* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2016, Intel Corporation.
+ * Copyright (c) 2011, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
/* file data for open files on MDS */
struct mdt_file_data {
/** portals handle must be first */
- struct portals_handle mfd_handle;
+ struct portals_handle mfd_open_handle;
/** open mode provided by client */
__u64 mfd_mode;
/** protected by med_open_lock */
/** xid of the open request */
__u64 mfd_xid;
/** old handle in replay case */
- struct lustre_handle mfd_old_handle;
+ struct lustre_handle mfd_open_handle_old;
/** point to opened object */
struct mdt_object *mfd_object;
};
struct proc_dir_entry *cdt_proc_dir; /**< cdt /proc directory */
__u64 cdt_policy; /**< policy flags */
enum cdt_states cdt_state; /**< state */
- spinlock_t cdt_state_lock; /**< cdt_state lock */
- atomic_t cdt_compound_id; /**< compound id
- * counter */
+ struct mutex cdt_state_lock; /**< cdt_state lock */
__u64 cdt_last_cookie; /**< last cookie
* allocated */
struct rw_semaphore cdt_llog_lock; /**< protect llog
* list */
struct mutex cdt_restore_lock; /**< protect restore
* list */
- cfs_time_t cdt_loop_period; /**< llog scan period */
- cfs_time_t cdt_grace_delay; /**< request grace
+ time64_t cdt_loop_period; /**< llog scan period */
+ time64_t cdt_grace_delay; /**< request grace
* delay */
- cfs_time_t cdt_active_req_timeout; /**< request timeout */
+ time64_t cdt_active_req_timeout; /**< request timeout */
__u32 cdt_default_archive_id; /**< archive id used
* when none are
* specified */
__u64 cdt_max_requests; /**< max count of started
* requests */
- atomic_t cdt_request_count; /**< current count of
- * started requests */
+ /** Current count of active requests */
+ atomic_t cdt_request_count; /** total */
+ atomic_t cdt_archive_count;
+ atomic_t cdt_restore_count;
+ atomic_t cdt_remove_count;
+
/* started requests (struct cdt_agent_req:car_cookie_hash)
* indexed by cookie */
struct cfs_hash *cdt_request_cookie_hash;
struct list_head cdt_request_list;
struct list_head cdt_agents; /**< list of register
* agents */
- struct list_head cdt_restore_hdl; /**< list of restore lock
- * handles */
+ struct list_head cdt_restore_handle_list;
/* Hash of cookies to locations of record locations in agent
* request log. */
/* Remove archive on last unlink policy */
bool cdt_remove_archive_on_last_unlink;
+
+ bool cdt_wakeup_coordinator;
};
/* mdt state flag bits */
#define MDT_FL_CFGLOG 0
#define MDT_FL_SYNCED 1
+/* possible values for mo_dom_lock */
+enum {
+ NO_DOM_LOCK_ON_OPEN = 0,
+ TRYLOCK_DOM_ON_OPEN = 1,
+ ALWAYS_DOM_LOCK_ON_OPEN = 2,
+ NUM_DOM_LOCK_ON_OPEN_MODES
+};
+
+struct mdt_statfs_cache {
+ struct obd_statfs msf_osfs;
+ __u64 msf_age;
+};
+
struct mdt_device {
/* super-class */
struct lu_device mdt_lu_dev;
struct md_device *mdt_child;
struct dt_device *mdt_bottom;
struct obd_export *mdt_bottom_exp;
+ struct local_oid_storage *mdt_los;
/** target device */
struct lu_target mdt_lut;
/*
unsigned int mo_user_xattr:1,
mo_acl:1,
mo_cos:1,
- mo_evict_tgt_nids:1;
+ mo_evict_tgt_nids:1,
+ mo_dom_read_open:1,
+ mo_migrate_hsm_allowed:1;
+ unsigned int mo_dom_lock;
} mdt_opts;
/* mdt state flags */
unsigned long mdt_state;
int mdt_max_ea_size;
+ /* preferred BRW size, decided by storage type and capability */
+ __u32 mdt_brw_size;
+
struct upcall_cache *mdt_identity_cache;
unsigned int mdt_capa_conf:1,
spinlock_t mdt_lock;
/* statfs optimization: we cache a bit */
- struct obd_statfs mdt_osfs;
- __u64 mdt_osfs_age;
+ struct mdt_statfs_cache mdt_sum_osfs;
+ struct mdt_statfs_cache mdt_osfs;
/* root squash */
struct root_squash_info mdt_squash;
#define MDT_SERVICE_WATCHDOG_FACTOR (2)
#define MDT_COS_DEFAULT (0)
+#define ENOENT_VERSION 1 /** 'virtual' version of non-existent object */
+
struct mdt_object {
struct lu_object_header mot_header;
struct lu_object mot_obj;
* attribute cache */
int mot_write_count;
spinlock_t mot_write_lock;
+ /* Lock to protect object's SOM update. */
+ struct mutex mot_som_mutex;
/* Lock to protect create_data */
struct mutex mot_lov_mutex;
/* lock to protect read/write stages for Data-on-MDT files */
#define MDT_EREMOTE_OPEN (EREMOTE + 1024)
struct mdt_reint_record {
- mdt_reint_t rr_opcode;
- const struct lustre_handle *rr_handle;
+ enum mds_reint_op rr_opcode;
+ const struct lustre_handle *rr_open_handle;
+ const struct lustre_handle *rr_lease_handle;
const struct lu_fid *rr_fid1;
const struct lu_fid *rr_fid2;
struct lu_name rr_name;
/*
* Object attributes.
*/
- struct md_attr mti_attr;
+ struct md_attr mti_attr;
+ struct md_attr mti_attr2; /* mdt_lvb.c */
/*
* Body for "habeo corpus" operations.
*/
mti_cross_ref:1,
/* big_lmm buffer was used and must be used in reply */
mti_big_lmm_used:1,
- mti_big_acl_used:1;
+ mti_big_acl_used:1,
+ mti_som_valid:1;
/* opdata for mdt_reint_open(), has the same as
* ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
struct {
struct md_attr attr;
} hsm;
- } mti_u;
+ struct {
+ struct md_attr attr;
+ } som;
+ } mti_u;
- struct lustre_handle mti_close_handle;
- loff_t mti_off;
- struct lu_buf mti_buf;
- struct lu_buf mti_big_buf;
+ struct lustre_handle mti_open_handle;
+ loff_t mti_off;
+ struct lu_buf mti_buf;
+ struct lu_buf mti_big_buf;
/* Ops object filename */
struct lu_name mti_name;
+ char mti_filename[NAME_MAX + 1];
/* per-thread values, can be re-used, may be vmalloc'd */
void *mti_big_lmm;
void *mti_big_acl;
int mti_big_aclsize;
/* should be enough to fit lustre_mdt_attrs */
char mti_xattr_buf[128];
- struct ldlm_enqueue_info mti_einfo;
+ struct ldlm_enqueue_info mti_einfo[2];
+ /* einfo used by mdt_remote_object_lock_try() */
+ struct ldlm_enqueue_info mti_remote_einfo;
struct tg_reply_data *mti_reply_data;
+
+ /* FLR: layout change API */
+ struct md_layout_change mti_layout;
};
extern struct lu_context_key mdt_thread_key;
struct hlist_node car_cookie_hash; /**< find req by cookie */
struct list_head car_request_list; /**< to chain all the req. */
atomic_t car_refcount; /**< reference counter */
- __u64 car_compound_id; /**< compound id */
__u64 car_flags; /**< request original flags */
struct obd_uuid car_uuid; /**< agent doing the req. */
__u32 car_archive_id; /**< archive id */
int car_canceled; /**< request was canceled */
- cfs_time_t car_req_start; /**< start time */
- cfs_time_t car_req_update; /**< last update time */
+ time64_t car_req_start; /**< start time */
+ time64_t car_req_update; /**< last update time */
struct hsm_action_item *car_hai; /**< req. to the agent */
struct cdt_req_progress car_progress; /**< track data mvt
* progress */
return exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE;
}
+enum {
+ LMM_NO_DOM,
+ LMM_DOM_ONLY,
+ LMM_DOM_OST
+};
+
+/* XXX Look into layout in MDT layer. This must be done in LOD. */
+static inline int mdt_lmm_dom_entry(struct lov_mds_md *lmm)
+{
+ struct lov_comp_md_v1 *comp_v1;
+ struct lov_mds_md *v1;
+ __u32 off;
+ int i;
+
+ if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
+ return LMM_NO_DOM;
+
+ comp_v1 = (struct lov_comp_md_v1 *)lmm;
+ off = le32_to_cpu(comp_v1->lcm_entries[0].lcme_offset);
+ v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
+
+ /* DoM entry is the first entry always */
+ if (lov_pattern(le32_to_cpu(v1->lmm_pattern)) != LOV_PATTERN_MDT)
+ return LMM_NO_DOM;
+
+ for (i = 1; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
+ int j;
+
+ off = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
+ v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
+
+ for (j = 0; j < le16_to_cpu(v1->lmm_stripe_count); j++) {
+ /* if there is any object on OST */
+ if (le32_to_cpu(v1->lmm_objects[j].l_ost_idx) !=
+ (__u32)-1UL)
+ return LMM_DOM_OST;
+ }
+ }
+ return LMM_DOM_ONLY;
+}
+
+static inline bool mdt_lmm_is_flr(struct lov_mds_md *lmm)
+{
+ struct lov_comp_md_v1 *lcm = (typeof(lcm))lmm;
+
+ return le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1 &&
+ le16_to_cpu(lcm->lcm_mirror_count) > 0;
+}
+
+static inline bool mdt_is_sum_statfs_client(struct obd_export *exp)
+{
+ return exp_connect_flags(exp) & OBD_CONNECT_FLAGS2 &&
+ exp_connect_flags2(exp) & OBD_CONNECT2_SUM_STATFS;
+}
+
__u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag);
void mdt_set_disposition(struct mdt_thread_info *info,
struct ldlm_reply *rep, __u64 op_flag);
struct mdt_object *o, const struct lu_fid *fid,
struct lustre_handle *lh,
enum ldlm_mode mode, __u64 ibits, bool cache);
+int mdt_reint_striped_lock(struct mdt_thread_info *info,
+ struct mdt_object *o,
+ struct mdt_lock_handle *lh,
+ __u64 ibits,
+ struct ldlm_enqueue_info *einfo,
+ bool cos_incompat);
+void mdt_reint_striped_unlock(struct mdt_thread_info *info,
+ struct mdt_object *o,
+ struct mdt_lock_handle *lh,
+ struct ldlm_enqueue_info *einfo, int decref);
enum mdt_name_flags {
MNF_FIX_ANON = 1,
struct mdt_object *o, struct lu_nodemap *nodemap);
#endif
void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
- const struct lu_attr *attr, const struct lu_fid *fid);
-
+ const struct lu_attr *attr, const struct lu_fid *fid);
+int mdt_pack_size2body(struct mdt_thread_info *info,
+ const struct lu_fid *fid, struct lustre_handle *lh);
int mdt_getxattr(struct mdt_thread_info *info);
int mdt_reint_setxattr(struct mdt_thread_info *info,
struct mdt_lock_handle *lh);
__u64 mode);
int mdt_reint_open(struct mdt_thread_info *info,
- struct mdt_lock_handle *lhc);
+ struct mdt_lock_handle *lhc);
-struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
- const struct lustre_handle *handle,
- bool is_replay);
+struct mdt_file_data *mdt_open_handle2mfd(struct mdt_export_data *med,
+ const struct lustre_handle *open_handle,
+ bool is_replay);
int mdt_get_info(struct tgt_session_info *tsi);
int mdt_attr_get_complex(struct mdt_thread_info *info,
const char *name);
int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
struct md_attr *ma, const char *name);
+int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
+ struct lu_fid *pfid);
int mdt_write_get(struct mdt_object *o);
void mdt_write_put(struct mdt_object *o);
int mdt_write_read(struct mdt_object *o);
int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
struct md_attr *);
void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
+int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
+ struct md_layout_change *spec);
+int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt);
struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
const struct lu_buf *mdt_buf_const(const struct lu_env *env,
struct mdt_thread_info *mti);
void mdt_thread_info_fini(struct mdt_thread_info *mti);
struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
+void mdt_intent_fixup_resent(struct mdt_thread_info *info,
+ struct ldlm_lock *new_lock,
+ struct mdt_lock_handle *lh, __u64 flags);
+int mdt_intent_lock_replace(struct mdt_thread_info *info,
+ struct ldlm_lock **lockp,
+ struct mdt_lock_handle *lh,
+ __u64 flags, int result);
int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
const struct md_hsm *mh);
struct linkea_data *ldata);
int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
struct mdt_body *repbody);
+int mdt_remote_dir_permission(struct mdt_thread_info *info);
static inline struct mdt_device *mdt_dev(struct lu_device *d)
{
llog_cb_t cb, void *data, u32 start_cat_idx,
u32 start_rec_idx, int rw);
int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
- __u64 compound_id, __u32 archive_id,
- __u64 flags, struct hsm_action_item *hai);
+ __u32 archive_id, __u64 flags,
+ struct hsm_action_item *hai);
int mdt_agent_record_update(const struct lu_env *env, struct mdt_device *mdt,
struct hsm_record_update *updates,
unsigned int updates_count);
struct obd_uuid *uuid);
int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
bool purge);
-int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
- struct hsm_progress_kernel *pgs);
/* mdt/mdt_hsm_cdt_client.c */
int mdt_hsm_add_actions(struct mdt_thread_info *info,
struct hsm_action_list *hal);
-int mdt_hsm_get_actions(struct mdt_thread_info *mti,
- struct hsm_action_list *hal);
+int mdt_hsm_get_action(struct mdt_thread_info *mti,
+ const struct lu_fid *fid,
+ enum hsm_copytool_action *action,
+ enum agent_req_status *status,
+ struct hsm_extent *extent);
bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
const struct lu_fid *fid);
/* mdt/mdt_hsm_cdt_requests.c */
extern struct cfs_hash_ops cdt_agent_record_hash_ops;
extern const struct file_operations mdt_hsm_active_requests_fops;
void dump_requests(char *prefix, struct coordinator *cdt);
-struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
- __u64 flags, struct obd_uuid *uuid,
+struct cdt_agent_req *mdt_cdt_alloc_request(__u32 archive_id, __u64 flags,
+ struct obd_uuid *uuid,
struct hsm_action_item *hai);
void mdt_cdt_free_request(struct cdt_agent_req *car);
int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
/* mdt/mdt_coordinator.c */
void mdt_hsm_dump_hal(int level, const char *prefix,
struct hsm_action_list *hal);
-struct cdt_restore_handle *mdt_hsm_restore_hdl_find(struct coordinator *cdt,
- const struct lu_fid *fid);
+int cdt_restore_handle_add(struct mdt_thread_info *mti, struct coordinator *cdt,
+ const struct lu_fid *fid,
+ const struct hsm_extent *he);
+struct cdt_restore_handle *cdt_restore_handle_find(struct coordinator *cdt,
+ const struct lu_fid *fid);
+void cdt_restore_handle_del(struct mdt_thread_info *mti,
+ struct coordinator *cdt, const struct lu_fid *fid);
/* coordinator management */
int mdt_hsm_cdt_init(struct mdt_device *mdt);
int mdt_hsm_cdt_stop(struct mdt_device *mdt);
u32 archive_id, u64 rq_flags,
const struct md_hsm *hsm);
int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
- struct hsm_progress_kernel *pgs,
- const int update_record);
+ struct hsm_progress_kernel *pgs);
int mdt_close_swap_layouts(struct mdt_thread_info *info,
struct mdt_object *o, struct md_attr *ma);
int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
+static int mdt_dom_glimpse_ast(struct ldlm_lock *lock, void *reqp)
+{
+ return -ELDLM_NO_LOCK_DATA;
+}
+
/* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
-static inline int mdt_fid_lock(struct ldlm_namespace *ns,
+static inline int mdt_fid_lock(const struct lu_env *env,
+ struct ldlm_namespace *ns,
struct lustre_handle *lh, enum ldlm_mode mode,
union ldlm_policy_data *policy,
const struct ldlm_res_id *res_id,
__u64 flags, const __u64 *client_cookie)
{
int rc;
+ bool glimpse = policy->l_inodebits.bits & MDS_INODELOCK_DOM;
LASSERT(ns != NULL);
LASSERT(lh != NULL);
- rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
+ rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS, policy,
mode, &flags, mdt_blocking_ast,
- ldlm_completion_ast, NULL, NULL, 0,
- LVB_T_NONE, client_cookie, lh);
+ ldlm_completion_ast,
+ glimpse ? mdt_dom_glimpse_ast : NULL,
+ NULL, 0, LVB_T_NONE, client_cookie, lh);
return rc == ELDLM_OK ? 0 : -EIO;
}
return mdt_dlm_lock_modes[mode];
}
+/* mdt_som.c */
+int mdt_set_som(struct mdt_thread_info *info, struct mdt_object *obj,
+ enum lustre_som_flags flag, __u64 size, __u64 blocks);
+int mdt_get_som(struct mdt_thread_info *info, struct mdt_object *obj,
+ struct md_attr *ma);
+int mdt_lsom_downgrade(struct mdt_thread_info *info, struct mdt_object *obj);
+int mdt_lsom_update(struct mdt_thread_info *info, struct mdt_object *obj,
+ bool truncate);
+
/* mdt_lvb.c */
extern struct ldlm_valblock_ops mdt_lvbo;
+int mdt_dom_lvb_is_valid(struct ldlm_resource *res);
+int mdt_dom_lvbo_update(const struct lu_env *env, struct ldlm_resource *res,
+ struct ldlm_lock *lock, struct ptlrpc_request *req,
+ bool increase_only);
-void mdt_enable_cos(struct mdt_device *, int);
+void mdt_enable_cos(struct mdt_device *dev, bool enable);
int mdt_cos_is_enabled(struct mdt_device *);
/* lprocfs stuff */
-enum {
- LPROC_MDT_OPEN = 0,
+enum mdt_stat_idx {
+ LPROC_MDT_OPEN,
LPROC_MDT_CLOSE,
LPROC_MDT_MKNOD,
LPROC_MDT_LINK,
LPROC_MDT_IO_PUNCH,
LPROC_MDT_LAST,
};
+
void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
void mdt_stats_counter_init(struct lprocfs_stats *stats);
int mdt_procfs_init(struct mdt_device *mdt, const char *name);
struct niobuf_remote *rnb, int npages,
struct niobuf_local *lnb, int old_rc);
int mdt_punch_hdl(struct tgt_session_info *tsi);
+int mdt_glimpse_enqueue(struct mdt_thread_info *mti, struct ldlm_namespace *ns,
+ struct ldlm_lock **lockp, __u64 flags);
+int mdt_brw_enqueue(struct mdt_thread_info *info, struct ldlm_namespace *ns,
+ struct ldlm_lock **lockp, __u64 flags);
+int mdt_dom_read_on_open(struct mdt_thread_info *mti, struct mdt_device *mdt,
+ struct lustre_handle *lh);
+void mdt_dom_discard_data(struct mdt_thread_info *info,
+ const struct lu_fid *fid);
+int mdt_dom_disk_lvbo_update(const struct lu_env *env, struct mdt_object *mo,
+ struct ldlm_resource *res, bool increase_only);
+void mdt_dom_obj_lvb_update(const struct lu_env *env, struct mdt_object *mo,
+ bool increase_only);
+int mdt_dom_lvb_alloc(struct ldlm_resource *res);
+
+static inline bool mdt_dom_check_for_discard(struct mdt_thread_info *mti,
+ struct mdt_object *mo)
+{
+ return lu_object_is_dying(&mo->mot_header) &&
+ S_ISREG(lu_object_attr(&mo->mot_obj));
+}
+
+int mdt_dom_object_size(const struct lu_env *env, struct mdt_device *mdt,
+ const struct lu_fid *fid, struct mdt_body *mb,
+ bool dom_lock);
+bool mdt_dom_client_has_lock(struct mdt_thread_info *info,
+ const struct lu_fid *fid);
+void mdt_hp_brw(struct tgt_session_info *tsi);
+void mdt_hp_punch(struct tgt_session_info *tsi);
+int mdt_data_version_get(struct tgt_session_info *tsi);
/* grants */
long mdt_grant_connect(const struct lu_env *env, struct obd_export *exp,
u64 want, bool conservative);
+extern struct kmem_cache *ldlm_glimpse_work_kmem;
#endif /* _MDT_INTERNAL_H */