X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fofd%2Fofd_internal.h;h=28c7d18e34903307fab4fb40e8618c6bf4c13c85;hb=6e6357dbf9a14aaea459f460dbe4f93e52c814d4;hp=827834fddbd037d30363d70a568a23507b447e48;hpb=b58a81067237f398f735f85b2211431cfc37493a;p=fs%2Flustre-release.git

diff --git a/lustre/ofd/ofd_internal.h b/lustre/ofd/ofd_internal.h
index 827834f..28c7d18e 100644
--- a/lustre/ofd/ofd_internal.h
+++ b/lustre/ofd/ofd_internal.h
@@ -27,7 +27,7 @@
  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2012, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -48,9 +48,13 @@
 #define OFD_ROCOMPAT_SUPP (0)
 #define OFD_INCOMPAT_SUPP (OBD_INCOMPAT_GROUPS | OBD_INCOMPAT_OST | \
 			   OBD_INCOMPAT_COMMON_LR)
-#define OFD_MAX_GROUPS	256
 #define OFD_PRECREATE_BATCH_DEFAULT (FILTER_SUBDIR_COUNT * 4)
 
+/* on small filesystems we should not precreate too many objects in
+ * a single transaction, otherwise we can overflow transactions */
+#define OFD_PRECREATE_SMALL_FS		(1024ULL * 1024 * 1024)
+#define OFD_PRECREATE_BATCH_SMALL	8
+
 /* Limit the returned fields marked valid to those that we actually might set */
 #define OFD_VALID_FLAGS (LA_TYPE | LA_MODE | LA_SIZE | LA_BLOCKS | \
 			 LA_BLKSIZE | LA_ATIME | LA_MTIME | LA_CTIME)
@@ -87,10 +91,31 @@ static inline void ofd_counter_incr(struct obd_export *exp, int opcode,
 				    char *jobid, long amount)
 {
 	if (exp->exp_obd && exp->exp_obd->u.obt.obt_jobstats.ojs_hash &&
-	    (exp->exp_connect_flags & OBD_CONNECT_JOBSTATS))
+	    (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS))
 		lprocfs_job_stats_log(exp->exp_obd, jobid, opcode, amount);
+
+	if (exp->exp_nid_stats != NULL &&
+	    exp->exp_nid_stats->nid_stats != NULL) {
+		if (opcode == LPROC_OFD_STATS_READ)
+			lprocfs_counter_add(exp->exp_nid_stats->nid_stats,
+					    LPROC_OFD_READ_BYTES, amount);
+		else if (opcode == LPROC_OFD_STATS_WRITE)
+			lprocfs_counter_add(exp->exp_nid_stats->nid_stats,
+					    LPROC_OFD_WRITE_BYTES, amount);
+	}
 }
 
+struct ofd_seq {
+	cfs_list_t		os_list;
+	obd_id			os_last_oid;
+	obd_seq			os_seq;
+	spinlock_t		os_last_oid_lock;
+	struct mutex		os_create_lock;
+	cfs_atomic_t		os_refc;
+	struct dt_object	*os_lastid_obj;
+	unsigned long		os_destroys_in_progress:1;
+};
+
 struct ofd_device {
 	struct dt_device	 ofd_dt_dev;
 	struct dt_device	*ofd_osd;
@@ -104,21 +129,18 @@ struct ofd_device {
 
 	/* last_rcvd file */
 	struct lu_target	 ofd_lut;
-	struct dt_object	*ofd_last_group_file;
 	struct dt_object	*ofd_health_check_file;
 
 	int			 ofd_subdir_count;
 
-	int			 ofd_max_group;
-	obd_id			 ofd_last_objids[OFD_MAX_GROUPS];
-	cfs_mutex_t		 ofd_create_locks[OFD_MAX_GROUPS];
-	struct dt_object	*ofd_lastid_obj[OFD_MAX_GROUPS];
-	cfs_spinlock_t		 ofd_objid_lock;
-	unsigned long		 ofd_destroys_in_progress;
-	int			 ofd_precreate_batch;
+	cfs_list_t		ofd_seq_list;
+	rwlock_t		ofd_seq_list_lock;
+	int			ofd_seq_count;
+	int			ofd_precreate_batch;
+	spinlock_t		ofd_batch_lock;
 
 	/* protect all statfs-related counters */
-	cfs_spinlock_t		 ofd_osfs_lock;
+	spinlock_t		 ofd_osfs_lock;
 	/* statfs optimization: we cache a bit  */
 	struct obd_statfs	 ofd_osfs;
 	__u64			 ofd_osfs_age;
@@ -135,7 +157,7 @@ struct ofd_device {
 
 	/* grants: all values in bytes */
 	/* grant lock to protect all grant counters */
-	cfs_spinlock_t		 ofd_grant_lock;
+	spinlock_t		 ofd_grant_lock;
 	/* total amount of dirty data reported by clients in incoming obdo */
 	obd_size		 ofd_tot_dirty;
 	/* sum of filesystem space granted to clients for async writes */
@@ -154,7 +176,7 @@ struct ofd_device {
 	int			 ofd_fmd_max_num; /* per ofd ofd_mod_data */
 	cfs_duration_t		 ofd_fmd_max_age; /* time to fmd expiry */
 
-	cfs_spinlock_t		 ofd_flags_lock;
+	spinlock_t		 ofd_flags_lock;
 	unsigned long		 ofd_raid_degraded:1,
 				 /* sync journal on writes */
 				 ofd_syncjournal:1,
@@ -163,6 +185,7 @@ struct ofd_device {
 				 /* shall we grant space to clients not
 				  * supporting OBD_CONNECT_GRANT_PARAM? */
 				 ofd_grant_compat_disable:1;
+	struct seq_server_site	 ofd_seq_site;
 };
 
 static inline struct ofd_device *ofd_dev(struct lu_device *d)
@@ -295,6 +318,7 @@ struct ofd_thread_info {
 
 	/* Space used by the I/O, used by grant code */
 	unsigned long			 fti_used;
+	struct ost_lvb			 fti_lvb;
 };
 
 extern void target_recovery_fini(struct obd_device *obd);
@@ -317,15 +341,21 @@ int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
 			int *from_cache);
 
 /* ofd_fs.c */
-obd_id ofd_last_id(struct ofd_device *ofd, obd_seq seq);
-void ofd_last_id_set(struct ofd_device *ofd, obd_id id, obd_seq seq);
-int ofd_last_id_write(const struct lu_env *env, struct ofd_device *ofd,
-		      obd_seq seq);
-int ofd_group_load(const struct lu_env *env, struct ofd_device *ofd, int);
+obd_id ofd_seq_last_oid(struct ofd_seq *oseq);
+void ofd_seq_last_oid_set(struct ofd_seq *oseq, obd_id id);
+int ofd_seq_last_oid_write(const struct lu_env *env, struct ofd_device *ofd,
+			   struct ofd_seq *oseq);
+int ofd_seqs_init(const struct lu_env *env, struct ofd_device *ofd);
+struct ofd_seq *ofd_seq_get(struct ofd_device *ofd, obd_seq seq);
+void ofd_seq_put(const struct lu_env *env, struct ofd_seq *oseq);
+
 int ofd_fs_setup(const struct lu_env *env, struct ofd_device *ofd,
 		 struct obd_device *obd);
 void ofd_fs_cleanup(const struct lu_env *env, struct ofd_device *ofd);
 int ofd_precreate_batch(struct ofd_device *ofd, int batch);
+struct ofd_seq *ofd_seq_load(const struct lu_env *env, struct ofd_device *ofd,
+			     obd_seq seq);
+void ofd_seqs_fini(const struct lu_env *env, struct ofd_device *ofd);
 
 /* ofd_io.c */
 int ofd_preprw(const struct lu_env *env,int cmd, struct obd_export *exp,
@@ -375,7 +405,7 @@ struct ofd_object *ofd_object_find_or_create(const struct lu_env *env,
 					     struct lu_attr *attr);
 int ofd_object_ff_check(const struct lu_env *env, struct ofd_object *fo);
 int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
-			  obd_id id, obd_seq group, int nr);
+			  obd_id id, struct ofd_seq *oseq, int nr, int sync);
 
 void ofd_object_put(const struct lu_env *env, struct ofd_object *fo);
 int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo,
@@ -403,7 +433,7 @@ static inline int ofd_grant_ratio_conv(int percentage)
 
 static inline int ofd_grant_param_supp(struct obd_export *exp)
 {
-	return !!(exp->exp_connect_flags & OBD_CONNECT_GRANT_PARAM);
+	return !!(exp_connect_flags(exp) & OBD_CONNECT_GRANT_PARAM);
 }
 
 /* Blocksize used for client not supporting OBD_CONNECT_GRANT_PARAM.
@@ -459,6 +489,12 @@ void ofd_fmd_drop(struct obd_export *exp, struct lu_fid *fid);
 #define ofd_fmd_drop(exp, fid) do {} while (0)
 #endif
 
+/* ofd_dev.c */
+int ofd_fid_set_index(const struct lu_env *env, struct ofd_device *ofd,
+		      int index);
+int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd);
+int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd);
+
 /* ofd_lvb.c */
 extern struct ldlm_valblock_ops ofd_lvbo;
 
@@ -501,33 +537,28 @@ static inline struct ofd_thread_info * ofd_info_init(const struct lu_env *env,
 static inline void ofd_build_resid(const struct lu_fid *fid,
 				   struct ldlm_res_id *resname)
 {
+	struct ost_id oid;
+
 	if (fid_is_idif(fid)) {
-		/* get id/seq like ostid_idif_pack() does */
-		osc_build_res_name(fid_idif_id(fid_seq(fid), fid_oid(fid),
-					       fid_ver(fid)),
-				   FID_SEQ_OST_MDT0, resname);
+		oid.oi_id = fid_idif_id(fid_seq(fid), fid_oid(fid),
+					fid_ver(fid));
+		oid.oi_seq = FID_SEQ_OST_MDT0;
 	} else {
-		/* In the future, where OSTs have FID sequences allocated. */
-		fid_build_reg_res_name(fid, resname);
+		oid.oi_id = fid_oid(fid);
+		oid.oi_seq = fid_seq(fid);
 	}
+	ostid_build_res_name(&oid, resname);
 }
 
 static inline void ofd_fid_from_resid(struct lu_fid *fid,
 				      const struct ldlm_res_id *name)
 {
-	/* if seq is FID_SEQ_OST_MDT0 then we have IDIF and resid was built
-	 * using osc_build_res_name function. */
-	if (fid_seq_is_mdt0(name->name[LUSTRE_RES_ID_VER_OID_OFF])) {
-		struct ost_id ostid;
-
-		ostid.oi_id = name->name[LUSTRE_RES_ID_SEQ_OFF];
-		ostid.oi_seq = name->name[LUSTRE_RES_ID_VER_OID_OFF];
-		fid_ostid_unpack(fid, &ostid, 0);
-	} else {
-		fid->f_seq = name->name[LUSTRE_RES_ID_SEQ_OFF];
-		fid->f_oid = (__u32)name->name[LUSTRE_RES_ID_VER_OID_OFF];
-		fid->f_ver = name->name[LUSTRE_RES_ID_VER_OID_OFF] >> 32;
-	}
+	/* To keep compatiblity, res[0] = oi_id, res[1] = oi_seq. */
+	struct ost_id ostid;
+
+	ostid.oi_id = name->name[LUSTRE_RES_ID_SEQ_OFF];
+	ostid.oi_seq = name->name[LUSTRE_RES_ID_VER_OID_OFF];
+	fid_ostid_unpack(fid, &ostid, 0);
 }
 
 static inline void ofd_oti2info(struct ofd_thread_info *info,