#include <lustre_import.h>
#include <lustre_handles.h>
#include <lustre_export.h> /* for obd_export, for LDLM_DEBUG */
+#include <interval_tree.h> /* for interval_node{}, ldlm_extent */
struct obd_ops;
struct obd_device;
#define LDLM_DEFAULT_LRU_SIZE (100 * num_online_cpus())
#define LDLM_DEFAULT_MAX_ALIVE (cfs_time_seconds(36000))
+#define LDLM_CTIME_AGE_LIMIT (10)
typedef enum {
ELDLM_OK = 0,
* w/o involving separate thread. in order to decrease cs rate */
#define LDLM_FL_ATOMIC_CB 0x4000000
+/* Cancel lock asynchronously. See ldlm_cli_cancel_unused_resource. */
+#define LDLM_FL_ASYNC 0x8000000
+
/* It may happen that a client initiate 2 operations, e.g. unlink and mkdir,
* such that server send blocking ast for conflict locks to this client for
* the 1st operation, whereas the 2nd operation has canceled this lock and
#define LDLM_FL_BL_AST 0x10000000
#define LDLM_FL_BL_DONE 0x20000000
-/* Cancel lock asynchronously. See ldlm_cli_cancel_unused_resource. */
-#define LDLM_FL_ASYNC 0x40000000
+/* measure lock contention and return -EUSERS if locking contention is high */
+#define LDLM_FL_DENY_ON_CONTENTION 0x40000000
/* The blocking callback is overloaded to perform two functions. These flags
* indicate which operation should be performed. */
#define LDLM_CB_BLOCKING 1
#define LDLM_CB_CANCELING 2
-/* position flag of skip list pointers */
-#define LDLM_SL_HEAD(skip_list) ((skip_list)->next != NULL)
-#define LDLM_SL_TAIL(skip_list) ((skip_list)->prev != NULL)
-#define LDLM_SL_EMPTY(skip_list) ((skip_list)->next == NULL && \
- (skip_list)->prev == NULL)
-
/* compatibility matrix */
#define LCK_COMPAT_EX LCK_NL
#define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
}
-static inline int lockmode_compat(ldlm_mode_t exist, ldlm_mode_t new)
+static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
{
- return (lck_compat_array[exist] & new);
+ return (lck_compat_array[exist_mode] & new_mode);
}
/*
#define LDLM_POOLS_FAST_SLV_CHANGE (50)
struct ldlm_pool {
- /* Common pool fields */
- cfs_proc_dir_entry_t *pl_proc_dir; /* Pool proc directory. */
- char pl_name[100]; /* Pool name, should be long
- * enough to contain complex
- * proc entry name. */
- spinlock_t pl_lock; /* Lock for protecting slv/clv
- * updates. */
- atomic_t pl_limit; /* Number of allowed locks in
- * in pool, both, client and
- * server side. */
- atomic_t pl_granted; /* Number of granted locks. */
- atomic_t pl_grant_rate; /* Grant rate per T. */
- atomic_t pl_cancel_rate; /* Cancel rate per T. */
- atomic_t pl_grant_speed; /* Grant speed (GR-CR) per T. */
- __u64 pl_server_lock_volume; /* Server lock volume.
- * Protected by pl_lock */
- atomic_t pl_lock_volume_factor; /* Lock volume factor. */
-
- time_t pl_recalc_time; /* Time when last slv from
- * server was obtained. */
- struct ldlm_pool_ops *pl_ops; /* Recalc and shrink ops. */
-
- int pl_grant_plan; /* Planned number of granted
- * locks for next T. */
- int pl_grant_step; /* Grant plan step for next
- * T. */
-
- struct lprocfs_stats *pl_stats; /* Pool statistics. */
+ /**
+ * Pool proc directory.
+ */
+ cfs_proc_dir_entry_t *pl_proc_dir;
+ /**
+ * Pool name, should be long enough to contain compound proc entry name.
+ */
+ char pl_name[100];
+ /**
+ * Lock for protecting slv/clv updates.
+ */
+ spinlock_t pl_lock;
+ /**
+ * Number of allowed locks in in pool, both, client and server side.
+ */
+ atomic_t pl_limit;
+ /**
+ * Number of granted locks in
+ */
+ atomic_t pl_granted;
+ /**
+ * Grant rate per T.
+ */
+ atomic_t pl_grant_rate;
+ /**
+ * Cancel rate per T.
+ */
+ atomic_t pl_cancel_rate;
+ /**
+ * Grant speed (GR-CR) per T.
+ */
+ atomic_t pl_grant_speed;
+ /**
+ * Server lock volume. Protected by pl_lock.
+ */
+ __u64 pl_server_lock_volume;
+ /**
+ * Current biggest client lock volume. Protected by pl_lock.
+ */
+ __u64 pl_client_lock_volume;
+ /**
+ * Lock volume factor. SLV on client is calculated as following:
+ * server_slv * lock_volume_factor.
+ */
+ atomic_t pl_lock_volume_factor;
+ /**
+ * Time when last slv from server was obtained.
+ */
+ time_t pl_recalc_time;
+ /**
+ * Recalc and shrink ops.
+ */
+ struct ldlm_pool_ops *pl_ops;
+ /**
+ * Planned number of granted locks for next T.
+ */
+ int pl_grant_plan;
+ /**
+ * Grant plan step for next T.
+ */
+ int pl_grant_step;
+ /**
+ * Pool statistics.
+ */
+ struct lprocfs_stats *pl_stats;
};
typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
LDLM_NAMESPACE_MODEST = 1 << 1
} ldlm_appetite_t;
-/* Default value for ->ns_shrink_thumb. If lock is not extent one its cost
+/*
+ * Default value for ->ns_shrink_thumb. If lock is not extent one its cost
* is one page. Here we have 256 pages which is 1M on i386. Thus by default
* all extent locks which have more than 1M long extent will be kept in lru,
- * others (including ibits locks) will be canceled on memory pressure event. */
+ * others (including ibits locks) will be canceled on memory pressure event.
+ */
#define LDLM_LOCK_SHRINK_THUMB 256
+/*
+ * Default values for the "max_nolock_size", "contention_time" and
+ * "contended_locks" namespace tunables.
+ */
+#define NS_DEFAULT_MAX_NOLOCK_BYTES 0
+#define NS_DEFAULT_CONTENTION_SECONDS 2
+#define NS_DEFAULT_CONTENDED_LOCKS 32
+
struct ldlm_namespace {
+ /**
+ * Namespace name. Used for logging, etc.
+ */
char *ns_name;
- ldlm_side_t ns_client; /* is this a client-side lock tree? */
- __u64 ns_connect_flags; /* ns connect flags supported
- * by server (may be changed via proc,
- * lru resize may be disabled/enabled) */
- __u64 ns_orig_connect_flags; /* client side orig connect
- * flags supported by server */
- struct list_head *ns_hash; /* hash table for ns */
+
+ /**
+ * Is this a client-side lock tree?
+ */
+ ldlm_side_t ns_client;
+
+ /**
+ * Namespce connect flags supported by server (may be changed via proc,
+ * lru resize may be disabled/enabled).
+ */
+ __u64 ns_connect_flags;
+
+ /**
+ * Client side orig connect flags supported by server.
+ */
+ __u64 ns_orig_connect_flags;
+
+ /**
+ * Hash table for namespace.
+ */
+ struct list_head *ns_hash;
spinlock_t ns_hash_lock;
- __u32 ns_refcount; /* count of resources in the hash */
- struct list_head ns_root_list; /* all root resources in ns */
- struct list_head ns_list_chain; /* position in global NS list */
- struct list_head ns_unused_list; /* all root resources in ns */
+ /**
+ * Count of resources in the hash.
+ */
+ __u32 ns_refcount;
+
+ /**
+ * All root resources in namespace.
+ */
+ struct list_head ns_root_list;
+
+ /**
+ * Position in global namespace list.
+ */
+ struct list_head ns_list_chain;
+
+ /**
+ * All root resources in namespace.
+ */
+ struct list_head ns_unused_list;
int ns_nr_unused;
spinlock_t ns_unused_lock;
unsigned int ns_max_unused;
unsigned int ns_max_age;
+
+ /**
+ * Seconds.
+ */
+ unsigned int ns_ctime_age_limit;
- /* Lower limit to number of pages in lock to keep it in cache */
+ /**
+ * Lower limit to number of pages in lock to keep it in cache.
+ */
unsigned int ns_shrink_thumb;
- cfs_time_t ns_next_dump; /* next debug dump, jiffies */
+
+ /**
+ * Next debug dump, jiffies.
+ */
+ cfs_time_t ns_next_dump;
atomic_t ns_locks;
__u64 ns_resources;
cfs_waitq_t ns_waitq;
struct ldlm_pool ns_pool;
ldlm_appetite_t ns_appetite;
+
+ /**
+ * If more than @ns_contented_locks found, the resource considered
+ * as contended.
+ */
+ unsigned ns_contended_locks;
+
+ /**
+ * The resource remembers contended state during @ns_contention_time,
+ * in seconds.
+ */
+ unsigned ns_contention_time;
+
+ /**
+ * Limit size of nolock requests, in bytes.
+ */
+ unsigned ns_max_nolock_size;
+
+ /**
+ * Backward link to obd, required for ldlm pool to store new SLV.
+ */
+ struct obd_device *ns_obd;
+
+ struct adaptive_timeout ns_at_estimate;/* estimated lock callback time*/
};
static inline int ns_is_client(struct ldlm_namespace *ns)
void *data);
typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
+/* Interval node data for each LDLM_EXTENT lock */
+struct ldlm_interval {
+ struct interval_node li_node; /* node for tree mgmt */
+ struct list_head li_group; /* the locks which have the same
+ * policy - group of the policy */
+};
+#define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
+
+/* the interval tree must be accessed inside the resource lock. */
+struct ldlm_interval_tree {
+ /* tree size, this variable is used to count
+ * granted PW locks in ldlm_extent_policy()*/
+ int lit_size;
+ ldlm_mode_t lit_mode; /* lock mode */
+ struct interval_node *lit_root; /* actually ldlm_interval */
+};
+
struct ldlm_lock {
struct portals_handle l_handle; // must be first in the structure
atomic_t l_refc;
/* protected by ns_hash_lock. FIXME */
struct list_head l_lru;
- /* protected by lr_lock */
- struct list_head l_res_link; // position in one of three res lists
+ /* protected by lr_lock, linkage to resource's lock queues */
+ struct list_head l_res_link;
- struct list_head l_sl_mode; // skip pointer for request mode
- struct list_head l_sl_policy; // skip pointer for inodebits
+ struct ldlm_interval *l_tree_node; /* tree node for ldlm_extent */
/* protected by led_lock */
struct list_head l_export_chain; // per-export chain of locks
void *l_lvb_data; /* an LVB received during */
void *l_lvb_swabber; /* an enqueue */
void *l_ast_data;
+ spinlock_t l_extents_list_lock;
+ struct list_head l_extents_list;
+
+ struct list_head l_cache_locks_list;
/* Server-side-only members */
struct list_head l_cp_ast;
struct ldlm_lock *l_blocking_lock;
int l_bl_ast_run;
+
+ /* protected by lr_lock, linkages to "skip lists" */
+ struct list_head l_sl_mode;
+ struct list_head l_sl_policy;
};
struct ldlm_resource {
struct ldlm_res_id lr_name;
atomic_t lr_refcount;
+ struct ldlm_interval_tree lr_itree[LCK_MODE_NUM]; /* interval trees*/
+
/* Server-side-only lock value block elements */
struct semaphore lr_lvb_sem;
__u32 lr_lvb_len;
void *lr_lvb_data;
+
+ /* when the resource was considered as contended */
+ cfs_time_t lr_contention_time;
};
struct ldlm_ast_work {
struct ldlm_enqueue_info {
__u32 ei_type; /* Type of the lock being enqueued. */
__u32 ei_mode; /* Mode of the lock being enqueued. */
- void *ei_cb_bl; /* Different callbacks for lock handling (blocking, */
- void *ei_cb_cp; /* completion, glimpse) */
- void *ei_cb_gl;
+ void *ei_cb_bl; /* blocking lock callback */
+ void *ei_cb_cp; /* lock completion callback */
+ void *ei_cb_gl; /* lock glimpse callback */
void *ei_cbdata; /* Data to be passed into callbacks. */
};
int ldlm_refresh_waiting_lock(struct ldlm_lock *lock);
void ldlm_revoke_export_locks(struct obd_export *exp);
int ldlm_get_ref(void);
-void ldlm_put_ref(int force);
+void ldlm_put_ref(void);
/* ldlm_lock.c */
ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
void ldlm_lock_allow_match(struct ldlm_lock *lock);
+int ldlm_lock_fast_match(struct ldlm_lock *, int, obd_off, obd_off, void **);
+void ldlm_lock_fast_release(void *, int);
ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
const struct ldlm_res_id *, ldlm_type_t type,
ldlm_policy_data_t *, ldlm_mode_t mode,
void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
/* resource.c */
-struct ldlm_namespace *ldlm_namespace_new(char *name, ldlm_side_t client,
- ldlm_appetite_t apt);
+struct ldlm_namespace *
+ldlm_namespace_new(struct obd_device *obd, char *name,
+ ldlm_side_t client, ldlm_appetite_t apt);
int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
-int ldlm_namespace_free(struct ldlm_namespace *ns, int force);
-void ldlm_namespace_move(struct ldlm_namespace *ns, ldlm_side_t client);
-struct ldlm_namespace *ldlm_namespace_first(ldlm_side_t client);
+void ldlm_namespace_free(struct ldlm_namespace *ns,
+ struct obd_import *imp, int force);
+void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
+void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
+void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
+struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
+void ldlm_namespace_get_locked(struct ldlm_namespace *ns);
+void ldlm_namespace_put_locked(struct ldlm_namespace *ns, int wakeup);
void ldlm_namespace_get(struct ldlm_namespace *ns);
void ldlm_namespace_put(struct ldlm_namespace *ns, int wakeup);
-void ldlm_namespace_get_nolock(struct ldlm_namespace *ns);
-void ldlm_namespace_put_nolock(struct ldlm_namespace *ns, int wakeup);
int ldlm_proc_setup(void);
#ifdef LPROCFS
void ldlm_proc_cleanup(void);
void ldlm_pool_fini(struct ldlm_pool *pl);
int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
int ldlm_pool_recalc(struct ldlm_pool *pl);
+__u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
__u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
+__u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
__u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
+void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);