4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd/osd_compat.c
38 * on-disk structure for managing /O
40 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
43 /* LUSTRE_VERSION_CODE */
44 #include <lustre_ver.h>
45 /* prerequisite for linux/xattr.h */
46 #include <linux/types.h>
47 /* prerequisite for linux/xattr.h */
49 /* XATTR_{REPLACE,CREATE} */
50 #include <linux/xattr.h>
53 * struct OBD_{ALLOC,FREE}*()
56 #include <obd_support.h>
58 #include "osd_internal.h"
61 static void osd_push_ctxt(const struct osd_device *dev,
62 struct lvfs_run_ctxt *newctxt,
63 struct lvfs_run_ctxt *save)
65 OBD_SET_CTXT_MAGIC(newctxt);
66 newctxt->pwdmnt = dev->od_mnt;
67 newctxt->pwd = dev->od_mnt->mnt_root;
68 newctxt->fs = get_ds();
70 push_ctxt(save, newctxt, NULL);
73 static void osd_pop_ctxt(const struct osd_device *dev,
74 struct lvfs_run_ctxt *new,
75 struct lvfs_run_ctxt *save)
77 pop_ctxt(save, new, NULL);
80 /* utility to make a directory */
81 static struct dentry *simple_mkdir(struct dentry *dir, struct vfsmount *mnt,
82 const char *name, int mode, int fix)
84 struct dentry *dchild;
88 // ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
89 CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
90 dchild = ll_lookup_one_len(name, dir, strlen(name));
94 if (dchild->d_inode) {
95 int old_mode = dchild->d_inode->i_mode;
96 if (!S_ISDIR(old_mode)) {
97 CERROR("found %s (%lu/%u) is mode %o\n", name,
98 dchild->d_inode->i_ino,
99 dchild->d_inode->i_generation, old_mode);
100 GOTO(out_err, err = -ENOTDIR);
103 /* Fixup directory permissions if necessary */
104 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
106 "fixing permissions on %s from %o to %o\n",
107 name, old_mode, mode);
108 dchild->d_inode->i_mode = (mode & S_IALLUGO) |
109 (old_mode & ~S_IALLUGO);
110 mark_inode_dirty(dchild->d_inode);
112 GOTO(out_up, dchild);
115 err = ll_vfs_mkdir(dir->d_inode, dchild, mnt, mode);
123 dchild = ERR_PTR(err);
128 int osd_last_rcvd_subdir_count(struct osd_device *osd)
130 struct lr_server_data lsd;
131 struct dentry *dlast;
134 int count = FILTER_SUBDIR_COUNT;
138 dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
141 return PTR_ERR(dlast);
142 else if (dlast->d_inode == NULL)
146 rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
147 if (rc == sizeof(lsd)) {
148 CDEBUG(D_INFO, "read last_rcvd header, uuid = %s, "
149 "subdir count = %d\n", lsd.lsd_uuid,
150 lsd.lsd_subdir_count);
151 if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
152 count = le16_to_cpu(lsd.lsd_subdir_count);
153 } else if (rc != 0) {
154 CERROR("Can't read last_rcvd file, rc = %d\n", rc);
166 static const char remote_parent_dir[] = "REMOTE_PARENT_DIR";
167 static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
169 struct lvfs_run_ctxt new;
170 struct lvfs_run_ctxt save;
171 struct dentry *parent;
172 struct osd_mdobj_map *omm;
174 struct osd_thread_info *info = osd_oti_get(env);
175 struct lu_fid *fid = &info->oti_fid;
179 OBD_ALLOC_PTR(dev->od_mdt_map);
180 if (dev->od_mdt_map == NULL)
183 omm = dev->od_mdt_map;
185 LASSERT(dev->od_fsops);
187 parent = osd_sb(dev)->s_root;
188 osd_push_ctxt(dev, &new, &save);
190 d = simple_mkdir(parent, dev->od_mnt, remote_parent_dir,
193 GOTO(cleanup, rc = PTR_ERR(d));
195 omm->omm_remote_parent = d;
197 /* Set LMA for remote parent inode */
198 lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
199 rc = osd_ea_fid_set(info, d->d_inode, fid, 0);
203 pop_ctxt(&save, &new, NULL);
205 if (omm->omm_remote_parent != NULL)
206 dput(omm->omm_remote_parent);
208 dev->od_mdt_map = NULL;
213 static void osd_mdt_fini(struct osd_device *osd)
215 struct osd_mdobj_map *omm = osd->od_mdt_map;
220 if (omm->omm_remote_parent)
221 dput(omm->omm_remote_parent);
224 osd->od_ost_map = NULL;
227 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
228 struct osd_object *obj, struct osd_thandle *oh)
230 struct osd_mdobj_map *omm = osd->od_mdt_map;
231 struct osd_thread_info *oti = osd_oti_get(env);
232 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
233 char *name = oti->oti_name;
234 struct dentry *dentry;
235 struct dentry *parent;
238 /* Set REMOTE_PARENT in lma, so other process like unlink or lfsck
239 * can identify this object quickly */
240 rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
244 lma->lma_incompat |= LMAI_REMOTE_PARENT;
245 lustre_lma_swab(lma);
246 rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
247 sizeof(*lma), XATTR_REPLACE);
251 parent = omm->omm_remote_parent;
252 sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
253 dentry = osd_child_dentry_by_inode(env, parent->d_inode,
255 mutex_lock(&parent->d_inode->i_mutex);
256 rc = osd_ldiskfs_add_entry(oh->ot_handle, dentry, obj->oo_inode,
258 LASSERTF(parent->d_inode->i_nlink > 1, "%s: %lu nlink %d",
259 osd_name(osd), parent->d_inode->i_ino,
260 parent->d_inode->i_nlink);
261 parent->d_inode->i_nlink++;
262 mark_inode_dirty(parent->d_inode);
263 mutex_unlock(&parent->d_inode->i_mutex);
267 int osd_delete_from_remote_parent(const struct lu_env *env,
268 struct osd_device *osd,
269 struct osd_object *obj,
270 struct osd_thandle *oh)
272 struct osd_mdobj_map *omm = osd->od_mdt_map;
273 struct osd_thread_info *oti = osd_oti_get(env);
274 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
275 char *name = oti->oti_name;
276 struct dentry *dentry;
277 struct dentry *parent;
278 struct ldiskfs_dir_entry_2 *de;
279 struct buffer_head *bh;
282 /* Check lma to see whether it is remote object */
283 rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
287 if (likely(!(lma->lma_incompat & LMAI_REMOTE_PARENT)))
290 parent = omm->omm_remote_parent;
291 sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
292 dentry = osd_child_dentry_by_inode(env, parent->d_inode,
294 mutex_lock(&parent->d_inode->i_mutex);
295 bh = osd_ldiskfs_find_entry(parent->d_inode, dentry, &de, NULL);
297 mutex_unlock(&parent->d_inode->i_mutex);
300 rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode, de, bh);
301 LASSERTF(parent->d_inode->i_nlink > 1, "%s: %lu nlink %d",
302 osd_name(osd), parent->d_inode->i_ino,
303 parent->d_inode->i_nlink);
304 parent->d_inode->i_nlink--;
305 mark_inode_dirty(parent->d_inode);
306 mutex_unlock(&parent->d_inode->i_mutex);
312 * directory structure on legacy OST:
314 * O/<seq>/d0-31/<objid>
321 static int osd_ost_init(struct osd_device *dev)
323 struct lvfs_run_ctxt new;
324 struct lvfs_run_ctxt save;
325 struct dentry *rootd = osd_sb(dev)->s_root;
330 OBD_ALLOC_PTR(dev->od_ost_map);
331 if (dev->od_ost_map == NULL)
334 /* to get subdir count from last_rcvd */
335 rc = osd_last_rcvd_subdir_count(dev);
337 OBD_FREE_PTR(dev->od_ost_map);
341 dev->od_ost_map->om_subdir_count = rc;
344 CFS_INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
345 rwlock_init(&dev->od_ost_map->om_seq_list_lock);
346 sema_init(&dev->od_ost_map->om_dir_init_sem, 1);
348 LASSERT(dev->od_fsops);
349 osd_push_ctxt(dev, &new, &save);
351 d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
353 GOTO(cleanup, rc = PTR_ERR(d));
355 ldiskfs_set_inode_state(d->d_inode, LDISKFS_STATE_LUSTRE_NO_OI);
356 dev->od_ost_map->om_root = d;
359 osd_pop_ctxt(dev, &new, &save);
361 OBD_FREE_PTR(dev->od_ost_map);
368 static void osd_seq_free(struct osd_obj_map *map,
369 struct osd_obj_seq *osd_seq)
373 cfs_list_del_init(&osd_seq->oos_seq_list);
375 if (osd_seq->oos_dirs) {
376 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
377 if (osd_seq->oos_dirs[j])
378 dput(osd_seq->oos_dirs[j]);
380 OBD_FREE(osd_seq->oos_dirs,
381 sizeof(struct dentry *) * osd_seq->oos_subdir_count);
384 if (osd_seq->oos_root)
385 dput(osd_seq->oos_root);
387 OBD_FREE_PTR(osd_seq);
390 static void osd_ost_fini(struct osd_device *osd)
392 struct osd_obj_seq *osd_seq;
393 struct osd_obj_seq *tmp;
394 struct osd_obj_map *map = osd->od_ost_map;
400 write_lock(&map->om_seq_list_lock);
401 cfs_list_for_each_entry_safe(osd_seq, tmp,
404 osd_seq_free(map, osd_seq);
406 write_unlock(&map->om_seq_list_lock);
410 osd->od_ost_map = NULL;
414 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
419 /* prepare structures for OST */
420 rc = osd_ost_init(dev);
424 /* prepare structures for MDS */
425 rc = osd_mdt_init(env, dev);
430 struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, obd_seq seq)
432 struct osd_obj_seq *osd_seq;
434 cfs_list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
435 if (osd_seq->oos_seq == seq)
441 struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, obd_seq seq)
443 struct osd_obj_seq *osd_seq;
445 read_lock(&map->om_seq_list_lock);
446 osd_seq = osd_seq_find_locked(map, seq);
447 read_unlock(&map->om_seq_list_lock);
451 void osd_obj_map_fini(struct osd_device *dev)
457 static int osd_obj_del_entry(struct osd_thread_info *info,
458 struct osd_device *osd,
459 struct dentry *dird, char *name,
462 struct ldiskfs_dir_entry_2 *de;
463 struct buffer_head *bh;
464 struct osd_thandle *oh;
465 struct dentry *child;
466 struct inode *dir = dird->d_inode;
470 oh = container_of(th, struct osd_thandle, ot_super);
471 LASSERT(oh->ot_handle != NULL);
472 LASSERT(oh->ot_handle->h_transaction != NULL);
475 child = &info->oti_child_dentry;
476 child->d_name.hash = 0;
477 child->d_name.name = name;
478 child->d_name.len = strlen(name);
479 child->d_parent = dird;
480 child->d_inode = NULL;
483 mutex_lock(&dir->i_mutex);
485 bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
487 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
490 mutex_unlock(&dir->i_mutex);
495 int osd_obj_add_entry(struct osd_thread_info *info,
496 struct osd_device *osd,
497 struct dentry *dir, char *name,
498 const struct osd_inode_id *id,
501 struct osd_thandle *oh;
502 struct dentry *child;
508 oh = container_of(th, struct osd_thandle, ot_super);
509 LASSERT(oh->ot_handle != NULL);
510 LASSERT(oh->ot_handle->h_transaction != NULL);
512 inode = &info->oti_inode;
513 inode->i_sb = osd_sb(osd);
514 osd_id_to_inode(inode, id);
515 inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
517 child = &info->oti_child_dentry;
518 child->d_name.hash = 0;
519 child->d_name.name = name;
520 child->d_name.len = strlen(name);
521 child->d_parent = dir;
522 child->d_inode = inode;
524 ll_vfs_dq_init(dir->d_inode);
525 mutex_lock(&dir->d_inode->i_mutex);
526 rc = osd_ldiskfs_add_entry(oh->ot_handle, child, inode, NULL);
527 mutex_unlock(&dir->d_inode->i_mutex);
533 * Use LPU64 for legacy OST sequences, but use LPX64i for new
534 * sequences names, so that the O/{seq}/dN/{oid} more closely
535 * follows the DFID/PFID format. This makes it easier to map from
536 * debug messages to objects in the future, and the legacy space
537 * of FID_SEQ_OST_MDT0 will be unused in the future.
539 static inline void osd_seq_name(char *seq_name, obd_seq seq)
541 sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
542 fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
543 fid_seq_is_idif(seq) ? 0 : seq);
546 static inline void osd_oid_name(char *name, const struct lu_fid *fid, obd_id id)
548 sprintf(name, (fid_seq_is_rsvd(fid_seq(fid)) ||
549 fid_seq_is_mdt0(fid_seq(fid)) ||
550 fid_seq_is_idif(fid_seq(fid))) ? LPU64 : LPX64i, id);
553 /* external locking is required */
554 static int osd_seq_load_locked(struct osd_device *osd,
555 struct osd_obj_seq *osd_seq)
557 struct osd_obj_map *map = osd->od_ost_map;
558 struct dentry *seq_dir;
564 if (osd_seq->oos_root != NULL)
568 LASSERT(map->om_root);
570 osd_seq_name(seq_name, osd_seq->oos_seq);
572 seq_dir = simple_mkdir(map->om_root, osd->od_mnt, seq_name, 0755, 1);
574 GOTO(out_err, rc = PTR_ERR(seq_dir));
575 else if (seq_dir->d_inode == NULL)
576 GOTO(out_put, rc = -EFAULT);
578 ldiskfs_set_inode_state(seq_dir->d_inode, LDISKFS_STATE_LUSTRE_NO_OI);
579 osd_seq->oos_root = seq_dir;
581 LASSERT(osd_seq->oos_dirs == NULL);
582 OBD_ALLOC(osd_seq->oos_dirs,
583 sizeof(seq_dir) * osd_seq->oos_subdir_count);
584 if (osd_seq->oos_dirs == NULL)
585 GOTO(out_put, rc = -ENOMEM);
587 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
591 sprintf(name, "d%u", i);
592 dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, name,
596 } else if (dir->d_inode) {
597 ldiskfs_set_inode_state(dir->d_inode,
598 LDISKFS_STATE_LUSTRE_NO_OI);
599 osd_seq->oos_dirs[i] = dir;
607 osd_seq_free(map, osd_seq);
611 osd_seq->oos_root = NULL;
617 struct osd_obj_seq *osd_seq_load(struct osd_device *osd, obd_seq seq)
619 struct osd_obj_map *map;
620 struct osd_obj_seq *osd_seq;
624 map = osd->od_ost_map;
626 LASSERT(map->om_root);
628 osd_seq = osd_seq_find(map, seq);
629 if (likely(osd_seq != NULL))
632 /* Serializing init process */
633 down(&map->om_dir_init_sem);
635 /* Check whether the seq has been added */
636 read_lock(&map->om_seq_list_lock);
637 osd_seq = osd_seq_find_locked(map, seq);
638 if (osd_seq != NULL) {
639 read_unlock(&map->om_seq_list_lock);
640 GOTO(cleanup, rc = 0);
642 read_unlock(&map->om_seq_list_lock);
644 OBD_ALLOC_PTR(osd_seq);
646 GOTO(cleanup, rc = -ENOMEM);
648 CFS_INIT_LIST_HEAD(&osd_seq->oos_seq_list);
649 osd_seq->oos_seq = seq;
650 /* Init subdir count to be 32, but each seq can have
651 * different subdir count */
652 osd_seq->oos_subdir_count = map->om_subdir_count;
653 rc = osd_seq_load_locked(osd, osd_seq);
657 write_lock(&map->om_seq_list_lock);
658 cfs_list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
659 write_unlock(&map->om_seq_list_lock);
662 up(&map->om_dir_init_sem);
665 OBD_FREE_PTR(osd_seq);
672 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
673 const struct lu_fid *fid, struct osd_inode_id *id)
675 struct osd_obj_map *map;
676 struct osd_obj_seq *osd_seq;
677 struct dentry *d_seq;
678 struct dentry *child;
679 struct ost_id *ostid = &info->oti_ostid;
682 struct ldiskfs_dir_entry_2 *de;
683 struct buffer_head *bh;
688 /* on the very first lookup we find and open directories */
690 map = dev->od_ost_map;
692 LASSERT(map->om_root);
694 fid_ostid_pack(fid, ostid);
695 osd_seq = osd_seq_load(dev, ostid->oi_seq);
697 RETURN(PTR_ERR(osd_seq));
699 dirn = ostid->oi_id & (osd_seq->oos_subdir_count - 1);
700 d_seq = osd_seq->oos_dirs[dirn];
703 osd_oid_name(name, fid, ostid->oi_id);
705 child = &info->oti_child_dentry;
706 child->d_parent = d_seq;
707 child->d_name.hash = 0;
708 child->d_name.name = name;
709 /* XXX: we can use rc from sprintf() instead of strlen() */
710 child->d_name.len = strlen(name);
712 dir = d_seq->d_inode;
713 mutex_lock(&dir->i_mutex);
714 bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
715 mutex_unlock(&dir->i_mutex);
720 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
723 inode = osd_iget(info, dev, id);
725 RETURN(PTR_ERR(inode));
731 int osd_obj_map_insert(struct osd_thread_info *info,
732 struct osd_device *osd,
733 const struct lu_fid *fid,
734 const struct osd_inode_id *id,
737 struct osd_obj_map *map;
738 struct osd_obj_seq *osd_seq;
740 struct ost_id *ostid = &info->oti_ostid;
745 map = osd->od_ost_map;
748 /* map fid to seq:objid */
749 fid_ostid_pack(fid, ostid);
751 osd_seq = osd_seq_load(osd, ostid->oi_seq);
753 RETURN(PTR_ERR(osd_seq));
755 dirn = ostid->oi_id & (osd_seq->oos_subdir_count - 1);
756 d = osd_seq->oos_dirs[dirn];
759 osd_oid_name(name, fid, ostid->oi_id);
760 rc = osd_obj_add_entry(info, osd, d, name, id, th);
765 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
766 const struct lu_fid *fid, struct thandle *th)
768 struct osd_obj_map *map;
769 struct osd_obj_seq *osd_seq;
771 struct ost_id *ostid = &info->oti_ostid;
776 map = osd->od_ost_map;
779 /* map fid to seq:objid */
780 fid_ostid_pack(fid, ostid);
782 osd_seq = osd_seq_load(osd, ostid->oi_seq);
784 GOTO(cleanup, rc = PTR_ERR(osd_seq));
786 dirn = ostid->oi_id & (osd_seq->oos_subdir_count - 1);
787 d = osd_seq->oos_dirs[dirn];
790 osd_oid_name(name, fid, ostid->oi_id);
791 rc = osd_obj_del_entry(info, osd, d, name, th);
796 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
797 const struct lu_fid *fid,
798 const struct osd_inode_id *id,
801 struct osd_obj_map *map = osd->od_ost_map;
802 struct dentry *root = osd_sb(osd)->s_root;
807 if (fid_is_last_id(fid)) {
808 struct osd_obj_seq *osd_seq;
810 /* on creation of LAST_ID we create O/<seq> hierarchy */
812 osd_seq = osd_seq_load(osd, fid_seq(fid));
814 RETURN(PTR_ERR(osd_seq));
815 rc = osd_obj_add_entry(info, osd, osd_seq->oos_root,
818 name = osd_lf_fid2name(fid);
820 CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
822 rc = osd_obj_add_entry(info, osd, root, name, id, th);
828 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
829 const struct lu_fid *fid, struct osd_inode_id *id)
832 struct dentry *dentry;
838 if (fid_is_last_id(fid)) {
839 struct osd_obj_seq *osd_seq;
841 osd_seq = osd_seq_load(osd, fid_seq(fid));
843 RETURN(PTR_ERR(osd_seq));
844 root = osd_seq->oos_root;
847 root = osd_sb(osd)->s_root;
848 name = osd_lf_fid2name(fid);
849 if (name == NULL || strlen(name) == 0)
853 dentry = ll_lookup_one_len(name, root, strlen(name));
854 if (!IS_ERR(dentry)) {
855 inode = dentry->d_inode;
857 if (is_bad_inode(inode)) {
860 osd_id_gen(id, inode->i_ino,
861 inode->i_generation);
865 /* if dentry is accessible after osd_compat_spec_insert it
866 * will still contain NULL inode, so don't keep it in cache */
867 d_invalidate(dentry);