*/
/*
* This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
*
* lustre/osd/osd_handler.c
*
struct dir_context ctx;
struct osd_thread_info *oclb_info;
struct osd_device *oclb_dev;
- struct osd_idmap_cache *oclb_oic;
int oclb_items;
bool oclb_found;
};
struct lu_fid *fid = &oti->oti_fid3;
struct osd_inode_id *id = &oti->oti_id3;
struct osd_device *dev = oclb->oclb_dev;
- struct osd_idmap_cache *oic = oclb->oclb_oic;
struct inode *inode;
oclb->oclb_items++;
iput(inode);
osd_add_oi_cache(oti, dev, id, fid);
- oic->oic_fid = *fid;
- oic->oic_lid = *id;
- oic->oic_dev = dev;
- osd_oii_insert(dev, oic, true);
+ osd_oii_insert(dev, fid, id, true);
oclb->oclb_found = true;
return 1;
* the correct OI mapping for the slave MDT-object.
*/
static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
- struct inode *inode, struct osd_idmap_cache *oic)
+ struct inode *inode)
{
struct lu_buf *buf = &oti->oti_big_buf;
struct dentry *dentry = &oti->oti_obj_dentry;
.ctx.actor = osd_stripe_dir_filldir,
.oclb_info = oti,
.oclb_dev = dev,
- .oclb_oic = oic,
.oclb_found = false,
};
int rc = 0;
out:
if (rc < 0)
CDEBUG(D_LFSCK,
- "%s: cannot check LMV, ino = %lu/%u "DFID": rc = %d\n",
+ "%s: cannot check LMV, ino = %lu/%u: rc = %d\n",
osd_ino2name(inode), inode->i_ino, inode->i_generation,
- PFID(&oic->oic_fid), rc);
+ rc);
else
rc = 0;
if (scrub->os_partial_scan && !scrub->os_in_join)
goto join;
- osd_add_oi_cache(info, dev, id, fid);
if (IS_ERR_OR_NULL(inode) || result) {
- osd_oii_insert(dev, oic, result == -ENOENT);
+ osd_oii_insert(dev, fid, id, result == -ENOENT);
GOTO(out, result = -EINPROGRESS);
}
LASSERT(remote);
LASSERT(obj->oo_inode == inode);
- osd_oii_insert(dev, oic, true);
+ osd_oii_insert(dev, fid, id, true);
goto found;
}
join:
rc1 = osd_scrub_start(env, dev, flags);
- LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
- "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
+ CDEBUG_LIMIT(D_LFSCK | D_CONSOLE | D_WARNING,
+ "%s: trigger OI scrub by RPC for "DFID"/%u with flags %#x: rc = %d\n",
+ osd_name(dev), PFID(fid), id->oii_ino, flags, rc1);
if (rc1 && rc1 != -EALREADY)
GOTO(out, result = -EREMCHG);
- osd_add_oi_cache(info, dev, id, fid);
if (IS_ERR_OR_NULL(inode) || result) {
- osd_oii_insert(dev, oic, result == -ENOENT);
+ osd_oii_insert(dev, fid, id, result == -ENOENT);
GOTO(out, result = -EINPROGRESS);
}
LASSERT(remote);
LASSERT(obj->oo_inode == inode);
- osd_oii_insert(dev, oic, true);
+ osd_oii_insert(dev, fid, id, true);
goto found;
check_lma:
if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
result = -EREMCHG;
+ osd_scrub_refresh_mapping(info, dev, fid, id, DTO_INDEX_DELETE,
+ true, 0, NULL);
goto trigger;
}
if (S_ISDIR(inode->i_mode) &&
(flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
- osd_check_lmv(info, dev, inode, oic);
+ osd_check_lmv(info, dev, inode);
result = osd_attach_jinode(inode);
if (result)
if (!rc)
rc = rc2;
- osd_process_truncates(&truncates);
+ osd_process_truncates(env, &truncates);
} else {
osd_trans_stop_cb(oh, th->th_result);
OBD_FREE_PTR(oh);
OBD_CKSUM_T10IP512 :
OBD_CKSUM_T10IP4K;
} else {
- CERROR("%s: unsupported checksum type of "
- "T10PI type '%s'",
+ CERROR("%s: unsupported checksum type of T10PI type '%s'\n",
d->od_svname, name);
}
} else {
- CERROR("%s: unsupported T10PI type '%s'",
+ CERROR("%s: unsupported T10PI type '%s'\n",
d->od_svname, name);
}
}
return result;
}
-
static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
struct dt_object *parent, struct dt_object *child,
umode_t child_mode)
struct thandle *handle)
{
struct osd_thandle *oh;
+ struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
+ int credits;
int rc;
ENTRY;
* vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
* leaf_node, could involves the block, block bitmap, groups, GDT
* change for each block, so add 4 * 2 credits in that case.
+ *
+ * The default ACL initialization may consume an additional 16 blocks
+ */
+ credits = osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
+ ((dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
+
+ /**
+ * While ldiskfs_new_inode() calls ldiskfs_init_acl() we have to add
+ * credits for possible default ACL creation in new inode
*/
- osd_trans_declare_op(env, oh, OSD_OT_CREATE,
- osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
- (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
+ if (hint && hint->dah_acl_len)
+ credits += osd_calc_bkmap_credits(sb, NULL, 0, -1,
+ (hint->dah_acl_len + sb->s_blocksize - 1) >>
+ sb->s_blocksize_bits);
+
+ osd_trans_declare_op(env, oh, OSD_OT_CREATE, credits);
+
/*
* Reuse idle OI block may cause additional one OI block
* to be changed.
dquot_initialize(inode);
rc = ll_vfs_removexattr(dentry, inode, XATTR_NAME_FID);
if (rc == -ENODATA) {
- if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
- RETURN(rc);
+ /* XATTR_NAME_FID is already absent */
+ rc = 0;
} else if (rc) {
RETURN(rc);
}
static int
osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
- struct osd_idmap_cache *oic)
+ const struct lu_fid *fid, struct osd_inode_id *id)
{
struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
- struct lu_fid *fid = &oic->oic_fid;
- struct osd_inode_id *id = &oic->oic_lid;
struct inode *inode = NULL;
int once = 0;
bool insert;
}
}
- rc = osd_oii_insert(dev, oic, insert);
+ rc = osd_oii_insert(dev, fid, id, insert);
/*
* There is race condition between osd_oi_lookup and OI scrub.
* The OI scrub finished just after osd_oi_lookup() failure.
if (!S_ISDIR(inode->i_mode))
rc = 0;
else
- rc = osd_check_lmv(oti, dev, inode, oic);
+ rc = osd_check_lmv(oti, dev, inode);
GOTO(out, rc);
}
if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
- CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
- "%s: trigger partial OI scrub for RPC inconsistency checking FID "DFID": rc = %d\n",
- osd_dev2name(dev), PFID(fid), rc);
+ CDEBUG_LIMIT(D_LFSCK | D_CONSOLE | D_WARNING,
+ "%s: trigger partial OI scrub for RPC inconsistency, checking FID "DFID"/%u: rc = %d\n",
+ osd_dev2name(dev), PFID(fid), id->oii_ino, rc);
if (rc == 0 || rc == -EALREADY)
goto again;
}
GOTO(out, rc);
out:
- if (inode)
- iput(inode);
+ iput(inode);
RETURN(rc);
}
static int osd_fail_fid_lookup(struct osd_thread_info *oti,
struct osd_device *dev,
- struct osd_idmap_cache *oic,
struct lu_fid *fid, __u32 ino)
{
struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
+ struct osd_idmap_cache *oic = &oti->oti_cache;
struct inode *inode;
int rc;
if (!IS_ERR(bh)) {
struct osd_thread_info *oti = osd_oti_get(env);
struct osd_inode_id *id = &oti->oti_id;
- struct osd_idmap_cache *oic = &oti->oti_cache;
struct osd_device *dev = osd_obj2dev(obj);
ino = le32_to_cpu(de->inode);
if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
brelse(bh);
- rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
+ rc = osd_fail_fid_lookup(oti, dev, fid, ino);
GOTO(out, rc);
}
osd_id_gen(id, ino, OSD_OII_NOGEN);
}
- if (rc != 0 || osd_remote_fid(env, dev, fid)) {
- fid_zero(&oic->oic_fid);
-
+ if (rc != 0 || osd_remote_fid(env, dev, fid))
GOTO(out, rc);
- }
- osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
- rc = osd_consistency_check(oti, dev, oic);
- if (rc == -ENOENT)
- fid_zero(&oic->oic_fid);
- else
+ rc = osd_consistency_check(oti, dev, fid, id);
+ if (rc != -ENOENT) {
/* Other error should not affect lookup result. */
rc = 0;
+
+ /* Normal file mapping should be added into OI cache
+ * after FID in LMA check, but for local files like
+ * hsm_actions, their FIDs are not stored in OI files,
+ * see osd_initial_OI_scrub(), and here is the only
+ * place to load mapping into OI cache.
+ */
+ if (!fid_is_namespace_visible(fid))
+ osd_add_oi_cache(osd_oti_get(env),
+ osd_obj2dev(obj), id, fid);
+ }
} else {
rc = PTR_ERR(bh);
}
}
};
-
-/**
- * Creates or initializes iterator context.
- *
- * \retval struct osd_it_ea, iterator structure on success
- *
- */
-static struct dt_it *osd_it_ea_init(const struct lu_env *env,
- struct dt_object *dt,
- __u32 attr)
+struct osd_it_ea *osd_it_dir_init(const struct lu_env *env,
+ struct inode *inode, __u32 attr)
{
- struct osd_object *obj = osd_dt_obj(dt);
struct osd_thread_info *info = osd_oti_get(env);
struct osd_it_ea *oie;
struct file *file;
- struct lu_object *lo = &dt->do_lu;
struct dentry *obj_dentry;
ENTRY;
- if (!dt_object_exists(dt) || obj->oo_destroyed)
- RETURN(ERR_PTR(-ENOENT));
-
OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
if (oie == NULL)
RETURN(ERR_PTR(-ENOMEM));
obj_dentry = &oie->oie_dentry;
- obj_dentry->d_inode = obj->oo_inode;
- obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
+ obj_dentry->d_inode = inode;
+ obj_dentry->d_sb = inode->i_sb;
obj_dentry->d_name.hash = 0;
oie->oie_rd_dirent = 0;
if (oie->oie_buf == NULL)
RETURN(ERR_PTR(-ENOMEM));
}
- oie->oie_obj = obj;
+ oie->oie_obj = NULL;
file = &oie->oie_file;
else
file->f_mode = FMODE_32BITHASH;
file->f_path.dentry = obj_dentry;
- file->f_mapping = obj->oo_inode->i_mapping;
- file->f_op = obj->oo_inode->i_fop;
- file->f_inode = obj->oo_inode;
+ file->f_mapping = inode->i_mapping;
+ file->f_op = inode->i_fop;
+ file->f_inode = inode;
+
+ RETURN(oie);
+}
+
+/**
+ * Creates or initializes iterator context.
+ *
+ * \retval struct osd_it_ea, iterator structure on success
+ *
+ */
+static struct dt_it *osd_it_ea_init(const struct lu_env *env,
+ struct dt_object *dt,
+ __u32 attr)
+{
+ struct osd_object *obj = osd_dt_obj(dt);
+ struct lu_object *lo = &dt->do_lu;
+ struct osd_it_ea *oie;
+
+ ENTRY;
+
+ if (!dt_object_exists(dt) || obj->oo_destroyed)
+ RETURN(ERR_PTR(-ENOENT));
+ oie = osd_it_dir_init(env, obj->oo_inode, attr);
+ if (IS_ERR(oie))
+ RETURN((struct dt_it *)oie);
+
+ oie->oie_obj = obj;
lu_object_get(lo);
RETURN((struct dt_it *)oie);
}
+void osd_it_dir_fini(const struct lu_env *env, struct osd_it_ea *oie,
+ struct inode *inode)
+{
+ struct osd_thread_info *info = osd_oti_get(env);
+
+ ENTRY;
+ oie->oie_file.f_op->release(inode, &oie->oie_file);
+ if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
+ OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
+ else
+ info->oti_it_ea_buf_used = 0;
+ OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
+ EXIT;
+}
+
/**
* Destroy or finishes iterator context.
*
*/
static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
{
- struct osd_thread_info *info = osd_oti_get(env);
struct osd_it_ea *oie = (struct osd_it_ea *)di;
struct osd_object *obj = oie->oie_obj;
struct inode *inode = obj->oo_inode;
ENTRY;
- oie->oie_file.f_op->release(inode, &oie->oie_file);
+ osd_it_dir_fini(env, (struct osd_it_ea *)di, inode);
osd_object_put(env, obj);
- if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
- OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
- else
- info->oti_it_ea_buf_used = 0;
- OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
EXIT;
}
/* "." is just the object itself. */
if (namelen == 1 && name[0] == '.') {
- *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
+ if (obj != NULL)
+ *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
} else if (d_type & LDISKFS_DIRENT_LUFID) {
rec = (struct osd_fid_pack *)(name + namelen + 1);
if (osd_fid_unpack(fid, rec) != 0)
d_type &= ~LDISKFS_DIRENT_LUFID;
/* NOT export local root. */
- if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
+ if (obj != NULL &&
+ unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
ino = obj->oo_inode->i_ino;
*fid = obj->oo_dt.do_lu.lo_header->loh_fid;
}
* \retval -ve on error
* \retval +1 reach the end of entry
*/
-static int osd_ldiskfs_it_fill(const struct lu_env *env,
- const struct dt_it *di)
+int osd_ldiskfs_it_fill(const struct lu_env *env, const struct dt_it *di)
{
struct osd_it_ea *it = (struct osd_it_ea *)di;
struct osd_object *obj = it->oie_obj;
- struct inode *inode = obj->oo_inode;
struct htree_lock *hlock = NULL;
struct file *filp = &it->oie_file;
int rc = 0;
it->oie_dirent = it->oie_buf;
it->oie_rd_dirent = 0;
- if (obj->oo_hl_head != NULL) {
- hlock = osd_oti_get(env)->oti_hlock;
- ldiskfs_htree_lock(hlock, obj->oo_hl_head,
- inode, LDISKFS_HLOCK_READDIR);
- } else {
- down_read(&obj->oo_ext_idx_sem);
+ if (obj) {
+ if (obj->oo_hl_head != NULL) {
+ hlock = osd_oti_get(env)->oti_hlock;
+ ldiskfs_htree_lock(hlock, obj->oo_hl_head,
+ obj->oo_inode,
+ LDISKFS_HLOCK_READDIR);
+ } else {
+ down_read(&obj->oo_ext_idx_sem);
+ }
}
filp->f_cred = current_cred();
it->oie_it_dirent = 1;
}
unlock:
- if (hlock != NULL)
- ldiskfs_htree_unlock(hlock);
- else
- up_read(&obj->oo_ext_idx_sem);
+ if (obj) {
+ if (hlock != NULL)
+ ldiskfs_htree_unlock(hlock);
+ else
+ up_read(&obj->oo_ext_idx_sem);
+ }
RETURN(rc);
}
EXIT;
}
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
+# ifndef LDISKFS_HAS_INCOMPAT_FEATURE
+/* Newer kernels provide the ldiskfs_set_feature_largedir() wrapper already,
+ * which calls ldiskfs_update_dynamic_rev() to update ancient filesystems.
+ * All ldiskfs filesystems are already v2, so it is a no-op and unnecessary.
+ * This avoids maintaining patches to export this otherwise-useless function.
+ */
+void ldiskfs_update_dynamic_rev(struct super_block *sb)
+{
+ /* do nothing */
+}
+# endif
+#endif
+
static int osd_mount(const struct lu_env *env,
struct osd_device *o, struct lustre_cfg *cfg)
{
"force_over_256tb",
"force_over_512tb",
"force_over_1024tb",
+ "resetoi",
NULL
};
strncat(options, opts, PAGE_SIZE);
}
if (lmd_flags & LMD_FLG_DEV_RDONLY) {
- LCONSOLE_WARN("%s: not support dev_rdonly on this device",
+ LCONSOLE_WARN("%s: not support dev_rdonly on this device\n",
name);
GOTO(out_mnt, rc = -EOPNOTSUPP);
GOTO(out_mnt, rc = -EINVAL);
}
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
#ifdef LDISKFS_MOUNT_DIRDATA
if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
"downgrade to Lustre-1.x again, you can enable it via "
"'tune2fs -O dirdata device'\n", name, dev);
#endif
+ /* enable large_dir on MDTs to avoid REMOTE_PARENT_DIR overflow,
+ * and on very large OSTs to avoid object directory overflow */
+ if (unlikely(!ldiskfs_has_feature_largedir(o->od_mnt->mnt_sb) &&
+ !strstr(name, "MGS"))) {
+ ldiskfs_set_feature_largedir(o->od_mnt->mnt_sb);
+ LCONSOLE_INFO("%s: enable 'large_dir' feature on device '%s'\n",
+ name, dev);
+ }
+#endif
inode = osd_sb(o)->s_root->d_inode;
lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
{
struct lu_device *l = osd2lu_dev(o);
struct osd_thread_info *info;
- int rc;
int cplen = 0;
+ char *opts = NULL;
+ bool restored = false;
+ int rc;
/* if the module was re-loaded, env can loose its keys */
rc = lu_env_refill((struct lu_env *)env);
if (rc != 0)
GOTO(out_site, rc);
+ opts = lustre_cfg_string(cfg, 3);
+ if (opts && strstr(opts, "resetoi"))
+ restored = true;
+
INIT_LIST_HEAD(&o->od_ios_list);
/* setup scrub, including OI files initialization */
o->od_in_init = 1;
- rc = osd_scrub_setup(env, o);
+ rc = osd_scrub_setup(env, o, restored);
o->od_in_init = 0;
if (rc < 0)
GOTO(out_site, rc);
#ifdef CONFIG_KALLSYMS
priv_security_file_alloc =
- (void *)kallsyms_lookup_name("security_file_alloc");
+ (void *)cfs_kallsyms_lookup_name("security_file_alloc");
#endif
rc = class_register_type(&osd_obd_device_ops, NULL, true,
git://git.whamcloud.com / fs / lustre-release.git / blobdiff