* Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
*/
-#define DEBUG_SUBSYSTEM S_MDS
+#define DEBUG_SUBSYSTEM S_OSD
#include <linux/module.h>
#include <lustre_quota.h>
#include <ldiskfs/xattr.h>
+#include <lustre_linkea.h>
int ldiskfs_pdo = 1;
CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
LINVRNT(osd_invariant(obj));
LASSERT(obj->oo_inode == NULL);
- LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
+ LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
dev = osd_dev(ldev);
scrub = &dev->od_scrub;
}
id = &info->oti_id;
- if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
+ if (!list_empty(&scrub->os_inconsistent_items)) {
/* Search order: 2. OI scrub pending list. */
result = osd_oii_lookup(dev, fid, id);
if (result == 0)
{
struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
- return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
+ return oh->ot_credits > osd_transaction_size(dev);
}
/*
dt_txn_hook_commit(th);
/* call per-transaction callbacks if any */
- cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
+ list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
"commit callback entry: magic=%x name='%s'\n",
dcb->dcb_magic, dcb->dcb_name);
- cfs_list_del_init(&dcb->dcb_linkage);
+ list_del_init(&dcb->dcb_linkage);
dcb->dcb_func(NULL, th, dcb, error);
}
atomic_set(&th->th_refc, 1);
th->th_alloc_size = sizeof(*oh);
oti->oti_dev = osd_dt_dev(d);
- CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
+ INIT_LIST_HEAD(&oh->ot_dcb_list);
osd_th_alloced(oh);
memset(oti->oti_declare_ops, 0,
*
* This should be removed when we can calculate the
* credits precisely. */
- oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
+ oh->ot_credits = osd_transaction_size(dev);
}
/*
int rc;
ENTRY;
- if (ss == NULL)
- RETURN(1);
+ LASSERT(ss != NULL);
+ LASSERT(ss->ss_server_fld != NULL);
rc = osd_fld_lookup(env, osd, seq, range);
if (rc != 0) {
*/
wait_event(iobuf->dr_wait,
atomic_read(&iobuf->dr_numreqs) == 0);
+ osd_fini_iobuf(oti->oti_dev, iobuf);
if (!rc)
rc = iobuf->dr_error;
LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
LASSERT(&dcb->dcb_func != NULL);
- cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
+ list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
return 0;
}
osd_index_fini(obj);
if (inode != NULL) {
struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
- qid_t uid = inode->i_uid;
- qid_t gid = inode->i_gid;
+ qid_t uid = i_uid_read(inode);
+ qid_t gid = i_gid_read(inode);
iput(inode);
obj->oo_inode = NULL;
static void osd_inode_getattr(const struct lu_env *env,
- struct inode *inode, struct lu_attr *attr)
-{
- attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
- LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
- LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
- LA_TYPE;
-
- attr->la_atime = LTIME_S(inode->i_atime);
- attr->la_mtime = LTIME_S(inode->i_mtime);
- attr->la_ctime = LTIME_S(inode->i_ctime);
- attr->la_mode = inode->i_mode;
- attr->la_size = i_size_read(inode);
- attr->la_blocks = inode->i_blocks;
- attr->la_uid = inode->i_uid;
- attr->la_gid = inode->i_gid;
- attr->la_flags = LDISKFS_I(inode)->i_flags;
- attr->la_nlink = inode->i_nlink;
- attr->la_rdev = inode->i_rdev;
- attr->la_blksize = 1 << inode->i_blkbits;
- attr->la_blkbits = inode->i_blkbits;
+ struct inode *inode, struct lu_attr *attr)
+{
+ attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
+ LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
+ LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
+ LA_TYPE;
+
+ attr->la_atime = LTIME_S(inode->i_atime);
+ attr->la_mtime = LTIME_S(inode->i_mtime);
+ attr->la_ctime = LTIME_S(inode->i_ctime);
+ attr->la_mode = inode->i_mode;
+ attr->la_size = i_size_read(inode);
+ attr->la_blocks = inode->i_blocks;
+ attr->la_uid = i_uid_read(inode);
+ attr->la_gid = i_gid_read(inode);
+ attr->la_flags = LDISKFS_I(inode)->i_flags;
+ attr->la_nlink = inode->i_nlink;
+ attr->la_rdev = inode->i_rdev;
+ attr->la_blksize = 1 << inode->i_blkbits;
+ attr->la_blkbits = inode->i_blkbits;
}
static int osd_attr_get(const struct lu_env *env,
struct osd_object *obj;
struct osd_thread_info *info = osd_oti_get(env);
struct lquota_id_info *qi = &info->oti_qi;
+ qid_t uid;
+ qid_t gid;
long long bspace;
int rc = 0;
- bool allocated;
+ bool enforce;
ENTRY;
LASSERT(dt != NULL);
* We still need to call the osd_declare_qid() to calculate the journal
* credits for updating quota accounting files and to trigger quota
* space adjustment once the operation is completed.*/
- if ((attr->la_valid & LA_UID) != 0 &&
- attr->la_uid != obj->oo_inode->i_uid) {
+ if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
+ /* USERQUOTA */
+ uid = i_uid_read(obj->oo_inode);
qi->lqi_type = USRQUOTA;
-
+ enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
/* inode accounting */
qi->lqi_is_blk = false;
- /* one more inode for the new owner ... */
+ /* one more inode for the new uid ... */
qi->lqi_id.qid_uid = attr->la_uid;
qi->lqi_space = 1;
- allocated = (attr->la_uid == 0) ? true : false;
- rc = osd_declare_qid(env, oh, qi, allocated, NULL);
+ /* Reserve credits for the new uid */
+ rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
if (rc == -EDQUOT || rc == -EINPROGRESS)
rc = 0;
if (rc)
RETURN(rc);
/* and one less inode for the current uid */
- qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
+ qi->lqi_id.qid_uid = uid;
qi->lqi_space = -1;
- rc = osd_declare_qid(env, oh, qi, true, NULL);
+ rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
if (rc == -EDQUOT || rc == -EINPROGRESS)
rc = 0;
if (rc)
/* block accounting */
qi->lqi_is_blk = true;
- /* more blocks for the new owner ... */
+ /* more blocks for the new uid ... */
qi->lqi_id.qid_uid = attr->la_uid;
qi->lqi_space = bspace;
- allocated = (attr->la_uid == 0) ? true : false;
- rc = osd_declare_qid(env, oh, qi, allocated, NULL);
+ /*
+ * Credits for the new uid has been reserved, re-use "obj"
+ * to save credit reservation.
+ */
+ rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
if (rc == -EDQUOT || rc == -EINPROGRESS)
rc = 0;
if (rc)
RETURN(rc);
- /* and finally less blocks for the current owner */
- qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
+ /* and finally less blocks for the current uid */
+ qi->lqi_id.qid_uid = uid;
qi->lqi_space = -bspace;
- rc = osd_declare_qid(env, oh, qi, true, NULL);
+ rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
if (rc == -EDQUOT || rc == -EINPROGRESS)
rc = 0;
if (rc)
RETURN(rc);
- }
- if (attr->la_valid & LA_GID &&
- attr->la_gid != obj->oo_inode->i_gid) {
+ /* GROUP QUOTA */
+ gid = i_gid_read(obj->oo_inode);
qi->lqi_type = GRPQUOTA;
+ enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
/* inode accounting */
qi->lqi_is_blk = false;
- /* one more inode for the new group owner ... */
+ /* one more inode for the new gid ... */
qi->lqi_id.qid_gid = attr->la_gid;
qi->lqi_space = 1;
- allocated = (attr->la_gid == 0) ? true : false;
- rc = osd_declare_qid(env, oh, qi, allocated, NULL);
+ rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
if (rc == -EDQUOT || rc == -EINPROGRESS)
rc = 0;
if (rc)
RETURN(rc);
/* and one less inode for the current gid */
- qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
+ qi->lqi_id.qid_gid = gid;
qi->lqi_space = -1;
- rc = osd_declare_qid(env, oh, qi, true, NULL);
+ rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
if (rc == -EDQUOT || rc == -EINPROGRESS)
rc = 0;
if (rc)
/* block accounting */
qi->lqi_is_blk = true;
- /* more blocks for the new owner ... */
+ /* more blocks for the new gid ... */
qi->lqi_id.qid_gid = attr->la_gid;
qi->lqi_space = bspace;
- allocated = (attr->la_gid == 0) ? true : false;
- rc = osd_declare_qid(env, oh, qi, allocated, NULL);
+ rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
if (rc == -EDQUOT || rc == -EINPROGRESS)
rc = 0;
if (rc)
RETURN(rc);
- /* and finally less blocks for the current owner */
- qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
+ /* and finally less blocks for the current gid */
+ qi->lqi_id.qid_gid = gid;
qi->lqi_space = -bspace;
- rc = osd_declare_qid(env, oh, qi, true, NULL);
+ rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
if (rc == -EDQUOT || rc == -EINPROGRESS)
rc = 0;
if (rc)
}
static int osd_inode_setattr(const struct lu_env *env,
- struct inode *inode, const struct lu_attr *attr)
+ struct inode *inode, const struct lu_attr *attr)
{
- __u64 bits;
+ __u64 bits = attr->la_valid;
+
+ /* Only allow set size for regular file */
+ if (!S_ISREG(inode->i_mode))
+ bits &= ~(LA_SIZE | LA_BLOCKS);
- bits = attr->la_valid;
+ if (bits == 0)
+ return 0;
if (bits & LA_ATIME)
inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
if (bits & LA_BLOCKS)
inode->i_blocks = attr->la_blocks;
#endif
- if (bits & LA_MODE)
- inode->i_mode = (inode->i_mode & S_IFMT) |
- (attr->la_mode & ~S_IFMT);
- if (bits & LA_UID)
- inode->i_uid = attr->la_uid;
- if (bits & LA_GID)
- inode->i_gid = attr->la_gid;
- if (bits & LA_NLINK)
+ if (bits & LA_MODE)
+ inode->i_mode = (inode->i_mode & S_IFMT) |
+ (attr->la_mode & ~S_IFMT);
+ if (bits & LA_UID)
+ i_uid_write(inode, attr->la_uid);
+ if (bits & LA_GID)
+ i_gid_write(inode, attr->la_gid);
+ if (bits & LA_NLINK)
set_nlink(inode, attr->la_nlink);
- if (bits & LA_RDEV)
- inode->i_rdev = attr->la_rdev;
+ if (bits & LA_RDEV)
+ inode->i_rdev = attr->la_rdev;
if (bits & LA_FLAGS) {
/* always keep S_NOCMTIME */
static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
{
- if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
- (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
+ if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
+ (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
struct iattr iattr;
int rc;
+ ll_vfs_dq_init(inode);
iattr.ia_valid = 0;
if (attr->la_valid & LA_UID)
iattr.ia_valid |= ATTR_UID;
if (attr->la_valid & LA_GID)
iattr.ia_valid |= ATTR_GID;
- iattr.ia_uid = attr->la_uid;
- iattr.ia_gid = attr->la_gid;
+ iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
+ iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
rc = ll_vfs_dq_transfer(inode, &iattr);
if (rc) {
CERROR("%s: quota transfer failed: rc = %d. Is quota "
- "enforcement enabled on the ldiskfs filesystem?",
- inode->i_sb->s_id, rc);
+ "enforcement enabled on the ldiskfs "
+ "filesystem?\n", inode->i_sb->s_id, rc);
return rc;
}
}
}
inode = obj->oo_inode;
- ll_vfs_dq_init(inode);
rc = osd_quota_transfer(inode, attr);
if (rc)
* should not happen since quota enforcement is no longer
* enabled on ldiskfs (lquota takes care of it).
*/
- LASSERTF(result == 0, "%d", result);
+ LASSERTF(result == 0, "%d\n", result);
ll_dirty_inode(inode, I_DIRTY_DATASYNC);
}
RETURN(0);
rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
- false, false, NULL, false);
+ osd_dt_obj(dt), false, NULL, false);
if (rc != 0)
RETURN(rc);
osd_trans_declare_op(env, oh, OSD_OT_DELETE,
osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
/* one less inode */
- rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
- false, true, NULL, false);
+ rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
+ -1, oh, obj, false, NULL, false);
if (rc)
RETURN(rc);
/* data to be truncated */
- rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
- true, true, NULL, false);
+ rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
+ 0, oh, obj, true, NULL, false);
RETURN(rc);
}
* its inmemory API.
*/
void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
- const struct dt_rec *fid)
+ const struct lu_fid *fid)
{
- if (!fid_is_namespace_visible((const struct lu_fid *)fid) ||
+ if (!fid_is_namespace_visible(fid) ||
OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
param->edp_magic = 0;
return;
static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
struct inode *dir,
- struct inode *parent_dir,
- const struct dt_rec *dot_fid,
- const struct dt_rec *dot_dot_fid,
+ struct inode *parent_dir,
+ const struct lu_fid *dot_fid,
+ const struct lu_fid *dot_dot_fid,
struct osd_thandle *oth)
{
struct ldiskfs_dentry_param *dot_ldp;
struct osd_device *osd,
struct osd_object *pobj,
const struct lu_fid *fid,
+ __u32 type,
struct thandle *th)
{
struct osd_thread_info *info = osd_oti_get(env);
oh = container_of(th, struct osd_thandle, ot_super);
LASSERT(oh->ot_handle->h_transaction != NULL);
- /* FIXME: Insert index api needs to know the mode of
- * the remote object. Just use S_IFDIR for now */
- local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
+ local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
if (IS_ERR(local)) {
CERROR("%s: create local error %d\n", osd_name(osd),
(int)PTR_ERR(local));
RETURN(ERR_PTR(rc));
}
+ if (!S_ISDIR(type))
+ RETURN(local);
+
rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
- (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
- (const struct dt_rec *)fid, oh);
+ lu_object_fid(&pobj->oo_dt.do_lu),
+ fid, oh);
if (rc != 0) {
CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
osd_name(osd), PFID(fid), rc);
* in case they are being linked into the PENDING directory
*/
spin_lock(&obj->oo_guard);
- ldiskfs_inc_count(oh->ot_handle, inode);
- LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
+ if (unlikely(inode->i_nlink == 0))
+ /* inc_nlink from 0 may cause WARN_ON */
+ set_nlink(inode, 1);
+ else {
+ ldiskfs_inc_count(oh->ot_handle, inode);
+ if (!S_ISDIR(inode->i_mode))
+ LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
+ }
spin_unlock(&obj->oo_guard);
ll_dirty_inode(inode, I_DIRTY_DATASYNC);
{
struct osd_thandle *oh;
int credits;
+ struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
LASSERT(handle != NULL);
} else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
credits = 1;
} else {
- struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
- struct super_block *sb = osd_sb(osd);
credits = osd_dto_credits_noquota[DTO_XATTR_SET];
if (buf && buf->lb_len > sb->s_blocksize) {
credits *= (buf->lb_len + sb->s_blocksize - 1) >>
sb->s_blocksize_bits;
}
+ /*
+ * xattr set may involve inode quota change, reserve credits for
+ * dquot_initialize()
+ */
+ oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
}
osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
if (fl & LU_XATTR_CREATE)
fs_flags |= XATTR_CREATE;
+ if (strcmp(name, XATTR_NAME_LMV) == 0) {
+ struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
+ int rc;
+
+ rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
+ if (rc != 0)
+ RETURN(rc);
+
+ lma->lma_incompat |= LMAI_STRIPED;
+ lustre_lma_swab(lma);
+ rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
+ sizeof(*lma), XATTR_REPLACE);
+ if (rc != 0)
+ RETURN(rc);
+ }
+
return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
fs_flags);
}
struct thandle *handle)
{
struct osd_thandle *oh;
+ struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
LASSERT(handle != NULL);
osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
osd_dto_credits_noquota[DTO_XATTR_SET]);
+ /*
+ * xattr del may involve inode quota change, reserve credits for
+ * dquot_initialize()
+ */
+ oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
return 0;
}
case LC_ID_NONE:
RETURN(NULL);
case LC_ID_PLAIN:
- capa->lc_uid = obj->oo_inode->i_uid;
- capa->lc_gid = obj->oo_inode->i_gid;
+ capa->lc_uid = i_uid_read(obj->oo_inode);
+ capa->lc_gid = i_gid_read(obj->oo_inode);
capa->lc_flags = LC_ID_PLAIN;
break;
case LC_ID_CONVERT: {
__u32 d[4], s[4];
- s[0] = obj->oo_inode->i_uid;
+ s[0] = i_uid_read(obj->oo_inode);
cfs_get_random_bytes(&(s[1]), sizeof(__u32));
- s[2] = obj->oo_inode->i_gid;
+ s[2] = i_uid_read(obj->oo_inode);
cfs_get_random_bytes(&(s[3]), sizeof(__u32));
rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
if (unlikely(rc))
* recheck under lock.
*/
if (!osd_has_index(obj))
- result = osd_iam_container_init(env, obj, dir);
+ result = osd_iam_container_init(env, obj,
+ obj->oo_dir);
else
result = 0;
up_write(&obj->oo_ext_idx_sem);
inode = osd_dt_obj(dt)->oo_inode;
LASSERT(inode);
- rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
- true, true, NULL, false);
+ rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
+ 0, oh, osd_dt_obj(dt), true, NULL, false);
RETURN(rc);
}
}
static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
- struct lu_fid *fid)
+ const struct lu_fid *fid)
{
+ struct seq_server_site *ss = osd_seq_site(osd);
ENTRY;
/* FID seqs not in FLDB, must be local seq */
if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
RETURN(0);
+ /* If FLD is not being initialized yet, it only happens during the
+ * initialization, likely during mgs initialization, and we assume
+ * this is local FID. */
+ if (ss == NULL || ss->ss_server_fld == NULL)
+ RETURN(0);
+
+ /* Only check the local FLDB here */
if (osd_seq_exists(env, osd, fid_seq(fid)))
RETURN(0);
* \retval -ve, on error
*/
static int __osd_ea_add_rec(struct osd_thread_info *info,
- struct osd_object *pobj, struct inode *cinode,
- const char *name, const struct dt_rec *fid,
- struct htree_lock *hlock, struct thandle *th)
+ struct osd_object *pobj, struct inode *cinode,
+ const char *name, const struct lu_fid *fid,
+ struct htree_lock *hlock, struct thandle *th)
{
struct ldiskfs_dentry_param *ldp;
struct dentry *child;
* \retval -ve, on error
*/
static int osd_add_dot_dotdot(struct osd_thread_info *info,
- struct osd_object *dir,
- struct inode *parent_dir, const char *name,
- const struct dt_rec *dot_fid,
- const struct dt_rec *dot_dot_fid,
- struct thandle *th)
+ struct osd_object *dir,
+ struct inode *parent_dir, const char *name,
+ const struct lu_fid *dot_fid,
+ const struct lu_fid *dot_dot_fid,
+ struct thandle *th)
{
struct inode *inode = dir->oo_inode;
struct osd_thandle *oth;
}
result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
- parent_dir, dot_fid,
- dot_dot_fid, oth);
+ parent_dir, dot_fid,
+ dot_dot_fid, oth);
if (result == 0)
dir->oo_compat_dotdot_created = 1;
}
* value, return by respective functions.
*/
static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
- struct inode *cinode, const char *name,
- const struct dt_rec *fid, struct thandle *th)
+ struct inode *cinode, const char *name,
+ const struct lu_fid *fid, struct thandle *th)
{
struct osd_thread_info *info = osd_oti_get(env);
struct htree_lock *hlock;
pobj->oo_inode, 0);
} else {
down_write(&pobj->oo_ext_idx_sem);
- }
- rc = osd_add_dot_dotdot(info, pobj, cinode, name,
- (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
+ }
+
+ rc = osd_add_dot_dotdot(info, pobj, cinode, name,
+ lu_object_fid(&pobj->oo_dt.do_lu),
fid, th);
} else {
if (hlock != NULL) {
if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
struct lu_fid *tfid = &info->oti_fid;
- *tfid = *(const struct lu_fid *)fid;
+ *tfid = *fid;
tfid->f_ver = ~0;
rc = __osd_ea_add_rec(info, pobj, cinode, name,
- (const struct dt_rec *)tfid,
- hlock, th);
+ tfid, hlock, th);
} else {
rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
hlock, th);
}
/**
+ * Get parent FID from the linkEA.
+ *
+ * For a directory which parent resides on remote MDT, to satisfy the
+ * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
+ * the other hand, to make the lookup(..) on the directory can return
+ * the real parent FID, we append the real parent FID after its ".."
+ * name entry in the /REMOTE_PARENT_DIR.
+ *
+ * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
+ * backup. So after the restore, we cannot get the right parent FID from
+ * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
+ * we have stored the real parent FID in the directory object's linkEA,
+ * we can parse the linkEA for the real parent FID.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] obj pointer to the object to be handled
+ * \param[out]fid pointer to the buffer to hold the parent FID
+ *
+ * \retval 0 for getting the real parent FID successfully
+ * \retval negative error number on failure
+ */
+static int osd_get_pfid_from_linkea(const struct lu_env *env,
+ struct osd_object *obj,
+ struct lu_fid *fid)
+{
+ struct osd_thread_info *oti = osd_oti_get(env);
+ struct lu_buf *buf = &oti->oti_big_buf;
+ struct dentry *dentry = &oti->oti_obj_dentry;
+ struct inode *inode = obj->oo_inode;
+ struct linkea_data ldata = { 0 };
+ int rc;
+ ENTRY;
+
+ fid_zero(fid);
+ if (!S_ISDIR(inode->i_mode))
+ RETURN(-EIO);
+
+again:
+ rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
+ buf->lb_buf, buf->lb_len);
+ if (rc == -ERANGE) {
+ rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
+ NULL, 0);
+ if (rc > 0) {
+ lu_buf_realloc(buf, rc);
+ if (buf->lb_buf == NULL)
+ RETURN(-ENOMEM);
+
+ goto again;
+ }
+ }
+
+ if (unlikely(rc == 0))
+ RETURN(-ENODATA);
+
+ if (rc < 0)
+ RETURN(rc);
+
+ if (unlikely(buf->lb_buf == NULL)) {
+ lu_buf_realloc(buf, rc);
+ if (buf->lb_buf == NULL)
+ RETURN(-ENOMEM);
+
+ goto again;
+ }
+
+ ldata.ld_buf = buf;
+ rc = linkea_init(&ldata);
+ if (rc == 0) {
+ linkea_first_entry(&ldata);
+ linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
+ }
+
+ RETURN(rc);
+}
+
+/**
* Calls ->lookup() to find dentry. From dentry get inode and
* read inode's ea to get fid. This is required for interoperability
* mode (b11826)
/* done with de, release bh */
brelse(bh);
- if (rc != 0)
- rc = osd_ea_fid_get(env, obj, ino, fid, id);
- else
+ if (rc != 0) {
+ if (unlikely(ino == osd_remote_parent_ino(dev)))
+ /* If the parent is on remote MDT, and there
+ * is no FID-in-dirent, then we have to get
+ * the parent FID from the linkEA. */
+ rc = osd_get_pfid_from_linkea(env, obj, fid);
+ else
+ rc = osd_ea_fid_get(env, obj, ino, fid, id);
+ } else {
osd_id_gen(id, ino, OSD_OII_NOGEN);
+ }
+
if (rc != 0) {
fid_zero(&oic->oic_fid);
GOTO(out, rc);
/* We ignore block quota on meta pool (MDTs), so needn't
* calculate how many blocks will be consumed by this index
* insert */
- rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0,
- oh, true, true, NULL, false);
+ rc = osd_declare_inode_qid(env, i_uid_read(inode),
+ i_gid_read(inode), 0, oh,
+ osd_dt_obj(dt), true, NULL, false);
}
if (fid == NULL)
{
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
- struct lu_fid *fid = (struct lu_fid *) rec;
+ struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
+ const struct lu_fid *fid = rec1->rec_fid;
const char *name = (const char *)key;
struct osd_thread_info *oti = osd_oti_get(env);
struct osd_inode_id *id = &oti->oti_id;
int rc;
ENTRY;
- LASSERT(osd_invariant(obj));
+ LASSERT(osd_invariant(obj));
LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
- LASSERT(th != NULL);
+ LASSERT(th != NULL);
osd_trans_exec_op(env, th, OSD_OT_INSERT);
- if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
- RETURN(-EACCES);
+ if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
+ RETURN(-EACCES);
- LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid));
+ LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
rc = osd_remote_fid(env, osd, fid);
if (rc < 0) {
child_inode = igrab(omm->omm_remote_parent->d_inode);
} else {
child_inode = osd_create_local_agent_inode(env, osd,
- obj, fid,
- th);
+ obj, fid, rec1->rec_type & S_IFMT, th);
if (IS_ERR(child_inode))
RETURN(PTR_ERR(child_inode));
}
child_inode = igrab(child->oo_inode);
}
- rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
+ rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
rc = ldiskfs_journal_get_write_access(jh, bh);
- if (rc != 0) {
- CERROR("%.16s: fail to write access for update dirent: "
- "name = %.*s, rc = %d\n",
- LDISKFS_SB(sb)->s_es->s_volume_name,
- ent->oied_namelen, ent->oied_name, rc);
+ if (rc != 0)
RETURN(rc);
- }
rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
rc = ldiskfs_journal_dirty_metadata(jh, bh);
- if (rc != 0)
- CERROR("%.16s: fail to dirty metadata for update dirent: "
- "name = %.*s, rc = %d\n",
- LDISKFS_SB(sb)->s_es->s_volume_name,
- ent->oied_namelen, ent->oied_name, rc);
RETURN(rc);
}
if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
dir->i_sb->s_blocksize)) {
rc = ldiskfs_journal_get_write_access(jh, bh);
- if (rc != 0) {
- CERROR("%.16s: fail to write access for reinsert "
- "dirent: name = %.*s, rc = %d\n",
- LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
- ent->oied_namelen, ent->oied_name, rc);
+ if (rc != 0)
RETURN(rc);
- }
de->name[de->name_len] = 0;
rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
de->file_type |= LDISKFS_DIRENT_LUFID;
rc = ldiskfs_journal_dirty_metadata(jh, bh);
- if (rc != 0)
- CERROR("%.16s: fail to dirty metadata for reinsert "
- "dirent: name = %.*s, rc = %d\n",
- LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
- ent->oied_namelen, ent->oied_name, rc);
RETURN(rc);
}
rc = ldiskfs_delete_entry(jh, dir, de, bh);
- if (rc != 0) {
- CERROR("%.16s: fail to delete entry for reinsert dirent: "
- "name = %.*s, rc = %d\n",
- LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
- ent->oied_namelen, ent->oied_name, rc);
+ if (rc != 0)
RETURN(rc);
- }
dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
ent->oied_namelen);
ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
- osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid);
+ osd_get_ldiskfs_dirent_param(ldp, fid);
dentry->d_fsdata = (void *)ldp;
ll_vfs_dq_init(dir);
rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
/* It is too bad, we cannot reinsert the name entry back.
* That means we lose it! */
if (rc != 0)
- CERROR("%.16s: fail to insert entry for reinsert dirent: "
- "name = %.*s, rc = %d\n",
+ CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
+ "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
- ent->oied_namelen, ent->oied_name, rc);
+ dir->i_ino, dir->i_generation,
+ ent->oied_namelen, ent->oied_name, PFID(fid), rc);
RETURN(rc);
}
jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
if (IS_ERR(jh)) {
rc = PTR_ERR(jh);
- CERROR("%.16s: fail to start trans for dirent "
- "check_repair: credits %d, name %.*s, rc %d\n",
- devname, credits, ent->oied_namelen,
- ent->oied_name, rc);
+ CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
+ "check_repair, dir = %lu/%u, credits = %d, "
+ "name = %.*s: rc = %d\n",
+ devname, dir->i_ino, dir->i_generation, credits,
+ ent->oied_namelen, ent->oied_name, rc);
RETURN(rc);
}
if (rc == -ENOENT || rc == -ESTALE) {
*attr |= LUDA_IGNORE;
rc = 0;
+ } else {
+ CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
+ "check_repair, dir = %lu/%u, name = %.*s: "
+ "rc = %d\n",
+ devname, dir->i_ino, dir->i_generation,
+ ent->oied_namelen, ent->oied_name, rc);
}
GOTO(out_journal, rc);
}
- /* skip the REMOTE_PARENT_DIR. */
- if (inode == dev->od_mdt_map->omm_remote_parent->d_inode)
- GOTO(out_inode, rc = 0);
-
rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
if (rc == 0) {
LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
if (rc == 0)
*attr |= LUDA_REPAIR;
+ else
+ CDEBUG(D_LFSCK, "%.16s: fail to update FID "
+ "in the dirent, dir = %lu/%u, "
+ "name = %.*s, "DFID": rc = %d\n",
+ devname, dir->i_ino, dir->i_generation,
+ ent->oied_namelen, ent->oied_name,
+ PFID(fid), rc);
} else {
/* Do not repair under dryrun mode. */
if (*attr & LUDA_VERIFY_DRYRUN) {
fid, bh, de, hlock);
if (rc == 0)
*attr |= LUDA_REPAIR;
+ else
+ CDEBUG(D_LFSCK, "%.16s: fail to append FID "
+ "after the dirent, dir = %lu/%u, "
+ "name = %.*s, "DFID": rc = %d\n",
+ devname, dir->i_ino, dir->i_generation,
+ ent->oied_namelen, ent->oied_name,
+ PFID(fid), rc);
}
} else if (rc == -ENODATA) {
/* Do not repair under dryrun mode. */
rc = osd_ea_fid_set(info, inode, fid, 0, 0);
if (rc == 0)
*attr |= LUDA_REPAIR;
+ else
+ CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
+ "update dirent, dir = %lu/%u, "
+ "name = %.*s, "DFID": rc = %d\n",
+ devname, dir->i_ino, dir->i_generation,
+ ent->oied_namelen, ent->oied_name,
+ PFID(fid), rc);
} else {
lu_igif_build(fid, inode->i_ino, inode->i_generation);
/* It is probably IGIF object. Only aappend the
fid, bh, de, hlock);
if (rc == 0)
*attr |= LUDA_UPGRADE;
+ else
+ CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
+ "after the dirent, dir = %lu/%u, "
+ "name = %.*s, "DFID": rc = %d\n",
+ devname, dir->i_ino, dir->i_generation,
+ ent->oied_namelen, ent->oied_name,
+ PFID(fid), rc);
}
}
int rc = 0;
ENTRY;
+ LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
+
if (attr & LUDA_VERIFY) {
- attr |= LUDA_TYPE;
- if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) {
+ if (unlikely(ino == osd_remote_parent_ino(dev))) {
attr |= LUDA_IGNORE;
- rc = 0;
+ /* If the parent is on remote MDT, and there
+ * is no FID-in-dirent, then we have to get
+ * the parent FID from the linkEA. */
+ osd_get_pfid_from_linkea(env, obj, fid);
} else {
rc = osd_dirent_check_repair(env, obj, it, fid, id,
&attr);
it->oie_dirent->oied_name[1] != '.'))
RETURN(-ENOENT);
- rc = osd_ea_fid_get(env, obj, ino, fid, id);
+ if (unlikely(ino == osd_remote_parent_ino(dev)))
+ /* If the parent is on remote MDT, and there
+ * is no FID-in-dirent, then we have to get
+ * the parent FID from the linkEA. */
+ rc = osd_get_pfid_from_linkea(env, obj, fid);
+ else
+ rc = osd_ea_fid_get(env, obj, ino, fid, id);
} else {
osd_id_gen(id, ino, OSD_OII_NOGEN);
}
* \param[in] env execution environment
* \param[in] di iterator's in memory structure
* \param[in] attr attribute of the entry, only requires LUDA_TYPE to
- * calculate the lu_dirent size.
+ * calculate the lu_dirent size.
*
* \retval record size(in bytes & in memory) of the current lu_dirent
- * entry.
+ * entry.
*/
static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
__u32 attr)
OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
lu_buf_free(&info->oti_iobuf.dr_pg_buf);
lu_buf_free(&info->oti_iobuf.dr_bl_buf);
+ lu_buf_free(&info->oti_big_buf);
OBD_FREE_PTR(info);
}
/* self-repair LMA by default */
o->od_lma_self_repair = 1;
- CFS_INIT_LIST_HEAD(&o->od_ios_list);
+ INIT_LIST_HEAD(&o->od_ios_list);
/* setup scrub, including OI files initialization */
rc = osd_scrub_setup(env, o);
if (rc < 0)
break;
case LCFG_PARAM:
LASSERT(&o->od_dt_dev);
- rc = class_process_proc_seq_param(PARAM_OSD,
- lprocfs_osd_obd_vars,
- cfg, &o->od_dt_dev);
+ rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
+ cfg, &o->od_dt_dev);
if (rc > 0 || rc == -ENOSYS)
- rc = class_process_proc_seq_param(PARAM_OST,
- lprocfs_osd_obd_vars,
- cfg, &o->od_dt_dev);
+ rc = class_process_proc_param(PARAM_OST,
+ lprocfs_osd_obd_vars,
+ cfg, &o->od_dt_dev);
break;
default:
rc = -ENOSYS;
rc = class_register_type(&osd_obd_device_ops, NULL, true,
lprocfs_osd_module_vars,
-#ifndef HAVE_ONLY_PROCFS_SEQ
- NULL,
-#endif
LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
if (rc)
lu_kmem_fini(ldiskfs_caches);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff