* Top-level entry points into osd module
*
* Author: Nikita Danilov <nikita@clusterfs.com>
+ * Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
*/
#ifndef EXPORT_SYMTAB
if (bits & LA_RDEV)
inode->i_rdev = attr->la_rdev;
- if (bits & LA_FLAGS) {
- struct ldiskfs_inode_info *li = LDISKFS_I(inode);
-
- li->i_flags = (li->i_flags & ~LDISKFS_FL_USER_MODIFIABLE) |
- (attr->la_flags & LDISKFS_FL_USER_MODIFIABLE);
- }
+ if (bits & LA_FLAGS)
+ inode->i_flags = ll_ext_to_inode_flags(attr->la_flags);
return 0;
}
}
/**
- * Helper function to pack the fid
+ * Helper function to pack the fid, ldiskfs stores fid in packed format.
*/
void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
struct lu_fid *befider)
pack->fp_len = sizeof(*befider) + 1;
}
+/**
+ * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
+ * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
+ * To have compatilibility with 1.8 ldiskfs driver we need to have
+ * magic number at start of fid data.
+ * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
+ * its inmemory API.
+ */
+void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
+ const struct dt_rec *fid)
+{
+ param->edp_magic = LDISKFS_LUFID_MAGIC;
+ param->edp_len = sizeof(struct lu_fid) + 1;
+
+ fid_cpu_to_be((struct lu_fid *)param->edp_data,
+ (struct lu_fid *)fid);
+}
+
int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
{
int result;
rc = 0;
}
iput(inode);
-
out:
RETURN(rc);
}
struct osd_object *obj = osd_dt_obj(dt);
struct osd_thread_info *info = osd_oti_get(env);
int result;
- int is_root = 0;
ENTRY;
result = __osd_object_create(info, obj, attr, hint, dof, th);
- if (hint && hint->dah_parent)
- is_root = osd_object_is_root(osd_dt_obj(hint->dah_parent));
-
/* objects under osd root shld have igif fid, so dont add fid EA */
- if (result == 0 && is_root == 0)
+ if (result == 0 && fid_seq(fid) >= FID_SEQ_DISTRIBUTED_START)
result = osd_ea_fid_set(env, dt, fid);
if (result == 0)
*
* which doesn't work for globally shared files like /last-received.
*/
-int fsfilt_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
-int fsfilt_ldiskfs_write_handle(struct inode *inode, void *buf, int bufsize,
- loff_t *offs, handle_t *handle);
+static int osd_ldiskfs_readlink(struct inode *inode, char *buffer, int buflen)
+{
+ struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
+
+ memcpy(buffer, (char*)ei->i_data, buflen);
+
+ return buflen;
+}
+
+static int osd_ldiskfs_read(struct inode *inode, void *buf, int size,
+ loff_t *offs)
+{
+ struct buffer_head *bh;
+ unsigned long block;
+ int osize = size;
+ int blocksize;
+ int csize;
+ int boffs;
+ int err;
+
+ /* prevent reading after eof */
+ spin_lock(&inode->i_lock);
+ if (i_size_read(inode) < *offs + size) {
+ size = i_size_read(inode) - *offs;
+ spin_unlock(&inode->i_lock);
+ if (size < 0) {
+ CDEBUG(D_EXT2, "size %llu is too short to read @%llu\n",
+ i_size_read(inode), *offs);
+ return -EBADR;
+ } else if (size == 0) {
+ return 0;
+ }
+ } else {
+ spin_unlock(&inode->i_lock);
+ }
+
+ blocksize = 1 << inode->i_blkbits;
+
+ while (size > 0) {
+ block = *offs >> inode->i_blkbits;
+ boffs = *offs & (blocksize - 1);
+ csize = min(blocksize - boffs, size);
+ bh = ldiskfs_bread(NULL, inode, block, 0, &err);
+ if (!bh) {
+ CERROR("can't read block: %d\n", err);
+ return err;
+ }
+
+ memcpy(buf, bh->b_data + boffs, csize);
+ brelse(bh);
+
+ *offs += csize;
+ buf += csize;
+ size -= csize;
+ }
+ return osize;
+}
static ssize_t osd_read(const struct lu_env *env, struct dt_object *dt,
struct lu_buf *buf, loff_t *pos,
struct lustre_capa *capa)
{
- struct inode *inode = osd_dt_obj(dt)->oo_inode;
+ struct osd_object *obj = osd_dt_obj(dt);
+ struct inode *inode = obj->oo_inode;
+ int rc;
if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
RETURN(-EACCES);
- return fsfilt_ldiskfs_read(inode, buf->lb_buf, buf->lb_len, pos);
+ /* Read small symlink from inode body as we need to maintain correct
+ * on-disk symlinks for ldiskfs.
+ */
+ if (S_ISLNK(obj->oo_dt.do_lu.lo_header->loh_attr) &&
+ (buf->lb_len <= sizeof (LDISKFS_I(inode)->i_data)))
+ rc = osd_ldiskfs_readlink(inode, buf->lb_buf, buf->lb_len);
+ else
+ rc = osd_ldiskfs_read(inode, buf->lb_buf, buf->lb_len, pos);
+
+ return rc;
+}
+
+static int osd_ldiskfs_writelink(struct inode *inode, char *buffer, int buflen)
+{
+
+ memcpy((char*)&LDISKFS_I(inode)->i_data, (char *)buffer,
+ buflen);
+ LDISKFS_I(inode)->i_disksize = buflen;
+ i_size_write(inode, buflen);
+ inode->i_sb->s_op->dirty_inode(inode);
+
+ return 0;
+}
+
+static int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
+ loff_t *offs, handle_t *handle)
+{
+ struct buffer_head *bh = NULL;
+ loff_t offset = *offs;
+ loff_t new_size = i_size_read(inode);
+ unsigned long block;
+ int blocksize = 1 << inode->i_blkbits;
+ int err = 0;
+ int size;
+ int boffs;
+ int dirty_inode = 0;
+
+ while (bufsize > 0) {
+ if (bh != NULL)
+ brelse(bh);
+
+ block = offset >> inode->i_blkbits;
+ boffs = offset & (blocksize - 1);
+ size = min(blocksize - boffs, bufsize);
+ bh = ldiskfs_bread(handle, inode, block, 1, &err);
+ if (!bh) {
+ CERROR("can't read/create block: %d\n", err);
+ break;
+ }
+
+ err = ldiskfs_journal_get_write_access(handle, bh);
+ if (err) {
+ CERROR("journal_get_write_access() returned error %d\n",
+ err);
+ break;
+ }
+ LASSERTF(boffs + size <= bh->b_size,
+ "boffs %d size %d bh->b_size %lu",
+ boffs, size, (unsigned long)bh->b_size);
+ memcpy(bh->b_data + boffs, buf, size);
+ err = ldiskfs_journal_dirty_metadata(handle, bh);
+ if (err)
+ break;
+
+ if (offset + size > new_size)
+ new_size = offset + size;
+ offset += size;
+ bufsize -= size;
+ buf += size;
+ }
+ if (bh)
+ brelse(bh);
+
+ /* correct in-core and on-disk sizes */
+ if (new_size > i_size_read(inode)) {
+ spin_lock(&inode->i_lock);
+ if (new_size > i_size_read(inode))
+ i_size_write(inode, new_size);
+ if (i_size_read(inode) > LDISKFS_I(inode)->i_disksize) {
+ LDISKFS_I(inode)->i_disksize = i_size_read(inode);
+ dirty_inode = 1;
+ }
+ spin_unlock(&inode->i_lock);
+ if (dirty_inode)
+ inode->i_sb->s_op->dirty_inode(inode);
+ }
+
+ if (err == 0)
+ *offs = offset;
+ return err;
}
static ssize_t osd_write(const struct lu_env *env, struct dt_object *dt,
struct thandle *handle, struct lustre_capa *capa,
int ignore_quota)
{
- struct inode *inode = osd_dt_obj(dt)->oo_inode;
+ struct osd_object *obj = osd_dt_obj(dt);
+ struct inode *inode = obj->oo_inode;
struct osd_thandle *oh;
- ssize_t result;
+ ssize_t result = 0;
#ifdef HAVE_QUOTA_SUPPORT
cfs_cap_t save = current->cap_effective;
#endif
else
current->cap_effective &= ~CFS_CAP_SYS_RESOURCE_MASK;
#endif
- result = fsfilt_ldiskfs_write_handle(inode, buf->lb_buf, buf->lb_len,
- pos, oh->ot_handle);
+ /* Write small symlink to inode body as we need to maintain correct
+ * on-disk symlinks for ldiskfs.
+ */
+ if(S_ISLNK(obj->oo_dt.do_lu.lo_header->loh_attr) &&
+ (buf->lb_len < sizeof (LDISKFS_I(inode)->i_data)))
+ result = osd_ldiskfs_writelink(inode, buf->lb_buf, buf->lb_len);
+ else
+ result = osd_ldiskfs_write_record(inode, buf->lb_buf,
+ buf->lb_len, pos,
+ oh->ot_handle);
#ifdef HAVE_QUOTA_SUPPORT
current->cap_effective = save;
#endif
RETURN(rc);
}
+static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
+ struct dt_rec *fid)
+{
+ struct osd_fid_pack *rec;
+ int rc = -ENODATA;
+
+ if (de->file_type & LDISKFS_DIRENT_LUFID) {
+ rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
+ rc = osd_fid_unpack((struct lu_fid *)fid, rec);
+ }
+ RETURN(rc);
+}
+
/**
* Index delete function for interoperability mode (b11826).
* It will remove the directory entry added by osd_index_ea_insert().
rc = iam_it_get(it, (struct iam_key *)key);
if (rc >= 0) {
if (S_ISDIR(obj->oo_inode->i_mode))
- iam_rec = (struct iam_rec *)oti->oti_fid_packed;
+ iam_rec = (struct iam_rec *)oti->oti_ldp;
else
iam_rec = (struct iam_rec *) rec;
cfs_cap_t save = current->cap_effective;
#endif
struct osd_thread_info *oti = osd_oti_get(env);
- struct iam_rec *iam_rec = (struct iam_rec *)oti->oti_fid_packed;
+ struct iam_rec *iam_rec = (struct iam_rec *)oti->oti_ldp;
int rc;
ENTRY;
*/
static int __osd_ea_add_rec(struct osd_thread_info *info,
struct osd_object *pobj,
- struct osd_object *cobj,
+ struct inode *cinode,
const char *name,
+ const struct dt_rec *fid,
struct thandle *th)
{
+ struct ldiskfs_dentry_param *ldp;
struct dentry *child;
struct osd_thandle *oth;
- struct inode *cinode = cobj->oo_inode;
int rc;
oth = container_of(th, struct osd_thandle, ot_super);
LASSERT(oth->ot_handle->h_transaction != NULL);
child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
+
+ if (fid_is_igif((struct lu_fid *)fid) ||
+ fid_seq((struct lu_fid *)fid) >= FID_SEQ_DISTRIBUTED_START) {
+ ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
+ osd_get_ldiskfs_dirent_param(ldp, fid);
+ child->d_fsdata = (void*) ldp;
+ } else
+ child->d_fsdata = NULL;
rc = ldiskfs_add_entry(oth->ot_handle, child, cinode);
RETURN(rc);
*/
static int osd_add_dot_dotdot(struct osd_thread_info *info,
struct osd_object *dir,
- struct osd_object *obj, const char *name,
+ struct inode *parent_dir, const char *name,
+ const struct dt_rec *dot_fid,
+ const struct dt_rec *dot_dot_fid,
struct thandle *th)
{
- struct inode *parent_dir = obj->oo_inode;
struct inode *inode = dir->oo_inode;
+ struct ldiskfs_dentry_param *dot_ldp;
+ struct ldiskfs_dentry_param *dot_dot_ldp;
struct osd_thandle *oth;
int result = 0;
if (dir->oo_compat_dot_created) {
result = -EEXIST;
} else {
- LASSERT(obj == dir);
+ LASSERT(inode == parent_dir);
dir->oo_compat_dot_created = 1;
result = 0;
}
} else if(strcmp(name, dotdot) == 0) {
+ dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
+ dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
+
if (!dir->oo_compat_dot_created)
return -EINVAL;
- if (dir->oo_compat_dotdot_created)
- return __osd_ea_add_rec(info, dir, obj, name, th);
+ if (fid_seq((struct lu_fid *) dot_fid) >= FID_SEQ_DISTRIBUTED_START) {
+ osd_get_ldiskfs_dirent_param(dot_ldp, dot_fid);
+ osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
+ } else {
+ dot_ldp = NULL;
+ dot_dot_ldp = NULL;
+ }
+ /* in case of rename, dotdot is already created */
+ if (dir->oo_compat_dotdot_created) {
+ return __osd_ea_add_rec(info, dir, parent_dir, name,
+ dot_dot_fid, th);
+ }
- result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir, inode);
+ result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir, inode,
+ dot_ldp, dot_dot_ldp);
if (result == 0)
dir->oo_compat_dotdot_created = 1;
}
*/
static int osd_ea_add_rec(const struct lu_env *env,
struct osd_object *pobj,
- struct osd_object *cobj,
+ struct inode *cinode,
const char *name,
+ const struct dt_rec *fid,
struct thandle *th)
{
struct osd_thread_info *info = osd_oti_get(env);
if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
name[2] =='\0')))
- rc = osd_add_dot_dotdot(info, pobj, cobj, name, th);
+ rc = osd_add_dot_dotdot(info, pobj, cinode, name,
+ (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
+ fid, th);
else
- rc = __osd_ea_add_rec(info, pobj, cobj, name, th);
+ rc = __osd_ea_add_rec(info, pobj, cinode, name, fid, th);
return rc;
}
bh = ll_ldiskfs_find_entry(dir, dentry, &de);
if (bh) {
ino = le32_to_cpu(de->inode);
+ rc = osd_get_fid_from_dentry(de, rec);
+
+ /* done with de, release bh */
brelse(bh);
- rc = osd_ea_fid_get(env, obj, ino, fid);
+ if (rc != 0)
+ rc = osd_ea_fid_get(env, obj, ino, fid);
} else
rc = -ENOENT;
current->cap_effective &= ~CFS_CAP_SYS_RESOURCE_MASK;
#endif
cfs_down_write(&obj->oo_ext_idx_sem);
- rc = osd_ea_add_rec(env, obj, child, name, th);
+ rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th);
cfs_up_write(&obj->oo_ext_idx_sem);
#ifdef HAVE_QUOTA_SUPPORT
current->cap_effective = save;
loff_t offset, __u64 ino,
unsigned d_type)
{
- struct osd_it_ea *it = (struct osd_it_ea *)buf;
- struct osd_it_ea_dirent *ent = it->oie_dirent;
+ struct osd_it_ea *it = (struct osd_it_ea *)buf;
+ struct osd_it_ea_dirent *ent = it->oie_dirent;
+ struct lu_fid *fid = &ent->oied_fid;
+ struct osd_fid_pack *rec;
ENTRY;
/* this should never happen */
OSD_IT_EA_BUFSIZE)
RETURN(1);
+ if (d_type & LDISKFS_DIRENT_LUFID) {
+ rec = (struct osd_fid_pack*) (name + namelen + 1);
+
+ if (osd_fid_unpack(fid, rec) != 0)
+ fid_zero(fid);
+
+ d_type &= ~LDISKFS_DIRENT_LUFID;
+ } else {
+ fid_zero(fid);
+ }
+
ent->oied_ino = ino;
ent->oied_off = offset;
ent->oied_namelen = namelen;
{
struct osd_it_ea *it = (struct osd_it_ea *)di;
struct osd_object *obj = it->oie_obj;
- struct osd_thread_info *info = osd_oti_get(env);
- struct lu_fid *fid = &info->oti_fid;
- int rc;
+ struct lu_fid *fid = &it->oie_dirent->oied_fid;
+ int rc = 0;
ENTRY;
- rc = osd_ea_fid_get(env, obj, it->oie_dirent->oied_ino, fid);
+ if (!fid_is_sane(fid))
+ rc = osd_ea_fid_get(env, obj, it->oie_dirent->oied_ino, fid);
if (rc == 0)
osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
git://git.whamcloud.com / fs / lustre-release.git / blobdiff