t> client_reconnect_denied; // clients whose reconnect msg have been denied .
+
+ feature_bitset_t supported_features;
+ feature_bitset_t supported_metric_spec;
+ feature_bitset_t required_client_features;
+
+ bool forward_all_requests_to_auth = false;
+ bool replay_unsafe_with_closed_session = false;
+ double cap_revoke_eviction_timeout = 0;
+ uint64_t max_snaps_per_dir = 100;
+ // long snapshot names have the following format: "_<SNAPSHOT-NAME>_<INODE-NUMBER>"
+ uint64_t snapshot_name_max = NAME_MAX - 1 - 1 - 13;
+ unsigned delegate_inos_pct = 0;
+ uint64_t dir_max_entries = 0;
+ int64_t bal_fragment_size_max = 0;
+
+ double inject_rename_corrupt_dentry_first = 0.0;
DecayCounter recall_throttle;
time last_recall_state;
+
+ MetricsHandler *metrics_handler;
+
+ // Cache cap acquisition throttle configs
+ uint64_t max_caps_per_client;
+ uint64_t cap_acquisition_throttle;
+ double max_caps_throttle_ratio;
+ double caps_throttle_retry_request_timeout;
+
+ size_t alternate_name_max = g_conf().get_val<Option::size_t>("mds_alternate_name_max");
+ size_t fscrypt_last_block_max_size = g_conf().get_val<Option::size_t>("mds_fscrypt_last_block_max_size");
};
static inline constexpr auto operator|(Server::RecallFlags a, Server::RecallFlags b) {
using T = std::underlying_type<Server::RecallFlags>::type;
return static_cast<T>(f) == static_cast<T>(0);
}
-
#endif
projects / ceph.git / blobdiff