* Copyright 2009 Sun Microsystems, Inc. All rights reserved
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2016, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define LMVEA_DELETE_VALUES(count, offset) \
((count) == 0 && (offset) == (typeof(offset))(-1))
-#define LOV_OFFSET_DEFAULT ((__u16)-1)
-
-struct lod_qos_rr {
- spinlock_t lqr_alloc; /* protect allocation index */
- __u32 lqr_start_idx; /* start index of new inode */
- __u32 lqr_offset_idx;/* aliasing for start_idx */
- int lqr_start_count;/* reseed counter */
- struct ost_pool lqr_pool; /* round-robin optimized list */
- unsigned long lqr_dirty:1; /* recalc round-robin list */
-};
-
struct pool_desc {
char pool_name[LOV_MAXPOOLNAME + 1];
struct ost_pool pool_obds; /* pool members */
atomic_t pool_refcount;
- struct lod_qos_rr pool_rr;
+ struct lu_qos_rr pool_rr;
struct hlist_node pool_hash; /* access by poolname */
struct list_head pool_list;
struct proc_dir_entry *pool_proc_entry;
#define pool_tgt_array(p) ((p)->pool_obds.op_array)
#define pool_tgt_rw_sem(p) ((p)->pool_obds.op_rw_sem)
-struct lod_qos {
- struct list_head lq_oss_list;
- struct rw_semaphore lq_rw_sem;
- __u32 lq_active_oss_count;
- unsigned int lq_prio_free; /* priority for free space */
- unsigned int lq_threshold_rr;/* priority for rr */
- struct lod_qos_rr lq_rr; /* round robin qos data */
- bool lq_dirty:1, /* recalc qos data */
- lq_same_space:1,/* the ost's all have approx.
- the same space avail */
- lq_reset:1; /* zero current penalties */
-};
-
-struct lod_qos_oss {
- struct obd_uuid lqo_uuid; /* ptlrpc's c_remote_uuid */
- struct list_head lqo_oss_list; /* link to lov_qos */
- __u64 lqo_bavail; /* total bytes avail on OSS */
- __u64 lqo_penalty; /* current penalty */
- __u64 lqo_penalty_per_obj; /* penalty decrease
- every obj*/
- time_t lqo_used; /* last used time, seconds */
- __u32 lqo_ost_count; /* number of osts on this oss */
-};
-
-struct ltd_qos {
- struct lod_qos_oss *ltq_oss; /* oss info */
- __u64 ltq_penalty; /* current penalty */
- __u64 ltq_penalty_per_obj; /* penalty decrease
- every obj*/
- __u64 ltq_weight; /* net weighting */
- time_t ltq_used; /* last used time, seconds */
- bool ltq_usable:1; /* usable for striping */
-};
-
-struct lod_tgt_desc {
- struct dt_device *ltd_tgt;
- struct list_head ltd_kill;
- struct obd_export *ltd_exp;
- struct obd_uuid ltd_uuid;
- __u32 ltd_gen;
- __u32 ltd_index;
- struct ltd_qos ltd_qos; /* qos info per target */
- struct obd_statfs ltd_statfs;
- struct ptlrpc_thread *ltd_recovery_thread;
- unsigned long ltd_active:1,/* is this target up for requests */
- ltd_activate:1,/* should target be activated */
- ltd_reap:1, /* should this target be deleted */
- ltd_got_update_log:1, /* Already got update log */
- ltd_connecting:1; /* target is connecting */
-};
+#define lod_tgt_desc lu_tgt_desc
#define TGT_PTRS 256 /* number of pointers at 1st level */
#define TGT_PTRS_PER_BLOCK 256 /* number of pointers at 2nd level */
struct rw_semaphore ltd_rw_sem;
};
+struct lod_avoid_guide {
+ /* ids of OSSs avoid guidance */
+ __u32 *lag_oss_avoid_array;
+ /* number of filled array items */
+ unsigned int lag_oaa_count;
+ /* number of allocated array items */
+ unsigned int lag_oaa_size;
+ /* bitmap of OSTs avoid guidance */
+ struct cfs_bitmap *lag_ost_avoid_bitmap;
+ /* how many OSTs are available for alloc */
+ __u32 lag_ost_avail;
+};
+
struct lod_device {
struct dt_device lod_dt_dev;
struct obd_export *lod_child_exp;
struct dt_device *lod_child;
- struct proc_dir_entry *lod_proc_entry;
struct lprocfs_stats *lod_stats;
spinlock_t lod_connects_lock;
int lod_connects;
/* maximum EA size underlied OSD may have */
unsigned int lod_osd_max_easize;
+ /* maximum size of MDT stripe for Data-on-MDT files. */
+ unsigned int lod_dom_max_stripesize;
/*FIXME: When QOS and pool is implemented for MDT, probably these
* structure should be moved to lod_tgt_descs as well.
*/
/* QoS info per LOD */
- struct lod_qos lod_qos; /* qos info per lod */
+ struct lu_qos lod_qos; /* qos info per lod */
/* OST pool data */
struct ost_pool lod_pool_info; /* all OSTs in a packed array */
enum lustre_sec_part lod_sp_me;
struct proc_dir_entry *lod_symlink;
+ struct dentry *lod_debugfs;
/* ROOT object, used to fetch FS default striping */
struct lod_object *lod_md_root;
#define lod_ostnr lod_ost_descs.ltd_tgtnr
#define lod_osts_size lod_ost_descs.ltd_tgts_size
#define ltd_ost ltd_tgt
-#define lod_ost_desc lod_tgt_desc
+#define lod_ost_desc lu_tgt_desc
#define lod_mdts lod_mdt_descs.ltd_tgts
#define lod_mdt_bitmap lod_mdt_descs.ltd_tgt_bitmap
#define lod_remote_mdt_count lod_mdt_descs.ltd_tgtnr
#define lod_mdts_size lod_mdt_descs.ltd_tgts_size
#define ltd_mdt ltd_tgt
-#define lod_mdt_desc lod_tgt_desc
+#define lod_mdt_desc lu_tgt_desc
struct lod_layout_component {
struct lu_extent llc_extent;
__u16 llc_stripe_offset;
__u16 llc_stripe_count;
__u16 llc_stripes_allocated;
+ __u64 llc_timestamp; /* snapshot time */
char *llc_pool;
/* ost list specified with LOV_USER_MAGIC_SPECIFIC lum */
struct ost_pool llc_ostlist;
struct dt_object **llc_stripe;
+ __u32 *llc_ost_indices;
};
struct lod_default_striping {
/* default LOV */
/* current layout component count */
__u16 lds_def_comp_cnt;
+ __u16 lds_def_mirror_cnt;
/* the largest comp count ever used */
__u32 lds_def_comp_size_cnt;
struct lod_layout_component *lds_def_comp_entries;
lds_dir_def_striping_set:1;
};
+struct lod_mirror_entry {
+ __u16 lme_stale:1,
+ lme_primary:1;
+ /* mirror id */
+ __u16 lme_id;
+ /* start,end index of this mirror in ldo_comp_entries */
+ __u16 lme_start;
+ __u16 lme_end;
+};
+
struct lod_object {
/* common fields for both files and directories */
struct dt_object ldo_obj;
+ struct mutex ldo_layout_mutex;
union {
/* file stripe (LOV) */
struct {
/* Layout component count for a regular file.
* It equals to 1 for non-composite layout. */
__u16 ldo_comp_cnt;
+ /* Layout mirror count for a PFLR file.
+ * It's 0 for files with non-composite layout. */
+ __u16 ldo_mirror_count;
+ struct lod_mirror_entry *ldo_mirrors;
__u32 ldo_is_composite:1,
- ldo_comp_cached:1;
+ ldo_flr_state:2,
+ ldo_comp_cached:1,
+ ldo_is_foreign:1;
};
/* directory stripe (LMV) */
struct {
__u16 ldo_dir_stripes_allocated;
__u32 ldo_dir_stripe_offset;
__u32 ldo_dir_hash_type;
+ __u32 ldo_dir_migrate_offset;
+ __u32 ldo_dir_migrate_hash;
/* Is a slave stripe of striped directory? */
__u32 ldo_dir_slave_stripe:1,
ldo_dir_striped:1,
/* the stripe has been loaded */
- ldo_dir_stripe_loaded:1;
+ ldo_dir_stripe_loaded:1,
+ /* foreign directory */
+ ldo_dir_is_foreign;
/*
- * default striping is not cached, so this field is
- * invalid after create, make sure it's used by
+ * This default LMV is parent default LMV, which will be
+ * used in child creation, and it's not cached, so this
+ * field is invalid after create, make sure it's used by
* lod_dir_striping_create_internal() only.
*/
struct lod_default_striping *ldo_def_striping;
};
};
- /* file stripe (LOV) */
- struct lod_layout_component *ldo_comp_entries;
- /* slave stripes of striped directory (LMV) */
- struct dt_object **ldo_stripe;
+ union {
+ struct {
+ /* foreign/raw format LOV */
+ char *ldo_foreign_lov;
+ size_t ldo_foreign_lov_size;
+ };
+ struct {
+ /* foreign/raw format LMV */
+ char *ldo_foreign_lmv;
+ size_t ldo_foreign_lmv_size;
+ };
+ struct {
+ /* file stripe (LOV) */
+ struct lod_layout_component *ldo_comp_entries;
+ /* slave stripes of striped directory (LMV) */
+ struct dt_object **ldo_stripe;
+ };
+ };
};
+#define lod_foreach_mirror_comp(comp, lo, mirror_idx) \
+for (comp = &lo->ldo_comp_entries[lo->ldo_mirrors[mirror_idx].lme_start]; \
+ comp <= &lo->ldo_comp_entries[lo->ldo_mirrors[mirror_idx].lme_end]; \
+ comp++)
+
+static inline bool lod_is_flr(const struct lod_object *lo)
+{
+ if (!lo->ldo_is_composite)
+ return false;
+
+ return (lo->ldo_flr_state & LCM_FL_FLR_MASK) != LCM_FL_NONE;
+}
+
static inline int lod_set_pool(char **pool, const char *new_pool)
{
int len;
static inline int lod_set_def_pool(struct lod_default_striping *lds,
int i, const char *new_pool)
{
- return lod_set_pool(&lds->lds_def_comp_entries[i].llc_pool,
- new_pool);
+ return lod_set_pool(&lds->lds_def_comp_entries[i].llc_pool, new_pool);
}
static inline int lod_obj_set_pool(struct lod_object *lo, int i,
const char *new_pool)
{
- return lod_set_pool(&lo->ldo_comp_entries[i].llc_pool,
- new_pool);
+ return lod_set_pool(&lo->ldo_comp_entries[i].llc_pool, new_pool);
}
/**
struct lu_attr lti_attr;
struct lod_it lti_it;
struct ldlm_res_id lti_res_id;
- struct ost_pool lti_inuse_osts;
/* used to hold lu_dirent, sizeof(struct lu_dirent) + NAME_MAX */
char lti_key[sizeof(struct lu_dirent) +
NAME_MAX];
struct lustre_cfg lti_lustre_cfg;
/* used to store parent default striping in create */
struct lod_default_striping lti_def_striping;
- struct filter_fid lti_ff;
+ struct filter_fid lti_ff;
+ __u32 *lti_comp_idx;
+ size_t lti_comp_size;
+ size_t lti_count;
+ struct lu_attr lti_layout_attr;
+ /* object allocation avoid guide info */
+ struct lod_avoid_guide lti_avoid;
};
extern const struct lu_device_operations lod_lu_ops;
return &obj->ldo_obj.do_lu;
}
+static inline const struct lu_fid *lod_object_fid(struct lod_object *obj)
+{
+ return lu_object_fid(lod2lu_obj(obj));
+}
+
static inline struct lod_object *lod_obj(const struct lu_object *o)
{
LASSERT(lu_device_is_lod(o->lo_dev));
if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
return lo->ldo_dir_stripe_count != 0;
+ if (lo->ldo_is_foreign)
+ return false;
+
for (i = 0; i < lo->ldo_comp_cnt; i++) {
if (lo->ldo_comp_entries[i].llc_stripe == NULL)
continue;
return lname;
}
+static inline struct lod_default_striping *
+lod_lds_buf_get(const struct lu_env *env)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+
+ info->lti_def_striping.lds_def_striping_set = 0;
+ info->lti_def_striping.lds_dir_def_striping_set = 0;
+ return &info->lti_def_striping;
+}
+
+static inline void lod_layout_get_pool(struct lod_layout_component *entries,
+ int count, char *pool, int len)
+{
+ int i;
+
+ for (i = 0; i < count; i++) {
+ if (entries[i].llc_pool != NULL) {
+ strlcpy(pool, entries[i].llc_pool, len);
+ break;
+ }
+ }
+}
+
#define lod_foreach_ost(__dev, index) \
if ((__dev)->lod_osts_size > 0) \
cfs_foreach_bit((__dev)->lod_ost_bitmap, (index))
unsigned gen, bool for_ost);
int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
struct lod_tgt_descs *ltd, bool for_ost);
-int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo);
-int lod_load_striping(const struct lu_env *env, struct lod_object *lo);
+int lod_striping_load(const struct lu_env *env, struct lod_object *lo);
+int lod_striping_reload(const struct lu_env *env, struct lod_object *lo,
+ const struct lu_buf *buf);
int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
const char *name);
const struct lu_buf *buf);
int lod_initialize_objects(const struct lu_env *env, struct lod_object *mo,
struct lov_ost_data_v1 *objs, int index);
-int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
- bool is_from_disk, __u64 start);
+int lod_verify_striping(struct lod_device *d, struct lod_object *lo,
+ const struct lu_buf *buf, bool is_from_disk);
int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
struct lov_mds_md *lmm, int *lmm_size, bool is_dir);
int lod_ea_store_resize(struct lod_thread_info *info, size_t size);
int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count);
void lod_free_def_comp_entries(struct lod_default_striping *lds);
void lod_free_comp_entries(struct lod_object *lo);
-int lod_alloc_comp_entries(struct lod_object *lo, int cnt);
+int lod_alloc_comp_entries(struct lod_object *lo, int mirror_cnt, int comp_cnt);
+int lod_fill_mirrors(struct lod_object *lo);
/* lod_pool.c */
int lod_ost_pool_add(struct ost_pool *op, __u32 idx, unsigned int min_count);
int lod_pool_add(struct obd_device *obd, char *poolname, char *ostname);
int lod_pool_remove(struct obd_device *obd, char *poolname, char *ostname);
+struct lod_obj_stripe_cb_data;
+typedef int (*lod_obj_stripe_cb_t)(const struct lu_env *env,
+ struct lod_object *lo, struct dt_object *dt,
+ struct thandle *th,
+ int comp_idx, int stripe_idx,
+ struct lod_obj_stripe_cb_data *data);
+typedef bool (*lod_obj_comp_skip_cb_t)(const struct lu_env *env,
+ struct lod_object *lo, int comp_idx,
+ struct lod_obj_stripe_cb_data *data);
+typedef int (*lod_obj_comp_cb_t)(const struct lu_env *env,
+ struct lod_object *lo, int comp_idx,
+ struct lod_obj_stripe_cb_data *data);
struct lod_obj_stripe_cb_data {
union {
const struct lu_attr *locd_attr;
- struct ost_pool *locd_inuse;
+ int locd_ost_index;
+ const struct lu_buf *locd_buf;
};
- bool locd_declare;
+ lod_obj_stripe_cb_t locd_stripe_cb;
+ lod_obj_comp_skip_cb_t locd_comp_skip_cb;
+ lod_obj_comp_cb_t locd_comp_cb;
+ bool locd_declare;
};
-typedef int (*lod_obj_stripe_cb_t)(const struct lu_env *env,
- struct lod_object *lo, struct dt_object *dt,
- struct thandle *th, int stripe_idx,
- struct lod_obj_stripe_cb_data *data);
/* lod_qos.c */
-int lod_prepare_inuse(const struct lu_env *env, struct lod_object *lo);
int lod_prepare_create(const struct lu_env *env, struct lod_object *lo,
struct lu_attr *attr, const struct lu_buf *buf,
struct thandle *th);
-int qos_add_tgt(struct lod_device*, struct lod_tgt_desc *);
-int qos_del_tgt(struct lod_device *, struct lod_tgt_desc *);
-void lod_qos_rr_init(struct lod_qos_rr *lqr);
int lod_use_defined_striping(const struct lu_env *, struct lod_object *,
const struct lu_buf *);
-int lod_obj_stripe_set_inuse_cb(const struct lu_env *env, struct lod_object *lo,
- struct dt_object *dt, struct thandle *th,
- int stripe_idx,
- struct lod_obj_stripe_cb_data *data);
int lod_qos_parse_config(const struct lu_env *env, struct lod_object *lo,
const struct lu_buf *buf);
int lod_qos_prep_create(const struct lu_env *env, struct lod_object *lo,
struct lu_attr *attr, struct thandle *th,
- int comp_idx, struct ost_pool *inuse);
+ int comp_idx);
__u16 lod_comp_entry_stripe_count(struct lod_object *lo,
struct lod_layout_component *entry,
bool is_dir);
__u16 lod_get_stripe_count(struct lod_device *lod, struct lod_object *lo,
- __u16 stripe_count);
+ __u16 stripe_count, bool overstriping);
+void lod_qos_statfs_update(const struct lu_env *env, struct lod_device *lod);
+int lod_statfs_and_check(const struct lu_env *env, struct lod_device *d,
+ int index, struct obd_statfs *sfs,
+ struct obd_statfs_info *info);
/* lproc_lod.c */
int lod_procfs_init(struct lod_device *lod);
int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
struct lu_attr *attr, struct dt_object_format *dof,
struct thandle *th);
-void lod_object_free_striping(const struct lu_env *env, struct lod_object *lo);
+int lod_alloc_foreign_lov(struct lod_object *lo, size_t size);
+void lod_free_foreign_lov(struct lod_object *lo);
+void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo);
+void lod_striping_free(const struct lu_env *env, struct lod_object *lo);
int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
- struct thandle *th, lod_obj_stripe_cb_t cb,
+ struct thandle *th,
struct lod_obj_stripe_cb_data *data);
+int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
+ struct lov_user_md_v3 *v3);
/* lod_sub_object.c */
struct thandle *lod_sub_get_thandle(const struct lu_env *env,
struct thandle *th);
int lod_sub_insert(const struct lu_env *env, struct dt_object *dt,
const struct dt_rec *rec, const struct dt_key *key,
- struct thandle *th, int ign);
+ struct thandle *th);
int lod_sub_declare_delete(const struct lu_env *env, struct dt_object *dt,
const struct dt_key *key, struct thandle *th);
int lod_sub_delete(const struct lu_env *env, struct dt_object *dt,
struct thandle *th);
ssize_t lod_sub_write(const struct lu_env *env, struct dt_object *dt,
const struct lu_buf *buf, loff_t *pos,
- struct thandle *th, int rq);
+ struct thandle *th);
int lod_sub_declare_punch(const struct lu_env *env, struct dt_object *dt,
__u64 start, __u64 end, struct thandle *th);
int lod_sub_punch(const struct lu_env *env, struct dt_object *dt,
git://git.whamcloud.com / fs / lustre-release.git / blobdiff