* Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, Intel Corporation.
+ * Copyright (c) 2012, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include <obd.h>
#include <obd_class.h>
#include <dt_object.h>
+#include <md_object.h>
#include <lustre_fid.h>
-#include <obd_ost.h>
#include <lustre_capa.h>
#define OFD_INIT_OBJID 0
-#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)
+#define OFD_PRECREATE_BATCH_DEFAULT (OBJ_SUBDIR_COUNT * 4)
/* on small filesystems we should not precreate too many objects in
* a single transaction, otherwise we can overflow transactions */
};
#define OFD_FMD_MAX_NUM_DEFAULT 128
-#define OFD_FMD_MAX_AGE_DEFAULT ((obd_timeout + 10) * CFS_HZ)
+#define OFD_FMD_MAX_AGE_DEFAULT ((obd_timeout + 10) * HZ)
-enum {
- LPROC_OFD_READ_BYTES = 0,
- LPROC_OFD_WRITE_BYTES = 1,
- LPROC_OFD_LAST,
-};
+#define OFD_SOFT_SYNC_LIMIT_DEFAULT 16
-/* for job stats */
+/* request stats */
enum {
LPROC_OFD_STATS_READ = 0,
- LPROC_OFD_STATS_WRITE = 1,
- LPROC_OFD_STATS_SETATTR = 2,
- LPROC_OFD_STATS_PUNCH = 3,
- LPROC_OFD_STATS_SYNC = 4,
+ LPROC_OFD_STATS_WRITE,
+ LPROC_OFD_STATS_GETATTR,
+ LPROC_OFD_STATS_SETATTR,
+ LPROC_OFD_STATS_PUNCH,
+ LPROC_OFD_STATS_SYNC,
+ LPROC_OFD_STATS_DESTROY,
+ LPROC_OFD_STATS_CREATE,
+ LPROC_OFD_STATS_STATFS,
+ LPROC_OFD_STATS_GET_INFO,
+ LPROC_OFD_STATS_SET_INFO,
+ LPROC_OFD_STATS_QUOTACTL,
LPROC_OFD_STATS_LAST,
};
static inline void ofd_counter_incr(struct obd_export *exp, int opcode,
char *jobid, long amount)
{
+ if (exp->exp_obd && exp->exp_obd->obd_stats)
+ lprocfs_counter_add(exp->exp_obd->obd_stats, opcode, 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);
+ lprocfs_counter_add(exp->exp_nid_stats->nid_stats, opcode,
+ amount);
}
}
+struct ofd_seq {
+ cfs_list_t os_list;
+ struct ost_id os_oi;
+ spinlock_t os_last_oid_lock;
+ struct mutex os_create_lock;
+ 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;
/* DLM name-space for meta-data locks maintained by this server */
struct ldlm_namespace *ofd_namespace;
- /* transaction callbacks */
- struct dt_txn_callback ofd_txn_cb;
-
/* 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;
+ __u64 ofd_inconsistency_self_detected;
+ __u64 ofd_inconsistency_self_repaired;
- int ofd_max_group;
- obd_id ofd_last_objids[OFD_MAX_GROUPS];
- struct mutex ofd_create_locks[OFD_MAX_GROUPS];
- struct dt_object *ofd_lastid_obj[OFD_MAX_GROUPS];
- 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 */
spinlock_t ofd_osfs_lock;
unsigned long ofd_raid_degraded:1,
/* sync journal on writes */
ofd_syncjournal:1,
- /* sync on lock cancel */
- ofd_sync_lock_cancel:2,
/* shall we grant space to clients not
* supporting OBD_CONNECT_GRANT_PARAM? */
- ofd_grant_compat_disable:1;
+ ofd_grant_compat_disable:1,
+ /* Protected by ofd_lastid_rwsem. */
+ ofd_lastid_rebuilding:1,
+ ofd_record_fid_accessed:1,
+ ofd_lfsck_verify_pfid:1;
+ struct seq_server_site ofd_seq_site;
+ /* the limit of SOFT_SYNC RPCs that will trigger a soft sync */
+ unsigned int ofd_soft_sync_limit;
+ /* Protect ::ofd_lastid_rebuilding */
+ struct rw_semaphore ofd_lastid_rwsem;
+ __u64 ofd_lastid_gen;
+ struct ptlrpc_thread ofd_inconsistency_thread;
+ struct list_head ofd_inconsistency_list;
+ spinlock_t ofd_inconsistency_lock;
};
static inline struct ofd_device *ofd_dev(struct lu_device *d)
struct ofd_object {
struct lu_object_header ofo_header;
struct dt_object ofo_obj;
- int ofo_ff_exists;
+ struct lu_fid ofo_pfid;
+ unsigned int ofo_pfid_checking:1,
+ ofo_pfid_verified:1;
};
static inline struct ofd_object *ofd_obj(struct lu_object *o)
struct obd_export *fti_exp;
__u64 fti_xid;
- __u64 fti_transno;
__u64 fti_pre_version;
- __u32 fti_has_trans:1, /* has txn already */
- fti_mult_trans:1;
struct lu_fid fti_fid;
struct lu_attr fti_attr;
struct lu_attr fti_attr2;
struct ldlm_res_id fti_resid;
- struct filter_fid fti_mds_fid;
- struct filter_fid fti_mds_fid2;
+ union {
+ struct filter_fid fti_mds_fid;
+ struct filter_fid_old fti_mds_fid_old;
+ };
struct ost_id fti_ostid;
struct ofd_object *fti_obj;
union {
/* Space used by the I/O, used by grant code */
unsigned long fti_used;
struct ost_lvb fti_lvb;
+ struct lfsck_request fti_lr;
};
extern void target_recovery_fini(struct obd_device *obd);
/* ofd_capa.c */
int ofd_update_capa_key(struct ofd_device *ofd, struct lustre_capa_key *key);
-int ofd_auth_capa(struct obd_export *exp, struct lu_fid *fid, obd_seq seq,
- struct lustre_capa *capa, __u64 opc);
+int ofd_auth_capa(struct obd_export *exp, const struct lu_fid *fid,
+ obd_seq seq, struct lustre_capa *capa, __u64 opc);
void ofd_free_capa_keys(struct ofd_device *ofd);
/* ofd_dev.c */
extern struct lu_context_key ofd_thread_key;
int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd);
+int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
+ struct lu_fid *fid, struct ll_user_fiemap *fiemap);
/* ofd_obd.c */
extern struct obd_ops ofd_obd_ops;
int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
struct obd_statfs *osfs, __u64 max_age,
int *from_cache);
+int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
+ const struct lu_fid *fid, int orphan);
+int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
+ struct obd_statfs *osfs, __u64 max_age, __u32 flags);
+int ofd_obd_disconnect(struct obd_export *exp);
/* 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);
+void ofd_seqs_free(const struct lu_env *env, struct ofd_device *ofd);
/* ofd_io.c */
+int ofd_start_inconsistency_verification_thread(struct ofd_device *ofd);
+int ofd_stop_inconsistency_verification_thread(struct ofd_device *ofd);
+int ofd_verify_ff(const struct lu_env *env, struct ofd_object *fo,
+ struct obdo *oa);
int ofd_preprw(const struct lu_env *env,int cmd, struct obd_export *exp,
struct obdo *oa, int objcount, struct obd_ioobj *obj,
struct niobuf_remote *rnb, int *nr_local,
/* lproc_ofd.c */
#ifdef LPROCFS
-void lprocfs_ofd_init_vars(struct lprocfs_static_vars *lvars);
-int lproc_ofd_attach_seqstat(struct obd_device *dev);
-extern struct file_operations ofd_per_nid_stats_fops;
+extern struct lprocfs_seq_vars lprocfs_ofd_obd_vars[];
void ofd_stats_counter_init(struct lprocfs_stats *stats);
#else
-static void lprocfs_ofd_init_vars(struct lprocfs_static_vars *lvars)
-{
- memset(lvars, 0, sizeof(*lvars));
-}
-static inline int lproc_ofd_attach_seqstat(struct obd_device *dev) {}
static inline void ofd_stats_counter_init(struct lprocfs_stats *stats) {}
#endif
struct ofd_object *ofd_object_find(const struct lu_env *env,
struct ofd_device *ofd,
const struct lu_fid *fid);
-struct ofd_object *ofd_object_find_or_create(const struct lu_env *env,
- struct ofd_device *ofd,
- const struct lu_fid *fid,
- struct lu_attr *attr);
-int ofd_object_ff_check(const struct lu_env *env, struct ofd_object *fo);
+int ofd_object_ff_load(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,
struct lu_attr *la, struct filter_fid *ff);
int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
__u64 start, __u64 end, struct lu_attr *la,
- struct filter_fid *ff);
+ struct filter_fid *ff, struct obdo *oa);
int ofd_object_destroy(const struct lu_env *, struct ofd_object *, int);
int ofd_attr_get(const struct lu_env *env, struct ofd_object *fo,
struct lu_attr *la);
int ofd_attr_handle_ugid(const struct lu_env *env, struct ofd_object *fo,
struct lu_attr *la, int is_setattr);
+static inline
+struct ofd_object *ofd_object_find_exists(const struct lu_env *env,
+ struct ofd_device *ofd,
+ const struct lu_fid *fid)
+{
+ struct ofd_object *fo;
+
+ fo = ofd_object_find(env, ofd, fid);
+ if (!IS_ERR(fo) && !ofd_object_exists(fo)) {
+ ofd_object_put(env, fo);
+ fo = ERR_PTR(-ENOENT);
+ }
+ return fo;
+}
+
/* ofd_grants.c */
#define OFD_GRANT_RATIO_SHIFT 8
static inline __u64 ofd_grant_reserved(struct ofd_device *ofd, obd_size bavail)
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.
struct ofd_device *ofd)
{
/* Clients which don't support OBD_CONNECT_GRANT_PARAM cannot handle
- * a block size > page size and consume CFS_PAGE_SIZE of grant when
+ * a block size > page size and consume PAGE_CACHE_SIZE of grant when
* dirtying a page regardless of the block size */
return !!(ofd_obd(ofd)->obd_self_export != exp &&
ofd->ofd_blockbits > COMPAT_BSIZE_SHIFT &&
/* When ofd_grant_compat_disable is set, we don't grant any space to
* clients not supporting OBD_CONNECT_GRANT_PARAM.
* Otherwise, space granted to such a client is inflated since it
- * consumes CFS_PAGE_SIZE of grant space per block */
+ * consumes PAGE_CACHE_SIZE of grant space per block */
return !!(ofd_grant_compat(exp, ofd) && ofd->ofd_grant_compat_disable);
}
void ofd_grant_sanity_check(struct obd_device *obd, const char *func);
long ofd_grant_connect(const struct lu_env *env, struct obd_export *exp,
- obd_size want);
+ obd_size want, bool new_conn);
void ofd_grant_discard(struct obd_export *exp);
void ofd_grant_prepare_read(const struct lu_env *env, struct obd_export *exp,
struct obdo *oa);
int ofd_fmd_init(void);
void ofd_fmd_exit(void);
struct ofd_mod_data *ofd_fmd_find(struct obd_export *exp,
- struct lu_fid *fid);
+ const struct lu_fid *fid);
struct ofd_mod_data *ofd_fmd_get(struct obd_export *exp,
- struct lu_fid *fid);
+ const struct lu_fid *fid);
void ofd_fmd_put(struct obd_export *exp, struct ofd_mod_data *fmd);
void ofd_fmd_expire(struct obd_export *exp);
void ofd_fmd_cleanup(struct obd_export *exp);
#ifdef DO_FMD_DROP
-void ofd_fmd_drop(struct obd_export *exp, struct lu_fid *fid);
+void ofd_fmd_drop(struct obd_export *exp, const struct lu_fid *fid);
#else
#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;
void *req_cookie, ldlm_mode_t mode, __u64 flags,
void *data);
-static inline struct ofd_thread_info * ofd_info(const struct lu_env *env)
+static inline struct ofd_thread_info *ofd_info(const struct lu_env *env)
{
struct ofd_thread_info *info;
+ lu_env_refill((void *)env);
info = lu_context_key_get(&env->le_ctx, &ofd_thread_key);
LASSERT(info);
- LASSERT(info->fti_env);
- LASSERT(info->fti_env == env);
return info;
}
-static inline struct ofd_thread_info * ofd_info_init(const struct lu_env *env,
- struct obd_export *exp)
+static inline struct ofd_thread_info *ofd_info_init(const struct lu_env *env,
+ struct obd_export *exp)
{
struct ofd_thread_info *info;
- info = lu_context_key_get(&env->le_ctx, &ofd_thread_key);
- LASSERT(info);
+ info = ofd_info(env);
LASSERT(info->fti_exp == NULL);
LASSERT(info->fti_env == NULL);
LASSERT(info->fti_attr.la_valid == 0);
info->fti_env = env;
info->fti_exp = exp;
info->fti_pre_version = 0;
- info->fti_transno = 0;
- info->fti_has_trans = 0;
return info;
}
-/* The same as osc_build_res_name() */
-static inline void ofd_build_resid(const struct lu_fid *fid,
- struct ldlm_res_id *resname)
+static inline struct ofd_thread_info *tsi2ofd_info(struct tgt_session_info *tsi)
{
- 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);
- } else {
- /* In the future, where OSTs have FID sequences allocated. */
- fid_build_reg_res_name(fid, resname);
- }
-}
+ struct ptlrpc_request *req = tgt_ses_req(tsi);
+ struct ofd_thread_info *info;
-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;
+ info = ofd_info(tsi->tsi_env);
+ LASSERT(info->fti_exp == NULL);
+ LASSERT(info->fti_env == NULL);
+ LASSERT(info->fti_attr.la_valid == 0);
+
+ info->fti_env = tsi->tsi_env;
+ info->fti_exp = tsi->tsi_exp;
+
+ info->fti_xid = req->rq_xid;
+ /** VBR: take versions from request */
+ if (req->rq_reqmsg != NULL &&
+ lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
+ __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
+
+ info->fti_pre_version = pre_version ? pre_version[0] : 0;
}
+ return info;
}
static inline void ofd_oti2info(struct ofd_thread_info *info,
struct obd_trans_info *oti)
{
info->fti_xid = oti->oti_xid;
- info->fti_transno = oti->oti_transno;
info->fti_pre_version = oti->oti_pre_version;
}
struct obd_trans_info *oti)
{
oti->oti_xid = info->fti_xid;
- LASSERTF(ergo(oti->oti_transno > 0,
- oti->oti_transno == info->fti_transno),
- "Overwrite replay transno "LPX64" by "LPX64"\n",
- oti->oti_transno, info->fti_transno);
- oti->oti_transno = info->fti_transno;
oti->oti_pre_version = info->fti_pre_version;
}
static inline void ofd_slc_set(struct ofd_device *ofd)
{
if (ofd->ofd_syncjournal == 1)
- ofd->ofd_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
- else if (ofd->ofd_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
- ofd->ofd_sync_lock_cancel = ALWAYS_SYNC_ON_CANCEL;
+ ofd->ofd_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
+ else if (ofd->ofd_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
+ ofd->ofd_lut.lut_sync_lock_cancel = ALWAYS_SYNC_ON_CANCEL;
}
-static inline void ofd_prepare_fidea(struct filter_fid *ff, struct obdo *oa)
+static inline void ofd_prepare_fidea(struct filter_fid *ff,
+ const struct obdo *oa)
{
- if (!(oa->o_valid & OBD_MD_FLGROUP))
- oa->o_seq = 0;
/* packing fid and converting it to LE for storing into EA.
* Here ->o_stripe_idx should be filled by LOV and rest of
* fields - by client. */
/* XXX: we are ignoring o_parent_ver here, since this should
* be the same for all objects in this fileset. */
ff->ff_parent.f_ver = cpu_to_le32(oa->o_stripe_idx);
- ff->ff_objid = cpu_to_le64(oa->o_id);
- ff->ff_seq = cpu_to_le64(oa->o_seq);
}
/* niobuf_remote has no rnb_ prefix in master */
#define rnb_offset offset
#define rnb_flags flags
#define rnb_len len
-/* the same for niobuf_local */
-#define lnb_flags flags
-#define lnb_rc rc
#endif /* _OFD_INTERNAL_H */
git://git.whamcloud.com / fs / lustre-release.git / blobdiff