#include <lustre/lustre_lfsck_user.h>
#include <lustre/lustre_user.h>
#include <lustre/lustre_idl.h>
+#include <lustre_lfsck.h>
#include <obd.h>
#include <lu_object.h>
#include <dt_object.h>
+#include <md_object.h>
#include <lustre_net.h>
#include <lustre_dlm.h>
#include <lustre_fid.h>
+#include <md_object.h>
#define HALF_SEC (HZ >> 1)
#define LFSCK_CHECKPOINT_INTERVAL 60
#define LFSCK_NAMEENTRY_REMOVED 2 /* The entry has been removed. */
#define LFSCK_NAMEENTRY_RECREATED 3 /* The entry has been recreated. */
-enum lfsck_status {
- /* The lfsck file is new created, for new MDT, upgrading from old disk,
- * or re-creating the lfsck file manually. */
- LS_INIT = 0,
-
- /* The first-step system scanning. */
- LS_SCANNING_PHASE1 = 1,
-
- /* The second-step system scanning. */
- LS_SCANNING_PHASE2 = 2,
-
- /* The LFSCK processing has completed for all objects. */
- LS_COMPLETED = 3,
-
- /* The LFSCK exited automatically for failure, will not auto restart. */
- LS_FAILED = 4,
-
- /* The LFSCK is stopped manually, will not auto restart. */
- LS_STOPPED = 5,
-
- /* LFSCK is paused automatically when umount,
- * will be restarted automatically when remount. */
- LS_PAUSED = 6,
-
- /* System crashed during the LFSCK,
- * will be restarted automatically after recovery. */
- LS_CRASHED = 7,
-};
-
enum lfsck_flags {
/* Finish the first cycle scanning. */
- LF_SCANNED_ONCE = 0x00000001ULL,
+ LF_SCANNED_ONCE = 0x00000001ULL,
/* There is some namespace inconsistency. */
- LF_INCONSISTENT = 0x00000002ULL,
+ LF_INCONSISTENT = 0x00000002ULL,
/* The device is upgraded from 1.8 format. */
- LF_UPGRADE = 0x00000004ULL,
+ LF_UPGRADE = 0x00000004ULL,
+
+ /* The server ever restarted during the LFSCK, and may miss to process
+ * some objects check/repair. */
+ LF_INCOMPLETE = 0x00000008ULL,
+
+ /* The LAST_ID (file) crashed. */
+ LF_CRASHED_LASTID = 0x00000010ULL,
};
struct lfsck_position {
/* How many items can be scanned at most per second. */
__u32 lb_speed_limit;
+ /* The windows size for async requests pipeline. */
+ __u16 lb_async_windows;
+
/* For 64-bits aligned. */
- __u32 lb_padding;
+ __u16 lb_padding;
/* For future using. */
__u64 lb_reserved[6];
__u64 ln_reserved[2];
};
+enum lfsck_layout_inconsistency_type {
+ LLIT_NONE = 0,
+ LLIT_DANGLING = 1,
+ LLIT_UNMATCHED_PAIR = 2,
+ LLIT_MULTIPLE_REFERENCED = 3,
+ LLIT_ORPHAN = 4,
+ LLIT_INCONSISTENT_OWNER = 5,
+ LLIT_OTHERS = 6,
+ LLIT_MAX = LLIT_OTHERS
+};
+
+struct lfsck_layout {
+ /* Magic number to detect that this struct contains valid data. */
+ __u32 ll_magic;
+
+ /* See 'enum lfsck_status'. */
+ __u32 ll_status;
+
+ /* See 'enum lfsck_flags'. */
+ __u32 ll_flags;
+
+ /* How many completed LFSCK runs on the device. */
+ __u32 ll_success_count;
+
+ /* How long the LFSCK phase1 has run in seconds. */
+ __u32 ll_run_time_phase1;
+
+ /* How long the LFSCK phase2 has run in seconds. */
+ __u32 ll_run_time_phase2;
+
+ /* Time for the last LFSCK completed in seconds since epoch. */
+ __u64 ll_time_last_complete;
+
+ /* Time for the latest LFSCK ran in seconds since epoch. */
+ __u64 ll_time_latest_start;
+
+ /* Time for the last LFSCK checkpoint in seconds since epoch. */
+ __u64 ll_time_last_checkpoint;
+
+ /* Position for the latest LFSCK started from. */
+ __u64 ll_pos_latest_start;
+
+ /* Position for the last LFSCK checkpoint. */
+ __u64 ll_pos_last_checkpoint;
+
+ /* Position for the first should be updated object. */
+ __u64 ll_pos_first_inconsistent;
+
+ /* How many objects have been checked. */
+ __u64 ll_objs_checked_phase1;
+
+ /* How many objects failed to be processed. */
+ __u64 ll_objs_failed_phase1;
+
+ /* How many objects have been double scanned. */
+ __u64 ll_objs_checked_phase2;
+
+ /* How many objects failed to be processed during double scan. */
+ __u64 ll_objs_failed_phase2;
+
+ /* kinds of inconsistency have been repaired.
+ * ll_objs_repaired[type - 1] is the count for the given @type. */
+ __u64 ll_objs_repaired[LLIT_MAX];
+
+ /* How many objects have been skipped because of related
+ * MDT(s)/OST(s) do not participate in the LFSCK */
+ __u64 ll_objs_skipped;
+
+ /* For further using. 256-bytes aligned now. */
+ __u64 ll_reserved[12];
+};
+
struct lfsck_component;
struct lfsck_operations {
int (*lfsck_double_scan)(const struct lu_env *env,
struct lfsck_component *com);
+
+ void (*lfsck_data_release)(const struct lu_env *env,
+ struct lfsck_component *com);
+
+ void (*lfsck_quit)(const struct lu_env *env,
+ struct lfsck_component *com);
};
+#define TGT_PTRS 256 /* number of pointers at 1st level */
+#define TGT_PTRS_PER_BLOCK 256 /* number of pointers at 2nd level */
+
+struct lfsck_tgt_desc {
+ struct list_head ltd_orphan_list;
+ struct dt_device *ltd_tgt;
+ struct dt_device *ltd_key;
+ struct obd_export *ltd_exp;
+ struct list_head ltd_layout_list;
+ atomic_t ltd_ref;
+ __u32 ltd_index;
+};
+
+struct lfsck_tgt_desc_idx {
+ struct lfsck_tgt_desc *ldi_tgts[TGT_PTRS_PER_BLOCK];
+};
+
+struct lfsck_tgt_descs {
+ /* list of known TGTs */
+ struct lfsck_tgt_desc_idx *ltd_tgts_idx[TGT_PTRS];
+
+ /* bitmap of TGTs available */
+ cfs_bitmap_t *ltd_tgts_bitmap;
+
+ /* for lfsck_tgt_desc::ltd_xxx_list */
+ spinlock_t ltd_lock;
+
+ /* for tgts table accessing and changes */
+ struct rw_semaphore ltd_rw_sem;
+
+ /* Temporary list for orphan targets. */
+ struct list_head ltd_orphan;
+
+ /* number of registered TGTs */
+ int ltd_tgtnr;
+};
+
+#define LTD_TGT(ltd, index) \
+ ((ltd)->ltd_tgts_idx[(index) / TGT_PTRS_PER_BLOCK]->\
+ ldi_tgts[(index) % TGT_PTRS_PER_BLOCK])
+
+#define OST_TGT(lfsck, index) LTD_TGT(&lfsck->li_ost_descs, index)
+#define MDT_TGT(lfsck, index) LTD_TGT(&lfsck->li_mdt_descs, index)
+
struct lfsck_component {
/* into lfsck_instance::li_list_(scan,double_scan,idle} */
cfs_list_t lc_link;
/* into lfsck_instance::li_list_dir */
cfs_list_t lc_link_dir;
+
struct rw_semaphore lc_sem;
- cfs_atomic_t lc_ref;
+ atomic_t lc_ref;
struct lfsck_position lc_pos_start;
struct lfsck_instance *lc_lfsck;
struct lfsck_operations *lc_ops;
void *lc_file_ram;
void *lc_file_disk;
+ void *lc_data;
+
+ /* The time for last checkpoint, jiffies */
+ cfs_time_t lc_time_last_checkpoint;
+
+ /* The time for next checkpoint, jiffies */
+ cfs_time_t lc_time_next_checkpoint;
+
__u32 lc_file_size;
/* How many objects have been checked since last checkpoint. */
__u32 lc_new_checked;
+
+ /* How many objects have been scanned since last sleep. */
+ __u32 lc_new_scanned;
+
unsigned int lc_journal:1;
__u16 lc_type;
};
/* For the components those are not scanning now. */
cfs_list_t li_list_idle;
- cfs_atomic_t li_ref;
+ atomic_t li_ref;
+ atomic_t li_double_scan_count;
struct ptlrpc_thread li_thread;
/* The time for last checkpoint, jiffies */
/* The time for next checkpoint, jiffies */
cfs_time_t li_time_next_checkpoint;
+ lfsck_out_notify li_out_notify;
+ void *li_out_notify_data;
struct dt_device *li_next;
struct dt_device *li_bottom;
struct ldlm_namespace *li_namespace;
/* It for directory traversal */
struct dt_it *li_di_dir;
+ /* Description of OST */
+ struct lfsck_tgt_descs li_ost_descs;
+
+ /* Description of MDT */
+ struct lfsck_tgt_descs li_mdt_descs;
+
/* namespace-based directory traversal position. */
__u64 li_cookie_dir;
LLF_REPAIR_FAILED = 0x02,
};
+struct lfsck_thread_args {
+ struct lu_env lta_env;
+ struct lfsck_instance *lta_lfsck;
+ struct lfsck_component *lta_com;
+};
+
struct lfsck_thread_info {
struct lu_name lti_name;
struct lu_buf lti_buf;
struct lu_fid lti_fid;
struct lu_fid lti_fid2;
struct lu_attr lti_la;
+ struct ost_id lti_oi;
+ union {
+ struct lustre_mdt_attrs lti_lma;
+ /* old LMA for compatibility */
+ char lti_lma_old[LMA_OLD_SIZE];
+ };
+ struct dt_object_format lti_dof;
/* lti_ent and lti_key must be conjoint,
* then lti_ent::lde_name will be lti_key. */
struct lu_dirent lti_ent;
char lti_key[NAME_MAX + 16];
+ struct lfsck_request lti_lr;
};
/* lfsck_lib.c */
+const char *lfsck_status2names(enum lfsck_status status);
void lfsck_component_cleanup(const struct lu_env *env,
struct lfsck_component *com);
+void lfsck_instance_cleanup(const struct lu_env *env,
+ struct lfsck_instance *lfsck);
int lfsck_bits_dump(char **buf, int *len, int bits, const char *names[],
const char *prefix);
int lfsck_time_dump(char **buf, int *len, __u64 time, const char *prefix);
void lfsck_pos_fill(const struct lu_env *env, struct lfsck_instance *lfsck,
struct lfsck_position *pos, bool init);
void lfsck_control_speed(struct lfsck_instance *lfsck);
+void lfsck_control_speed_by_self(struct lfsck_component *com);
int lfsck_reset(const struct lu_env *env, struct lfsck_instance *lfsck,
bool init);
+struct lfsck_thread_args *lfsck_thread_args_init(struct lfsck_instance *lfsck,
+ struct lfsck_component *com);
+void lfsck_thread_args_fini(struct lfsck_thread_args *lta);
void lfsck_fail(const struct lu_env *env, struct lfsck_instance *lfsck,
bool new_checked);
int lfsck_checkpoint(const struct lu_env *env, struct lfsck_instance *lfsck);
int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck,
int result);
int lfsck_double_scan(const struct lu_env *env, struct lfsck_instance *lfsck);
+void lfsck_quit(const struct lu_env *env, struct lfsck_instance *lfsck);
/* lfsck_engine.c */
int lfsck_master_engine(void *args);
int lfsck_namespace_setup(const struct lu_env *env,
struct lfsck_instance *lfsck);
-extern const char *lfsck_status_names[];
+/* lfsck_layout.c */
+int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck);
+
extern const char *lfsck_flags_names[];
extern const char *lfsck_param_names[];
extern struct lu_context_key lfsck_thread_key;
lu_object_put(env, &obj->do_lu);
}
+static inline struct lfsck_tgt_desc *lfsck_tgt_get(struct lfsck_tgt_descs *ltds,
+ __u32 index)
+{
+ struct lfsck_tgt_desc *ltd;
+
+ ltd = LTD_TGT(ltds, index);
+ if (ltd != NULL)
+ atomic_inc(<d->ltd_ref);
+
+ return ltd;
+}
+
+static inline void lfsck_tgt_put(struct lfsck_tgt_desc *ltd)
+{
+ if (atomic_dec_and_test(<d->ltd_ref))
+ OBD_FREE_PTR(ltd);
+}
+
+static inline struct lfsck_component *
+lfsck_component_get(struct lfsck_component *com)
+{
+ atomic_inc(&com->lc_ref);
+
+ return com;
+}
+
+static inline void lfsck_component_put(const struct lu_env *env,
+ struct lfsck_component *com)
+{
+ if (atomic_dec_and_test(&com->lc_ref)) {
+ if (com->lc_obj != NULL)
+ lu_object_put_nocache(env, &com->lc_obj->do_lu);
+ if (com->lc_file_ram != NULL)
+ OBD_FREE(com->lc_file_ram, com->lc_file_size);
+ if (com->lc_file_disk != NULL)
+ OBD_FREE(com->lc_file_disk, com->lc_file_size);
+ if (com->lc_data != NULL) {
+ LASSERT(com->lc_ops->lfsck_data_release != NULL);
+
+ com->lc_ops->lfsck_data_release(env, com);
+ }
+
+ OBD_FREE_PTR(com);
+ }
+}
+
+static inline struct lfsck_instance *
+lfsck_instance_get(struct lfsck_instance *lfsck)
+{
+ atomic_inc(&lfsck->li_ref);
+
+ return lfsck;
+}
+
+static inline void lfsck_instance_put(const struct lu_env *env,
+ struct lfsck_instance *lfsck)
+{
+ if (atomic_dec_and_test(&lfsck->li_ref))
+ lfsck_instance_cleanup(env, lfsck);
+}
+
+static inline mdsno_t lfsck_dev_idx(struct dt_device *dev)
+{
+ return dev->dd_lu_dev.ld_site->ld_seq_site->ss_node_id;
+}
+
#endif /* _LFSCK_INTERNAL_H */
git://git.whamcloud.com / fs / lustre-release.git / blobdiff