* GPL HEADER END
*/
/*
- * Copyright (c) 2012, 2013, Intel Corporation.
+ * Copyright (c) 2013, 2014, Intel Corporation.
*/
/*
* lustre/lfsck/lfsck_lib.c
#define DEBUG_SUBSYSTEM S_LFSCK
+#include <linux/kthread.h>
#include <libcfs/list.h>
#include <lu_object.h>
#include <dt_object.h>
"broadcast",
"orphan",
"create_ostobj",
+ "create_mdtobj",
NULL
};
}
cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
- ltd = LTD_TGT(ltds, idx);
+ ltd = lfsck_ltd2tgt(ltds, idx);
if (likely(ltd != NULL)) {
LASSERT(list_empty(<d->ltd_layout_list));
LASSERT(list_empty(<d->ltd_layout_phase_list));
ltds->ltd_tgtnr--;
cfs_bitmap_clear(ltds->ltd_tgts_bitmap, idx);
- LTD_TGT(ltds, idx) = NULL;
+ lfsck_assign_tgt(ltds, NULL, idx);
lfsck_tgt_put(ltd);
}
}
GOTO(unlock, rc = -ENOMEM);
}
- LTD_TGT(ltds, index) = ltd;
+ lfsck_assign_tgt(ltds, ltd, index);
cfs_bitmap_set(ltds->ltd_tgts_bitmap, index);
ltds->ltd_tgtnr++;
RETURN(rc);
}
-/**
- * Request the specified ibits lock for the given object.
- *
- * Before the LFSCK modifying on the namespace visible object,
- * it needs to acquire related ibits ldlm lock.
- *
- * \param[in] env pointer to the thread context
- * \param[in] lfsck pointer to the lfsck instance
- * \param[in] obj pointer to the dt_object to be locked
- * \param[out] lh pointer to the lock handle
- * \param[in] ibits the bits for the ldlm lock to be acquired
- * \param[in] mode the mode for the ldlm lock to be acquired
- *
- * \retval 0 for success
- * \retval negative error number on failure
- */
-int lfsck_ibits_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
- struct dt_object *obj, struct lustre_handle *lh,
- __u64 bits, ldlm_mode_t mode)
+static int __lfsck_ibits_lock(const struct lu_env *env,
+ struct lfsck_instance *lfsck,
+ struct dt_object *obj,
+ struct ldlm_res_id *resid,
+ struct lustre_handle *lh,
+ __u64 bits, ldlm_mode_t mode)
{
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;
__u64 flags = LDLM_FL_ATOMIC_CB;
int rc;
memset(policy, 0, sizeof(*policy));
policy->l_inodebits.bits = bits;
- fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
if (dt_object_remote(obj)) {
struct ldlm_enqueue_info *einfo = &info->lti_einfo;
}
/**
+ * Request the specified ibits lock for the given object.
+ *
+ * Before the LFSCK modifying on the namespace visible object,
+ * it needs to acquire related ibits ldlm lock.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] lfsck pointer to the lfsck instance
+ * \param[in] obj pointer to the dt_object to be locked
+ * \param[out] lh pointer to the lock handle
+ * \param[in] bits the bits for the ldlm lock to be acquired
+ * \param[in] mode the mode for the ldlm lock to be acquired
+ *
+ * \retval 0 for success
+ * \retval negative error number on failure
+ */
+int lfsck_ibits_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
+ struct dt_object *obj, struct lustre_handle *lh,
+ __u64 bits, ldlm_mode_t mode)
+{
+ struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
+
+ LASSERT(!lustre_handle_is_used(lh));
+
+ fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
+ return __lfsck_ibits_lock(env, lfsck, obj, resid, lh, bits, mode);
+}
+
+/**
* Release the the specified ibits lock.
*
* If the lock has been acquired before, release it
}
}
+/**
+ * Request compound ibits locks for the given <obj, name> pairs.
+ *
+ * Before the LFSCK modifying on the namespace visible object, it needs to
+ * acquire related ibits ldlm lock. Usually, we can use lfsck_ibits_lock for
+ * the lock purpose. But the simple lfsck_ibits_lock for directory-based
+ * modificationis (such as insert name entry to the directory) may be too
+ * coarse-grained and not efficient.
+ *
+ * The lfsck_lock() will request compound ibits locks on the specified
+ * <obj, name> pairs: the PDO (Parallel Directory Operations) ibits (UPDATE)
+ * lock on the directory object, and the regular ibits lock on the name hash.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] lfsck pointer to the lfsck instance
+ * \param[in] obj pointer to the dt_object to be locked
+ * \param[in] name used for building the PDO lock resource
+ * \param[out] llh pointer to the lfsck_lock_handle
+ * \param[in] bits the bits for the ldlm lock to be acquired
+ * \param[in] mode the mode for the ldlm lock to be acquired
+ *
+ * \retval 0 for success
+ * \retval negative error number on failure
+ */
+int lfsck_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
+ struct dt_object *obj, const char *name,
+ struct lfsck_lock_handle *llh, __u64 bits, ldlm_mode_t mode)
+{
+ struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
+ int rc;
+
+ LASSERT(S_ISDIR(lfsck_object_type(obj)));
+ LASSERT(name != NULL);
+ LASSERT(!lustre_handle_is_used(&llh->llh_pdo_lh));
+ LASSERT(!lustre_handle_is_used(&llh->llh_reg_lh));
+
+ switch (mode) {
+ case LCK_EX:
+ llh->llh_pdo_mode = LCK_EX;
+ break;
+ case LCK_PW:
+ llh->llh_pdo_mode = LCK_CW;
+ break;
+ case LCK_PR:
+ llh->llh_pdo_mode = LCK_CR;
+ break;
+ default:
+ CDEBUG(D_LFSCK, "%s: unexpected PDO lock mode %u on the obj "
+ DFID"\n", lfsck_lfsck2name(lfsck), mode,
+ PFID(lfsck_dto2fid(obj)));
+ LBUG();
+ }
+
+ fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
+ rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_pdo_lh,
+ MDS_INODELOCK_UPDATE, llh->llh_pdo_mode);
+ if (rc != 0)
+ return rc;
+
+ llh->llh_reg_mode = mode;
+ resid->name[LUSTRE_RES_ID_HSH_OFF] = full_name_hash(name, strlen(name));
+ rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_reg_lh,
+ bits, llh->llh_reg_mode);
+ if (rc != 0)
+ lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode);
+
+ return rc;
+}
+
+/**
+ * Release the the compound ibits locks.
+ *
+ * \param[in] llh pointer to the lfsck_lock_handle to be released
+ */
+void lfsck_unlock(struct lfsck_lock_handle *llh)
+{
+ lfsck_ibits_unlock(&llh->llh_reg_lh, llh->llh_reg_mode);
+ lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode);
+}
+
int lfsck_find_mdt_idx_by_fid(const struct lu_env *env,
struct lfsck_instance *lfsck,
const struct lu_fid *fid)
{
- struct seq_server_site *ss =
- lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
+ struct seq_server_site *ss = lfsck_dev_site(lfsck);
struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
int rc;
static const char dotlustre[] = ".lustre";
static const char lostfound[] = "lost+found";
+/**
+ * Remove the name entry from the .lustre/lost+found directory.
+ *
+ * No need to care about the object referenced by the name entry,
+ * either the name entry is invalid or redundant, or the referenced
+ * object has been processed or will be handled by others.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] lfsck pointer to the lfsck instance
+ * \param[in] name the name for the name entry to be removed
+ *
+ * \retval 0 for success
+ * \retval negative error number on failure
+ */
+static int lfsck_lpf_remove_name_entry(const struct lu_env *env,
+ struct lfsck_instance *lfsck,
+ const char *name)
+{
+ struct dt_object *parent = lfsck->li_lpf_root_obj;
+ struct dt_device *dev = lfsck_obj2dev(parent);
+ struct thandle *th;
+ struct lfsck_lock_handle *llh = &lfsck_env_info(env)->lti_llh;
+ int rc;
+ ENTRY;
+
+ rc = lfsck_lock(env, lfsck, parent, name, llh,
+ MDS_INODELOCK_UPDATE, LCK_PW);
+ if (rc != 0)
+ RETURN(rc);
+
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ GOTO(unlock, rc = PTR_ERR(th));
+
+ rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_declare_ref_del(env, parent, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_delete(env, parent, (const struct dt_key *)name, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ dt_write_lock(env, parent, 0);
+ rc = dt_ref_del(env, parent, th);
+ dt_write_unlock(env, parent);
+
+ GOTO(stop, rc);
+
+stop:
+ dt_trans_stop(env, dev, th);
+
+unlock:
+ lfsck_unlock(llh);
+
+ CDEBUG(D_LFSCK, "%s: remove name entry "DFID"/%s: rc = %d\n",
+ lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(parent)), name, rc);
+
+ return rc;
+}
+
static int lfsck_create_lpf_local(const struct lu_env *env,
struct lfsck_instance *lfsck,
- struct dt_object *parent,
struct dt_object *child,
struct lu_attr *la,
struct dt_object_format *dof,
const char *name)
{
struct dt_insert_rec *rec = &lfsck_env_info(env)->lti_dt_rec;
- struct dt_device *dev = lfsck->li_bottom;
+ struct dt_object *parent = lfsck->li_lpf_root_obj;
+ struct dt_device *dev = lfsck_obj2dev(child);
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct dt_object *bk_obj = lfsck->li_bookmark_obj;
const struct lu_fid *cfid = lfsck_dto2fid(child);
struct thandle *th = NULL;
- struct linkea_data ldata = { 0 };
+ struct linkea_data ldata = { NULL };
struct lu_buf linkea_buf;
const struct lu_name *cname;
loff_t pos = 0;
/* 1b.2. insert dot into child dir */
rec->rec_fid = cfid;
rc = dt_insert(env, child, (const struct dt_rec *)rec,
- (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
+ (const struct dt_key *)dot, th, 1);
if (rc != 0)
GOTO(unlock, rc);
/* 1b.3. insert dotdot into child dir */
rec->rec_fid = &LU_LPF_FID;
rc = dt_insert(env, child, (const struct dt_rec *)rec,
- (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1);
+ (const struct dt_key *)dotdot, th, 1);
if (rc != 0)
GOTO(unlock, rc);
/* 3b. insert linkEA for child. */
rc = dt_xattr_set(env, child, &linkea_buf,
- XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
+ XATTR_NAME_LINK, 0, th);
dt_write_unlock(env, child);
if (rc != 0)
GOTO(stop, rc);
/* 4b. insert name into parent dir */
rec->rec_fid = cfid;
rc = dt_insert(env, parent, (const struct dt_rec *)rec,
- (const struct dt_key *)name, th, BYPASS_CAPA, 1);
+ (const struct dt_key *)name, th, 1);
if (rc != 0)
GOTO(stop, rc);
static int lfsck_create_lpf_remote(const struct lu_env *env,
struct lfsck_instance *lfsck,
- struct dt_object *parent,
struct dt_object *child,
struct lu_attr *la,
struct dt_object_format *dof,
const char *name)
{
struct dt_insert_rec *rec = &lfsck_env_info(env)->lti_dt_rec;
+ struct dt_object *parent = lfsck->li_lpf_root_obj;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct dt_object *bk_obj = lfsck->li_bookmark_obj;
const struct lu_fid *cfid = lfsck_dto2fid(child);
struct thandle *th = NULL;
- struct linkea_data ldata = { 0 };
+ struct linkea_data ldata = { NULL };
struct lu_buf linkea_buf;
const struct lu_name *cname;
struct dt_device *dev;
/* Transaction I: locally */
- dev = lfsck->li_bottom;
+ dev = lfsck_obj2dev(child);
th = dt_trans_create(env, dev);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
rec->rec_type = S_IFDIR;
rec->rec_fid = cfid;
rc = dt_insert(env, child, (const struct dt_rec *)rec,
- (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
+ (const struct dt_key *)dot, th, 1);
if (rc != 0)
GOTO(unlock, rc);
/* 1b.3. insert dotdot into child dir */
rec->rec_fid = &LU_LPF_FID;
rc = dt_insert(env, child, (const struct dt_rec *)rec,
- (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1);
+ (const struct dt_key *)dotdot, th, 1);
if (rc != 0)
GOTO(unlock, rc);
/* 3b. insert linkEA for child */
rc = dt_xattr_set(env, child, &linkea_buf,
- XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
+ XATTR_NAME_LINK, 0, th);
if (rc != 0)
GOTO(unlock, rc);
/* Transaction II: remotely */
- dev = lfsck->li_next;
+ dev = lfsck_obj2dev(parent);
th = dt_trans_create(env, dev);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
+ th->th_sync = 1;
/* 5a. insert name into parent dir */
rec->rec_fid = cfid;
rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
if (rc != 0)
GOTO(stop, rc);
- rc = dt_trans_start(env, dev, th);
+ rc = dt_trans_start_local(env, dev, th);
if (rc != 0)
GOTO(stop, rc);
/* 5b. insert name into parent dir */
rc = dt_insert(env, parent, (const struct dt_rec *)rec,
- (const struct dt_key *)name, th, BYPASS_CAPA, 1);
+ (const struct dt_key *)name, th, 1);
if (rc != 0)
GOTO(stop, rc);
stop:
dt_trans_stop(env, dev, th);
- if (rc != 0 && dev == lfsck->li_next)
+ if (rc != 0 && dev == lfsck_obj2dev(parent))
CDEBUG(D_LFSCK, "%s: partially created the object "DFID
"for orphans, but failed to insert the name %s "
"to the .lustre/lost+found/. Such inconsistency "
return rc;
}
-/* Do NOT create .lustre/lost+found/MDTxxxx when register the lfsck instance,
- * because the MDT0 maybe not reaady for sequence allocation yet. We do that
- * only when it is required, such as orphan OST-objects repairing. */
-int lfsck_create_lpf(const struct lu_env *env, struct lfsck_instance *lfsck)
+/**
+ * Create the MDTxxxx directory under /ROOT/.lustre/lost+found/
+ *
+ * The /ROOT/.lustre/lost+found/MDTxxxx/ directory is used for holding
+ * orphans and other uncertain inconsistent objects found during the
+ * LFSCK. Such directory will be created by the LFSCK engine on the
+ * local MDT before the LFSCK scanning.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] lfsck pointer to the lfsck instance
+ *
+ * \retval 0 for success
+ * \retval negative error number on failure
+ */
+static int lfsck_create_lpf(const struct lu_env *env,
+ struct lfsck_instance *lfsck)
{
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct lfsck_thread_info *info = lfsck_env_info(env);
struct lu_fid *cfid = &info->lti_fid2;
struct lu_attr *la = &info->lti_la;
struct dt_object_format *dof = &info->lti_dof;
- struct dt_object *parent = NULL;
+ struct dt_object *parent = lfsck->li_lpf_root_obj;
struct dt_object *child = NULL;
- struct lustre_handle lh = { 0 };
+ struct lfsck_lock_handle *llh = &info->lti_llh;
char name[8];
- int node = lfsck_dev_idx(lfsck->li_bottom);
+ int node = lfsck_dev_idx(lfsck);
int rc = 0;
ENTRY;
LASSERT(lfsck->li_master);
+ LASSERT(parent != NULL);
+ LASSERT(lfsck->li_lpf_obj == NULL);
- sprintf(name, "MDT%04x", node);
- if (node == 0) {
- parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
- &LU_LPF_FID);
- } else {
- struct lfsck_tgt_desc *ltd;
-
- ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, 0);
- if (unlikely(ltd == NULL))
- RETURN(-ENXIO);
-
- parent = lfsck_object_find_by_dev(env, ltd->ltd_tgt,
- &LU_LPF_FID);
- lfsck_tgt_put(ltd);
- }
- if (IS_ERR(parent))
- RETURN(PTR_ERR(parent));
-
- if (lfsck->li_lpf_obj != NULL)
- GOTO(out, rc = 0);
-
- if (unlikely(!dt_try_as_dir(env, parent)))
- GOTO(out, rc = -ENOTDIR);
-
- rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
- MDS_INODELOCK_UPDATE, LCK_EX);
+ rc = lfsck_lock(env, lfsck, parent, name, llh,
+ MDS_INODELOCK_UPDATE, LCK_PW);
if (rc != 0)
- GOTO(out, rc);
-
- mutex_lock(&lfsck->li_mutex);
- if (lfsck->li_lpf_obj != NULL)
- GOTO(unlock, rc = 0);
+ RETURN(rc);
+ snprintf(name, 8, "MDT%04x", node);
if (fid_is_zero(&bk->lb_lpf_fid)) {
/* There is corner case that: in former LFSCK scanning we have
* created the .lustre/lost+found/MDTxxxx but failed to update
* the lfsck_bookmark::lb_lpf_fid successfully. So need lookup
* it from MDT0 firstly. */
rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
- (const struct dt_key *)name, BYPASS_CAPA);
+ (const struct dt_key *)name);
if (rc != 0 && rc != -ENOENT)
GOTO(unlock, rc);
*cfid = bk->lb_lpf_fid;
}
- child = lfsck_object_find_by_dev(env, lfsck->li_bottom, cfid);
+ child = lfsck_object_find_bottom(env, lfsck, cfid);
if (IS_ERR(child))
GOTO(unlock, rc = PTR_ERR(child));
dof->dof_type = dt_mode_to_dft(S_IFDIR);
if (node == 0)
- rc = lfsck_create_lpf_local(env, lfsck, parent, child, la,
- dof, name);
+ rc = lfsck_create_lpf_local(env, lfsck, child, la, dof, name);
else
- rc = lfsck_create_lpf_remote(env, lfsck, parent, child, la,
- dof, name);
+ rc = lfsck_create_lpf_remote(env, lfsck, child, la, dof, name);
if (rc == 0)
lfsck->li_lpf_obj = child;
GOTO(unlock, rc);
unlock:
- mutex_unlock(&lfsck->li_mutex);
- lfsck_ibits_unlock(&lh, LCK_EX);
+ lfsck_unlock(llh);
if (rc != 0 && child != NULL && !IS_ERR(child))
- lu_object_put(env, &child->do_lu);
-out:
- if (parent != NULL && !IS_ERR(parent))
- lu_object_put(env, &parent->do_lu);
+ lfsck_object_put(env, child);
return rc;
}
*
* \param[in] env pointer to the thread context
* \param[in] lfsck pointer to the lfsck instance
- * \param[in] parent pointer to the lost+found object
*
* \retval 0 for success
* \retval negative error number on failure
*/
static int lfsck_scan_lpf_bad_entries(const struct lu_env *env,
- struct lfsck_instance *lfsck,
- struct dt_object *parent)
+ struct lfsck_instance *lfsck)
{
+ struct dt_object *parent = lfsck->li_lpf_root_obj;
struct lu_dirent *ent =
(struct lu_dirent *)lfsck_env_info(env)->lti_key;
const struct dt_it_ops *iops = &parent->do_index_ops->dio_it;
int rc;
ENTRY;
- it = iops->init(env, parent, LUDA_64BITHASH, BYPASS_CAPA);
+ it = iops->init(env, parent, LUDA_64BITHASH);
if (IS_ERR(it))
RETURN(PTR_ERR(it));
if (off != 7) {
remove:
- rc = lfsck_remove_name_entry(env, lfsck, parent,
- ent->lde_name, S_IFDIR);
+ rc = lfsck_lpf_remove_name_entry(env, lfsck,
+ ent->lde_name);
if (rc != 0)
break;
}
*
* \param[in] env pointer to the thread context
* \param[in] lfsck pointer to the lfsck instance
- * \param[in] parent pointer to the lost+found object
* \param[in] child pointer to the lost+found sub-directory object
* \param[in] name the name for lost+found sub-directory object
* \param[out] fid pointer to the buffer to hold the FID of the object
*/
static int lfsck_verify_lpf_pairs(const struct lu_env *env,
struct lfsck_instance *lfsck,
- struct dt_object *parent,
struct dt_object *child, const char *name,
struct lu_fid *fid,
enum lfsck_verify_lpf_types type)
{
+ struct dt_object *parent = lfsck->li_lpf_root_obj;
struct lfsck_thread_info *info = lfsck_env_info(env);
char *name2 = info->lti_key;
struct lu_fid *fid2 = &info->lti_fid3;
fid_zero(fid);
rc = dt_lookup(env, child, (struct dt_rec *)fid,
- (const struct dt_key *)dotdot, BYPASS_CAPA);
+ (const struct dt_key *)dotdot);
if (rc != 0)
GOTO(linkea, rc);
}
cname = lfsck_name_get_const(env, name, strlen(name));
- rc = lfsck_verify_linkea(env, lfsck->li_bottom, child, cname,
- &LU_LPF_FID);
+ rc = lfsck_verify_linkea(env, child, cname, &LU_LPF_FID);
if (rc == 0)
rc = lfsck_update_lpf_entry(env, lfsck, parent, child,
name, type);
GOTO(out_done, rc);
}
- parent2 = lfsck_object_find_by_dev(env, lfsck->li_next, fid);
+ parent2 = lfsck_object_find_bottom(env, lfsck, fid);
if (IS_ERR(parent2))
GOTO(linkea, parent2);
if (!dt_object_exists(parent2)) {
- lu_object_put(env, &parent2->do_lu);
+ lfsck_object_put(env, parent2);
GOTO(linkea, parent2 = ERR_PTR(-ENOENT));
}
if (!dt_try_as_dir(env, parent2)) {
- lu_object_put(env, &parent2->do_lu);
+ lfsck_object_put(env, parent2);
GOTO(linkea, parent2 = ERR_PTR(-ENOTDIR));
}
}
rc = dt_lookup(env, parent2, (struct dt_rec *)fid,
- (const struct dt_key *)name2, BYPASS_CAPA);
+ (const struct dt_key *)name2);
dt_read_unlock(env, child);
lfsck_ibits_unlock(&lh, LCK_PR);
if (rc != 0 && rc != -ENOENT)
lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), rc);
} else /* if (type == LVLT_BY_NAMEENTRY) */ {
/* The name entry is wrong, remove it. */
- rc = lfsck_remove_name_entry(env, lfsck, parent, name, S_IFDIR);
+ rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
}
GOTO(out_put, rc);
out_put:
if (parent2 != NULL && !IS_ERR(parent2))
- lu_object_put(env, &parent2->do_lu);
+ lfsck_object_put(env, parent2);
out_done:
return rc;
struct lu_fid *pfid = &info->lti_fid;
struct lu_fid *cfid = &info->lti_fid2;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
- struct dt_object *parent = NULL;
+ struct dt_object *parent;
/* child1's FID is in the bookmark file. */
struct dt_object *child1 = NULL;
/* child2's FID is in the name entry MDTxxxx. */
struct dt_object *child2 = NULL;
- struct dt_device *dev = lfsck->li_bottom;
const struct lu_name *cname;
char name[8];
- int node = lfsck_dev_idx(dev);
+ int node = lfsck_dev_idx(lfsck);
int rc = 0;
ENTRY;
LASSERT(lfsck->li_master);
+ if (lfsck->li_lpf_root_obj != NULL)
+ RETURN(0);
+
if (node == 0) {
- parent = lfsck_object_find_by_dev(env, dev, &LU_LPF_FID);
+ parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
+ &LU_LPF_FID);
} else {
struct lfsck_tgt_desc *ltd;
LASSERT(dt_object_exists(parent));
- if (unlikely(!dt_try_as_dir(env, parent)))
+ if (unlikely(!dt_try_as_dir(env, parent))) {
+ lfsck_object_put(env, parent);
+
GOTO(put, rc = -ENOTDIR);
+ }
+ lfsck->li_lpf_root_obj = parent;
if (node == 0) {
- rc = lfsck_scan_lpf_bad_entries(env, lfsck, parent);
+ rc = lfsck_scan_lpf_bad_entries(env, lfsck);
if (rc != 0)
CDEBUG(D_LFSCK, "%s: scan .lustre/lost+found/ "
"for bad sub-directories: rc = %d\n",
if (rc != 0)
GOTO(put, rc);
} else {
- child1 = lfsck_object_find_by_dev(env, dev,
+ child1 = lfsck_object_find_bottom(env, lfsck,
&bk->lb_lpf_fid);
- if (IS_ERR(child1))
- GOTO(put, rc = PTR_ERR(child1));
+ if (IS_ERR(child1)) {
+ child1 = NULL;
+ goto find_child2;
+ }
if (unlikely(!dt_object_exists(child1) ||
dt_object_remote(child1)) ||
if (rc != 0)
GOTO(put, rc);
- lu_object_put(env, &child1->do_lu);
+ lfsck_object_put(env, child1);
child1 = NULL;
} else if (unlikely(!dt_try_as_dir(env, child1))) {
GOTO(put, rc = -ENOTDIR);
}
}
+find_child2:
snprintf(name, 8, "MDT%04x", node);
rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
- (const struct dt_key *)name, BYPASS_CAPA);
+ (const struct dt_key *)name);
if (rc == -ENOENT) {
if (!fid_is_zero(&bk->lb_lpf_fid))
goto check_child1;
/* Invalid FID in the name entry, remove the name entry. */
if (!fid_is_norm(cfid)) {
- rc = lfsck_remove_name_entry(env, lfsck, parent, name, S_IFDIR);
+ rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
if (rc != 0)
GOTO(put, rc);
goto check_child1;
}
- child2 = lfsck_object_find_by_dev(env, dev, cfid);
+ child2 = lfsck_object_find_bottom(env, lfsck, cfid);
if (IS_ERR(child2))
GOTO(put, rc = PTR_ERR(child2));
if (unlikely(!dt_object_exists(child2) ||
dt_object_remote(child2)) ||
!S_ISDIR(lfsck_object_type(child2))) {
- rc = lfsck_remove_name_entry(env, lfsck, parent, name,
- S_IFDIR);
+ rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
if (rc != 0)
GOTO(put, rc);
GOTO(put, rc = -ENOTDIR);
if (child1 == NULL) {
- rc = lfsck_verify_lpf_pairs(env, lfsck, parent, child2, name,
+ rc = lfsck_verify_lpf_pairs(env, lfsck, child2, name,
pfid, LVLT_BY_NAMEENTRY);
} else if (!lu_fid_eq(cfid, &bk->lb_lpf_fid)) {
- rc = lfsck_verify_lpf_pairs(env, lfsck, parent, child1, name,
+ rc = lfsck_verify_lpf_pairs(env, lfsck, child1, name,
pfid, LVLT_BY_BOOKMARK);
if (!lu_fid_eq(pfid, &LU_LPF_FID))
- rc = lfsck_verify_lpf_pairs(env, lfsck, parent, child2,
+ rc = lfsck_verify_lpf_pairs(env, lfsck, child2,
name, pfid,
LVLT_BY_NAMEENTRY);
} else {
}
cname = lfsck_name_get_const(env, name, strlen(name));
- rc = lfsck_verify_linkea(env, dev, child2, cname, &LU_LPF_FID);
+ rc = lfsck_verify_linkea(env, child2, cname, &LU_LPF_FID);
}
GOTO(put, rc);
check_child1:
if (child1 != NULL)
- rc = lfsck_verify_lpf_pairs(env, lfsck, parent, child1, name,
+ rc = lfsck_verify_lpf_pairs(env, lfsck, child1, name,
pfid, LVLT_BY_BOOKMARK);
GOTO(put, rc);
put:
- if (lfsck->li_lpf_obj != NULL &&
- unlikely(!dt_try_as_dir(env, lfsck->li_lpf_obj)))
- rc = -ENOTDIR;
+ if (lfsck->li_lpf_obj != NULL) {
+ if (unlikely(!dt_try_as_dir(env, lfsck->li_lpf_obj))) {
+ lfsck_object_put(env, lfsck->li_lpf_obj);
+ lfsck->li_lpf_obj = NULL;
+ rc = -ENOTDIR;
+ }
+ } else if (rc == 0) {
+ rc = lfsck_create_lpf(env, lfsck);
+ }
if (child2 != NULL && !IS_ERR(child2))
- lu_object_put(env, &child2->do_lu);
+ lfsck_object_put(env, child2);
if (child1 != NULL && !IS_ERR(child1))
- lu_object_put(env, &child1->do_lu);
- if (parent != NULL && !IS_ERR(parent))
- lu_object_put(env, &parent->do_lu);
+ lfsck_object_put(env, child1);
return rc;
}
static int lfsck_fid_init(struct lfsck_instance *lfsck)
{
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
- struct seq_server_site *ss;
+ struct seq_server_site *ss = lfsck_dev_site(lfsck);
char *prefix;
int rc = 0;
ENTRY;
- ss = lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
if (unlikely(ss == NULL))
RETURN(-ENXIO);
LASSERT(thread_is_init(thread) || thread_is_stopped(thread));
if (lfsck->li_obj_oit != NULL) {
- lu_object_put_nocache(env, &lfsck->li_obj_oit->do_lu);
+ lfsck_object_put(env, lfsck->li_obj_oit);
lfsck->li_obj_oit = NULL;
}
lfsck_tgt_descs_fini(&lfsck->li_ost_descs);
lfsck_tgt_descs_fini(&lfsck->li_mdt_descs);
+ if (lfsck->li_lfsck_dir != NULL) {
+ lfsck_object_put(env, lfsck->li_lfsck_dir);
+ lfsck->li_lfsck_dir = NULL;
+ }
+
if (lfsck->li_bookmark_obj != NULL) {
- lu_object_put_nocache(env, &lfsck->li_bookmark_obj->do_lu);
+ lfsck_object_put(env, lfsck->li_bookmark_obj);
lfsck->li_bookmark_obj = NULL;
}
if (lfsck->li_lpf_obj != NULL) {
- lu_object_put(env, &lfsck->li_lpf_obj->do_lu);
+ lfsck_object_put(env, lfsck->li_lpf_obj);
lfsck->li_lpf_obj = NULL;
}
+ if (lfsck->li_lpf_root_obj != NULL) {
+ lfsck_object_put(env, lfsck->li_lpf_root_obj);
+ lfsck->li_lpf_root_obj = NULL;
+ }
+
if (lfsck->li_los != NULL) {
local_oid_storage_fini(env, lfsck->li_los);
lfsck->li_los = NULL;
int flag;
int i;
bool newline = (bits != 0 ? false : true);
+ int rc;
- seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
+ rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
+ if (rc < 0)
+ return rc;
for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
if (flag & bits) {
if (bits == 0)
newline = true;
- seq_printf(m, "%s%c", names[i],
- newline ? '\n' : ',');
+ rc = seq_printf(m, "%s%c", names[i],
+ newline ? '\n' : ',');
+ if (rc < 0)
+ return rc;
}
}
}
if (!newline)
- seq_printf(m, "\n");
- return 0;
+ rc = seq_printf(m, "\n");
+
+ return rc;
}
-int lfsck_time_dump(struct seq_file *m, __u64 time, const char *prefix)
+int lfsck_time_dump(struct seq_file *m, __u64 time, const char *name)
{
- if (time != 0)
- seq_printf(m, "%s: "LPU64" seconds\n", prefix,
- cfs_time_current_sec() - time);
- else
- seq_printf(m, "%s: N/A\n", prefix);
- return 0;
+ int rc;
+
+ if (time == 0) {
+ rc = seq_printf(m, "%s_time: N/A\n", name);
+ if (rc == 0)
+ rc = seq_printf(m, "time_since_%s: N/A\n", name);
+
+ return rc;
+ }
+
+ rc = seq_printf(m, "%s_time: "LPU64"\n", name, time);
+ if (rc == 0)
+ rc = seq_printf(m, "time_since_%s: "LPU64" seconds\n",
+ name, cfs_time_current_sec() - time);
+
+ return rc;
}
int lfsck_pos_dump(struct seq_file *m, struct lfsck_position *pos,
{
if (fid_is_zero(&pos->lp_dir_parent)) {
if (pos->lp_oit_cookie == 0)
- seq_printf(m, "%s: N/A, N/A, N/A\n",
- prefix);
- else
- seq_printf(m, "%s: "LPU64", N/A, N/A\n",
- prefix, pos->lp_oit_cookie);
- } else {
- seq_printf(m, "%s: "LPU64", "DFID", "LPX64"\n",
- prefix, pos->lp_oit_cookie,
- PFID(&pos->lp_dir_parent), pos->lp_dir_cookie);
+ return seq_printf(m, "%s: N/A, N/A, N/A\n", prefix);
+
+ return seq_printf(m, "%s: "LPU64", N/A, N/A\n",
+ prefix, pos->lp_oit_cookie);
}
- return 0;
+
+ return seq_printf(m, "%s: "LPU64", "DFID", "LPX64"\n",
+ prefix, pos->lp_oit_cookie,
+ PFID(&pos->lp_dir_parent), pos->lp_dir_cookie);
}
void lfsck_pos_fill(const struct lu_env *env, struct lfsck_instance *lfsck,
bool dirty = false;
if (limit != LFSCK_SPEED_NO_LIMIT) {
- if (limit > HZ) {
- lfsck->li_sleep_rate = limit / HZ;
+ if (limit > msecs_to_jiffies(MSEC_PER_SEC)) {
+ lfsck->li_sleep_rate = limit /
+ msecs_to_jiffies(MSEC_PER_SEC);
lfsck->li_sleep_jif = 1;
} else {
lfsck->li_sleep_rate = 1;
- lfsck->li_sleep_jif = HZ / limit;
+ lfsck->li_sleep_jif = msecs_to_jiffies(MSEC_PER_SEC) /
+ limit;
}
} else {
lfsck->li_sleep_jif = 0;
return lad;
}
+struct lfsck_assistant_object *
+lfsck_assistant_object_init(const struct lu_env *env, const struct lu_fid *fid,
+ const struct lu_attr *attr, __u64 cookie,
+ bool is_dir)
+{
+ struct lfsck_assistant_object *lso;
+
+ OBD_ALLOC_PTR(lso);
+ if (lso == NULL)
+ return ERR_PTR(-ENOMEM);
+
+ lso->lso_fid = *fid;
+ if (attr != NULL)
+ lso->lso_attr = *attr;
+
+ atomic_set(&lso->lso_ref, 1);
+ lso->lso_oit_cookie = cookie;
+ if (is_dir)
+ lso->lso_is_dir = 1;
+
+ return lso;
+}
+
+struct dt_object *
+lfsck_assistant_object_load(const struct lu_env *env,
+ struct lfsck_instance *lfsck,
+ struct lfsck_assistant_object *lso)
+{
+ struct dt_object *obj;
+
+ obj = lfsck_object_find_bottom(env, lfsck, &lso->lso_fid);
+ if (IS_ERR(obj))
+ return obj;
+
+ if (unlikely(!dt_object_exists(obj) || lfsck_is_dead_obj(obj))) {
+ lso->lso_dead = 1;
+ lfsck_object_put(env, obj);
+
+ return ERR_PTR(-ENOENT);
+ }
+
+ if (lso->lso_is_dir && unlikely(!dt_try_as_dir(env, obj))) {
+ lfsck_object_put(env, obj);
+
+ return ERR_PTR(-ENOTDIR);
+ }
+
+ return obj;
+}
+
/**
* Generic LFSCK asynchronous communication interpretor function.
* The LFSCK RPC reply for both the event notification and status
spin_unlock(<ds->ltd_lock);
memset(lr, 0, sizeof(*lr));
- lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
+ lr->lr_index = lfsck_dev_idx(lfsck);
lr->lr_event = LE_PEER_EXIT;
lr->lr_active = type;
lr->lr_status = LS_CO_PAUSED;
lad->lad_to_double_scan = 0;
lad->lad_in_double_scan = 0;
lad->lad_exit = 0;
+ lad->lad_advance_lock = false;
thread_set_flags(athread, 0);
lta = lfsck_thread_args_init(lfsck, com, lsp);
struct ptlrpc_thread *athread = &lad->lad_thread;
struct l_wait_info lwi = { 0 };
- if (com->lc_new_checked == 0)
- return LFSCK_CHECKPOINT_SKIP;
-
l_wait_event(mthread->t_ctl_waitq,
list_empty(&lad->lad_req_list) ||
!thread_is_running(mthread) ||
lfsck = lfsck_instance_find(key, true, false);
if (likely(lfsck != NULL)) {
- if (val > LFSCK_ASYNC_WIN_MAX) {
- CWARN("%s: Too large async window size, which "
- "may cause memory issues. The valid range "
- "is [0 - %u]. If you do not want to restrict "
- "the window size for async requests pipeline, "
- "just set it as 0.\n",
+ if (val < 1 || val > LFSCK_ASYNC_WIN_MAX) {
+ CWARN("%s: invalid async windows size that may "
+ "cause memory issues. The valid range is "
+ "[1 - %u].\n",
lfsck_lfsck2name(lfsck), LFSCK_ASYNC_WIN_MAX);
rc = -EINVAL;
} else if (lfsck->li_bookmark_ram.lb_async_windows != val) {
memset(lr, 0, sizeof(*lr));
lr->lr_event = LE_STOP;
- lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
+ lr->lr_index = lfsck_dev_idx(lfsck);
lr->lr_status = stop->ls_status;
lr->lr_version = bk->lb_version;
lr->lr_active = LFSCK_TYPES_ALL;
memset(lr, 0, sizeof(*lr));
lr->lr_event = LE_START;
- lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
+ lr->lr_index = lfsck_dev_idx(lfsck);
lr->lr_speed = bk->lb_speed_limit;
lr->lr_version = bk->lb_version;
lr->lr_active = start->ls_active;
lr->lr_param = start->ls_flags;
lr->lr_async_windows = bk->lb_async_windows;
lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
- LSV_ASYNC_WINDOWS;
+ LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ |
+ LSV_CREATE_MDTOBJ;
laia->laia_com = NULL;
laia->laia_ltds = ltds;
laia->laia_ltd = ltd;
ltd->ltd_layout_done = 0;
ltd->ltd_namespace_done = 0;
+ ltd->ltd_synced_failures = 0;
rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
lfsck_async_interpret, laia,
LFSCK_NOTIFY);
case LE_FID_ACCESSED:
case LE_PEER_EXIT:
case LE_CONDITIONAL_DESTROY:
- case LE_CREATE_ORPHAN:
case LE_SKIP_NLINK_DECLARE:
case LE_SKIP_NLINK:
case LE_SET_LMV_MASTER:
RETURN(rc);
}
-EXPORT_SYMBOL(lfsck_query);
int lfsck_register_namespace(const struct lu_env *env, struct dt_device *key,
struct ldlm_namespace *ns)
lfsck->li_local_root_fid = *fid;
if (master) {
lfsck->li_master = 1;
- if (lfsck_dev_idx(key) == 0) {
+ if (lfsck_dev_idx(lfsck) == 0) {
struct lu_fid *pfid = &lfsck_env_info(env)->lti_fid2;
const struct lu_name *cname;
rc = dt_lookup(env, root,
(struct dt_rec *)(&lfsck->li_global_root_fid),
- (const struct dt_key *)"ROOT", BYPASS_CAPA);
+ (const struct dt_key *)"ROOT");
if (rc != 0)
GOTO(out, rc);
if (IS_ERR(obj))
GOTO(out, rc = PTR_ERR(obj));
+ if (unlikely(!dt_try_as_dir(env, obj)))
+ GOTO(out, rc = -ENOTDIR);
+
rc = dt_lookup(env, obj, (struct dt_rec *)fid,
- (const struct dt_key *)dotlustre, BYPASS_CAPA);
+ (const struct dt_key *)dotlustre);
if (rc != 0)
GOTO(out, rc);
- lu_object_put(env, &obj->do_lu);
+ lfsck_object_put(env, obj);
obj = dt_locate(env, key, fid);
if (IS_ERR(obj))
GOTO(out, rc = PTR_ERR(obj));
cname = lfsck_name_get_const(env, dotlustre,
strlen(dotlustre));
- rc = lfsck_verify_linkea(env, key, obj, cname,
+ rc = lfsck_verify_linkea(env, obj, cname,
&lfsck->li_global_root_fid);
if (rc != 0)
GOTO(out, rc);
+ if (unlikely(!dt_try_as_dir(env, obj)))
+ GOTO(out, rc = -ENOTDIR);
+
*pfid = *fid;
rc = dt_lookup(env, obj, (struct dt_rec *)fid,
- (const struct dt_key *)lostfound,
- BYPASS_CAPA);
+ (const struct dt_key *)lostfound);
if (rc != 0)
GOTO(out, rc);
- lu_object_put(env, &obj->do_lu);
+ lfsck_object_put(env, obj);
obj = dt_locate(env, key, fid);
if (IS_ERR(obj))
GOTO(out, rc = PTR_ERR(obj));
cname = lfsck_name_get_const(env, lostfound,
strlen(lostfound));
- rc = lfsck_verify_linkea(env, key, obj, cname, pfid);
+ rc = lfsck_verify_linkea(env, obj, cname, pfid);
if (rc != 0)
GOTO(out, rc);
- lu_object_put(env, &obj->do_lu);
+ lfsck_object_put(env, obj);
obj = NULL;
}
}
if (IS_ERR(obj))
GOTO(out, rc = PTR_ERR(obj));
- lu_object_get(&obj->do_lu);
- lfsck->li_obj_oit = obj;
rc = obj->do_ops->do_index_try(env, obj, &dt_otable_features);
if (rc != 0)
GOTO(out, rc);
+ lfsck->li_obj_oit = obj;
+ obj = local_file_find_or_create(env, lfsck->li_los, root, LFSCK_DIR,
+ S_IFDIR | S_IRUGO | S_IWUSR);
+ if (IS_ERR(obj))
+ GOTO(out, rc = PTR_ERR(obj));
+
+ lu_object_get(&obj->do_lu);
+ lfsck->li_lfsck_dir = obj;
rc = lfsck_bookmark_setup(env, lfsck);
if (rc != 0)
GOTO(out, rc);
rc = lfsck_add_target_from_orphan(env, lfsck);
out:
if (obj != NULL && !IS_ERR(obj))
- lu_object_put(env, &obj->do_lu);
+ lfsck_object_put(env, obj);
if (root != NULL && !IS_ERR(root))
- lu_object_put(env, &root->do_lu);
+ lfsck_object_put(env, root);
if (rc != 0)
lfsck_instance_cleanup(env, lfsck);
return rc;
if (unlikely(index >= ltds->ltd_tgts_bitmap->size))
goto unlock;
- ltd = LTD_TGT(ltds, index);
+ ltd = lfsck_ltd2tgt(ltds, index);
if (unlikely(ltd == NULL))
goto unlock;
ltds->ltd_tgtnr--;
cfs_bitmap_clear(ltds->ltd_tgts_bitmap, index);
- LTD_TGT(ltds, index) = NULL;
+ lfsck_assign_tgt(ltds, NULL, index);
unlock:
if (ltd == NULL) {
MODULE_AUTHOR("Intel Corporation <http://www.intel.com/>");
MODULE_DESCRIPTION("LFSCK");
+MODULE_VERSION(LUSTRE_VERSION_STRING);
MODULE_LICENSE("GPL");
-cfs_module(lfsck, LUSTRE_VERSION_STRING, lfsck_init, lfsck_exit);
+module_init(lfsck_init);
+module_exit(lfsck_exit);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff