* Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2015, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
*
* lustre/osd/osd_compat.c
*
* Author: Alex Zhuravlev <bzzz@whamcloud.com>
*/
-/* LUSTRE_VERSION_CODE */
-#include <lustre_ver.h>
/* prerequisite for linux/xattr.h */
#include <linux/types.h>
/* prerequisite for linux/xattr.h */
#include "osd_oi.h"
static void osd_push_ctxt(const struct osd_device *dev,
- struct lvfs_run_ctxt *newctxt,
- struct lvfs_run_ctxt *save)
+ struct lvfs_run_ctxt *newctxt,
+ struct lvfs_run_ctxt *save)
{
OBD_SET_CTXT_MAGIC(newctxt);
newctxt->pwdmnt = dev->od_mnt;
newctxt->pwd = dev->od_mnt->mnt_root;
- newctxt->fs = get_ds();
newctxt->umask = current_umask();
newctxt->dt = NULL;
push_ctxt(save, newctxt);
}
+struct dentry *osd_lookup_one_len_common(struct osd_device *dev,
+ const char *name,
+ struct dentry *base, int len,
+ enum oi_check_flags flags)
+{
+ struct dentry *dchild;
+
+ /*
+ * We can't use inode_is_locked() directly since we can't know
+ * if the current thread context took the lock earlier or if
+ * another thread context took the lock. OI_LOCKED tells us
+ * if the current thread context has already taken the lock.
+ */
+ if (!(flags & OI_LOCKED)) {
+ /* If another thread took this lock already we will
+ * just have to wait until the other thread is done.
+ */
+ inode_lock(base->d_inode);
+ dchild = lookup_one_len(name, base, len);
+ inode_unlock(base->d_inode);
+ } else {
+ /* This thread context already has taken the lock.
+ * Other threads will have to wait until we are done.
+ */
+ dchild = lookup_one_len(name, base, len);
+ }
+ if (IS_ERR(dchild))
+ return dchild;
+
+ if (dchild->d_inode && unlikely(is_bad_inode(dchild->d_inode))) {
+ CERROR("%s: bad inode returned %lu/%u: rc = -ENOENT\n",
+ osd_name(dev), dchild->d_inode->i_ino,
+ dchild->d_inode->i_generation);
+ dput(dchild);
+ dchild = ERR_PTR(-ENOENT);
+ }
+
+ return dchild;
+}
+
+/**
+ * osd_lookup_one_len_unlocked
+ *
+ * @dev: obd device we are searching
+ * @name: pathname component to lookup
+ * @base: base directory to lookup from
+ * @len: maximum length @len should be interpreted to
+ *
+ * Unlike osd_lookup_one_len, this should be called without the parent
+ * i_mutex held, and will take the i_mutex itself.
+ */
+struct dentry *osd_lookup_one_len_unlocked(struct osd_device *dev,
+ const char *name,
+ struct dentry *base, int len)
+{
+ return osd_lookup_one_len_common(dev, name, base, len, ~OI_LOCKED);
+}
+
+/**
+ * osd_lookup_one_len - lookup single pathname component
+ *
+ * @dev: obd device we are searching
+ * @name: pathname component to lookup
+ * @base: base directory to lookup from
+ * @len: maximum length @len should be interpreted to
+ *
+ * The caller must hold inode lock
+ */
+struct dentry *osd_lookup_one_len(struct osd_device *dev, const char *name,
+ struct dentry *base, int len)
+{
+ return osd_lookup_one_len_common(dev, name, base, len, OI_LOCKED);
+}
+
/* utility to make a directory */
-static struct dentry *simple_mkdir(struct dentry *dir, struct vfsmount *mnt,
- const char *name, int mode, int fix)
+static struct dentry *
+simple_mkdir(const struct lu_env *env, struct osd_device *osd,
+ struct dentry *dir, const struct lu_fid *fid,
+ const char *name, __u32 compat, int mode, bool *created)
{
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct lu_fid *tfid = &info->oti_fid3;
+ struct inode *inode;
struct dentry *dchild;
int err = 0;
+
ENTRY;
// ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
- dchild = ll_lookup_one_len(name, dir, strlen(name));
+ dchild = osd_lookup_one_len_unlocked(osd, name, dir, strlen(name));
if (IS_ERR(dchild))
- GOTO(out_up, dchild);
+ RETURN(dchild);
+
+ inode = dchild->d_inode;
+ if (inode) {
+ struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
+ int old_mode = inode->i_mode;
+
+ if (created)
+ *created = false;
- if (dchild->d_inode) {
- int old_mode = dchild->d_inode->i_mode;
if (!S_ISDIR(old_mode)) {
CERROR("found %s (%lu/%u) is mode %o\n", name,
- dchild->d_inode->i_ino,
- dchild->d_inode->i_generation, old_mode);
+ inode->i_ino, inode->i_generation, old_mode);
GOTO(out_err, err = -ENOTDIR);
}
+ if (unlikely(osd->od_dt_dev.dd_rdonly))
+ RETURN(dchild);
+
/* Fixup directory permissions if necessary */
- if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
+ if ((old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
CDEBUG(D_CONFIG,
"fixing permissions on %s from %o to %o\n",
name, old_mode, mode);
- dchild->d_inode->i_mode = (mode & S_IALLUGO) |
- (old_mode & ~S_IALLUGO);
- mark_inode_dirty(dchild->d_inode);
+ inode->i_mode = (mode & S_IALLUGO) |
+ (old_mode & ~S_IALLUGO);
+ mark_inode_dirty(inode);
}
- GOTO(out_up, dchild);
+
+ err = osd_get_lma(info, inode, &info->oti_obj_dentry,
+ &info->oti_ost_attrs);
+ if (err == -ENODATA)
+ goto set_fid;
+
+ if (err)
+ GOTO(out_err, err);
+
+ if ((fid && !lu_fid_eq(fid, &lma->lma_self_fid)) ||
+ lma->lma_compat != compat)
+ goto set_fid;
+
+ RETURN(dchild);
}
- err = vfs_mkdir(dir->d_inode, dchild, mode);
+ err = vfs_mkdir(&init_user_ns, dir->d_inode, dchild, mode);
+ if (err)
+ GOTO(out_err, err);
+
+ inode = dchild->d_inode;
+ if (created)
+ *created = true;
+
+set_fid:
+ if (fid)
+ *tfid = *fid;
+ else
+ lu_igif_build(tfid, inode->i_ino, inode->i_generation);
+ err = osd_ea_fid_set(info, inode, tfid, compat, 0);
if (err)
GOTO(out_err, err);
out_err:
dput(dchild);
- dchild = ERR_PTR(err);
-out_up:
- return dchild;
+ return ERR_PTR(err);
}
static int osd_last_rcvd_subdir_count(struct osd_device *osd)
{
- struct lr_server_data lsd;
- struct dentry *dlast;
- loff_t off;
- int rc = 0;
- int count = OBJ_SUBDIR_COUNT;
-
- ENTRY;
-
- dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
- strlen(LAST_RCVD));
- if (IS_ERR(dlast))
- return PTR_ERR(dlast);
- else if (dlast->d_inode == NULL)
- goto out;
-
- off = 0;
- rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
- if (rc == sizeof(lsd)) {
- CDEBUG(D_INFO, "read last_rcvd header, uuid = %s, "
- "subdir count = %d\n", lsd.lsd_uuid,
- lsd.lsd_subdir_count);
+ struct lr_server_data lsd;
+ struct dentry *dlast;
+ loff_t off;
+ int rc = 0;
+ int count = OBJ_SUBDIR_COUNT;
+
+ ENTRY;
+
+ dlast = osd_lookup_one_len_unlocked(osd, LAST_RCVD, osd_sb(osd)->s_root,
+ strlen(LAST_RCVD));
+ if (IS_ERR(dlast))
+ return PTR_ERR(dlast);
+ else if (dlast->d_inode == NULL)
+ goto out;
+
+ off = 0;
+ rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
+ if (rc == sizeof(lsd)) {
+ CDEBUG(D_INFO,
+ "read last_rcvd header, uuid = %s, subdir count = %d\n",
+ lsd.lsd_uuid, lsd.lsd_subdir_count);
if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
count = le16_to_cpu(lsd.lsd_subdir_count);
} else if (rc != 0) {
return count;
}
-static const char remote_parent_dir[] = "REMOTE_PARENT_DIR";
static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
{
- struct lvfs_run_ctxt new;
- struct lvfs_run_ctxt save;
- struct dentry *parent;
- struct osd_mdobj_map *omm;
- struct dentry *d;
- struct osd_thread_info *info = osd_oti_get(env);
- struct lu_fid *fid = &info->oti_fid3;
- int rc = 0;
+ struct lvfs_run_ctxt new;
+ struct lvfs_run_ctxt save;
+ struct dentry *parent;
+ struct osd_mdobj_map *omm;
+ struct dentry *d;
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct lu_fid *fid = &info->oti_fid3;
+ int rc = 0;
+
ENTRY;
OBD_ALLOC_PTR(dev->od_mdt_map);
parent = osd_sb(dev)->s_root;
osd_push_ctxt(dev, &new, &save);
- d = simple_mkdir(parent, dev->od_mnt, remote_parent_dir,
- 0755, 1);
+ lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
+ d = simple_mkdir(env, dev, parent, fid, REMOTE_PARENT_DIR,
+ LMAC_NOT_IN_OI, 0755, NULL);
if (IS_ERR(d))
GOTO(cleanup, rc = PTR_ERR(d));
omm->omm_remote_parent = d;
- /* Set LMA for remote parent inode */
- lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
- rc = osd_ea_fid_set(info, d->d_inode, fid, LMAC_NOT_IN_OI, 0);
-
- GOTO(cleanup, rc);
+ GOTO(cleanup, rc = 0);
cleanup:
pop_ctxt(&save, &new);
int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
struct osd_object *obj, struct osd_thandle *oh)
{
- struct osd_mdobj_map *omm = osd->od_mdt_map;
- struct osd_thread_info *oti = osd_oti_get(env);
- struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
- char *name = oti->oti_name;
- struct osd_thread_info *info = osd_oti_get(env);
- struct dentry *dentry;
- struct dentry *parent;
- int rc;
-
- /* Set REMOTE_PARENT in lma, so other process like unlink or lfsck
- * can identify this object quickly */
- rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
- if (rc != 0)
+ struct osd_mdobj_map *omm = osd->od_mdt_map;
+ struct osd_thread_info *oti = osd_oti_get(env);
+ struct lustre_mdt_attrs *lma = &oti->oti_ost_attrs.loa_lma;
+ char *name = oti->oti_name;
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct dentry *dentry;
+ struct dentry *parent;
+ int rc;
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AGENTENT))
+ RETURN(0);
+
+ /*
+ * Set REMOTE_PARENT in lma, so other process like unlink or lfsck
+ * can identify this object quickly
+ */
+ rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry,
+ &oti->oti_ost_attrs);
+ if (rc)
RETURN(rc);
lma->lma_incompat |= LMAI_REMOTE_PARENT;
lustre_lma_swab(lma);
rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
sizeof(*lma), XATTR_REPLACE);
- if (rc != 0)
+ if (rc)
RETURN(rc);
parent = omm->omm_remote_parent;
sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
dentry = osd_child_dentry_by_inode(env, parent->d_inode,
name, strlen(name));
- mutex_lock(&parent->d_inode->i_mutex);
+ inode_lock(parent->d_inode);
rc = osd_ldiskfs_add_entry(info, osd, oh->ot_handle, dentry,
obj->oo_inode, NULL);
- CDEBUG(D_INODE, "%s: add %s:%lu to remote parent %lu.\n", osd_name(osd),
- name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
- ldiskfs_inc_count(oh->ot_handle, parent->d_inode);
+ if (!rc && S_ISDIR(obj->oo_inode->i_mode))
+ ldiskfs_inc_count(oh->ot_handle, parent->d_inode);
+ else if (unlikely(rc == -EEXIST))
+ rc = 0;
+ if (!rc)
+ lu_object_set_agent_entry(&obj->oo_dt.do_lu);
+ CDEBUG(D_INODE, "%s: create agent entry for %s: rc = %d\n",
+ osd_name(osd), name, rc);
mark_inode_dirty(parent->d_inode);
- mutex_unlock(&parent->d_inode->i_mutex);
+ inode_unlock(parent->d_inode);
RETURN(rc);
}
int osd_delete_from_remote_parent(const struct lu_env *env,
struct osd_device *osd,
struct osd_object *obj,
- struct osd_thandle *oh)
+ struct osd_thandle *oh, bool destroy)
{
- struct osd_mdobj_map *omm = osd->od_mdt_map;
- struct osd_thread_info *oti = osd_oti_get(env);
- struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
- char *name = oti->oti_name;
- struct dentry *dentry;
- struct dentry *parent;
+ struct osd_mdobj_map *omm = osd->od_mdt_map;
+ struct osd_thread_info *oti = osd_oti_get(env);
+ struct lustre_mdt_attrs *lma = &oti->oti_ost_attrs.loa_lma;
+ char *name = oti->oti_name;
+ struct dentry *dentry;
+ struct dentry *parent;
struct ldiskfs_dir_entry_2 *de;
- struct buffer_head *bh;
- int rc;
-
- /* Check lma to see whether it is remote object */
- rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
- if (rc != 0)
- RETURN(rc);
-
- if (likely(!(lma->lma_incompat & LMAI_REMOTE_PARENT)))
- RETURN(0);
+ struct buffer_head *bh;
+ int rc;
parent = omm->omm_remote_parent;
sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
dentry = osd_child_dentry_by_inode(env, parent->d_inode,
name, strlen(name));
- mutex_lock(&parent->d_inode->i_mutex);
+ inode_lock(parent->d_inode);
bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
NULL, NULL);
if (IS_ERR(bh)) {
- mutex_unlock(&parent->d_inode->i_mutex);
- RETURN(PTR_ERR(bh));
+ inode_unlock(parent->d_inode);
+ rc = PTR_ERR(bh);
+ if (unlikely(rc == -ENOENT))
+ rc = 0;
+ } else {
+ rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode,
+ de, bh);
+ if (!rc && S_ISDIR(obj->oo_inode->i_mode))
+ ldiskfs_dec_count(oh->ot_handle, parent->d_inode);
+ mark_inode_dirty(parent->d_inode);
+ inode_unlock(parent->d_inode);
+ brelse(bh);
+ CDEBUG(D_INODE, "%s: remove agent entry for %s: rc = %d\n",
+ osd_name(osd), name, rc);
}
- CDEBUG(D_INODE, "%s: el %s:%lu to remote parent %lu.\n", osd_name(osd),
- name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
- rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode, de, bh);
- ldiskfs_dec_count(oh->ot_handle, parent->d_inode);
- mark_inode_dirty(parent->d_inode);
- mutex_unlock(&parent->d_inode->i_mutex);
- brelse(bh);
+
+ if (destroy || rc) {
+ if (!rc)
+ lu_object_clear_agent_entry(&obj->oo_dt.do_lu);
+
+ RETURN(rc);
+ }
+
+ rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry,
+ &oti->oti_ost_attrs);
+ if (rc)
+ RETURN(rc);
/* Get rid of REMOTE_PARENT flag from incompat */
lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
lustre_lma_swab(lma);
rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
sizeof(*lma), XATTR_REPLACE);
+ if (!rc)
+ lu_object_clear_agent_entry(&obj->oo_dt.do_lu);
RETURN(rc);
}
const struct lu_fid *fid,
struct osd_inode_id *id)
{
- struct osd_mdobj_map *omm = osd->od_mdt_map;
- char *name = oti->oti_name;
- struct dentry *parent;
- struct dentry *dentry;
+ struct osd_mdobj_map *omm = osd->od_mdt_map;
+ char *name = oti->oti_name;
+ struct dentry *parent;
+ struct dentry *dentry;
struct ldiskfs_dir_entry_2 *de;
- struct buffer_head *bh;
- int rc;
+ struct buffer_head *bh;
+ int rc;
+
ENTRY;
+ if (unlikely(osd->od_is_ost))
+ RETURN(-ENOENT);
+
parent = omm->omm_remote_parent;
sprintf(name, DFID_NOBRACE, PFID(fid));
dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
name, strlen(name));
- mutex_lock(&parent->d_inode->i_mutex);
+ inode_lock(parent->d_inode);
bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
NULL, NULL);
if (IS_ERR(bh)) {
rc = 0;
}
}
- mutex_unlock(&parent->d_inode->i_mutex);
+ inode_unlock(parent->d_inode);
if (rc == 0)
osd_add_oi_cache(oti, osd, id, fid);
RETURN(rc);
*/
static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
{
- struct lvfs_run_ctxt new;
- struct lvfs_run_ctxt save;
- struct dentry *rootd = osd_sb(dev)->s_root;
- struct dentry *d;
- struct osd_thread_info *info = osd_oti_get(env);
- struct inode *inode;
- struct lu_fid *fid = &info->oti_fid3;
- int rc;
+ struct lvfs_run_ctxt new;
+ struct lvfs_run_ctxt save;
+ struct dentry *d;
+ int rc;
+ bool created = false;
+
ENTRY;
OBD_ALLOC_PTR(dev->od_ost_map);
GOTO(cleanup_alloc, rc);
dev->od_ost_map->om_subdir_count = rc;
- rc = 0;
-
INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
rwlock_init(&dev->od_ost_map->om_seq_list_lock);
mutex_init(&dev->od_ost_map->om_dir_init_mutex);
- osd_push_ctxt(dev, &new, &save);
-
- d = ll_lookup_one_len("O", rootd, strlen("O"));
+ osd_push_ctxt(dev, &new, &save);
+ d = simple_mkdir(env, dev, osd_sb(dev)->s_root, NULL, "O",
+ LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0755, &created);
if (IS_ERR(d))
GOTO(cleanup_ctxt, rc = PTR_ERR(d));
- if (d->d_inode == NULL) {
- dput(d);
- /* The lookup() may be called again inside simple_mkdir().
- * Since the repeated lookup() only be called for "/O" at
- * mount time, it will not affect the whole performance. */
- d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
- if (IS_ERR(d))
- GOTO(cleanup_ctxt, rc = PTR_ERR(d));
+ if (created)
/* It is quite probably that the device is new formatted. */
dev->od_maybe_new = 1;
- }
- inode = d->d_inode;
dev->od_ost_map->om_root = d;
- /* 'What the @fid is' is not imporatant, because the object
- * has no OI mapping, and only is visible inside the OSD.*/
- lu_igif_build(fid, inode->i_ino, inode->i_generation);
- rc = osd_ea_fid_set(info, inode, fid,
- LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
- if (rc)
- GOTO(cleanup_dentry, rc);
-
pop_ctxt(&save, &new);
RETURN(0);
-cleanup_dentry:
- dput(d);
cleanup_ctxt:
pop_ctxt(&save, &new);
cleanup_alloc:
for (j = 0; j < osd_seq->oos_subdir_count; j++) {
if (osd_seq->oos_dirs[j])
dput(osd_seq->oos_dirs[j]);
- }
- OBD_FREE(osd_seq->oos_dirs,
- sizeof(struct dentry *) * osd_seq->oos_subdir_count);
- }
+ }
+ OBD_FREE_PTR_ARRAY(osd_seq->oos_dirs,
+ osd_seq->oos_subdir_count);
+ }
if (osd_seq->oos_root)
dput(osd_seq->oos_root);
static void osd_ost_fini(struct osd_device *osd)
{
- struct osd_obj_seq *osd_seq;
- struct osd_obj_seq *tmp;
- struct osd_obj_map *map = osd->od_ost_map;
+ struct osd_obj_seq *osd_seq;
+ struct osd_obj_seq *tmp;
+ struct osd_obj_map *map = osd->od_ost_map;
+
ENTRY;
if (map == NULL)
EXIT;
}
+static int osd_index_backup_dir_init(const struct lu_env *env,
+ struct osd_device *dev)
+{
+ struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
+ struct lvfs_run_ctxt new;
+ struct lvfs_run_ctxt save;
+ struct dentry *dentry;
+ int rc = 0;
+
+ ENTRY;
+
+ lu_local_obj_fid(fid, INDEX_BACKUP_OID);
+ osd_push_ctxt(dev, &new, &save);
+ dentry = simple_mkdir(env, dev, osd_sb(dev)->s_root, fid,
+ INDEX_BACKUP_DIR, LMAC_NOT_IN_OI, 0755, NULL);
+ if (IS_ERR(dentry)) {
+ rc = PTR_ERR(dentry);
+ } else {
+ dev->od_index_backup_inode = igrab(dentry->d_inode);
+ dput(dentry);
+ }
+ pop_ctxt(&save, &new);
+
+ RETURN(rc);
+}
+
+static void osd_index_backup_dir_fini(struct osd_device *dev)
+{
+ iput(dev->od_index_backup_inode);
+ dev->od_index_backup_inode = NULL;
+}
+
int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
{
int rc;
+ bool mdt_init = false;
+
ENTRY;
- /* prepare structures for OST */
rc = osd_ost_init(env, dev);
if (rc)
RETURN(rc);
- /* prepare structures for MDS */
- rc = osd_mdt_init(env, dev);
- if (rc)
+ if (!dev->od_is_ost) {
+ rc = osd_mdt_init(env, dev);
+ if (rc) {
+ osd_ost_fini(dev);
+ RETURN(rc);
+ }
+
+ mdt_init = true;
+ }
+
+ rc = osd_index_backup_dir_init(env, dev);
+ if (rc) {
osd_ost_fini(dev);
+ if (mdt_init)
+ osd_mdt_fini(dev);
+ }
- RETURN(rc);
+ RETURN(rc);
}
static struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, u64 seq)
void osd_obj_map_fini(struct osd_device *dev)
{
+ osd_index_backup_dir_fini(dev);
osd_ost_fini(dev);
osd_mdt_fini(dev);
}
const struct osd_inode_id *id,
handle_t *th)
{
- struct inode *parent = dir->d_inode;
- struct dentry *child;
+ struct inode *parent = dir->d_inode;
+ struct dentry *child;
struct ldiskfs_dir_entry_2 *de;
- struct buffer_head *bh;
- struct inode *inode;
- struct dentry *dentry = &info->oti_obj_dentry;
- struct osd_inode_id *oi_id = &info->oti_id3;
- struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
- struct lu_fid *oi_fid = &lma->lma_self_fid;
- int rc;
+ struct buffer_head *bh;
+ struct inode *inode;
+ struct dentry *dentry = &info->oti_obj_dentry;
+ struct osd_inode_id *oi_id = &info->oti_id3;
+ struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
+ struct lu_fid *oi_fid = &lma->lma_self_fid;
+ int rc;
+
ENTRY;
LASSERT(th != NULL);
child->d_name.name = name;
child->d_name.len = strlen(name);
- ll_vfs_dq_init(parent);
- mutex_lock(&parent->i_mutex);
+ dquot_initialize(parent);
+ inode_lock(parent);
bh = osd_ldiskfs_find_entry(parent, &child->d_name, &de, NULL, NULL);
if (IS_ERR(bh))
GOTO(out, rc = PTR_ERR(bh));
GOTO(out, rc);
}
- /* The EA inode should NOT be in OI, old OI scrub may added
- * such OI mapping by wrong, replace it. */
+ /*
+ * The EA inode should NOT be in OI, old OI scrub may added
+ * such OI mapping by wrong, replace it.
+ */
if (unlikely(osd_is_ea_inode(inode))) {
iput(inode);
goto update;
}
- rc = osd_get_lma(info, inode, dentry, lma);
+ rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
if (rc == -ENODATA) {
rc = osd_get_idif(info, inode, dentry, oi_fid);
if (rc > 0 || rc == -ENODATA) {
if (rc != 0)
GOTO(out, rc);
- /* If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
- * either a crashed object or a uninitialized one. Replace it. */
+ /*
+ * If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
+ * either a crashed object or a uninitialized one. Replace it.
+ */
if (oi_fid != NULL && lu_fid_eq(fid, oi_fid)) {
CERROR("%s: the FID "DFID" is used by two objects: "
"%u/%u %u/%u\n", osd_name(osd), PFID(fid),
}
update:
- /* There may be temporary inconsistency: On one hand, the new
+ /*
+ * There may be temporary inconsistency: On one hand, the new
* object may be referenced by multiple entries, which is out
* of our control unless we traverse the whole /O completely,
* which is non-flat order and inefficient, should be avoided;
* On the other hand, the old object may become orphan if it
* is still valid. Since it was referenced by an invalid entry,
* making it as invisible temporary may be not worse. OI scrub
- * will process it later. */
+ * will process it later.
+ */
rc = ldiskfs_journal_get_write_access(th, bh);
if (rc != 0)
GOTO(out, rc);
out:
if (!IS_ERR(bh))
brelse(bh);
- mutex_unlock(&parent->i_mutex);
+ inode_unlock(parent);
return rc;
}
-static int osd_obj_del_entry(struct osd_thread_info *info,
- struct osd_device *osd,
- struct dentry *dird, char *name,
- handle_t *th)
+int osd_obj_del_entry(struct osd_thread_info *info, struct osd_device *osd,
+ struct dentry *dird, char *name, int namelen,
+ handle_t *th)
{
struct ldiskfs_dir_entry_2 *de;
- struct buffer_head *bh;
- struct dentry *child;
- struct inode *dir = dird->d_inode;
- int rc;
+ struct buffer_head *bh;
+ struct dentry *child;
+ struct inode *dir = dird->d_inode;
+ int rc;
+
ENTRY;
LASSERT(th != NULL);
LASSERT(th->h_transaction != NULL);
-
child = &info->oti_child_dentry;
child->d_name.hash = 0;
child->d_name.name = name;
- child->d_name.len = strlen(name);
+ child->d_name.len = namelen;
child->d_parent = dird;
child->d_inode = NULL;
- ll_vfs_dq_init(dir);
- mutex_lock(&dir->i_mutex);
+ dquot_initialize(dir);
+ inode_lock(dir);
bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
if (IS_ERR(bh)) {
rc = PTR_ERR(bh);
rc = ldiskfs_delete_entry(th, dir, de, bh);
brelse(bh);
}
- mutex_unlock(&dir->i_mutex);
+ inode_unlock(dir);
RETURN(rc);
}
inode = info->oti_inode;
if (unlikely(inode == NULL)) {
struct ldiskfs_inode_info *lii;
+
OBD_ALLOC_PTR(lii);
if (lii == NULL)
RETURN(-ENOMEM);
if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
inode->i_ino++;
- ll_vfs_dq_init(dir->d_inode);
- mutex_lock(&dir->d_inode->i_mutex);
+ dquot_initialize(dir->d_inode);
+ inode_lock(dir->d_inode);
rc = osd_ldiskfs_add_entry(info, osd, th, child, inode, NULL);
- mutex_unlock(&dir->d_inode->i_mutex);
+ inode_unlock(dir->d_inode);
RETURN(rc);
}
fid_seq_is_idif(seq) ? 0 : seq);
}
-static inline void osd_oid_name(char *name, size_t name_size,
- const struct lu_fid *fid, u64 id)
-{
- snprintf(name, name_size,
- (fid_seq_is_rsvd(fid_seq(fid)) ||
- fid_seq_is_mdt0(fid_seq(fid)) ||
- fid_seq_is_idif(fid_seq(fid))) ? "%llu" : "%llx", id);
-}
-
/* external locking is required */
static int osd_seq_load_locked(struct osd_thread_info *info,
struct osd_device *osd,
struct osd_obj_seq *osd_seq)
{
- struct osd_obj_map *map = osd->od_ost_map;
- struct dentry *seq_dir;
- struct inode *inode;
- struct lu_fid *fid = &info->oti_fid3;
- int rc = 0;
- int i;
- char dir_name[32];
+ struct osd_obj_map *map = osd->od_ost_map;
+ struct dentry *seq_dir;
+ int rc = 0;
+ int i;
+ char dir_name[32];
+
ENTRY;
if (osd_seq->oos_root != NULL)
osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
- seq_dir = simple_mkdir(map->om_root, osd->od_mnt, dir_name, 0755, 1);
+ seq_dir = simple_mkdir(info->oti_env, osd, map->om_root, NULL, dir_name,
+ LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0755, NULL);
if (IS_ERR(seq_dir))
GOTO(out_err, rc = PTR_ERR(seq_dir));
else if (seq_dir->d_inode == NULL)
GOTO(out_put, rc = -EFAULT);
- inode = seq_dir->d_inode;
osd_seq->oos_root = seq_dir;
- /* 'What the @fid is' is not imporatant, because the object
- * has no OI mapping, and only is visible inside the OSD.*/
- lu_igif_build(fid, inode->i_ino, inode->i_generation);
- rc = osd_ea_fid_set(info, inode, fid,
- LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
- if (rc != 0)
- GOTO(out_put, rc);
-
LASSERT(osd_seq->oos_dirs == NULL);
- OBD_ALLOC(osd_seq->oos_dirs,
- sizeof(seq_dir) * osd_seq->oos_subdir_count);
+ OBD_ALLOC_PTR_ARRAY(osd_seq->oos_dirs, osd_seq->oos_subdir_count);
if (osd_seq->oos_dirs == NULL)
GOTO(out_put, rc = -ENOMEM);
struct dentry *dir;
snprintf(dir_name, sizeof(dir_name), "d%u", i);
- dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, dir_name,
- 0700, 1);
+ dir = simple_mkdir(info->oti_env, osd, osd_seq->oos_root, NULL,
+ dir_name, LMAC_NOT_IN_OI | LMAC_FID_ON_OST,
+ 0700, NULL);
if (IS_ERR(dir)) {
GOTO(out_free, rc = PTR_ERR(dir));
} else if (dir->d_inode == NULL) {
GOTO(out_free, rc = -EFAULT);
}
- inode = dir->d_inode;
osd_seq->oos_dirs[i] = dir;
-
- /* 'What the @fid is' is not imporatant, because the object
- * has no OI mapping, and only is visible inside the OSD.*/
- lu_igif_build(fid, inode->i_ino, inode->i_generation);
- rc = osd_ea_fid_set(info, inode, fid,
- LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
- if (rc != 0)
- GOTO(out_free, rc);
}
if (rc != 0) {
if (osd_seq->oos_dirs[i] != NULL)
dput(osd_seq->oos_dirs[i]);
}
- OBD_FREE(osd_seq->oos_dirs,
- sizeof(seq_dir) * osd_seq->oos_subdir_count);
+ OBD_FREE_PTR_ARRAY(osd_seq->oos_dirs,
+ osd_seq->oos_subdir_count);
out_put:
dput(seq_dir);
osd_seq->oos_root = NULL;
RETURN(rc);
}
-static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
- struct osd_device *osd, u64 seq)
+struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
+ struct osd_device *osd, u64 seq)
{
- struct osd_obj_map *map;
- struct osd_obj_seq *osd_seq;
- int rc = 0;
+ struct osd_obj_map *map;
+ struct osd_obj_seq *osd_seq;
+ int rc = 0;
+
ENTRY;
map = osd->od_ost_map;
INIT_LIST_HEAD(&osd_seq->oos_seq_list);
osd_seq->oos_seq = seq;
- /* Init subdir count to be 32, but each seq can have
- * different subdir count */
+ /*
+ * Init subdir count to be 32, but each seq can have
+ * different subdir count
+ */
osd_seq->oos_subdir_count = map->om_subdir_count;
rc = osd_seq_load_locked(info, osd, osd_seq);
if (rc != 0)
int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
const struct lu_fid *fid, struct osd_inode_id *id)
{
- struct osd_obj_map *map;
- struct osd_obj_seq *osd_seq;
- struct dentry *d_seq;
- struct dentry *child;
- struct ost_id *ostid = &info->oti_ostid;
- int dirn;
- char name[32];
- struct ldiskfs_dir_entry_2 *de;
- struct buffer_head *bh;
- struct inode *dir;
- struct inode *inode;
- ENTRY;
-
- /* on the very first lookup we find and open directories */
-
- map = dev->od_ost_map;
- LASSERT(map);
+ struct osd_obj_map *map;
+ struct osd_obj_seq *osd_seq;
+ struct dentry *d_seq;
+ struct dentry *child;
+ struct ost_id *ostid = &info->oti_ostid;
+ int dirn;
+ char name[32];
+ struct ldiskfs_dir_entry_2 *de;
+ struct buffer_head *bh;
+ struct inode *dir;
+ struct inode *inode;
+
+ ENTRY;
+
+ /* on the very first lookup we find and open directories */
+ map = dev->od_ost_map;
+ LASSERT(map);
LASSERT(map->om_root);
- fid_to_ostid(fid, ostid);
+ fid_to_ostid(fid, ostid);
osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
if (IS_ERR(osd_seq))
RETURN(PTR_ERR(osd_seq));
child->d_name.len = strlen(name);
dir = d_seq->d_inode;
- mutex_lock(&dir->i_mutex);
+ inode_lock(dir);
bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
- mutex_unlock(&dir->i_mutex);
+ inode_unlock(dir);
if (IS_ERR(bh))
RETURN(PTR_ERR(bh));
const struct osd_inode_id *id,
handle_t *th)
{
- struct osd_obj_map *map;
- struct osd_obj_seq *osd_seq;
- struct dentry *d;
- struct ost_id *ostid = &info->oti_ostid;
- u64 oid;
- int dirn, rc = 0;
- char name[32];
+ struct osd_obj_map *map;
+ struct osd_obj_seq *osd_seq;
+ struct dentry *d;
+ struct ost_id *ostid = &info->oti_ostid;
+ u64 oid;
+ int dirn, rc = 0;
+ char name[32];
+
ENTRY;
map = osd->od_ost_map;
int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
const struct lu_fid *fid, handle_t *th)
{
- struct osd_obj_map *map;
- struct osd_obj_seq *osd_seq;
- struct dentry *d;
- struct ost_id *ostid = &info->oti_ostid;
- int dirn, rc = 0;
- char name[32];
- ENTRY;
+ struct osd_obj_map *map;
+ struct osd_obj_seq *osd_seq;
+ struct dentry *d;
+ struct ost_id *ostid = &info->oti_ostid;
+ int dirn, rc = 0;
+ char name[32];
- map = osd->od_ost_map;
- LASSERT(map);
+ ENTRY;
+
+ map = osd->od_ost_map;
+ LASSERT(map);
/* map fid to seq:objid */
- fid_to_ostid(fid, ostid);
+ fid_to_ostid(fid, ostid);
osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
if (IS_ERR(osd_seq))
LASSERT(d);
osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
- rc = osd_obj_del_entry(info, osd, d, name, th);
+ rc = osd_obj_del_entry(info, osd, d, name, strlen(name), th);
cleanup:
- RETURN(rc);
+ RETURN(rc);
}
int osd_obj_map_update(struct osd_thread_info *info,
const struct osd_inode_id *id,
handle_t *th)
{
- struct osd_obj_seq *osd_seq;
- struct dentry *d;
- struct ost_id *ostid = &info->oti_ostid;
- int dirn, rc = 0;
- char name[32];
+ struct osd_obj_seq *osd_seq;
+ struct dentry *d;
+ struct ost_id *ostid = &info->oti_ostid;
+ int dirn, rc = 0;
+ char name[32];
+
ENTRY;
fid_to_ostid(fid, ostid);
struct dentry *src_child,
const struct lu_fid *fid)
{
- struct osd_obj_seq *osd_seq;
- struct dentry *tgt_parent;
- struct dentry *tgt_child = &info->oti_child_dentry;
- struct inode *dir;
- struct inode *inode = src_child->d_inode;
- struct ost_id *ostid = &info->oti_ostid;
- handle_t *jh;
+ struct osd_obj_seq *osd_seq;
+ struct dentry *tgt_parent;
+ struct dentry *tgt_child = &info->oti_child_dentry;
+ struct inode *dir;
+ struct inode *inode = src_child->d_inode;
+ struct ost_id *ostid = &info->oti_ostid;
+ handle_t *jh;
struct ldiskfs_dir_entry_2 *de;
- struct buffer_head *bh;
- char name[32];
- int dirn;
- int rc = 0;
+ struct buffer_head *bh;
+ char name[32];
+ int dirn;
+ int rc = 0;
+
ENTRY;
if (fid_is_last_id(fid)) {
if (IS_ERR(jh))
RETURN(PTR_ERR(jh));
- ll_vfs_dq_init(src_parent);
- ll_vfs_dq_init(dir);
+ dquot_initialize(src_parent);
+ dquot_initialize(dir);
- mutex_lock(&src_parent->i_mutex);
- mutex_lock(&dir->i_mutex);
+ inode_lock(dir);
bh = osd_ldiskfs_find_entry(dir, &tgt_child->d_name, &de, NULL, NULL);
if (!IS_ERR(bh)) {
- /* XXX: If some other object occupied the same slot. And If such
- * inode is zero-sized and with SUID+SGID, then means it is
- * a new created one. Maybe we can remove it and insert the
- * original one back to the /O/<seq>/d<x>. But there are
- * something to be considered:
+ /*
+ * XXX: If some other object occupied the same slot. And If such
+ * inode is zero-sized and with SUID+SGID, then means it is
+ * a new created one. Maybe we can remove it and insert the
+ * original one back to the /O/<seq>/d<x>. But there are
+ * something to be considered:
*
- * 1) The OST-object under /lost+found has crashed LMA.
- * So it should not conflict with the current one.
+ * 1) The OST-object under /lost+found has crashed LMA.
+ * So it should not conflict with the current one.
*
* 2) There are race conditions that: someone may just want
* to modify the current one. Even if the OI scrub takes
* has been removed when the RPC service thread waiting
* for the lock.
*
- * So keep it there before we have suitable solution. */
+ * So keep it there before we have suitable solution.
+ */
brelse(bh);
- mutex_unlock(&dir->i_mutex);
- mutex_unlock(&src_parent->i_mutex);
+ inode_unlock(dir);
ldiskfs_journal_stop(jh);
rc = -EEXIST;
/* If the src object has never been modified, then remove it. */
if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
inode->i_mode & S_ISGID) {
- rc = ll_vfs_unlink(src_parent, src_child);
+ rc = vfs_unlink(&init_user_ns, src_parent, src_child);
if (unlikely(rc == -ENOENT))
rc = 0;
}
- RETURN(rc);
+ if (rc)
+ RETURN(rc);
}
bh = osd_ldiskfs_find_entry(src_parent, &src_child->d_name, &de,
GOTO(unlock, rc);
unlock:
- mutex_unlock(&dir->i_mutex);
- mutex_unlock(&src_parent->i_mutex);
+ inode_unlock(dir);
ldiskfs_journal_stop(jh);
return rc;
}
const struct lu_fid *fid, const struct osd_inode_id *id,
handle_t *th)
{
- struct dentry *root;
- char *name = NULL;
- int rc;
+ struct dentry *root;
+ char *name = NULL;
+ int rc;
+
ENTRY;
root = osd_object_spec_find(info, osd, fid, &name);
const struct lu_fid *fid, const struct osd_inode_id *id,
handle_t *th)
{
- struct dentry *root;
- char *name = NULL;
- int rc;
+ struct dentry *root;
+ char *name = NULL;
+ int rc;
+
ENTRY;
root = osd_object_spec_find(info, osd, fid, &name);
}
int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
- const struct lu_fid *fid, struct osd_inode_id *id)
+ const struct lu_fid *fid, struct osd_inode_id *id,
+ enum oi_check_flags flags)
{
- struct dentry *root;
- struct dentry *dentry;
- struct inode *inode;
- char *name = NULL;
- int rc = -ENOENT;
+ struct dentry *root;
+ struct dentry *dentry;
+ struct inode *inode;
+ char *name = NULL;
+ int rc = -ENOENT;
+
ENTRY;
if (fid_is_last_id(fid)) {
RETURN(-ENOENT);
}
- dentry = ll_lookup_one_len(name, root, strlen(name));
+ dentry = osd_lookup_one_len_common(osd, name, root, strlen(name),
+ flags);
if (!IS_ERR(dentry)) {
inode = dentry->d_inode;
if (inode) {
rc = 0;
}
}
- /* if dentry is accessible after osd_compat_spec_insert it
- * will still contain NULL inode, so don't keep it in cache */
+ /*
+ * if dentry is accessible after osd_compat_spec_insert it
+ * will still contain NULL inode, so don't keep it in cache
+ */
d_invalidate(dentry);
dput(dentry);
}
RETURN(rc);
}
+
+#ifndef HAVE_BIO_INTEGRITY_ENABLED
+bool bio_integrity_enabled(struct bio *bio)
+{
+ struct blk_integrity *bi = blk_get_integrity(bio_get_disk(bio));
+
+ if (bio_op(bio) != REQ_OP_READ && bio_op(bio) != REQ_OP_WRITE)
+ return false;
+
+ if (!bio_sectors(bio))
+ return false;
+
+ /* Already protected? */
+ if (bio_integrity(bio))
+ return false;
+
+ if (bi == NULL)
+ return false;
+
+ if (bio_data_dir(bio) == READ && bi->profile->verify_fn != NULL &&
+ (bi->flags & BLK_INTEGRITY_VERIFY))
+ return true;
+
+ if (bio_data_dir(bio) == WRITE && bi->profile->generate_fn != NULL &&
+ (bi->flags & BLK_INTEGRITY_GENERATE))
+ return true;
+
+ return false;
+}
+#endif