#include "lfsck_internal.h"
-#define LFSCK_LAYOUT_MAGIC 0xB173AE14
+#define LFSCK_LAYOUT_MAGIC_V1 0xB173AE14
+#define LFSCK_LAYOUT_MAGIC_V2 0xB1734D76
+
+#define LFSCK_LAYOUT_MAGIC LFSCK_LAYOUT_MAGIC_V2
static const char lfsck_layout_name[] = "lfsck_layout";
struct dt_object *llo_obj;
struct lu_attr llo_attr;
atomic_t llo_ref;
+ __u64 llo_cookie;
__u16 llo_gen;
};
struct lfsck_layout_req {
- struct list_head llr_list;
+ struct lfsck_assistant_req llr_lar;
struct lfsck_layout_object *llr_parent;
struct dt_object *llr_child;
__u32 llr_ost_idx;
__u32 llr_lov_idx; /* offset in LOV EA */
};
-struct lfsck_layout_master_data {
- spinlock_t llmd_lock;
- struct list_head llmd_req_list;
-
- /* list for the ost targets involve layout verification. */
- struct list_head llmd_ost_list;
-
- /* list for the ost targets in phase1 scanning. */
- struct list_head llmd_ost_phase1_list;
-
- /* list for the ost targets in phase1 scanning. */
- struct list_head llmd_ost_phase2_list;
-
- /* list for the mdt targets involve layout verification. */
- struct list_head llmd_mdt_list;
-
- /* list for the mdt targets in phase1 scanning. */
- struct list_head llmd_mdt_phase1_list;
-
- /* list for the mdt targets in phase1 scanning. */
- struct list_head llmd_mdt_phase2_list;
-
- struct ptlrpc_thread llmd_thread;
- __u32 llmd_touch_gen;
- int llmd_prefetched;
- int llmd_assistant_status;
- int llmd_post_result;
- unsigned int llmd_to_post:1,
- llmd_to_double_scan:1,
- llmd_in_double_scan:1,
- llmd_exit:1;
-};
-
struct lfsck_layout_slave_async_args {
struct obd_export *llsaa_exp;
struct lfsck_component *llsaa_com;
static struct lfsck_layout_object *
lfsck_layout_object_init(const struct lu_env *env, struct dt_object *obj,
- __u16 gen)
+ __u64 cookie, __u16 gen)
{
struct lfsck_layout_object *llo;
int rc;
lu_object_get(&obj->do_lu);
llo->llo_obj = obj;
+ llo->llo_cookie = cookie;
/* The gen can be used to check whether some others have changed the
* file layout after LFSCK pre-fetching but before real verification. */
llo->llo_gen = gen;
}
static struct lfsck_layout_req *
-lfsck_layout_req_init(struct lfsck_layout_object *parent,
- struct dt_object *child, __u32 ost_idx, __u32 lov_idx)
+lfsck_layout_assistant_req_init(struct lfsck_layout_object *parent,
+ struct dt_object *child, __u32 ost_idx,
+ __u32 lov_idx)
{
struct lfsck_layout_req *llr;
if (llr == NULL)
return ERR_PTR(-ENOMEM);
- INIT_LIST_HEAD(&llr->llr_list);
+ INIT_LIST_HEAD(&llr->llr_lar.lar_list);
atomic_inc(&parent->llo_ref);
llr->llr_parent = parent;
llr->llr_child = child;
return llr;
}
-static inline void lfsck_layout_req_fini(const struct lu_env *env,
- struct lfsck_layout_req *llr)
+static void lfsck_layout_assistant_req_fini(const struct lu_env *env,
+ struct lfsck_assistant_req *lar)
{
+ struct lfsck_layout_req *llr =
+ container_of0(lar, struct lfsck_layout_req, llr_lar);
+
lu_object_put(env, &llr->llr_child->do_lu);
lfsck_layout_object_put(env, llr->llr_parent);
OBD_FREE_PTR(llr);
}
-static inline bool lfsck_layout_req_empty(struct lfsck_layout_master_data *llmd)
+static int
+lfsck_layout_assistant_sync_failures_interpret(const struct lu_env *env,
+ struct ptlrpc_request *req,
+ void *args, int rc)
{
- bool empty = false;
+ struct lfsck_async_interpret_args *laia = args;
- spin_lock(&llmd->llmd_lock);
- if (list_empty(&llmd->llmd_req_list))
- empty = true;
- spin_unlock(&llmd->llmd_lock);
+ if (rc == 0)
+ atomic_dec(laia->laia_count);
- return empty;
+ return 0;
+}
+
+/**
+ * Notify remote LFSCK instances about former failures.
+ *
+ * The local LFSCK instance has recorded which OSTs have ever failed to respond
+ * some LFSCK verification requests (maybe because of network issues or the OST
+ * itself trouble). During the respond gap, the OST may missed some OST-objects
+ * verification, then the OST cannot know whether related OST-objects have been
+ * referenced by related MDT-objects or not, then in the second-stage scanning,
+ * these OST-objects will be regarded as orphan, if the OST-object contains bad
+ * parent FID for back reference, then it will misguide the LFSCK to make wrong
+ * fixing for the fake orphan.
+ *
+ * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
+ * it will scan the bitmap for the ever failed OTs, and notify them that it has
+ * ever missed some OST-object verification and should skip orphan handling for
+ * all MDTs that are in layout LFSCK.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] lr pointer to the lfsck request
+ */
+static void lfsck_layout_assistant_sync_failures(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct lfsck_request *lr)
+{
+ struct lfsck_async_interpret_args *laia =
+ &lfsck_env_info(env)->lti_laia2;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_layout *lo = com->lc_file_ram;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
+ struct lfsck_tgt_desc *ltd;
+ struct ptlrpc_request_set *set;
+ atomic_t count;
+ __u32 idx;
+ int rc = 0;
+ ENTRY;
+
+ if (!lad->lad_incomplete || lo->ll_flags & LF_INCOMPLETE)
+ RETURN_EXIT;
+
+ /* If the MDT has ever failed to verfiy some OST-objects,
+ * then sync failures with them firstly. */
+ lr->lr_flags2 = lo->ll_flags | LF_INCOMPLETE;
+
+ atomic_set(&count, 0);
+ memset(laia, 0, sizeof(*laia));
+ laia->laia_count = &count;
+ set = ptlrpc_prep_set();
+ if (set == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ down_read(<ds->ltd_rw_sem);
+ cfs_foreach_bit(lad->lad_bitmap, idx) {
+ ltd = LTD_TGT(ltds, idx);
+ LASSERT(ltd != NULL);
+
+ spin_lock(<ds->ltd_lock);
+ list_del_init(<d->ltd_layout_phase_list);
+ list_del_init(<d->ltd_layout_list);
+ spin_unlock(<ds->ltd_lock);
+
+ rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
+ lfsck_layout_assistant_sync_failures_interpret,
+ laia, LFSCK_NOTIFY);
+ if (rc != 0) {
+ CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
+ "notify target %x for %s phase1 done: "
+ "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
+ ltd->ltd_index, lad->lad_name, rc);
+
+ break;
+ }
+
+ atomic_inc(&count);
+ }
+ up_read(<ds->ltd_rw_sem);
+
+ if (rc == 0 && atomic_read(&count) > 0)
+ rc = ptlrpc_set_wait(set);
+
+ ptlrpc_set_destroy(set);
+
+ if (rc == 0 && atomic_read(&count) > 0)
+ rc = -EINVAL;
+
+ GOTO(out, rc);
+
+out:
+ if (rc != 0)
+ /* If failed to sync failures with the OSTs, then have to
+ * mark the whole LFSCK as LF_INCOMPLETE to skip the whole
+ * subsequent orphan OST-object handling. */
+ lo->ll_flags |= LF_INCOMPLETE;
+
+ lr->lr_flags2 = lo->ll_flags;
}
static int lfsck_layout_get_lovea(const struct lu_env *env,
- struct dt_object *obj,
- struct lu_buf *buf, ssize_t *buflen)
+ struct dt_object *obj, struct lu_buf *buf)
{
int rc;
return rc;
lu_buf_realloc(buf, rc);
- if (buflen != NULL)
- *buflen = buf->lb_len;
-
if (buf->lb_buf == NULL)
return -ENOMEM;
if (unlikely(buf->lb_buf == NULL)) {
lu_buf_alloc(buf, rc);
- if (buflen != NULL)
- *buflen = buf->lb_len;
-
if (buf->lb_buf == NULL)
return -ENOMEM;
struct ost_id oi;
int rc;
- lmm_oi_cpu_to_le(&oi, &lmm->lmm_oi);
+ lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
rc = -EOPNOTSUPP;
else
if (lov_pattern(pattern) != LOV_PATTERN_RAID0) {
struct ost_id oi;
- lmm_oi_cpu_to_le(&oi, &lmm->lmm_oi);
+ lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u on "DOSTID"\n",
pattern, POSTID(&oi));
des->ll_objs_repaired[i] =
le64_to_cpu(src->ll_objs_repaired[i]);
des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
+ des->ll_bitmap_size = le32_to_cpu(src->ll_bitmap_size);
}
static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
des->ll_objs_repaired[i] =
cpu_to_le64(src->ll_objs_repaired[i]);
des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
+ des->ll_bitmap_size = cpu_to_le32(src->ll_bitmap_size);
}
/**
- * \retval +ve: the lfsck_layout is broken, the caller should reset it.
- * \retval 0: succeed.
- * \retval -ve: failed cases.
+ * Load the OST bitmap from the lfsck_layout tracing file.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ *
+ * \retval positive number for data corruption
+ * \retval 0 for success
+ * \retval negative error number on failure
+ */
+static int lfsck_layout_load_bitmap(const struct lu_env *env,
+ struct lfsck_component *com)
+{
+ struct dt_object *obj = com->lc_obj;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_layout *lo = com->lc_file_ram;
+ const struct dt_body_operations *dbo = obj->do_body_ops;
+ cfs_bitmap_t *bitmap = lad->lad_bitmap;
+ loff_t pos = com->lc_file_size;
+ ssize_t size;
+ __u32 nbits;
+ int rc;
+ ENTRY;
+
+ if (com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size >
+ lo->ll_bitmap_size)
+ nbits = com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size;
+ else
+ nbits = lo->ll_bitmap_size;
+
+ if (unlikely(nbits < BITS_PER_LONG))
+ nbits = BITS_PER_LONG;
+
+ if (nbits > bitmap->size) {
+ __u32 new_bits = bitmap->size;
+ cfs_bitmap_t *new_bitmap;
+
+ while (new_bits < nbits)
+ new_bits <<= 1;
+
+ new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
+ if (new_bitmap == NULL)
+ RETURN(-ENOMEM);
+
+ lad->lad_bitmap = new_bitmap;
+ CFS_FREE_BITMAP(bitmap);
+ bitmap = new_bitmap;
+ }
+
+ if (lo->ll_bitmap_size == 0) {
+ lad->lad_incomplete = 0;
+ CFS_RESET_BITMAP(bitmap);
+
+ RETURN(0);
+ }
+
+ size = (lo->ll_bitmap_size + 7) >> 3;
+ rc = dbo->dbo_read(env, obj,
+ lfsck_buf_get(env, bitmap->data, size), &pos,
+ BYPASS_CAPA);
+ if (rc == 0) {
+ RETURN(-ENOENT);
+ } else if (rc != size) {
+ CDEBUG(D_LFSCK, "%s: lfsck_layout bitmap size %u != %u\n",
+ lfsck_lfsck2name(com->lc_lfsck),
+ (unsigned int)size, rc);
+
+ RETURN(rc);
+ }
+
+ if (cfs_bitmap_check_empty(bitmap))
+ lad->lad_incomplete = 0;
+ else
+ lad->lad_incomplete = 1;
+
+ RETURN(0);
+}
+
+/**
+ * Load the layout LFSCK tracing file from disk.
+ *
+ * The layout LFSCK tracing file records the layout LFSCK status information
+ * and other statistics, such as how many objects have been scanned, and how
+ * many objects have been repaired, and etc. It also contains the bitmap for
+ * failed OSTs during the layout LFSCK. All these information will be loaded
+ * from disk to RAM when the layout LFSCK component setup.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ *
+ * \retval positive number for file data corruption, the caller
+ * should reset the layout LFSCK tracing file
+ * \retval 0 for success
+ * \retval negative error number on failure
*/
static int lfsck_layout_load(const struct lu_env *env,
struct lfsck_component *com)
return 0;
}
+/**
+ * Store the layout LFSCK tracing file on disk.
+ *
+ * The layout LFSCK tracing file records the layout LFSCK status information
+ * and other statistics, such as how many objects have been scanned, and how
+ * many objects have been repaired, and etc. It also contains the bitmap for
+ * failed OSTs during the layout LFSCK. All these information will be synced
+ * from RAM to disk periodically.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ *
+ * \retval 0 for success
+ * \retval negative error number on failure
+ */
static int lfsck_layout_store(const struct lu_env *env,
struct lfsck_component *com)
{
- struct dt_object *obj = com->lc_obj;
- struct lfsck_instance *lfsck = com->lc_lfsck;
- struct lfsck_layout *lo = com->lc_file_disk;
- struct thandle *handle;
- ssize_t size = com->lc_file_size;
- loff_t pos = 0;
- int rc;
+ struct dt_object *obj = com->lc_obj;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_layout *lo_ram = com->lc_file_ram;
+ struct lfsck_layout *lo = com->lc_file_disk;
+ struct thandle *th;
+ struct dt_device *dev = lfsck->li_bottom;
+ cfs_bitmap_t *bitmap = NULL;
+ loff_t pos;
+ ssize_t size = com->lc_file_size;
+ __u32 nbits = 0;
+ int rc;
ENTRY;
- lfsck_layout_cpu_to_le(lo, com->lc_file_ram);
- handle = dt_trans_create(env, lfsck->li_bottom);
- if (IS_ERR(handle))
- GOTO(log, rc = PTR_ERR(handle));
+ if (lfsck->li_master) {
+ struct lfsck_assistant_data *lad = com->lc_data;
+
+ bitmap = lad->lad_bitmap;
+ nbits = bitmap->size;
+
+ LASSERT(nbits > 0);
+ LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
+ }
+
+ lo_ram->ll_bitmap_size = nbits;
+ lfsck_layout_cpu_to_le(lo, lo_ram);
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ GOTO(log, rc = PTR_ERR(th));
rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
- pos, handle);
+ (loff_t)0, th);
if (rc != 0)
GOTO(out, rc);
- rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
+ if (bitmap != NULL) {
+ rc = dt_declare_record_write(env, obj,
+ lfsck_buf_get(env, bitmap->data, nbits >> 3),
+ (loff_t)size, th);
+ if (rc != 0)
+ GOTO(out, rc);
+ }
+
+ rc = dt_trans_start_local(env, dev, th);
if (rc != 0)
GOTO(out, rc);
- rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos,
- handle);
+ pos = 0;
+ rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos, th);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ if (bitmap != NULL) {
+ pos = size;
+ rc = dt_record_write(env, obj,
+ lfsck_buf_get(env, bitmap->data, nbits >> 3),
+ &pos, th);
+ }
GOTO(out, rc);
out:
- dt_trans_stop(env, lfsck->li_bottom, handle);
+ dt_trans_stop(env, dev, th);
log:
if (rc != 0)
CDEBUG(D_LFSCK, "%s: fail to store lfsck_layout: rc = %d\n",
lfsck_lfsck2name(lfsck), rc);
+
return rc;
}
GOTO(stop, rc);
dt_write_lock(env, obj, 0);
- if (likely(!dt_object_exists(obj))) {
+ if (likely(dt_object_exists(obj) == 0)) {
rc = dt_create(env, obj, la, NULL, dof, th);
if (rc == 0)
rc = dt_record_write(env, obj,
lfsck->li_out_notify(env, lfsck->li_out_notify_data,
LE_LASTID_REBUILDING);
lo->ll_flags |= LF_CRASHED_LASTID;
+
+ CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
+ "LAST_ID file (1) for the sequence "LPX64
+ ", old value "LPU64", known value "LPU64"\n",
+ lfsck_lfsck2name(lfsck), lls->lls_seq,
+ lastid, lls->lls_lastid);
}
} else if (lastid >= lls->lls_lastid) {
lls->lls_lastid = lastid;
list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
loff_t pos = 0;
- /* XXX: Add the code back if we really found related
- * inconsistent cases in the future. */
-#if 0
- if (!lls->lls_dirty) {
- /* In OFD, before the pre-creation, the LAST_ID
- * file will be updated firstly, which may hide
- * some potential crashed cases. For example:
- *
- * The old obj1's ID is higher than old LAST_ID
- * but lower than the new LAST_ID, but the LFSCK
- * have not touch the obj1 until the OFD updated
- * the LAST_ID. So the LFSCK does not regard it
- * as crashed case. But when OFD does not create
- * successfully, it will set the LAST_ID as the
- * real created objects' ID, then LFSCK needs to
- * found related inconsistency. */
- rc = lfsck_layout_lastid_reload(env, com, lls);
- if (likely(!lls->lls_dirty))
- continue;
- }
-#endif
+ if (!lls->lls_dirty)
+ continue;
CDEBUG(D_LFSCK, "%s: layout LFSCK will sync the LAST_ID for "
"<seq> "LPX64" as <oid> "LPU64"\n",
RETURN(PTR_ERR(obj));
/* LAST_ID crashed, to be rebuilt */
- if (!dt_object_exists(obj)) {
+ if (dt_object_exists(obj) == 0) {
if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
LASSERT(lfsck->li_out_notify != NULL);
LE_LASTID_REBUILDING);
lo->ll_flags |= LF_CRASHED_LASTID;
+ CDEBUG(D_LFSCK, "%s: layout LFSCK cannot find the "
+ "LAST_ID file for sequence "LPX64"\n",
+ lfsck_lfsck2name(lfsck), lls->lls_seq);
+
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
cfs_fail_val > 0) {
struct l_wait_info lwi = LWI_TIMEOUT(
lfsck->li_out_notify(env, lfsck->li_out_notify_data,
LE_LASTID_REBUILDING);
lo->ll_flags |= LF_CRASHED_LASTID;
+
+ CDEBUG(D_LFSCK, "%s: layout LFSCK finds invalid "
+ "LAST_ID file for the sequence "LPX64
+ ": rc = %d\n",
+ lfsck_lfsck2name(lfsck), lls->lls_seq, rc);
}
lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
}
static void lfsck_layout_record_failure(const struct lu_env *env,
- struct lfsck_instance *lfsck,
- struct lfsck_layout *lo)
+ struct lfsck_instance *lfsck,
+ struct lfsck_layout *lo)
{
+ __u64 cookie;
+
lo->ll_objs_failed_phase1++;
- if (unlikely(lo->ll_pos_first_inconsistent == 0)) {
- lo->ll_pos_first_inconsistent =
- lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
+ cookie = lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
lfsck->li_di_oit);
+ if (lo->ll_pos_first_inconsistent == 0 ||
+ lo->ll_pos_first_inconsistent < cookie) {
+ lo->ll_pos_first_inconsistent = cookie;
CDEBUG(D_LFSCK, "%s: layout LFSCK hit first non-repaired "
"inconsistency at the pos ["LPU64"]\n",
}
}
-static int lfsck_layout_master_async_interpret(const struct lu_env *env,
- struct ptlrpc_request *req,
- void *args, int rc)
-{
- struct lfsck_async_interpret_args *laia = args;
- struct lfsck_component *com = laia->laia_com;
- struct lfsck_layout_master_data *llmd = com->lc_data;
- struct lfsck_tgt_descs *ltds = laia->laia_ltds;
- struct lfsck_tgt_desc *ltd = laia->laia_ltd;
- struct lfsck_request *lr = laia->laia_lr;
-
- switch (lr->lr_event) {
- case LE_START:
- if (rc != 0) {
- struct lfsck_layout *lo = com->lc_file_ram;
-
- CDEBUG(D_LFSCK, "%s: fail to notify %s %x for layout "
- "start: rc = %d\n",
- lfsck_lfsck2name(com->lc_lfsck),
- (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
- ltd->ltd_index, rc);
- lo->ll_flags |= LF_INCOMPLETE;
- break;
- }
-
- spin_lock(<ds->ltd_lock);
- if (ltd->ltd_dead || ltd->ltd_layout_done) {
- spin_unlock(<ds->ltd_lock);
- break;
- }
-
- if (lr->lr_flags & LEF_TO_OST) {
- if (list_empty(<d->ltd_layout_list))
- list_add_tail(<d->ltd_layout_list,
- &llmd->llmd_ost_list);
- if (list_empty(<d->ltd_layout_phase_list))
- list_add_tail(<d->ltd_layout_phase_list,
- &llmd->llmd_ost_phase1_list);
- } else {
- if (list_empty(<d->ltd_layout_list))
- list_add_tail(<d->ltd_layout_list,
- &llmd->llmd_mdt_list);
- if (list_empty(<d->ltd_layout_phase_list))
- list_add_tail(<d->ltd_layout_phase_list,
- &llmd->llmd_mdt_phase1_list);
- }
- spin_unlock(<ds->ltd_lock);
- break;
- case LE_STOP:
- case LE_PHASE1_DONE:
- case LE_PHASE2_DONE:
- case LE_PEER_EXIT:
- if (rc != 0 && rc != -EALREADY)
- CDEBUG(D_LFSCK, "%s: fail to notify %s %x for layout: "
- "event = %d, rc = %d\n",
- lfsck_lfsck2name(com->lc_lfsck),
- (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
- ltd->ltd_index, lr->lr_event, rc);
- break;
- case LE_QUERY: {
- struct lfsck_reply *reply;
-
- if (rc != 0) {
- spin_lock(<ds->ltd_lock);
- list_del_init(<d->ltd_layout_phase_list);
- list_del_init(<d->ltd_layout_list);
- spin_unlock(<ds->ltd_lock);
- break;
- }
-
- reply = req_capsule_server_get(&req->rq_pill,
- &RMF_LFSCK_REPLY);
- if (reply == NULL) {
- rc = -EPROTO;
- CDEBUG(D_LFSCK, "%s: invalid query reply: rc = %d\n",
- lfsck_lfsck2name(com->lc_lfsck), rc);
- spin_lock(<ds->ltd_lock);
- list_del_init(<d->ltd_layout_phase_list);
- list_del_init(<d->ltd_layout_list);
- spin_unlock(<ds->ltd_lock);
- break;
- }
-
- switch (reply->lr_status) {
- case LS_SCANNING_PHASE1:
- break;
- case LS_SCANNING_PHASE2:
- spin_lock(<ds->ltd_lock);
- list_del_init(<d->ltd_layout_phase_list);
- if (ltd->ltd_dead || ltd->ltd_layout_done) {
- spin_unlock(<ds->ltd_lock);
- break;
- }
-
- if (lr->lr_flags & LEF_TO_OST)
- list_add_tail(<d->ltd_layout_phase_list,
- &llmd->llmd_ost_phase2_list);
- else
- list_add_tail(<d->ltd_layout_phase_list,
- &llmd->llmd_mdt_phase2_list);
- spin_unlock(<ds->ltd_lock);
- break;
- default:
- spin_lock(<ds->ltd_lock);
- list_del_init(<d->ltd_layout_phase_list);
- list_del_init(<d->ltd_layout_list);
- spin_unlock(<ds->ltd_lock);
- break;
- }
- break;
- }
- default:
- CDEBUG(D_LFSCK, "%s: layout LFSCK unexpected event: rc = %d\n",
- lfsck_lfsck2name(com->lc_lfsck), lr->lr_event);
- break;
- }
-
- if (!laia->laia_shared) {
- lfsck_tgt_put(ltd);
- lfsck_component_put(env, com);
- }
-
- return 0;
-}
-
-static int lfsck_layout_master_query_others(const struct lu_env *env,
- struct lfsck_component *com)
-{
- struct lfsck_thread_info *info = lfsck_env_info(env);
- struct lfsck_request *lr = &info->lti_lr;
- struct lfsck_async_interpret_args *laia = &info->lti_laia;
- struct lfsck_instance *lfsck = com->lc_lfsck;
- struct lfsck_layout_master_data *llmd = com->lc_data;
- struct ptlrpc_request_set *set;
- struct lfsck_tgt_descs *ltds;
- struct lfsck_tgt_desc *ltd;
- struct list_head *head;
- int rc = 0;
- int rc1 = 0;
- ENTRY;
-
- set = ptlrpc_prep_set();
- if (set == NULL)
- RETURN(-ENOMEM);
-
- llmd->llmd_touch_gen++;
- memset(lr, 0, sizeof(*lr));
- lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
- lr->lr_event = LE_QUERY;
- lr->lr_active = LFSCK_TYPE_LAYOUT;
- laia->laia_com = com;
- laia->laia_lr = lr;
- laia->laia_shared = 0;
-
- if (!list_empty(&llmd->llmd_mdt_phase1_list)) {
- ltds = &lfsck->li_mdt_descs;
- lr->lr_flags = 0;
- head = &llmd->llmd_mdt_phase1_list;
- } else {
-
-again:
- ltds = &lfsck->li_ost_descs;
- lr->lr_flags = LEF_TO_OST;
- head = &llmd->llmd_ost_phase1_list;
- }
-
- laia->laia_ltds = ltds;
- spin_lock(<ds->ltd_lock);
- while (!list_empty(head)) {
- ltd = list_entry(head->next,
- struct lfsck_tgt_desc,
- ltd_layout_phase_list);
- if (ltd->ltd_layout_gen == llmd->llmd_touch_gen)
- break;
-
- ltd->ltd_layout_gen = llmd->llmd_touch_gen;
- list_del(<d->ltd_layout_phase_list);
- list_add_tail(<d->ltd_layout_phase_list, head);
- atomic_inc(<d->ltd_ref);
- laia->laia_ltd = ltd;
- spin_unlock(<ds->ltd_lock);
- rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
- lfsck_layout_master_async_interpret,
- laia, LFSCK_QUERY);
- if (rc != 0) {
- CDEBUG(D_LFSCK, "%s: layout LFSCK fail to query %s %x: "
- "rc = %d\n", lfsck_lfsck2name(lfsck),
- (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
- ltd->ltd_index, rc);
- lfsck_tgt_put(ltd);
- rc1 = rc;
- }
- spin_lock(<ds->ltd_lock);
- }
- spin_unlock(<ds->ltd_lock);
-
- rc = ptlrpc_set_wait(set);
- if (rc < 0) {
- ptlrpc_set_destroy(set);
- RETURN(rc);
- }
-
- if (!(lr->lr_flags & LEF_TO_OST) &&
- list_empty(&llmd->llmd_mdt_phase1_list))
- goto again;
-
- ptlrpc_set_destroy(set);
-
- RETURN(rc1 != 0 ? rc1 : rc);
-}
-
-static inline bool
-lfsck_layout_master_to_orphan(struct lfsck_layout_master_data *llmd)
-{
- return list_empty(&llmd->llmd_mdt_phase1_list) &&
- (!list_empty(&llmd->llmd_ost_phase2_list) ||
- list_empty(&llmd->llmd_ost_phase1_list));
-}
-
-static int lfsck_layout_master_notify_others(const struct lu_env *env,
- struct lfsck_component *com,
- struct lfsck_request *lr)
-{
- struct lfsck_thread_info *info = lfsck_env_info(env);
- struct lfsck_async_interpret_args *laia = &info->lti_laia;
- struct lfsck_instance *lfsck = com->lc_lfsck;
- struct lfsck_layout_master_data *llmd = com->lc_data;
- struct lfsck_layout *lo = com->lc_file_ram;
- struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
- struct ptlrpc_request_set *set;
- struct lfsck_tgt_descs *ltds;
- struct lfsck_tgt_desc *ltd;
- struct lfsck_tgt_desc *next;
- struct list_head *head;
- __u32 idx;
- int rc = 0;
- ENTRY;
-
- set = ptlrpc_prep_set();
- if (set == NULL)
- RETURN(-ENOMEM);
-
- lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
- lr->lr_active = LFSCK_TYPE_LAYOUT;
- laia->laia_com = com;
- laia->laia_lr = lr;
- laia->laia_shared = 0;
- switch (lr->lr_event) {
- case LE_START:
- /* Notify OSTs firstly, then handle other MDTs if needed. */
- ltds = &lfsck->li_ost_descs;
- laia->laia_ltds = ltds;
- down_read(<ds->ltd_rw_sem);
- cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
- ltd = lfsck_tgt_get(ltds, idx);
- LASSERT(ltd != NULL);
-
- laia->laia_ltd = ltd;
- ltd->ltd_layout_done = 0;
- rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
- lfsck_layout_master_async_interpret,
- laia, LFSCK_NOTIFY);
- if (rc != 0) {
- CDEBUG(D_LFSCK, "%s: layout LFSCK fail to "
- "notify %s %x for start: rc = %d\n",
- lfsck_lfsck2name(lfsck),
- (lr->lr_flags & LEF_TO_OST) ? "OST" :
- "MDT", idx, rc);
- lfsck_tgt_put(ltd);
- lo->ll_flags |= LF_INCOMPLETE;
- }
- }
- up_read(<ds->ltd_rw_sem);
-
- /* Sync up */
- rc = ptlrpc_set_wait(set);
- if (rc < 0) {
- ptlrpc_set_destroy(set);
- RETURN(rc);
- }
-
- if (!(bk->lb_param & LPF_ALL_TGT))
- break;
-
- /* link other MDT targets locallly. */
- ltds = &lfsck->li_mdt_descs;
- spin_lock(<ds->ltd_lock);
- cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
- ltd = LTD_TGT(ltds, idx);
- LASSERT(ltd != NULL);
-
- if (!list_empty(<d->ltd_layout_list))
- continue;
-
- list_add_tail(<d->ltd_layout_list,
- &llmd->llmd_mdt_list);
- list_add_tail(<d->ltd_layout_phase_list,
- &llmd->llmd_mdt_phase1_list);
- }
- spin_unlock(<ds->ltd_lock);
- break;
- case LE_STOP:
- case LE_PHASE2_DONE:
- case LE_PEER_EXIT: {
- /* Handle other MDTs firstly if needed, then notify the OSTs. */
- if (bk->lb_param & LPF_ALL_TGT) {
- head = &llmd->llmd_mdt_list;
- ltds = &lfsck->li_mdt_descs;
- if (lr->lr_event == LE_STOP) {
- /* unlink other MDT targets locallly. */
- spin_lock(<ds->ltd_lock);
- list_for_each_entry_safe(ltd, next, head,
- ltd_layout_list) {
- list_del_init(<d->ltd_layout_phase_list);
- list_del_init(<d->ltd_layout_list);
- }
- spin_unlock(<ds->ltd_lock);
-
- lr->lr_flags |= LEF_TO_OST;
- head = &llmd->llmd_ost_list;
- ltds = &lfsck->li_ost_descs;
- } else {
- lr->lr_flags &= ~LEF_TO_OST;
- }
- } else {
- lr->lr_flags |= LEF_TO_OST;
- head = &llmd->llmd_ost_list;
- ltds = &lfsck->li_ost_descs;
- }
-
-again:
- laia->laia_ltds = ltds;
- spin_lock(<ds->ltd_lock);
- while (!list_empty(head)) {
- ltd = list_entry(head->next, struct lfsck_tgt_desc,
- ltd_layout_list);
- if (!list_empty(<d->ltd_layout_phase_list))
- list_del_init(<d->ltd_layout_phase_list);
- list_del_init(<d->ltd_layout_list);
- atomic_inc(<d->ltd_ref);
- laia->laia_ltd = ltd;
- spin_unlock(<ds->ltd_lock);
- rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
- lfsck_layout_master_async_interpret,
- laia, LFSCK_NOTIFY);
- if (rc != 0) {
- CDEBUG(D_LFSCK, "%s: layout LFSCK fail to "
- "notify %s %x for stop/phase2_done/"
- "peer_exit: rc = %d\n",
- lfsck_lfsck2name(lfsck),
- (lr->lr_flags & LEF_TO_OST) ? "OST" :
- "MDT", ltd->ltd_index, rc);
- lfsck_tgt_put(ltd);
- }
- spin_lock(<ds->ltd_lock);
- }
- spin_unlock(<ds->ltd_lock);
-
- rc = ptlrpc_set_wait(set);
- if (rc < 0) {
- ptlrpc_set_destroy(set);
- RETURN(rc);
- }
-
- if (!(lr->lr_flags & LEF_TO_OST)) {
- lr->lr_flags |= LEF_TO_OST;
- head = &llmd->llmd_ost_list;
- ltds = &lfsck->li_ost_descs;
- goto again;
- }
- break;
- }
- case LE_PHASE1_DONE:
- llmd->llmd_touch_gen++;
- ltds = &lfsck->li_mdt_descs;
- laia->laia_ltds = ltds;
- spin_lock(<ds->ltd_lock);
- while (!list_empty(&llmd->llmd_mdt_phase1_list)) {
- ltd = list_entry(llmd->llmd_mdt_phase1_list.next,
- struct lfsck_tgt_desc,
- ltd_layout_phase_list);
- if (ltd->ltd_layout_gen == llmd->llmd_touch_gen)
- break;
-
- ltd->ltd_layout_gen = llmd->llmd_touch_gen;
- list_del_init(<d->ltd_layout_phase_list);
- list_add_tail(<d->ltd_layout_phase_list,
- &llmd->llmd_mdt_phase1_list);
- atomic_inc(<d->ltd_ref);
- laia->laia_ltd = ltd;
- spin_unlock(<ds->ltd_lock);
- rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
- lfsck_layout_master_async_interpret,
- laia, LFSCK_NOTIFY);
- if (rc != 0) {
- CDEBUG(D_LFSCK, "%s: layout LFSCK fail to "
- "notify MDT %x for phase1_done: "
- "rc = %d\n", lfsck_lfsck2name(lfsck),
- ltd->ltd_index, rc);
- lfsck_tgt_put(ltd);
- }
- spin_lock(<ds->ltd_lock);
- }
- spin_unlock(<ds->ltd_lock);
- break;
- default:
- CDEBUG(D_LFSCK, "%s: layout LFSCK unexpected event: rc = %d\n",
- lfsck_lfsck2name(lfsck), lr->lr_event);
- rc = -EINVAL;
- break;
- }
-
- rc = ptlrpc_set_wait(set);
- ptlrpc_set_destroy(set);
-
- RETURN(rc);
-}
-
static int lfsck_layout_double_scan_result(const struct lu_env *env,
struct lfsck_component *com,
int rc)
{
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_layout *lo = com->lc_file_ram;
- struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
down_write(&com->lc_sem);
lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
if (rc > 0) {
com->lc_journal = 0;
- if (lo->ll_flags & LF_INCOMPLETE)
+ if (lo->ll_flags & LF_INCOMPLETE) {
lo->ll_status = LS_PARTIAL;
- else
- lo->ll_status = LS_COMPLETED;
- if (!(bk->lb_param & LPF_DRYRUN))
+ } else {
+ if (lfsck->li_master) {
+ struct lfsck_assistant_data *lad = com->lc_data;
+
+ if (lad->lad_incomplete)
+ lo->ll_status = LS_PARTIAL;
+ else
+ lo->ll_status = LS_COMPLETED;
+ } else {
+ lo->ll_status = LS_COMPLETED;
+ }
+ }
+ if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
lo->ll_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
lo->ll_success_count++;
return rc;
}
-static int lfsck_layout_lock(const struct lu_env *env,
- struct lfsck_component *com,
- struct dt_object *obj,
- struct lustre_handle *lh, __u64 bits)
-{
- struct lfsck_thread_info *info = lfsck_env_info(env);
- ldlm_policy_data_t *policy = &info->lti_policy;
- struct ldlm_res_id *resid = &info->lti_resid;
- struct lfsck_instance *lfsck = com->lc_lfsck;
- __u64 flags = LDLM_FL_ATOMIC_CB;
- int rc;
-
- LASSERT(lfsck->li_namespace != NULL);
-
- memset(policy, 0, sizeof(*policy));
- policy->l_inodebits.bits = bits;
- fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
- rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_IBITS,
- policy, LCK_EX, &flags, ldlm_blocking_ast,
- ldlm_completion_ast, NULL, NULL, 0,
- LVB_T_NONE, NULL, lh);
- if (rc == ELDLM_OK) {
- rc = 0;
- } else {
- memset(lh, 0, sizeof(*lh));
- rc = -EIO;
- }
-
- return rc;
-}
-
-static void lfsck_layout_unlock(struct lustre_handle *lh)
-{
- if (lustre_handle_is_used(lh)) {
- ldlm_lock_decref(lh, LCK_EX);
- memset(lh, 0, sizeof(*lh));
- }
-}
-
static int lfsck_layout_trans_stop(const struct lu_env *env,
struct dt_device *dev,
struct thandle *handle, int result)
{
struct ost_id *oi = &lfsck_env_info(env)->lti_oi;
struct lov_mds_md_v1 *lmm = buf->lb_buf;
+ struct lu_buf ea_buf;
int rc;
+ __u32 magic;
+ __u16 count;
+
+ magic = le32_to_cpu(lmm->lmm_magic);
+ count = le16_to_cpu(lmm->lmm_stripe_count);
fid_to_ostid(cfid, oi);
ostid_cpu_to_le(oi, &slot->l_ost_oi);
if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE) {
struct lov_ost_data_v1 *objs;
int i;
- __u16 count;
- count = le16_to_cpu(lmm->lmm_stripe_count);
- if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
+ if (magic == LOV_MAGIC_V1)
objs = &lmm->lmm_objects[0];
else
objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
lmm->lmm_pattern &= ~cpu_to_le32(LOV_PATTERN_F_HOLE);
}
- rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV, fl, handle,
+ lfsck_buf_init(&ea_buf, lmm, lov_mds_md_size(count, magic));
+ rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle,
BYPASS_CAPA);
if (rc == 0)
rc = 1;
__u32 pattern = LOV_PATTERN_RAID0;
count = ea_off + 1;
- LASSERT(buf->lb_len == lov_mds_md_size(count, LOV_MAGIC_V1));
+ LASSERT(buf->lb_len >= lov_mds_md_size(count, LOV_MAGIC_V1));
if (ea_off != 0 || reset) {
pattern |= LOV_PATTERN_F_HOLE;
gap = ea_off - count;
if (gap >= 0)
count = ea_off + 1;
- LASSERT(buf->lb_len == lov_mds_md_size(count, magic));
+ LASSERT(buf->lb_len >= lov_mds_md_size(count, magic));
if (gap > 0) {
memset(objs, 0, gap * sizeof(*objs));
*
* type "R": The orphan OST-object knows its parent MDT-object FID,
* but does not know the position (the file name) in the
- * namespace.
+ * layout.
*
* The orphan name will be like:
* ${FID}-${infix}-${type}-${conflict_version}
struct dt_object *pobj = NULL;
struct dt_object *cobj = NULL;
struct thandle *th = NULL;
- struct lu_buf *pbuf = NULL;
+ struct lu_buf pbuf = { 0 };
struct lu_buf *ea_buf = &info->lti_big_buf;
+ struct lu_buf lov_buf;
struct lustre_handle lh = { 0 };
struct linkea_data ldata = { 0 };
struct lu_buf linkea_buf;
const struct lu_name *pname;
- int buflen = ea_buf->lb_len;
+ int size = 0;
int idx = 0;
int rc = 0;
ENTRY;
* real parent MDT-object's FID::f_ver, instead it is the
* OST-object index in its parent MDT-object's layout EA. */
ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
- pbuf = lfsck_buf_get(env, ff, sizeof(struct filter_fid));
+ lfsck_buf_init(&pbuf, ff, sizeof(struct filter_fid));
cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
if (IS_ERR(cobj))
GOTO(log, rc = PTR_ERR(cobj));
memset(dof, 0, sizeof(*dof));
dof->dof_type = dt_mode_to_dft(S_IFREG);
- rc = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
- if (buflen < rc) {
- lu_buf_realloc(ea_buf, rc);
- buflen = ea_buf->lb_len;
+ size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
+ if (ea_buf->lb_len < size) {
+ lu_buf_realloc(ea_buf, size);
if (ea_buf->lb_buf == NULL)
GOTO(put, rc = -ENOMEM);
- } else {
- ea_buf->lb_len = rc;
}
/* Hold update lock on the .lustre/lost+found/MDTxxxx/.
* because creating MDT-object for orphan OST-object is rare, we
* do not much care about the performance. It can be improved in
* the future when needed. */
- rc = lfsck_layout_lock(env, com, lfsck->li_lpf_obj, &lh,
- MDS_INODELOCK_UPDATE);
+ rc = lfsck_ibits_lock(env, lfsck, lfsck->li_lpf_obj, &lh,
+ MDS_INODELOCK_UPDATE, LCK_EX);
if (rc != 0)
GOTO(put, rc);
* If other subsequent modifications failed, then next LFSCK scanning
* will process the OST-object as orphan again with known parent FID. */
if (cobj != NULL) {
- rc = dt_declare_xattr_set(env, cobj, pbuf, XATTR_NAME_FID, 0, th);
+ rc = dt_declare_xattr_set(env, cobj, &pbuf, XATTR_NAME_FID,
+ 0, th);
if (rc != 0)
GOTO(stop, rc);
}
GOTO(stop, rc);
/* 3a. Add layout EA for the MDT-object. */
- rc = dt_declare_xattr_set(env, pobj, ea_buf, XATTR_NAME_LOV,
+ lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
+ rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
LU_XATTR_CREATE, th);
if (rc != 0)
GOTO(stop, rc);
GOTO(stop, rc);
/* 5a. insert linkEA for parent. */
- linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
- linkea_buf.lb_len = ldata.ld_leh->leh_len;
+ lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
+ ldata.ld_leh->leh_len);
rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
XATTR_NAME_LINK, 0, th);
if (rc != 0)
/* 1b. Update OST-object's parent information remotely. */
if (cobj != NULL) {
- rc = dt_xattr_set(env, cobj, pbuf, XATTR_NAME_FID, 0, th,
+ rc = dt_xattr_set(env, cobj, &pbuf, XATTR_NAME_FID, 0, th,
BYPASS_CAPA);
if (rc != 0)
GOTO(stop, rc);
if (rc == 0)
/* 3b. Add layout EA for the MDT-object. */
rc = lfsck_layout_extend_lovea(env, lfsck, th, pobj, cfid,
- ea_buf, LU_XATTR_CREATE,
+ &lov_buf, LU_XATTR_CREATE,
ltd->ltd_index, ea_off, false);
dt_write_unlock(env, pobj);
if (rc < 0)
dt_trans_stop(env, next, th);
unlock:
- lfsck_layout_unlock(&lh);
+ lfsck_ibits_unlock(&lh, LCK_EX);
put:
if (cobj != NULL && !IS_ERR(cobj))
lu_object_put(env, &cobj->do_lu);
if (pobj != NULL && !IS_ERR(pobj))
lu_object_put(env, &pobj->do_lu);
- ea_buf->lb_len = buflen;
log:
if (rc < 0)
RETURN(PTR_ERR(obj));
dt_read_lock(env, obj, 0);
- if (dt_object_exists(obj) == 0) {
+ if (dt_object_exists(obj) == 0 ||
+ lfsck_is_dead_obj(obj)) {
dt_read_unlock(env, obj);
GOTO(put, rc = -ENOENT);
struct lu_fid *cfid,
struct lu_buf *ea_buf,
struct lov_ost_data_v1 *slot,
- __u32 ea_off, __u32 ori_len)
+ __u32 ea_off)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
struct lu_fid *cfid2 = &info->lti_fid2;
ENTRY;
ostid_le_to_cpu(&slot->l_ost_oi, oi);
- ostid_to_fid(cfid2, oi, ost_idx2);
+ rc = ostid_to_fid(cfid2, oi, ost_idx2);
+ if (rc != 0)
+ GOTO(out, rc);
/* Hold layout lock on the parent to prevent others to access. */
- rc = lfsck_layout_lock(env, com, parent, &lh,
- MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
+ rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
+ MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
+ LCK_EX);
if (rc != 0)
GOTO(out, rc);
* a new MDT-object for the orphan OST-object. */
if (rc == -ETXTBSY) {
/* No need the layout lock on the original parent. */
- lfsck_layout_unlock(&lh);
- ea_buf->lb_len = ori_len;
+ lfsck_ibits_unlock(&lh, LCK_EX);
fid_zero(&rec->lor_fid);
snprintf(infix, LFSCK_TMPBUF_LEN, "-"DFID"-%x",
dt_trans_stop(env, dev, th);
unlock:
- lfsck_layout_unlock(&lh);
+ lfsck_ibits_unlock(&lh, LCK_EX);
out:
- ea_buf->lb_len = ori_len;
-
CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
"OST-object "DFID" on the OST %x with the orphan "DFID" on "
"the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
struct dt_device *dt = lfsck->li_bottom;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct thandle *handle = NULL;
- size_t buflen = buf->lb_len;
size_t lovea_size;
struct lov_mds_md_v1 *lmm;
struct lov_ost_data_v1 *objs;
bool locked = false;
ENTRY;
- rc = lfsck_layout_lock(env, com, parent, &lh,
- MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
+ rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
+ MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
+ LCK_EX);
if (rc != 0) {
CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
"LOV EA for "DFID": parent "DFID", OST-index %u, "
lovea_size = rc;
if (buf->lb_len < lovea_size) {
lu_buf_realloc(buf, lovea_size);
- buflen = buf->lb_len;
if (buf->lb_buf == NULL)
GOTO(unlock_layout, rc = -ENOMEM);
}
LASSERT(buf->lb_len >= lovea_size);
- buf->lb_len = lovea_size;
rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
buf, fl, ost_idx, ea_off, false);
LASSERT(buf->lb_len >= lovea_size);
- buf->lb_len = lovea_size;
- memset(lmm, 0, buf->lb_len);
rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
buf, fl, ost_idx, ea_off, true);
rc = lovea_size;
goto again;
}
- buf->lb_len = lovea_size;
+
rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
buf, fl, ost_idx, ea_off, false);
LASSERTF(rc > 0, "invalid rc = %d\n", rc);
- buf->lb_len = lovea_size;
for (i = 0; i < count; i++, objs++) {
/* The MDT-object was created via lfsck_layout_recover_create()
* by others before, and we fill the dummy layout EA. */
}
ostid_le_to_cpu(&objs->l_ost_oi, oi);
- ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
+ rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
+ if (rc != 0) {
+ CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
+ "invalid layout EA at the slot %d, index %u\n",
+ lfsck_lfsck2name(lfsck),
+ PFID(lfsck_dto2fid(parent)), i,
+ le32_to_cpu(objs->l_ost_idx));
+
+ GOTO(unlock_parent, rc);
+ }
+
/* It should be rare case, the slot is there, but the LFSCK
* does not handle it during the first-phase cycle scanning. */
if (unlikely(lu_fid_eq(fid, cfid))) {
dt_write_unlock(env, parent);
if (handle != NULL)
dt_trans_stop(env, dt, handle);
- lfsck_layout_unlock(&lh);
- buf->lb_len = buflen;
+ lfsck_ibits_unlock(&lh, LCK_EX);
rc = lfsck_layout_update_pfid(env, com, parent,
cfid, ltd->ltd_tgt, i);
dt_write_unlock(env, parent);
if (handle != NULL)
dt_trans_stop(env, dt, handle);
- lfsck_layout_unlock(&lh);
+ lfsck_ibits_unlock(&lh, LCK_EX);
if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
objs = &lmm->lmm_objects[ea_off];
else
objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
- buf, objs, ea_off, buflen);
+ buf, objs, ea_off);
RETURN(rc);
dt_trans_stop(env, dt, handle);
unlock_layout:
- lfsck_layout_unlock(&lh);
- buf->lb_len = buflen;
+ lfsck_ibits_unlock(&lh, LCK_EX);
return rc;
}
struct lfsck_component *com,
struct lfsck_tgt_desc *ltd)
{
- struct lfsck_layout *lo = com->lc_file_ram;
+ struct lfsck_assistant_data *lad = com->lc_data;
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct lfsck_thread_info *info = lfsck_env_info(env);
"scanning for OST%04x\n",
lfsck_lfsck2name(lfsck), ltd->ltd_index);
+ if (lad->lad_incomplete &&
+ cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
+ CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
+ "scanning for OST%04x\n",
+ lfsck_lfsck2name(lfsck), ltd->ltd_index);
+
+ RETURN(0);
+ }
+
ostid_set_seq(oi, FID_SEQ_IDIF);
ostid_set_id(oi, 0);
- ostid_to_fid(fid, oi, ltd->ltd_index);
+ rc = ostid_to_fid(fid, oi, ltd->ltd_index);
+ if (rc != 0)
+ GOTO(log, rc);
+
obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
if (unlikely(IS_ERR(obj)))
GOTO(log, rc = PTR_ERR(obj));
if (rc == -ESRCH) {
/* -ESRCH means that the orphan OST-objects rbtree has been
* cleanup because of the OSS server restart or other errors. */
- lo->ll_flags |= LF_INCOMPLETE;
+ lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
GOTO(fini, rc);
}
cla->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
LA_ATIME | LA_MTIME | LA_CTIME;
- rc = lfsck_layout_lock(env, com, parent, &lh,
- MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
+ rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
+ MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
+ LCK_EX);
if (rc != 0)
GOTO(log, rc);
GOTO(stop, rc);
dt_read_lock(env, parent, 0);
- if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
+ if (unlikely(lfsck_is_dead_obj(parent)))
GOTO(unlock2, rc = 1);
rc = dt_create(env, child, cla, hint, NULL, handle);
rc = lfsck_layout_trans_stop(env, dev, handle, rc);
unlock1:
- lfsck_layout_unlock(&lh);
+ lfsck_ibits_unlock(&lh, LCK_EX);
log:
CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found dangling "
int rc;
ENTRY;
- rc = lfsck_layout_lock(env, com, parent, &lh,
- MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
+ rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
+ MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
+ LCK_EX);
if (rc != 0)
GOTO(log, rc);
GOTO(stop, rc);
dt_write_lock(env, parent, 0);
- if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
+ if (unlikely(lfsck_is_dead_obj(parent)))
GOTO(unlock2, rc = 1);
rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle,
rc = lfsck_layout_trans_stop(env, dev, handle, rc);
unlock1:
- lfsck_layout_unlock(&lh);
+ lfsck_ibits_unlock(&lh, LCK_EX);
log:
CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired unmatched "
struct lov_mds_md_v1 *lmm;
struct lov_ost_data_v1 *objs;
struct lustre_handle lh = { 0 };
+ struct lu_buf ea_buf;
__u32 magic;
int rc;
ENTRY;
- rc = lfsck_layout_lock(env, com, parent, &lh,
- MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
+ rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
+ MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
+ LCK_EX);
if (rc != 0)
GOTO(log, rc);
GOTO(stop, rc);
dt_write_lock(env, parent, 0);
- if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
+ if (unlikely(lfsck_is_dead_obj(parent)))
GOTO(unlock2, rc = 0);
rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV, BYPASS_CAPA);
ostid_cpu_to_le(oi, &objs[llr->llr_lov_idx].l_ost_oi);
objs[llr->llr_lov_idx].l_ost_gen = cpu_to_le32(0);
objs[llr->llr_lov_idx].l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
- rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV,
+ lfsck_buf_init(&ea_buf, lmm,
+ lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
+ magic));
+ rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV,
LU_XATTR_REPLACE, handle, BYPASS_CAPA);
GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
dt_trans_stop(env, pdev, handle);
unlock1:
- lfsck_layout_unlock(&lh);
+ lfsck_ibits_unlock(&lh, LCK_EX);
log:
CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired multiple "
/* Use the dt_object lock to serialize with destroy and attr_set. */
dt_read_lock(env, parent, 0);
- if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
+ if (unlikely(lfsck_is_dead_obj(parent)))
GOTO(unlock, rc = 1);
/* Get the latest parent's owner. */
}
tobj = lfsck_object_find(env, com->lc_lfsck, pfid);
- if (tobj == NULL)
- RETURN(LLIT_UNMATCHED_PAIR);
-
if (IS_ERR(tobj))
RETURN(PTR_ERR(tobj));
- if (!dt_object_exists(tobj))
+ dt_read_lock(env, tobj, 0);
+ if (dt_object_exists(tobj) == 0 ||
+ lfsck_is_dead_obj(tobj))
+ GOTO(out, rc = LLIT_UNMATCHED_PAIR);
+
+ if (!S_ISREG(lfsck_object_type(tobj)))
GOTO(out, rc = LLIT_UNMATCHED_PAIR);
/* Load the tobj's layout EA, in spite of it is a local MDT-object or
* remote one on another MDT. Then check whether the given OST-object
* is in such layout. If yes, it is multiple referenced, otherwise it
* is unmatched referenced case. */
- rc = lfsck_layout_get_lovea(env, tobj, buf, NULL);
- if (rc == 0)
+ rc = lfsck_layout_get_lovea(env, tobj, buf);
+ if (rc == 0 || rc == -ENOENT)
GOTO(out, rc = LLIT_UNMATCHED_PAIR);
if (rc < 0)
for (i = 0; i < count; i++, objs++) {
struct lu_fid *tfid = &info->lti_fid2;
struct ost_id *oi = &info->lti_oi;
+ __u32 idx2;
if (lovea_slot_is_dummy(objs))
continue;
ostid_le_to_cpu(&objs->l_ost_oi, oi);
- ostid_to_fid(tfid, oi, le32_to_cpu(objs->l_ost_idx));
+ idx2 = le32_to_cpu(objs->l_ost_idx);
+ rc = ostid_to_fid(tfid, oi, idx2);
+ if (rc != 0) {
+ CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
+ "invalid layout EA at the slot %d, index %u\n",
+ lfsck_lfsck2name(com->lc_lfsck),
+ PFID(pfid), i, idx2);
+
+ GOTO(out, rc = LLIT_UNMATCHED_PAIR);
+ }
+
if (lu_fid_eq(cfid, tfid)) {
*lov_ea = *buf;
GOTO(out, rc = LLIT_UNMATCHED_PAIR);
out:
+ dt_read_unlock(env, tobj);
lfsck_object_put(env, tobj);
return rc;
}
-static int lfsck_layout_assistant_handle_one(const struct lu_env *env,
+static int lfsck_layout_assistant_handler_p1(const struct lu_env *env,
struct lfsck_component *com,
- struct lfsck_layout_req *llr)
+ struct lfsck_assistant_req *lar)
{
+ struct lfsck_layout_req *llr =
+ container_of0(lar, struct lfsck_layout_req, llr_lar);
struct lfsck_layout *lo = com->lc_file_ram;
struct lfsck_thread_info *info = lfsck_env_info(env);
struct filter_fid_old *pea = &info->lti_old_pfid;
struct lu_fid *pfid = &info->lti_fid;
- struct lu_buf *buf = NULL;
+ struct lu_buf buf = { 0 };
struct dt_object *parent = llr->llr_parent->llo_obj;
struct dt_object *child = llr->llr_child;
struct lu_attr *pla = &info->lti_la;
int rc;
ENTRY;
- rc = dt_attr_get(env, parent, pla, BYPASS_CAPA);
- if (rc != 0) {
- if (lu_object_is_dying(parent->do_lu.lo_header))
- RETURN(0);
+ if (unlikely(lfsck_is_dead_obj(parent)))
+ RETURN(0);
+ rc = dt_attr_get(env, parent, pla, BYPASS_CAPA);
+ if (rc != 0)
GOTO(out, rc);
- }
rc = dt_attr_get(env, child, cla, BYPASS_CAPA);
if (rc == -ENOENT) {
- if (lu_object_is_dying(parent->do_lu.lo_header))
+ if (unlikely(lfsck_is_dead_obj(parent)))
RETURN(0);
type = LLIT_DANGLING;
if (rc != 0)
GOTO(out, rc);
- buf = lfsck_buf_get(env, pea, sizeof(struct filter_fid_old));
- rc= dt_xattr_get(env, child, buf, XATTR_NAME_FID, BYPASS_CAPA);
+ lfsck_buf_init(&buf, pea, sizeof(struct filter_fid_old));
+ rc = dt_xattr_get(env, child, &buf, XATTR_NAME_FID, BYPASS_CAPA);
if (unlikely(rc >= 0 && rc != sizeof(struct filter_fid_old) &&
rc != sizeof(struct filter_fid))) {
type = LLIT_UNMATCHED_PAIR;
rc = lfsck_layout_check_parent(env, com, parent, pfid,
lu_object_fid(&child->do_lu),
- pla, cla, llr, buf, idx);
+ pla, cla, llr, &buf, idx);
if (rc > 0) {
type = rc;
goto repair;
break;
case LLIT_MULTIPLE_REFERENCED:
rc = lfsck_layout_repair_multiple_references(env, com, llr,
- pla, buf);
+ pla, &buf);
break;
case LLIT_INCONSISTENT_OWNER:
rc = lfsck_layout_repair_owner(env, com, llr, pla);
out:
down_write(&com->lc_sem);
if (rc < 0) {
- struct lfsck_layout_master_data *llmd = com->lc_data;
+ struct lfsck_assistant_data *lad = com->lc_data;
- if (unlikely(llmd->llmd_exit)) {
+ if (unlikely(lad->lad_exit)) {
rc = 0;
} else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
"talk with OST %x: rc = %d\n",
lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
- lo->ll_flags |= LF_INCOMPLETE;
+ lfsck_lad_set_bitmap(env, com, llr->llr_ost_idx);
lo->ll_objs_skipped++;
rc = 0;
} else {
return rc;
}
-static int lfsck_layout_assistant(void *args)
+static int lfsck_layout_assistant_handler_p2(const struct lu_env *env,
+ struct lfsck_component *com)
{
- struct lfsck_thread_args *lta = args;
- struct lu_env *env = <a->lta_env;
- struct lfsck_component *com = lta->lta_com;
- struct lfsck_instance *lfsck = lta->lta_lfsck;
- struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
- struct lfsck_position *pos = &com->lc_pos_start;
- struct lfsck_thread_info *info = lfsck_env_info(env);
- struct lfsck_request *lr = &info->lti_lr;
- struct lfsck_layout_master_data *llmd = com->lc_data;
- struct ptlrpc_thread *mthread = &lfsck->li_thread;
- struct ptlrpc_thread *athread = &llmd->llmd_thread;
- struct lfsck_layout_req *llr;
- struct l_wait_info lwi = { 0 };
- int rc = 0;
- int rc1 = 0;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
+ struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
+ struct lfsck_tgt_desc *ltd;
+ int rc = 0;
ENTRY;
- memset(lr, 0, sizeof(*lr));
- lr->lr_event = LE_START;
- lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
- LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ;
- lr->lr_speed = bk->lb_speed_limit;
- lr->lr_version = bk->lb_version;
- lr->lr_param = bk->lb_param;
- lr->lr_async_windows = bk->lb_async_windows;
- lr->lr_flags = LEF_TO_OST;
- if (pos->lp_oit_cookie <= 1)
- lr->lr_param |= LPF_RESET;
-
- rc = lfsck_layout_master_notify_others(env, com, lr);
- if (rc != 0) {
- CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to notify "
- "others for LFSCK start: rc = %d\n",
- lfsck_lfsck2name(lfsck), rc);
- GOTO(fini, rc);
- }
-
- spin_lock(&llmd->llmd_lock);
- thread_set_flags(athread, SVC_RUNNING);
- spin_unlock(&llmd->llmd_lock);
- wake_up_all(&mthread->t_ctl_waitq);
-
- while (1) {
- while (!list_empty(&llmd->llmd_req_list)) {
- bool wakeup = false;
-
- if (unlikely(llmd->llmd_exit ||
- !thread_is_running(mthread)))
- GOTO(cleanup1, rc = llmd->llmd_post_result);
-
- llr = list_entry(llmd->llmd_req_list.next,
- struct lfsck_layout_req,
- llr_list);
- /* Only the lfsck_layout_assistant thread itself can
- * remove the "llr" from the head of the list, LFSCK
- * engine thread only inserts other new "lld" at the
- * end of the list. So it is safe to handle current
- * "llr" without the spin_lock. */
- rc = lfsck_layout_assistant_handle_one(env, com, llr);
- spin_lock(&llmd->llmd_lock);
- list_del_init(&llr->llr_list);
- llmd->llmd_prefetched--;
- /* Wake up the main engine thread only when the list
- * is empty or half of the prefetched items have been
- * handled to avoid too frequent thread schedule. */
- if (llmd->llmd_prefetched == 0 ||
- (bk->lb_async_windows != 0 &&
- bk->lb_async_windows / 2 ==
- llmd->llmd_prefetched))
- wakeup = true;
- spin_unlock(&llmd->llmd_lock);
- if (wakeup)
- wake_up_all(&mthread->t_ctl_waitq);
-
- lfsck_layout_req_fini(env, llr);
- if (rc < 0 && bk->lb_param & LPF_FAILOUT)
- GOTO(cleanup1, rc);
- }
-
- l_wait_event(athread->t_ctl_waitq,
- !lfsck_layout_req_empty(llmd) ||
- llmd->llmd_exit ||
- llmd->llmd_to_post ||
- llmd->llmd_to_double_scan,
- &lwi);
-
- if (unlikely(llmd->llmd_exit))
- GOTO(cleanup1, rc = llmd->llmd_post_result);
-
- if (!list_empty(&llmd->llmd_req_list))
- continue;
-
- if (llmd->llmd_to_post) {
- llmd->llmd_to_post = 0;
- LASSERT(llmd->llmd_post_result > 0);
-
- memset(lr, 0, sizeof(*lr));
- lr->lr_event = LE_PHASE1_DONE;
- lr->lr_status = llmd->llmd_post_result;
- rc = lfsck_layout_master_notify_others(env, com, lr);
- if (rc != 0)
- CDEBUG(D_LFSCK, "%s: layout LFSCK assistant "
- "failed to notify others for LFSCK "
- "post: rc = %d\n",
- lfsck_lfsck2name(lfsck), rc);
-
- /* Wakeup the master engine to go ahead. */
- wake_up_all(&mthread->t_ctl_waitq);
- }
-
- if (llmd->llmd_to_double_scan) {
- llmd->llmd_to_double_scan = 0;
- atomic_inc(&lfsck->li_double_scan_count);
- llmd->llmd_in_double_scan = 1;
- wake_up_all(&mthread->t_ctl_waitq);
-
- CDEBUG(D_LFSCK, "%s: layout LFSCK assistant phase2 "
- "scan start\n", lfsck_lfsck2name(lfsck));
-
- com->lc_new_checked = 0;
- com->lc_new_scanned = 0;
- com->lc_time_last_checkpoint = cfs_time_current();
- com->lc_time_next_checkpoint =
- com->lc_time_last_checkpoint +
- cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
-
- /* flush all async updating before handling orphan. */
- dt_sync(env, lfsck->li_next);
-
- while (llmd->llmd_in_double_scan) {
- struct lfsck_tgt_descs *ltds =
- &lfsck->li_ost_descs;
- struct lfsck_tgt_desc *ltd;
-
- rc = lfsck_layout_master_query_others(env, com);
- if (lfsck_layout_master_to_orphan(llmd))
- goto orphan;
-
- if (rc < 0)
- GOTO(cleanup2, rc);
-
- /* Pull LFSCK status on related targets once
- * per 30 seconds if we are not notified. */
- lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
- cfs_time_seconds(1),
- NULL, NULL);
- rc = l_wait_event(athread->t_ctl_waitq,
- lfsck_layout_master_to_orphan(llmd) ||
- llmd->llmd_exit ||
- !thread_is_running(mthread),
- &lwi);
-
- if (unlikely(llmd->llmd_exit ||
- !thread_is_running(mthread)))
- GOTO(cleanup2, rc = 0);
-
- if (rc == -ETIMEDOUT)
- continue;
-
- if (rc < 0)
- GOTO(cleanup2, rc);
-
-orphan:
- spin_lock(<ds->ltd_lock);
- while (!list_empty(
- &llmd->llmd_ost_phase2_list)) {
- ltd = list_entry(
- llmd->llmd_ost_phase2_list.next,
- struct lfsck_tgt_desc,
- ltd_layout_phase_list);
- list_del_init(
- <d->ltd_layout_phase_list);
- spin_unlock(<ds->ltd_lock);
-
- if (bk->lb_param & LPF_ALL_TGT) {
- rc = lfsck_layout_scan_orphan(
- env, com, ltd);
- if (rc != 0 &&
- bk->lb_param & LPF_FAILOUT)
- GOTO(cleanup2, rc);
- }
-
- if (unlikely(llmd->llmd_exit ||
- !thread_is_running(mthread)))
- GOTO(cleanup2, rc = 0);
-
- spin_lock(<ds->ltd_lock);
- }
-
- if (list_empty(&llmd->llmd_ost_phase1_list)) {
- spin_unlock(<ds->ltd_lock);
- GOTO(cleanup2, rc = 1);
- }
- spin_unlock(<ds->ltd_lock);
- }
- }
- }
-
-cleanup1:
- /* Cleanup the unfinished requests. */
- spin_lock(&llmd->llmd_lock);
- if (rc < 0)
- llmd->llmd_assistant_status = rc;
-
- while (!list_empty(&llmd->llmd_req_list)) {
- llr = list_entry(llmd->llmd_req_list.next,
- struct lfsck_layout_req,
- llr_list);
- list_del_init(&llr->llr_list);
- llmd->llmd_prefetched--;
- spin_unlock(&llmd->llmd_lock);
- lfsck_layout_req_fini(env, llr);
- spin_lock(&llmd->llmd_lock);
- }
- spin_unlock(&llmd->llmd_lock);
-
- LASSERTF(llmd->llmd_prefetched == 0, "unmatched prefeteched objs %d\n",
- llmd->llmd_prefetched);
+ spin_lock(<ds->ltd_lock);
+ while (!list_empty(&lad->lad_ost_phase2_list)) {
+ ltd = list_entry(lad->lad_ost_phase2_list.next,
+ struct lfsck_tgt_desc,
+ ltd_layout_phase_list);
+ list_del_init(<d->ltd_layout_phase_list);
+ if (bk->lb_param & LPF_ALL_TGT) {
+ spin_unlock(<ds->ltd_lock);
+ rc = lfsck_layout_scan_orphan(env, com, ltd);
+ if (rc != 0 && bk->lb_param & LPF_FAILOUT)
+ RETURN(rc);
-cleanup2:
- memset(lr, 0, sizeof(*lr));
- if (rc > 0) {
- lr->lr_event = LE_PHASE2_DONE;
- lr->lr_status = rc;
- } else if (rc == 0) {
- if (lfsck->li_flags & LPF_ALL_TGT) {
- lr->lr_event = LE_STOP;
- lr->lr_status = LS_STOPPED;
- } else {
- lr->lr_event = LE_PEER_EXIT;
- switch (lfsck->li_status) {
- case LS_PAUSED:
- case LS_CO_PAUSED:
- lr->lr_status = LS_CO_PAUSED;
- break;
- case LS_STOPPED:
- case LS_CO_STOPPED:
- lr->lr_status = LS_CO_STOPPED;
- break;
- default:
- CDEBUG(D_LFSCK, "%s: unknown status: rc = %d\n",
- lfsck_lfsck2name(lfsck),
- lfsck->li_status);
- lr->lr_status = LS_CO_FAILED;
- break;
- }
- }
- } else {
- if (lfsck->li_flags & LPF_ALL_TGT) {
- lr->lr_event = LE_STOP;
- lr->lr_status = LS_FAILED;
- } else {
- lr->lr_event = LE_PEER_EXIT;
- lr->lr_status = LS_CO_FAILED;
+ if (unlikely(lad->lad_exit ||
+ !thread_is_running(&lfsck->li_thread)))
+ RETURN(0);
+ spin_lock(<ds->ltd_lock);
}
}
- rc1 = lfsck_layout_master_notify_others(env, com, lr);
- if (rc1 != 0) {
- CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to "
- "notify others for LFSCK quit: rc = %d\n",
- lfsck_lfsck2name(lfsck), rc1);
- rc = rc1;
- }
-
- /* flush all async updating before exit. */
- dt_sync(env, lfsck->li_next);
-
- /* Under force exit case, some requests may be just freed without
- * verification, those objects should be re-handled when next run.
- * So not update the on-disk tracing file under such case. */
- if (llmd->llmd_in_double_scan) {
- struct lfsck_layout *lo = com->lc_file_ram;
-
- if (!llmd->llmd_exit)
- rc1 = lfsck_layout_double_scan_result(env, com, rc);
-
- CDEBUG(D_LFSCK, "%s: layout LFSCK assistant phase2 scan "
- "finished, status %d: rc = %d\n",
- lfsck_lfsck2name(lfsck), lo->ll_status, rc1);
- }
-
-fini:
- if (llmd->llmd_in_double_scan)
- atomic_dec(&lfsck->li_double_scan_count);
-
- spin_lock(&llmd->llmd_lock);
- llmd->llmd_assistant_status = (rc1 != 0 ? rc1 : rc);
- thread_set_flags(athread, SVC_STOPPED);
- wake_up_all(&mthread->t_ctl_waitq);
- spin_unlock(&llmd->llmd_lock);
- lfsck_thread_args_fini(lta);
+ if (list_empty(&lad->lad_ost_phase1_list))
+ rc = 1;
+ else
+ rc = 0;
+ spin_unlock(<ds->ltd_lock);
- return rc;
+ RETURN(rc);
}
static int
break;
llst->llst_gen = llsd->llsd_touch_gen;
- list_del(&llst->llst_list);
- list_add_tail(&llst->llst_list,
- &llsd->llsd_master_list);
+ list_move_tail(&llst->llst_list,
+ &llsd->llsd_master_list);
atomic_inc(&llst->llst_ref);
spin_unlock(&llsd->llsd_lock);
struct lfsck_component *com,
enum lfsck_events event, int result)
{
+ struct lfsck_layout *lo = com->lc_file_ram;
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_layout_slave_data *llsd = com->lc_data;
struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
lr->lr_status = result;
lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
lr->lr_active = LFSCK_TYPE_LAYOUT;
+ lr->lr_flags2 = lo->ll_flags;
llsd->llsd_touch_gen++;
spin_lock(&llsd->llsd_lock);
while (!list_empty(&llsd->llsd_master_list)) {
break;
llst->llst_gen = llsd->llsd_touch_gen;
- list_del(&llst->llst_list);
- list_add_tail(&llst->llst_list,
- &llsd->llsd_master_list);
+ list_move_tail(&llst->llst_list,
+ &llsd->llsd_master_list);
atomic_inc(&llst->llst_ref);
spin_unlock(&llsd->llsd_lock);
RETURN(PTR_ERR(obj));
dt_read_lock(env, obj, 0);
- if (unlikely(!dt_object_exists(obj)))
+ if (unlikely(dt_object_exists(obj) == 0 ||
+ lfsck_is_dead_obj(obj)))
GOTO(unlock, rc = -ENOENT);
- rc = lfsck_layout_get_lovea(env, obj, buf, NULL);
+ if (!S_ISREG(lfsck_object_type(obj)))
+ GOTO(unlock, rc = -ENODATA);
+
+ rc = lfsck_layout_get_lovea(env, obj, buf);
if (rc < 0)
GOTO(unlock, rc);
fid_cpu_to_le(&ff->ff_parent, &lr->lr_fid2);
buf = lfsck_buf_get(env, ff, sizeof(*ff));
dt_write_lock(env, obj, 0);
- if (unlikely(!dt_object_exists(obj)))
+ if (unlikely(dt_object_exists(obj) == 0 ||
+ lfsck_is_dead_obj(obj)))
GOTO(unlock, rc = 0);
th = dt_trans_create(env, dev);
lo->ll_magic = LFSCK_LAYOUT_MAGIC;
lo->ll_status = LS_INIT;
+ if (com->lc_lfsck->li_master) {
+ struct lfsck_assistant_data *lad = com->lc_data;
+
+ lad->lad_incomplete = 0;
+ CFS_RESET_BITMAP(lad->lad_bitmap);
+ }
+
rc = lfsck_layout_store(env, com);
up_write(&com->lc_sem);
static int lfsck_layout_master_checkpoint(const struct lu_env *env,
struct lfsck_component *com, bool init)
{
- struct lfsck_instance *lfsck = com->lc_lfsck;
- struct lfsck_layout *lo = com->lc_file_ram;
- struct lfsck_layout_master_data *llmd = com->lc_data;
- struct ptlrpc_thread *mthread = &lfsck->li_thread;
- struct ptlrpc_thread *athread = &llmd->llmd_thread;
- struct l_wait_info lwi = { 0 };
- int rc;
-
- if (com->lc_new_checked == 0 && !init)
- return 0;
-
- l_wait_event(mthread->t_ctl_waitq,
- list_empty(&llmd->llmd_req_list) ||
- !thread_is_running(mthread) ||
- thread_is_stopped(athread),
- &lwi);
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_layout *lo = com->lc_file_ram;
+ int rc;
- if (!thread_is_running(mthread) || thread_is_stopped(athread))
- return 0;
+ if (!init) {
+ rc = lfsck_checkpoint_generic(env, com);
+ if (rc != 0)
+ return rc > 0 ? 0 : rc;
+ }
down_write(&com->lc_sem);
if (init) {
- lo->ll_pos_latest_start = lfsck->li_pos_current.lp_oit_cookie;
+ lo->ll_pos_latest_start =
+ lfsck->li_pos_checkpoint.lp_oit_cookie;
} else {
lo->ll_pos_last_checkpoint =
- lfsck->li_pos_current.lp_oit_cookie;
+ lfsck->li_pos_checkpoint.lp_oit_cookie;
lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
HALF_SEC - lfsck->li_time_last_checkpoint);
lo->ll_time_last_checkpoint = cfs_time_current_sec();
down_write(&com->lc_sem);
if (init) {
- lo->ll_pos_latest_start = lfsck->li_pos_current.lp_oit_cookie;
+ lo->ll_pos_latest_start =
+ lfsck->li_pos_checkpoint.lp_oit_cookie;
} else {
lo->ll_pos_last_checkpoint =
- lfsck->li_pos_current.lp_oit_cookie;
+ lfsck->li_pos_checkpoint.lp_oit_cookie;
lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
HALF_SEC - lfsck->li_time_last_checkpoint);
lo->ll_time_last_checkpoint = cfs_time_current_sec();
if (!lfsck->li_drop_dryrun ||
lo->ll_pos_first_inconsistent == 0) {
lo->ll_status = LS_SCANNING_PHASE2;
- list_del_init(&com->lc_link);
- list_add_tail(&com->lc_link,
- &lfsck->li_list_double_scan);
+ list_move_tail(&com->lc_link,
+ &lfsck->li_list_double_scan);
pos->lp_oit_cookie = 0;
} else {
int i;
struct lfsck_component *com,
struct lfsck_start_param *lsp)
{
- struct lfsck_instance *lfsck = com->lc_lfsck;
- struct lfsck_layout_master_data *llmd = com->lc_data;
- struct ptlrpc_thread *mthread = &lfsck->li_thread;
- struct ptlrpc_thread *athread = &llmd->llmd_thread;
- struct lfsck_thread_args *lta;
- struct task_struct *task;
- int rc;
+ int rc;
ENTRY;
+ rc = lfsck_layout_load_bitmap(env, com);
+ if (rc > 0) {
+ rc = lfsck_layout_reset(env, com, false);
+ if (rc == 0)
+ rc = lfsck_set_param(env, com->lc_lfsck,
+ lsp->lsp_start, true);
+ }
+
+ if (rc != 0)
+ GOTO(log, rc);
+
rc = lfsck_layout_prep(env, com, lsp->lsp_start);
if (rc != 0)
RETURN(rc);
- llmd->llmd_assistant_status = 0;
- llmd->llmd_post_result = 0;
- llmd->llmd_to_post = 0;
- llmd->llmd_to_double_scan = 0;
- llmd->llmd_in_double_scan = 0;
- llmd->llmd_exit = 0;
- thread_set_flags(athread, 0);
-
- lta = lfsck_thread_args_init(lfsck, com, lsp);
- if (IS_ERR(lta))
- RETURN(PTR_ERR(lta));
-
- task = kthread_run(lfsck_layout_assistant, lta, "lfsck_layout");
- if (IS_ERR(task)) {
- rc = PTR_ERR(task);
- CERROR("%s: cannot start LFSCK layout assistant thread: "
- "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
- lfsck_thread_args_fini(lta);
- } else {
- struct l_wait_info lwi = { 0 };
+ rc = lfsck_start_assistant(env, com, lsp);
- l_wait_event(mthread->t_ctl_waitq,
- thread_is_running(athread) ||
- thread_is_stopped(athread),
- &lwi);
- if (unlikely(!thread_is_running(athread)))
- rc = llmd->llmd_assistant_status;
- else
- rc = 0;
- }
+ GOTO(log, rc);
+log:
CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
- LPU64"\n", lfsck_lfsck2name(lfsck),
+ LPU64"\n", lfsck_lfsck2name(com->lc_lfsck),
com->lc_pos_start.lp_oit_cookie);
- RETURN(rc);
+ return 0;
}
/* Pre-fetch the attribute for each stripe in the given layout EA. */
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct lfsck_layout *lo = com->lc_file_ram;
- struct lfsck_layout_master_data *llmd = com->lc_data;
+ struct lfsck_assistant_data *lad = com->lc_data;
struct lfsck_layout_object *llo = NULL;
struct lov_ost_data_v1 *objs;
struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
struct ptlrpc_thread *mthread = &lfsck->li_thread;
- struct ptlrpc_thread *athread = &llmd->llmd_thread;
- struct l_wait_info lwi = { 0 };
- struct lu_buf *buf;
+ struct ptlrpc_thread *athread = &lad->lad_thread;
+ struct l_wait_info lwi = { 0 };
+ struct lu_buf buf;
int rc = 0;
int i;
__u32 magic;
__u16 gen;
ENTRY;
- buf = lfsck_buf_get(env, &info->lti_old_pfid,
- sizeof(struct filter_fid_old));
+ lfsck_buf_init(&buf, &info->lti_old_pfid,
+ sizeof(struct filter_fid_old));
count = le16_to_cpu(lmm->lmm_stripe_count);
gen = le16_to_cpu(lmm->lmm_layout_gen);
/* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
struct lfsck_layout_req *llr;
struct lfsck_tgt_desc *tgt = NULL;
struct dt_object *cobj = NULL;
- __u32 index =
- le32_to_cpu(objs->l_ost_idx);
+ __u32 index;
bool wakeup = false;
if (unlikely(lovea_slot_is_dummy(objs)))
l_wait_event(mthread->t_ctl_waitq,
bk->lb_async_windows == 0 ||
- llmd->llmd_prefetched < bk->lb_async_windows ||
+ lad->lad_prefetched < bk->lb_async_windows ||
!thread_is_running(mthread) ||
thread_is_stopped(athread),
&lwi);
thread_is_stopped(athread))
GOTO(out, rc = 0);
+ if (unlikely(lfsck_is_dead_obj(parent)))
+ GOTO(out, rc = 0);
+
ostid_le_to_cpu(&objs->l_ost_oi, oi);
- ostid_to_fid(fid, oi, index);
+ index = le32_to_cpu(objs->l_ost_idx);
+ rc = ostid_to_fid(fid, oi, index);
+ if (rc != 0) {
+ CDEBUG(D_LFSCK, "%s: get invalid layout EA for "DFID
+ ": "DOSTID", idx:%u\n", lfsck_lfsck2name(lfsck),
+ PFID(lfsck_dto2fid(parent)), POSTID(oi), index);
+ goto next;
+ }
+
tgt = lfsck_tgt_get(ltds, index);
if (unlikely(tgt == NULL)) {
CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
"did not join the layout LFSCK\n",
lfsck_lfsck2name(lfsck), index);
- lo->ll_flags |= LF_INCOMPLETE;
+ lfsck_lad_set_bitmap(env, com, index);
goto next;
}
- cobj = lfsck_object_find_by_dev(env, tgt->ltd_tgt, fid);
+ /* There is potential deadlock race condition between object
+ * destroy and layout LFSCK. Consider the following scenario:
+ *
+ * 1) The LFSCK thread obtained the parent object firstly, at
+ * that time, the parent object has not been destroyed yet.
+ *
+ * 2) One RPC service thread destroyed the parent and all its
+ * children objects. Because the LFSCK is referencing the
+ * parent object, then the parent object will be marked as
+ * dying in RAM. On the other hand, the parent object is
+ * referencing all its children objects, then all children
+ * objects will be marked as dying in RAM also.
+ *
+ * 3) The LFSCK thread tries to find some child object with
+ * the parent object referenced. Then it will find that the
+ * child object is dying. According to the object visibility
+ * rules: the object with dying flag cannot be returned to
+ * others. So the LFSCK thread has to wait until the dying
+ * object has been purged from RAM, then it can allocate a
+ * new object (with the same FID) in RAM. Unfortunately, the
+ * LFSCK thread itself is referencing the parent object, and
+ * cause the parent object cannot be purged, then cause the
+ * child object cannot be purged also. So the LFSCK thread
+ * will fall into deadlock.
+ *
+ * We introduce non-blocked version lu_object_find() to allow
+ * the LFSCK thread to return failure immediately (instead of
+ * wait) when it finds dying (child) object, then the LFSCK
+ * thread can check whether the parent object is dying or not.
+ * So avoid above deadlock. LU-5395 */
+ cobj = lfsck_object_find_by_dev_nowait(env, tgt->ltd_tgt, fid);
if (IS_ERR(cobj)) {
+ if (lfsck_is_dead_obj(parent)) {
+ lfsck_tgt_put(tgt);
+
+ GOTO(out, rc = 0);
+ }
+
rc = PTR_ERR(cobj);
goto next;
}
if (rc != 0)
goto next;
- rc = dt_declare_xattr_get(env, cobj, buf, XATTR_NAME_FID,
+ rc = dt_declare_xattr_get(env, cobj, &buf, XATTR_NAME_FID,
BYPASS_CAPA);
if (rc != 0)
goto next;
if (llo == NULL) {
- llo = lfsck_layout_object_init(env, parent, gen);
+ llo = lfsck_layout_object_init(env, parent,
+ lfsck->li_pos_current.lp_oit_cookie, gen);
if (IS_ERR(llo)) {
rc = PTR_ERR(llo);
goto next;
}
}
- llr = lfsck_layout_req_init(llo, cobj, index, i);
+ llr = lfsck_layout_assistant_req_init(llo, cobj, index, i);
if (IS_ERR(llr)) {
rc = PTR_ERR(llr);
goto next;
}
cobj = NULL;
- spin_lock(&llmd->llmd_lock);
- if (llmd->llmd_assistant_status < 0) {
- spin_unlock(&llmd->llmd_lock);
- lfsck_layout_req_fini(env, llr);
+ spin_lock(&lad->lad_lock);
+ if (lad->lad_assistant_status < 0) {
+ spin_unlock(&lad->lad_lock);
+ lfsck_layout_assistant_req_fini(env, &llr->llr_lar);
lfsck_tgt_put(tgt);
- RETURN(llmd->llmd_assistant_status);
+ RETURN(lad->lad_assistant_status);
}
- list_add_tail(&llr->llr_list, &llmd->llmd_req_list);
- if (llmd->llmd_prefetched == 0)
+ list_add_tail(&llr->llr_lar.lar_list, &lad->lad_req_list);
+ if (lad->lad_prefetched == 0)
wakeup = true;
- llmd->llmd_prefetched++;
- spin_unlock(&llmd->llmd_lock);
+ lad->lad_prefetched++;
+ spin_unlock(&lad->lad_lock);
if (wakeup)
wake_up_all(&athread->t_ctl_waitq);
/* For the given object, read its layout EA locally. For each stripe, pre-fetch
* the OST-object's attribute and generate an structure lfsck_layout_req on the
- * list ::llmd_req_list.
+ * list ::lad_req_list.
*
* For each request on above list, the lfsck_layout_assistant thread compares
* the OST side attribute with local attribute, if inconsistent, then repair it.
struct lfsck_thread_info *info = lfsck_env_info(env);
struct ost_id *oi = &info->lti_oi;
struct lfsck_layout *lo = com->lc_file_ram;
- struct lfsck_layout_master_data *llmd = com->lc_data;
+ struct lfsck_assistant_data *lad = com->lc_data;
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct thandle *handle = NULL;
struct lov_mds_md_v1 *lmm = NULL;
struct dt_device *dev = lfsck->li_bottom;
struct lustre_handle lh = { 0 };
- ssize_t buflen = buf->lb_len;
+ struct lu_buf ea_buf = { 0 };
int rc = 0;
+ int size = 0;
bool locked = false;
bool stripe = false;
bool bad_oi = false;
if (!S_ISREG(lfsck_object_type(obj)))
GOTO(out, rc = 0);
- if (llmd->llmd_assistant_status < 0)
+ if (lad->lad_assistant_status < 0)
GOTO(out, rc = -ESRCH);
fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
locked = true;
again:
- rc = lfsck_layout_get_lovea(env, obj, buf, &buflen);
+ if (dt_object_exists(obj) == 0 ||
+ lfsck_is_dead_obj(obj))
+ GOTO(out, rc = 0);
+
+ rc = lfsck_layout_get_lovea(env, obj, buf);
if (rc <= 0)
GOTO(out, rc);
- buf->lb_len = rc;
+ size = rc;
lmm = buf->lb_buf;
rc = lfsck_layout_verify_header(lmm);
/* If the LOV EA crashed, then it is possible to be rebuilt later
/* Inconsistent lmm_oi, should be repaired. */
bad_oi = true;
+ lmm->lmm_oi = *oi;
if (bk->lb_param & LPF_DRYRUN) {
down_write(&com->lc_sem);
if (!lustre_handle_is_used(&lh)) {
dt_read_unlock(env, obj);
locked = false;
- buf->lb_len = buflen;
- rc = lfsck_layout_lock(env, com, obj, &lh,
- MDS_INODELOCK_LAYOUT |
- MDS_INODELOCK_XATTR);
+ rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
+ MDS_INODELOCK_LAYOUT |
+ MDS_INODELOCK_XATTR, LCK_EX);
if (rc != 0)
GOTO(out, rc);
if (IS_ERR(handle))
GOTO(out, rc = PTR_ERR(handle));
- rc = dt_declare_xattr_set(env, obj, buf, XATTR_NAME_LOV,
+ lfsck_buf_init(&ea_buf, lmm, size);
+ rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
LU_XATTR_REPLACE, handle);
if (rc != 0)
GOTO(out, rc);
goto again;
}
- lmm->lmm_oi = *oi;
- rc = dt_xattr_set(env, obj, buf, XATTR_NAME_LOV,
+ rc = dt_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
LU_XATTR_REPLACE, handle, BYPASS_CAPA);
if (rc != 0)
GOTO(out, rc);
if (handle != NULL && !IS_ERR(handle))
dt_trans_stop(env, dev, handle);
- lfsck_layout_unlock(&lh);
+ lfsck_ibits_unlock(&lh, LCK_EX);
if (bad_oi)
CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
lfsck_layout_record_failure(env, lfsck, lo);
up_write(&com->lc_sem);
}
- buf->lb_len = buflen;
return rc;
}
if (unlikely(fid_is_last_id(fid)))
GOTO(unlock, rc = 0);
- oid = fid_oid(fid);
+ if (fid_is_idif(fid))
+ oid = fid_idif_id(fid_seq(fid), fid_oid(fid), fid_ver(fid));
+ else
+ oid = fid_oid(fid);
+
if (oid > lls->lls_lastid_known)
lls->lls_lastid_known = oid;
if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
/* OFD may create new objects during LFSCK scanning. */
rc = lfsck_layout_lastid_reload(env, com, lls);
- if (unlikely(rc != 0))
+ if (unlikely(rc != 0)) {
CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
"reload LAST_ID for "LPX64": rc = %d\n",
lfsck_lfsck2name(com->lc_lfsck),
lls->lls_seq, rc);
- if (oid <= lls->lls_lastid)
+
+ GOTO(unlock, rc);
+ }
+
+ if (oid <= lls->lls_lastid ||
+ lo->ll_flags & LF_CRASHED_LASTID)
GOTO(unlock, rc = 0);
LASSERT(lfsck->li_out_notify != NULL);
lfsck->li_out_notify(env, lfsck->li_out_notify_data,
LE_LASTID_REBUILDING);
lo->ll_flags |= LF_CRASHED_LASTID;
+
+ CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
+ "LAST_ID file (2) for the sequence "LPX64
+ ", old value "LPU64", known value "LPU64"\n",
+ lfsck_lfsck2name(lfsck), lls->lls_seq,
+ lls->lls_lastid, oid);
}
lls->lls_lastid = oid;
static int lfsck_layout_exec_dir(const struct lu_env *env,
struct lfsck_component *com,
- struct dt_object *obj,
- struct lu_dirent *ent)
+ struct lu_dirent *ent, __u16 type)
{
return 0;
}
struct lfsck_component *com,
int result, bool init)
{
- struct lfsck_instance *lfsck = com->lc_lfsck;
- struct lfsck_layout *lo = com->lc_file_ram;
- struct lfsck_layout_master_data *llmd = com->lc_data;
- struct ptlrpc_thread *mthread = &lfsck->li_thread;
- struct ptlrpc_thread *athread = &llmd->llmd_thread;
- struct l_wait_info lwi = { 0 };
- int rc;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_layout *lo = com->lc_file_ram;
+ int rc;
ENTRY;
-
- llmd->llmd_post_result = result;
- llmd->llmd_to_post = 1;
- if (llmd->llmd_post_result <= 0)
- llmd->llmd_exit = 1;
-
- wake_up_all(&athread->t_ctl_waitq);
- l_wait_event(mthread->t_ctl_waitq,
- (result > 0 && list_empty(&llmd->llmd_req_list)) ||
- thread_is_stopped(athread),
- &lwi);
-
- if (llmd->llmd_assistant_status < 0)
- result = llmd->llmd_assistant_status;
+ lfsck_post_generic(env, com, &result);
down_write(&com->lc_sem);
spin_lock(&lfsck->li_lock);
- /* When LFSCK failed, there may be some prefetched objects those are
- * not been processed yet, we do not know the exactly position, then
- * just restart from last check-point next time. */
- if (!init && !llmd->llmd_exit)
+ if (!init)
lo->ll_pos_last_checkpoint =
- lfsck->li_pos_current.lp_oit_cookie;
+ lfsck->li_pos_checkpoint.lp_oit_cookie;
if (result > 0) {
- lo->ll_status = LS_SCANNING_PHASE2;
+ if (lo->ll_flags & LF_INCOMPLETE)
+ lo->ll_status = LS_PARTIAL;
+ else
+ lo->ll_status = LS_SCANNING_PHASE2;
lo->ll_flags |= LF_SCANNED_ONCE;
lo->ll_flags &= ~LF_UPGRADE;
- list_del_init(&com->lc_link);
- list_add_tail(&com->lc_link, &lfsck->li_list_double_scan);
+ list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
} else if (result == 0) {
lo->ll_status = lfsck->li_status;
if (lo->ll_status == 0)
lo->ll_status = LS_STOPPED;
if (lo->ll_status != LS_PAUSED) {
- list_del_init(&com->lc_link);
- list_add_tail(&com->lc_link, &lfsck->li_list_idle);
+ list_move_tail(&com->lc_link, &lfsck->li_list_idle);
}
} else {
lo->ll_status = LS_FAILED;
- list_del_init(&com->lc_link);
- list_add_tail(&com->lc_link, &lfsck->li_list_idle);
+ list_move_tail(&com->lc_link, &lfsck->li_list_idle);
}
spin_unlock(&lfsck->li_lock);
spin_lock(&lfsck->li_lock);
if (!init)
lo->ll_pos_last_checkpoint =
- lfsck->li_pos_current.lp_oit_cookie;
+ lfsck->li_pos_checkpoint.lp_oit_cookie;
+
if (result > 0) {
- lo->ll_status = LS_SCANNING_PHASE2;
+ if (lo->ll_flags & LF_INCOMPLETE)
+ lo->ll_status = LS_PARTIAL;
+ else
+ lo->ll_status = LS_SCANNING_PHASE2;
lo->ll_flags |= LF_SCANNED_ONCE;
if (lo->ll_flags & LF_CRASHED_LASTID) {
done = true;
lo->ll_flags &= ~LF_CRASHED_LASTID;
+
+ CDEBUG(D_LFSCK, "%s: layout LFSCK has rebuilt "
+ "crashed LAST_ID files successfully\n",
+ lfsck_lfsck2name(lfsck));
}
lo->ll_flags &= ~LF_UPGRADE;
- list_del_init(&com->lc_link);
- list_add_tail(&com->lc_link, &lfsck->li_list_double_scan);
+ list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
} else if (result == 0) {
lo->ll_status = lfsck->li_status;
if (lo->ll_status == 0)
lo->ll_status = LS_STOPPED;
- if (lo->ll_status != LS_PAUSED) {
- list_del_init(&com->lc_link);
- list_add_tail(&com->lc_link, &lfsck->li_list_idle);
- }
+ if (lo->ll_status != LS_PAUSED)
+ list_move_tail(&com->lc_link, &lfsck->li_list_idle);
} else {
lo->ll_status = LS_FAILED;
- list_del_init(&com->lc_link);
- list_add_tail(&com->lc_link, &lfsck->li_list_idle);
+ list_move_tail(&com->lc_link, &lfsck->li_list_idle);
}
spin_unlock(&lfsck->li_lock);
static int lfsck_layout_master_double_scan(const struct lu_env *env,
struct lfsck_component *com)
{
- struct lfsck_layout_master_data *llmd = com->lc_data;
- struct ptlrpc_thread *mthread = &com->lc_lfsck->li_thread;
- struct ptlrpc_thread *athread = &llmd->llmd_thread;
- struct lfsck_layout *lo = com->lc_file_ram;
- struct l_wait_info lwi = { 0 };
-
- if (unlikely(lo->ll_status != LS_SCANNING_PHASE2))
- return 0;
-
- llmd->llmd_to_double_scan = 1;
- wake_up_all(&athread->t_ctl_waitq);
- l_wait_event(mthread->t_ctl_waitq,
- llmd->llmd_in_double_scan ||
- thread_is_stopped(athread),
- &lwi);
- if (llmd->llmd_assistant_status < 0)
- return llmd->llmd_assistant_status;
+ struct lfsck_layout *lo = com->lc_file_ram;
- return 0;
+ return lfsck_double_scan_generic(env, com, lo->ll_status);
}
static int lfsck_layout_slave_double_scan(const struct lu_env *env,
static void lfsck_layout_master_data_release(const struct lu_env *env,
struct lfsck_component *com)
{
- struct lfsck_layout_master_data *llmd = com->lc_data;
+ struct lfsck_assistant_data *lad = com->lc_data;
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_tgt_descs *ltds;
struct lfsck_tgt_desc *ltd;
struct lfsck_tgt_desc *next;
- LASSERT(llmd != NULL);
- LASSERT(thread_is_init(&llmd->llmd_thread) ||
- thread_is_stopped(&llmd->llmd_thread));
- LASSERT(list_empty(&llmd->llmd_req_list));
+ LASSERT(lad != NULL);
+ LASSERT(thread_is_init(&lad->lad_thread) ||
+ thread_is_stopped(&lad->lad_thread));
+ LASSERT(list_empty(&lad->lad_req_list));
com->lc_data = NULL;
ltds = &lfsck->li_ost_descs;
spin_lock(<ds->ltd_lock);
- list_for_each_entry_safe(ltd, next, &llmd->llmd_ost_phase1_list,
+ list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
ltd_layout_phase_list) {
list_del_init(<d->ltd_layout_phase_list);
}
- list_for_each_entry_safe(ltd, next, &llmd->llmd_ost_phase2_list,
+ list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
ltd_layout_phase_list) {
list_del_init(<d->ltd_layout_phase_list);
}
- list_for_each_entry_safe(ltd, next, &llmd->llmd_ost_list,
+ list_for_each_entry_safe(ltd, next, &lad->lad_ost_list,
ltd_layout_list) {
list_del_init(<d->ltd_layout_list);
}
ltds = &lfsck->li_mdt_descs;
spin_lock(<ds->ltd_lock);
- list_for_each_entry_safe(ltd, next, &llmd->llmd_mdt_phase1_list,
+ list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
ltd_layout_phase_list) {
list_del_init(<d->ltd_layout_phase_list);
}
- list_for_each_entry_safe(ltd, next, &llmd->llmd_mdt_phase2_list,
+ list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
ltd_layout_phase_list) {
list_del_init(<d->ltd_layout_phase_list);
}
- list_for_each_entry_safe(ltd, next, &llmd->llmd_mdt_list,
+ list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
ltd_layout_list) {
list_del_init(<d->ltd_layout_list);
}
spin_unlock(<ds->ltd_lock);
- OBD_FREE_PTR(llmd);
+ CFS_FREE_BITMAP(lad->lad_bitmap);
+
+ OBD_FREE_PTR(lad);
}
static void lfsck_layout_slave_data_release(const struct lu_env *env,
OBD_FREE_PTR(llsd);
}
-static void lfsck_layout_master_quit(const struct lu_env *env,
- struct lfsck_component *com)
-{
- struct lfsck_layout_master_data *llmd = com->lc_data;
- struct ptlrpc_thread *mthread = &com->lc_lfsck->li_thread;
- struct ptlrpc_thread *athread = &llmd->llmd_thread;
- struct l_wait_info lwi = { 0 };
-
- llmd->llmd_exit = 1;
- wake_up_all(&athread->t_ctl_waitq);
- l_wait_event(mthread->t_ctl_waitq,
- thread_is_init(athread) ||
- thread_is_stopped(athread),
- &lwi);
-}
-
static void lfsck_layout_slave_quit(const struct lu_env *env,
struct lfsck_component *com)
{
{
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_layout *lo = com->lc_file_ram;
- struct lfsck_layout_master_data *llmd = com->lc_data;
+ struct lfsck_assistant_data *lad = com->lc_data;
struct lfsck_tgt_descs *ltds;
struct lfsck_tgt_desc *ltd;
bool fail = false;
RETURN(rc);
}
- CDEBUG(D_LFSCK, "%s: layout LFSCK master handle notify %u "
+ CDEBUG(D_LFSCK, "%s: layout LFSCK master handles notify %u "
"from %s %x, status %d\n", lfsck_lfsck2name(lfsck),
lr->lr_event, (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
lr->lr_index, lr->lr_status);
list_del_init(<d->ltd_layout_phase_list);
switch (lr->lr_event) {
case LE_PHASE1_DONE:
- if (lr->lr_status <= 0) {
+ if (lr->lr_status <= 0 || lr->lr_flags2 & LF_INCOMPLETE) {
+ if (lr->lr_flags2 & LF_INCOMPLETE) {
+ if (lr->lr_flags & LEF_FROM_OST)
+ lfsck_lad_set_bitmap(env, com,
+ ltd->ltd_index);
+ else
+ lo->ll_flags |= LF_INCOMPLETE;
+ }
ltd->ltd_layout_done = 1;
list_del_init(<d->ltd_layout_list);
- lo->ll_flags |= LF_INCOMPLETE;
fail = true;
break;
}
if (lr->lr_flags & LEF_FROM_OST) {
if (list_empty(<d->ltd_layout_list))
list_add_tail(<d->ltd_layout_list,
- &llmd->llmd_ost_list);
+ &lad->lad_ost_list);
list_add_tail(<d->ltd_layout_phase_list,
- &llmd->llmd_ost_phase2_list);
+ &lad->lad_ost_phase2_list);
} else {
if (list_empty(<d->ltd_layout_list))
list_add_tail(<d->ltd_layout_list,
- &llmd->llmd_mdt_list);
+ &lad->lad_mdt_list);
list_add_tail(<d->ltd_layout_phase_list,
- &llmd->llmd_mdt_phase2_list);
+ &lad->lad_mdt_phase2_list);
}
break;
case LE_PHASE2_DONE:
fail = true;
ltd->ltd_layout_done = 1;
list_del_init(<d->ltd_layout_list);
- if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT))
- lo->ll_flags |= LF_INCOMPLETE;
+ if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) &&
+ !(lr->lr_flags & LEF_FROM_OST))
+ lo->ll_flags |= LF_INCOMPLETE;
break;
default:
break;
stop->ls_status = lr->lr_status;
stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
lfsck_stop(env, lfsck->li_bottom, stop);
- } else if (lfsck_layout_master_to_orphan(llmd)) {
- wake_up_all(&llmd->llmd_thread.t_ctl_waitq);
+ } else if (lfsck_phase2_next_ready(lad)) {
+ wake_up_all(&lad->lad_thread.t_ctl_waitq);
}
RETURN(0);
RETURN(rc);
}
+ case LE_PHASE1_DONE: {
+ if (lr->lr_flags2 & LF_INCOMPLETE) {
+ struct lfsck_layout *lo = com->lc_file_ram;
+
+ lo->ll_flags |= LF_INCOMPLETE;
+ llst = lfsck_layout_llst_find_and_del(llsd,
+ lr->lr_index,
+ true);
+ if (llst != NULL) {
+ lfsck_layout_llst_put(llst);
+ if (list_empty(&llsd->llsd_master_list))
+ wake_up_all(
+ &lfsck->li_thread.t_ctl_waitq);
+ }
+ }
+
+ RETURN(0);
+ }
case LE_PHASE2_DONE:
case LE_PEER_EXIT:
CDEBUG(D_LFSCK, "%s: layout LFSCK slave handle notify %u "
llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
if (llst == NULL)
- RETURN(-ENXIO);
+ RETURN(0);
lfsck_layout_llst_put(llst);
if (list_empty(&llsd->llsd_master_list))
wake_up_all(&lfsck->li_thread.t_ctl_waitq);
if (lr->lr_event == LE_PEER_EXIT &&
- lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
+ (lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT ||
+ (list_empty(&llsd->llsd_master_list) &&
+ (lr->lr_status == LS_STOPPED ||
+ lr->lr_status == LS_CO_STOPPED)))) {
struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
memset(stop, 0, sizeof(*stop));
return lo->ll_status;
}
-static int lfsck_layout_master_stop_notify(const struct lu_env *env,
- struct lfsck_component *com,
- struct lfsck_tgt_descs *ltds,
- struct lfsck_tgt_desc *ltd,
- struct ptlrpc_request_set *set)
-{
- struct lfsck_thread_info *info = lfsck_env_info(env);
- struct lfsck_async_interpret_args *laia = &info->lti_laia;
- struct lfsck_request *lr = &info->lti_lr;
- struct lfsck_instance *lfsck = com->lc_lfsck;
- int rc;
-
- spin_lock(<ds->ltd_lock);
- if (list_empty(<d->ltd_layout_list)) {
- LASSERT(list_empty(<d->ltd_layout_phase_list));
- spin_unlock(<ds->ltd_lock);
-
- return 0;
- }
-
- list_del_init(<d->ltd_layout_phase_list);
- list_del_init(<d->ltd_layout_list);
- spin_unlock(<ds->ltd_lock);
-
- memset(lr, 0, sizeof(*lr));
- lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
- lr->lr_event = LE_PEER_EXIT;
- lr->lr_active = LFSCK_TYPE_LAYOUT;
- lr->lr_status = LS_CO_PAUSED;
- if (ltds == &lfsck->li_ost_descs)
- lr->lr_flags = LEF_TO_OST;
-
- laia->laia_com = com;
- laia->laia_ltds = ltds;
- atomic_inc(<d->ltd_ref);
- laia->laia_ltd = ltd;
- laia->laia_lr = lr;
- laia->laia_shared = 0;
-
- rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
- lfsck_layout_master_async_interpret,
- laia, LFSCK_NOTIFY);
- if (rc != 0) {
- CDEBUG(D_LFSCK, "%s: layout LFSCK fail to notify %s %x "
- "for co-stop: rc = %d\n",
- lfsck_lfsck2name(lfsck),
- (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
- ltd->ltd_index, rc);
- lfsck_tgt_put(ltd);
- }
-
- return rc;
-}
-
/* with lfsck::li_lock held */
static int lfsck_layout_slave_join(const struct lu_env *env,
struct lfsck_component *com,
.lfsck_exec_oit = lfsck_layout_master_exec_oit,
.lfsck_exec_dir = lfsck_layout_exec_dir,
.lfsck_post = lfsck_layout_master_post,
- .lfsck_interpret = lfsck_layout_master_async_interpret,
.lfsck_dump = lfsck_layout_dump,
.lfsck_double_scan = lfsck_layout_master_double_scan,
.lfsck_data_release = lfsck_layout_master_data_release,
- .lfsck_quit = lfsck_layout_master_quit,
+ .lfsck_quit = lfsck_quit_generic,
.lfsck_in_notify = lfsck_layout_master_in_notify,
.lfsck_query = lfsck_layout_query,
- .lfsck_stop_notify = lfsck_layout_master_stop_notify,
};
static struct lfsck_operations lfsck_layout_slave_ops = {
.lfsck_join = lfsck_layout_slave_join,
};
+static void lfsck_layout_assistant_fill_pos(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct lfsck_position *pos)
+{
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_layout_req *llr;
+
+ if (list_empty(&lad->lad_req_list))
+ return;
+
+ llr = list_entry(lad->lad_req_list.next,
+ struct lfsck_layout_req,
+ llr_lar.lar_list);
+ pos->lp_oit_cookie = llr->llr_parent->llo_cookie - 1;
+}
+
+struct lfsck_assistant_operations lfsck_layout_assistant_ops = {
+ .la_handler_p1 = lfsck_layout_assistant_handler_p1,
+ .la_handler_p2 = lfsck_layout_assistant_handler_p2,
+ .la_fill_pos = lfsck_layout_assistant_fill_pos,
+ .la_double_scan_result = lfsck_layout_double_scan_result,
+ .la_req_fini = lfsck_layout_assistant_req_fini,
+ .la_sync_failures = lfsck_layout_assistant_sync_failures,
+};
+
int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
{
struct lfsck_component *com;
com->lc_lfsck = lfsck;
com->lc_type = LFSCK_TYPE_LAYOUT;
if (lfsck->li_master) {
- struct lfsck_layout_master_data *llmd;
-
com->lc_ops = &lfsck_layout_master_ops;
- OBD_ALLOC_PTR(llmd);
- if (llmd == NULL)
+ com->lc_data = lfsck_assistant_data_init(
+ &lfsck_layout_assistant_ops,
+ "lfsck_layout");
+ if (com->lc_data == NULL)
GOTO(out, rc = -ENOMEM);
-
- INIT_LIST_HEAD(&llmd->llmd_req_list);
- spin_lock_init(&llmd->llmd_lock);
- INIT_LIST_HEAD(&llmd->llmd_ost_list);
- INIT_LIST_HEAD(&llmd->llmd_ost_phase1_list);
- INIT_LIST_HEAD(&llmd->llmd_ost_phase2_list);
- INIT_LIST_HEAD(&llmd->llmd_mdt_list);
- INIT_LIST_HEAD(&llmd->llmd_mdt_phase1_list);
- INIT_LIST_HEAD(&llmd->llmd_mdt_phase2_list);
- init_waitqueue_head(&llmd->llmd_thread.t_ctl_waitq);
- com->lc_data = llmd;
} else {
struct lfsck_layout_slave_data *llsd;
* If the system crashed before the status stored,
* it will be loaded back when next time. */
lo->ll_status = LS_CRASHED;
- lo->ll_flags |= LF_INCOMPLETE;
+ if (!lfsck->li_master)
+ lo->ll_flags |= LF_INCOMPLETE;
/* fall through */
case LS_PAUSED:
case LS_CRASHED:
struct lfsck_component *com = NULL;
struct lfsck_layout_slave_data *llsd;
struct lfsck_orphan_it *it = NULL;
+ struct lfsck_layout *lo;
int rc = 0;
ENTRY;
if (unlikely(com == NULL))
GOTO(out, rc = -ENOENT);
+ lo = com->lc_file_ram;
+ if (lo->ll_flags & LF_INCOMPLETE)
+ GOTO(out, rc = -ESRCH);
+
llsd = com->lc_data;
if (!llsd->llsd_rbtree_valid)
GOTO(out, rc = -ESRCH);
}
dt_read_lock(env, obj, 0);
- if (!dt_object_exists(obj)) {
+ if (dt_object_exists(obj) == 0 ||
+ lfsck_is_dead_obj(obj)) {
dt_read_unlock(env, obj);
lfsck_object_put(env, obj);
pos++;