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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/mdd/mdd_object.c
34 * Lustre Metadata Server (mdd) routines
36 * Author: Wang Di <wangdi@clusterfs.com>
39 #define DEBUG_SUBSYSTEM S_MDS
41 #include <linux/module.h>
43 #include <obd_class.h>
44 #include <obd_support.h>
45 #include <lprocfs_status.h>
46 /* fid_be_cpu(), fid_cpu_to_be(). */
47 #include <lustre_fid.h>
48 #include <lustre_idmap.h>
49 #include <uapi/linux/lustre/lustre_param.h>
50 #include <lustre_mds.h>
52 #include "mdd_internal.h"
54 static const struct lu_object_operations mdd_lu_obj_ops;
56 struct mdd_object_user {
57 struct list_head mou_list; /**< linked off mod_users */
58 int mou_flags; /**< open mode by client */
59 __u64 mou_uidgid; /**< uid_gid on client */
60 int mou_opencount; /**< # opened */
61 ktime_t mou_deniednext; /**< time of next access denied
66 static int mdd_xattr_get(const struct lu_env *env,
67 struct md_object *obj, struct lu_buf *buf,
70 static int mdd_changelog_data_store_by_fid(const struct lu_env *env,
71 struct mdd_device *mdd,
72 enum changelog_rec_type type,
73 int flags, const struct lu_fid *fid,
74 const char *xattr_name,
75 struct thandle *handle);
77 static inline bool has_prefix(const char *str, const char *prefix);
80 static unsigned int flags_helper(int flags)
82 unsigned int mflags = 0;
84 if (flags & MDS_FMODE_EXEC) {
85 mflags = MDS_FMODE_EXEC;
87 if (flags & MDS_FMODE_READ)
88 mflags = MDS_FMODE_READ;
90 (MDS_FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
91 mflags |= MDS_FMODE_WRITE;
97 /** Allocate/init a user and its sub-structures.
102 * \retval mou [OUT] success valid structure
105 static struct mdd_object_user *mdd_obj_user_alloc(int flags,
106 uid_t uid, gid_t gid)
108 struct mdd_object_user *mou;
112 OBD_SLAB_ALLOC_PTR(mou, mdd_object_kmem);
114 RETURN(ERR_PTR(-ENOMEM));
116 mou->mou_flags = flags;
117 mou->mou_uidgid = ((__u64)uid << 32) | gid;
118 mou->mou_opencount = 0;
119 mou->mou_deniednext = ktime_set(0, 0);
125 * Free a user and its sub-structures.
127 * \param mou [IN] user to be freed.
129 static void mdd_obj_user_free(struct mdd_object_user *mou)
131 OBD_SLAB_FREE_PTR(mou, mdd_object_kmem);
135 * Find if UID/GID already has this file open
137 * Caller should have write-locked \param mdd_obj.
138 * \param mdd_obj [IN] mdd_obj
139 * \param uid [IN] client uid
140 * \param gid [IN] client gid
141 * \retval user pointer or NULL if not found
144 struct mdd_object_user *mdd_obj_user_find(struct mdd_object *mdd_obj,
145 uid_t uid, gid_t gid, int flags)
147 struct mdd_object_user *mou;
152 uidgid = ((__u64)uid << 32) | gid;
153 list_for_each_entry(mou, &mdd_obj->mod_users, mou_list) {
154 if (mou->mou_uidgid == uidgid &&
155 flags_helper(mou->mou_flags) == flags_helper(flags))
162 * Add a user to the list of openers for this file
164 * Caller should have write-locked \param mdd_obj.
165 * \param mdd_obj [IN] mdd_obj
166 * \param mou [IN] user
168 * \retval -ve failure
170 static int mdd_obj_user_add(struct mdd_object *mdd_obj,
171 struct mdd_object_user *mou,
174 struct mdd_device *mdd = mdd_obj2mdd_dev(mdd_obj);
175 struct mdd_object_user *tmp;
176 __u32 uid = mou->mou_uidgid >> 32;
177 __u32 gid = mou->mou_uidgid & ((1UL << 32) - 1);
181 tmp = mdd_obj_user_find(mdd_obj, uid,
182 gid, mou->mou_flags);
186 list_add_tail(&mou->mou_list, &mdd_obj->mod_users);
189 /* next 'access denied' notification cannot happen before
192 mou->mou_deniednext =
193 ktime_add(ktime_get(),
194 ktime_set(mdd->mdd_cl.mc_deniednext, 0));
196 mou->mou_opencount++;
201 * Remove UID from the list
203 * Caller should have write-locked \param mdd_obj.
204 * \param mdd_obj [IN] mdd_obj
205 * \param uid [IN] user
206 * \retval -ve failure
208 static int mdd_obj_user_remove(struct mdd_object *mdd_obj,
209 struct mdd_object_user *mou)
216 list_del_init(&mou->mou_list);
218 mdd_obj_user_free(mou);
222 int mdd_la_get(const struct lu_env *env, struct mdd_object *obj,
227 if (mdd_object_exists(obj) == 0)
230 rc = mdo_attr_get(env, obj, la);
231 if (unlikely(rc != 0)) {
233 obj->mod_flags |= DEAD_OBJ;
237 if (la->la_valid & LA_FLAGS &&
238 la->la_flags & LUSTRE_ORPHAN_FL)
239 obj->mod_flags |= ORPHAN_OBJ | DEAD_OBJ;
244 struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
246 struct mdd_thread_info *info;
248 lu_env_refill((struct lu_env *)env);
249 info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
250 LASSERT(info != NULL);
254 struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
258 buf = &mdd_env_info(env)->mti_buf[0];
264 const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
265 const void *area, ssize_t len)
269 buf = &mdd_env_info(env)->mti_buf[0];
270 buf->lb_buf = (void *)area;
275 struct lu_object *mdd_object_alloc(const struct lu_env *env,
276 const struct lu_object_header *hdr,
279 struct mdd_object *mdd_obj;
282 OBD_SLAB_ALLOC_PTR_GFP(mdd_obj, mdd_object_kmem, GFP_NOFS);
286 o = mdd2lu_obj(mdd_obj);
287 lu_object_init(o, NULL, d);
288 mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
289 mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
290 mdd_obj->mod_count = 0;
291 o->lo_ops = &mdd_lu_obj_ops;
292 INIT_LIST_HEAD(&mdd_obj->mod_users);
297 static int mdd_object_init(const struct lu_env *env, struct lu_object *o,
298 const struct lu_object_conf *unused)
300 struct mdd_device *d = lu2mdd_dev(o->lo_dev);
301 struct mdd_object *mdd_obj = lu2mdd_obj(o);
302 struct lu_object *below;
303 struct lu_device *under;
306 mdd_obj->mod_cltime = ktime_set(0, 0);
307 under = &d->mdd_child->dd_lu_dev;
308 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
310 RETURN(PTR_ERR(below));
312 lu_object_add(o, below);
317 static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
321 if (lu_object_exists(o)) {
322 struct mdd_object *mdd_obj = lu2mdd_obj(o);
323 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
325 rc = mdd_la_get(env, mdd_obj, attr);
331 static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
333 struct mdd_object *mdd = lu2mdd_obj(o);
334 struct mdd_object_user *mou, *tmp2;
337 list_for_each_entry_safe(mou, tmp2, &mdd->mod_users, mou_list) {
338 list_del(&mou->mou_list);
339 mdd_obj_user_free(mou);
343 OBD_SLAB_FREE_PTR(mdd, mdd_object_kmem);
346 static int mdd_object_print(const struct lu_env *env, void *cookie,
347 lu_printer_t p, const struct lu_object *o)
349 struct mdd_object *mdd = lu2mdd_obj((struct lu_object *)o);
351 return (*p)(env, cookie,
352 LUSTRE_MDD_NAME"-object@%p(open_count=%d, valid=%x, cltime=%lldns, flags=%lx)",
353 mdd, mdd->mod_count, mdd->mod_valid,
354 ktime_to_ns(mdd->mod_cltime), mdd->mod_flags);
357 static const struct lu_object_operations mdd_lu_obj_ops = {
358 .loo_object_init = mdd_object_init,
359 .loo_object_start = mdd_object_start,
360 .loo_object_free = mdd_object_free,
361 .loo_object_print = mdd_object_print,
364 struct mdd_object *mdd_object_find(const struct lu_env *env,
365 struct mdd_device *d,
366 const struct lu_fid *f)
368 return md2mdd_obj(md_object_find_slice(env, &d->mdd_md_dev, f));
372 * No permission check is needed.
374 int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
377 struct mdd_object *mdd_obj = md2mdd_obj(obj);
382 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr);
383 if ((ma->ma_need & MA_INODE) != 0 && mdd_is_dead_obj(mdd_obj))
384 ma->ma_attr.la_nlink = 0;
390 * No permission check is needed.
392 static int mdd_xattr_get(const struct lu_env *env,
393 struct md_object *obj, struct lu_buf *buf,
396 struct mdd_object *mdd_obj = md2mdd_obj(obj);
397 struct mdd_device *mdd;
402 if (mdd_object_exists(mdd_obj) == 0) {
403 CERROR("%s: object "DFID" not found: rc = -2\n",
404 mdd_obj_dev_name(mdd_obj),
405 PFID(mdd_object_fid(mdd_obj)));
409 /* If the object has been destroyed, then do not get LMVEA, because
410 * it needs to load stripes from the iteration of the master object,
411 * and it will cause problem if master object has been destroyed, see
413 if (unlikely((mdd_obj->mod_flags & DEAD_OBJ) &&
414 !(mdd_obj->mod_flags & ORPHAN_OBJ) &&
415 strcmp(name, XATTR_NAME_LMV) == 0))
418 /* If the object has been delete from the namespace, then
419 * get linkEA should return -ENOENT as well */
420 if (unlikely((mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ)) &&
421 strcmp(name, XATTR_NAME_LINK) == 0))
424 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
425 rc = mdo_xattr_get(env, mdd_obj, buf, name);
426 mdd_read_unlock(env, mdd_obj);
430 /* record only getting user xattrs and acls */
432 mdd_changelog_enabled(env, mdd, CL_GETXATTR) &&
433 (has_prefix(name, XATTR_USER_PREFIX) ||
434 has_prefix(name, XATTR_NAME_POSIX_ACL_ACCESS) ||
435 has_prefix(name, XATTR_NAME_POSIX_ACL_DEFAULT))) {
436 struct thandle *handle;
439 LASSERT(mdo2fid(mdd_obj) != NULL);
441 handle = mdd_trans_create(env, mdd);
443 RETURN(PTR_ERR(handle));
445 rc2 = mdd_declare_changelog_store(env, mdd, CL_GETXATTR, NULL,
450 rc2 = mdd_trans_start(env, mdd, handle);
454 rc2 = mdd_changelog_data_store_by_fid(env, mdd, CL_GETXATTR, 0,
455 mdo2fid(mdd_obj), name,
459 rc2 = mdd_trans_stop(env, mdd, rc2, handle);
468 * Permission check is done when open,
469 * no need check again.
471 int mdd_readlink(const struct lu_env *env, struct md_object *obj,
474 struct mdd_object *mdd_obj = md2mdd_obj(obj);
475 struct dt_object *next;
480 if (mdd_object_exists(mdd_obj) == 0) {
481 CERROR("%s: object "DFID" not found: rc = -2\n",
482 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
486 next = mdd_object_child(mdd_obj);
487 LASSERT(next != NULL);
488 LASSERT(next->do_body_ops != NULL);
489 LASSERT(next->do_body_ops->dbo_read != NULL);
490 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
491 rc = dt_read(env, next, buf, &pos);
492 mdd_read_unlock(env, mdd_obj);
497 * No permission check is needed.
499 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
502 struct mdd_object *mdd_obj = md2mdd_obj(obj);
507 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
508 rc = mdo_xattr_list(env, mdd_obj, buf);
509 mdd_read_unlock(env, mdd_obj);
511 /* If the buffer is NULL then we are only here to get the
512 * length of the xattr name list. */
513 if (rc < 0 || buf->lb_buf == NULL)
517 * Filter out XATTR_NAME_LINK if this is an orphan object. See
520 if (unlikely(mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ))) {
521 char *end = (char *)buf->lb_buf + rc;
522 char *p = buf->lb_buf;
525 char *next = p + strlen(p) + 1;
527 if (strcmp(p, XATTR_NAME_LINK) == 0) {
529 memmove(p, next, end - next);
531 CDEBUG(D_INFO, "Filtered out "XATTR_NAME_LINK
532 " of orphan "DFID"\n",
533 PFID(mdd_object_fid(mdd_obj)));
544 int mdd_invalidate(const struct lu_env *env, struct md_object *obj)
546 return mdo_invalidate(env, md2mdd_obj(obj));
549 int mdd_declare_create_object_internal(const struct lu_env *env,
550 struct mdd_object *p,
551 struct mdd_object *c,
552 struct lu_attr *attr,
553 struct thandle *handle,
554 const struct md_op_spec *spec,
555 struct dt_allocation_hint *hint)
557 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
558 const struct dt_index_features *feat = spec->sp_feat;
562 if (feat != &dt_directory_features && feat != NULL) {
563 dof->dof_type = DFT_INDEX;
564 dof->u.dof_idx.di_feat = feat;
566 dof->dof_type = dt_mode_to_dft(attr->la_mode);
567 if (dof->dof_type == DFT_REGULAR) {
568 dof->u.dof_reg.striped =
569 md_should_create(spec->sp_cr_flags);
570 if (spec->sp_cr_flags & MDS_OPEN_HAS_EA)
571 dof->u.dof_reg.striped = 0;
572 /* is this replay? */
574 dof->u.dof_reg.striped = 0;
578 rc = mdo_declare_create_object(env, c, attr, hint, dof, handle);
583 int mdd_create_object_internal(const struct lu_env *env, struct mdd_object *p,
584 struct mdd_object *c, struct lu_attr *attr,
585 struct thandle *handle,
586 const struct md_op_spec *spec,
587 struct dt_allocation_hint *hint)
589 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
593 LASSERT(!mdd_object_exists(c));
595 rc = mdo_create_object(env, c, attr, hint, dof, handle);
600 int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj,
601 const struct lu_attr *attr, struct thandle *handle,
607 rc = mdo_attr_set(env, obj, attr, handle);
608 #ifdef CONFIG_FS_POSIX_ACL
609 if (!rc && (attr->la_valid & LA_MODE) && needacl)
610 rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
615 int mdd_update_time(const struct lu_env *env, struct mdd_object *obj,
616 const struct lu_attr *oattr, struct lu_attr *attr,
617 struct thandle *handle)
622 LASSERT(attr->la_valid & LA_CTIME);
623 LASSERT(oattr != NULL);
625 /* Make sure the ctime is increased only, however, it's not strictly
626 * reliable at here because there is not guarantee to hold lock on
627 * object, so we just bypass some unnecessary cmtime setting first
628 * and OSD has to check it again. */
629 if (attr->la_ctime < oattr->la_ctime)
630 attr->la_valid &= ~(LA_MTIME | LA_CTIME);
631 else if (attr->la_valid == LA_CTIME &&
632 attr->la_ctime == oattr->la_ctime)
633 attr->la_valid &= ~LA_CTIME;
635 if (attr->la_valid != 0)
636 rc = mdd_attr_set_internal(env, obj, attr, handle, 0);
641 * This gives the same functionality as the code between
642 * sys_chmod and inode_setattr
643 * chown_common and inode_setattr
644 * utimes and inode_setattr
645 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
647 static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
648 const struct lu_attr *oattr, struct lu_attr *la,
649 const unsigned long flags)
658 /* Do not permit change file type */
659 if (la->la_valid & LA_TYPE)
662 /* They should not be processed by setattr */
663 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
666 LASSERT(oattr != NULL);
668 uc = lu_ucred_check(env);
672 if (la->la_valid == LA_CTIME) {
673 if (!(flags & MDS_PERM_BYPASS))
674 /* This is only for set ctime when rename's source is
676 rc = mdd_may_delete(env, NULL, NULL, obj, oattr, NULL,
678 if (rc == 0 && la->la_ctime <= oattr->la_ctime)
679 la->la_valid &= ~LA_CTIME;
683 if (la->la_valid == LA_ATIME) {
684 /* This is an atime-only attribute update for close RPCs. */
685 if (la->la_atime < (oattr->la_atime +
686 mdd_obj2mdd_dev(obj)->mdd_atime_diff))
687 la->la_valid &= ~LA_ATIME;
691 /* Check if flags change. */
692 if (la->la_valid & LA_FLAGS) {
693 unsigned int oldflags = oattr->la_flags &
694 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
695 unsigned int newflags = la->la_flags &
696 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
698 if ((uc->uc_fsuid != oattr->la_uid) &&
699 !md_capable(uc, CFS_CAP_FOWNER))
702 /* The IMMUTABLE and APPEND_ONLY flags can
703 * only be changed by the relevant capability. */
704 if ((oldflags ^ newflags) &&
705 !md_capable(uc, CFS_CAP_LINUX_IMMUTABLE))
708 if (!S_ISDIR(oattr->la_mode))
709 la->la_flags &= ~(LUSTRE_DIRSYNC_FL | LUSTRE_TOPDIR_FL);
712 if (oattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL) &&
713 (la->la_valid & ~LA_FLAGS) &&
714 !(flags & MDS_PERM_BYPASS))
717 /* Check for setting the obj time. */
718 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
719 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
720 if ((uc->uc_fsuid != oattr->la_uid) &&
721 !md_capable(uc, CFS_CAP_FOWNER)) {
722 rc = mdd_permission_internal(env, obj, oattr,
729 if (la->la_valid & LA_KILL_SUID) {
730 la->la_valid &= ~LA_KILL_SUID;
731 if ((oattr->la_mode & S_ISUID) &&
732 !(la->la_valid & LA_MODE)) {
733 la->la_mode = oattr->la_mode;
734 la->la_valid |= LA_MODE;
736 la->la_mode &= ~S_ISUID;
739 if (la->la_valid & LA_KILL_SGID) {
740 la->la_valid &= ~LA_KILL_SGID;
741 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
742 (S_ISGID | S_IXGRP)) &&
743 !(la->la_valid & LA_MODE)) {
744 la->la_mode = oattr->la_mode;
745 la->la_valid |= LA_MODE;
747 la->la_mode &= ~S_ISGID;
750 /* Make sure a caller can chmod. */
751 if (la->la_valid & LA_MODE) {
752 if (!(flags & MDS_PERM_BYPASS) &&
753 (uc->uc_fsuid != oattr->la_uid) &&
754 !md_capable(uc, CFS_CAP_FOWNER))
757 if (la->la_mode == (umode_t) -1)
758 la->la_mode = oattr->la_mode;
760 la->la_mode = (la->la_mode & S_IALLUGO) |
761 (oattr->la_mode & ~S_IALLUGO);
763 /* Also check the setgid bit! */
764 if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ?
765 la->la_gid : oattr->la_gid) &&
766 !md_capable(uc, CFS_CAP_FSETID))
767 la->la_mode &= ~S_ISGID;
769 la->la_mode = oattr->la_mode;
772 /* Make sure a caller can chown. */
773 if (la->la_valid & LA_UID) {
774 if (la->la_uid == (uid_t) -1)
775 la->la_uid = oattr->la_uid;
776 if (((uc->uc_fsuid != oattr->la_uid) ||
777 (la->la_uid != oattr->la_uid)) &&
778 !md_capable(uc, CFS_CAP_CHOWN))
781 /* If the user or group of a non-directory has been
782 * changed by a non-root user, remove the setuid bit.
783 * 19981026 David C Niemi <niemi@tux.org>
785 * Changed this to apply to all users, including root,
786 * to avoid some races. This is the behavior we had in
787 * 2.0. The check for non-root was definitely wrong
788 * for 2.2 anyway, as it should have been using
789 * CAP_FSETID rather than fsuid -- 19990830 SD. */
790 if (((oattr->la_mode & S_ISUID) == S_ISUID) &&
791 !S_ISDIR(oattr->la_mode)) {
792 la->la_mode &= ~S_ISUID;
793 la->la_valid |= LA_MODE;
797 /* Make sure caller can chgrp. */
798 if (la->la_valid & LA_GID) {
799 if (la->la_gid == (gid_t) -1)
800 la->la_gid = oattr->la_gid;
801 if (((uc->uc_fsuid != oattr->la_uid) ||
802 ((la->la_gid != oattr->la_gid) &&
803 !lustre_in_group_p(uc, la->la_gid))) &&
804 !md_capable(uc, CFS_CAP_CHOWN))
807 /* Likewise, if the user or group of a non-directory
808 * has been changed by a non-root user, remove the
809 * setgid bit UNLESS there is no group execute bit
810 * (this would be a file marked for mandatory
811 * locking). 19981026 David C Niemi <niemi@tux.org>
813 * Removed the fsuid check (see the comment above) --
815 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
816 (S_ISGID | S_IXGRP)) && !S_ISDIR(oattr->la_mode)) {
817 la->la_mode &= ~S_ISGID;
818 la->la_valid |= LA_MODE;
822 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
823 if (!((flags & MDS_OWNEROVERRIDE) &&
824 (uc->uc_fsuid == oattr->la_uid)) &&
825 !(flags & MDS_PERM_BYPASS)) {
826 rc = mdd_permission_internal(env, obj, oattr,
833 if (la->la_valid & LA_CTIME) {
834 /* The pure setattr, it has the priority over what is
835 * already set, do not drop it if ctime is equal. */
836 if (la->la_ctime < oattr->la_ctime)
837 la->la_valid &= ~(LA_ATIME | LA_MTIME | LA_CTIME);
843 static int mdd_changelog_data_store_by_fid(const struct lu_env *env,
844 struct mdd_device *mdd,
845 enum changelog_rec_type type,
846 int flags, const struct lu_fid *fid,
847 const char *xattr_name,
848 struct thandle *handle)
850 const struct lu_ucred *uc = lu_ucred(env);
851 struct llog_changelog_rec *rec;
854 int xflags = CLFE_INVALID;
857 flags = (flags & CLF_FLAGMASK) | CLF_VERSION | CLF_EXTRA_FLAGS;
860 if (uc->uc_jobid[0] != '\0')
862 xflags |= CLFE_UIDGID;
865 if (type == CL_OPEN || type == CL_DN_OPEN)
867 if (type == CL_SETXATTR || type == CL_GETXATTR)
868 xflags |= CLFE_XATTR;
870 reclen = llog_data_len(LLOG_CHANGELOG_HDR_SZ +
871 changelog_rec_offset(flags & CLF_SUPPORTED,
872 xflags & CLFE_SUPPORTED));
873 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
874 if (buf->lb_buf == NULL)
878 rec->cr_hdr.lrh_len = reclen;
879 rec->cr.cr_flags = flags;
880 rec->cr.cr_type = (__u32)type;
881 rec->cr.cr_tfid = *fid;
882 rec->cr.cr_namelen = 0;
884 if (flags & CLF_JOBID)
885 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
887 if (flags & CLF_EXTRA_FLAGS) {
888 mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
889 if (xflags & CLFE_UIDGID)
890 mdd_changelog_rec_extra_uidgid(&rec->cr,
891 uc->uc_uid, uc->uc_gid);
892 if (xflags & CLFE_NID)
893 mdd_changelog_rec_extra_nid(&rec->cr, uc->uc_nid);
894 if (xflags & CLFE_OPEN)
895 mdd_changelog_rec_extra_omode(&rec->cr, flags);
896 if (xflags & CLFE_XATTR) {
897 if (xattr_name == NULL)
899 mdd_changelog_rec_extra_xattr(&rec->cr, xattr_name);
903 rc = mdd_changelog_store(env, mdd, rec, handle);
908 /** Store a data change changelog record
909 * If this fails, we must fail the whole transaction; we don't
910 * want the change to commit without the log entry.
911 * \param mdd_obj - mdd_object of change
912 * \param handle - transaction handle
914 int mdd_changelog_data_store(const struct lu_env *env, struct mdd_device *mdd,
915 enum changelog_rec_type type, int flags,
916 struct mdd_object *mdd_obj, struct thandle *handle)
920 LASSERT(mdd_obj != NULL);
921 LASSERT(handle != NULL);
923 if (!mdd_changelog_enabled(env, mdd, type))
926 if (mdd_is_volatile_obj(mdd_obj))
929 if ((type >= CL_MTIME) && (type <= CL_ATIME) &&
930 ktime_before(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) {
931 /* Don't need multiple updates in this log */
932 /* Don't check under lock - no big deal if we get an extra
937 rc = mdd_changelog_data_store_by_fid(env, mdd, type, flags,
938 mdo2fid(mdd_obj), NULL, handle);
940 mdd_obj->mod_cltime = ktime_get();
945 static int mdd_changelog_data_store_xattr(const struct lu_env *env,
946 struct mdd_device *mdd,
947 enum changelog_rec_type type,
948 int flags, struct mdd_object *mdd_obj,
949 const char *xattr_name,
950 struct thandle *handle)
954 LASSERT(mdd_obj != NULL);
955 LASSERT(handle != NULL);
957 if (!mdd_changelog_enabled(env, mdd, type))
960 if (mdd_is_volatile_obj(mdd_obj))
963 if ((type >= CL_MTIME) && (type <= CL_ATIME) &&
964 ktime_before(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) {
965 /* Don't need multiple updates in this log */
966 /* Don't check under lock - no big deal if we get an extra
972 rc = mdd_changelog_data_store_by_fid(env, mdd, type, flags,
973 mdo2fid(mdd_obj), xattr_name,
976 mdd_obj->mod_cltime = ktime_get();
981 static int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type,
982 int flags, struct md_device *m, const struct lu_fid *fid)
984 struct thandle *handle;
985 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
989 LASSERT(fid != NULL);
991 /* We'll check this again below, but we check now before we
992 * start a transaction. */
993 if (!mdd_changelog_enabled(env, mdd, type))
996 handle = mdd_trans_create(env, mdd);
998 RETURN(PTR_ERR(handle));
1000 rc = mdd_declare_changelog_store(env, mdd, type, NULL, NULL, handle);
1004 rc = mdd_trans_start(env, mdd, handle);
1008 rc = mdd_changelog_data_store_by_fid(env, mdd, type, flags,
1012 rc = mdd_trans_stop(env, mdd, rc, handle);
1018 * Save LMA extended attributes with data from \a ma.
1020 * HSM and Size-On-MDS data will be extracted from \ma if they are valid, if
1021 * not, LMA EA will be first read from disk, modified and write back.
1024 /* Precedence for choosing record type when multiple
1025 * attributes change: setattr > mtime > ctime > atime
1026 * (ctime changes when mtime does, plus chmod/chown.
1027 * atime and ctime are independent.) */
1028 static int mdd_attr_set_changelog(const struct lu_env *env,
1029 struct md_object *obj, struct thandle *handle,
1032 struct mdd_device *mdd = mdo2mdd(obj);
1035 bits = (valid & LA_SIZE) ? 1 << CL_TRUNC : 0;
1036 bits |= (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0;
1037 bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0;
1038 bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0;
1039 bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0;
1040 bits = bits & mdd->mdd_cl.mc_mask;
1041 /* This is an implementation limit rather than a protocol limit */
1042 CLASSERT(CL_LAST <= sizeof(int) * 8);
1046 /* The record type is the lowest non-masked set bit */
1049 /* FYI we only store the first CLF_FLAGMASK bits of la_valid */
1050 return mdd_changelog_data_store(env, mdd, type, (int)valid,
1051 md2mdd_obj(obj), handle);
1054 static int mdd_declare_attr_set(const struct lu_env *env,
1055 struct mdd_device *mdd,
1056 struct mdd_object *obj,
1057 const struct lu_attr *attr,
1058 struct thandle *handle)
1062 rc = mdo_declare_attr_set(env, obj, attr, handle);
1066 #ifdef CONFIG_FS_POSIX_ACL
1067 if (attr->la_valid & LA_MODE) {
1068 mdd_read_lock(env, obj, MOR_TGT_CHILD);
1069 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL,
1070 XATTR_NAME_ACL_ACCESS);
1071 mdd_read_unlock(env, obj);
1072 if (rc == -EOPNOTSUPP || rc == -ENODATA)
1078 struct lu_buf *buf = mdd_buf_get(env, NULL, rc);
1079 rc = mdo_declare_xattr_set(env, obj, buf,
1080 XATTR_NAME_ACL_ACCESS, 0,
1088 rc = mdd_declare_changelog_store(env, mdd, CL_SETXATTR, NULL, NULL,
1096 * permission changes may require sync operation, to mitigate performance
1097 * impact, only do this for dir and when permission is reduced.
1099 * For regular files, version is updated with permission change (see VBR), async
1100 * permission won't cause any issue, while missing permission change on
1101 * directory may affect accessibility of other objects after recovery.
1103 static inline bool permission_needs_sync(const struct lu_attr *old,
1104 const struct lu_attr *new)
1106 if (!S_ISDIR(old->la_mode))
1109 if (new->la_valid & (LA_UID | LA_GID))
1112 if (new->la_valid & LA_MODE &&
1113 new->la_mode & (S_ISUID | S_ISGID | S_ISVTX))
1116 if ((new->la_valid & LA_MODE) &&
1117 ((new->la_mode & old->la_mode) & S_IRWXUGO) !=
1118 (old->la_mode & S_IRWXUGO))
1124 /* set attr and LOV EA at once, return updated attr */
1125 int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
1126 const struct md_attr *ma)
1128 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1129 struct mdd_device *mdd = mdo2mdd(obj);
1130 struct thandle *handle = NULL;
1131 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
1132 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1133 const struct lu_attr *la = &ma->ma_attr;
1134 struct lu_ucred *uc;
1138 /* we do not use ->attr_set() for LOV/HSM EA any more */
1139 LASSERT((ma->ma_valid & MA_LOV) == 0);
1140 LASSERT((ma->ma_valid & MA_HSM) == 0);
1142 rc = mdd_la_get(env, mdd_obj, attr);
1146 *la_copy = ma->ma_attr;
1147 rc = mdd_fix_attr(env, mdd_obj, attr, la_copy, ma->ma_attr_flags);
1151 /* no need to setattr anymore */
1152 if (la_copy->la_valid == 0) {
1153 CDEBUG(D_INODE, "%s: no valid attribute on "DFID", previous"
1154 "valid is %#llx\n", mdd2obd_dev(mdd)->obd_name,
1155 PFID(mdo2fid(mdd_obj)), la->la_valid);
1160 /* If an unprivileged user changes group of some file,
1161 * the setattr operation will be processed synchronously to
1162 * honor the quota limit of the corresponding group. see LU-5152 */
1163 uc = lu_ucred_check(env);
1164 if (S_ISREG(attr->la_mode) && la->la_valid & LA_GID &&
1165 la->la_gid != attr->la_gid && uc != NULL && uc->uc_fsuid != 0) {
1166 la_copy->la_valid |= LA_FLAGS;
1167 la_copy->la_flags |= LUSTRE_SET_SYNC_FL;
1169 /* Flush the possible existing sync requests to OSTs to
1170 * keep the order of sync for the current setattr operation
1171 * will be sent directly to OSTs. see LU-5152 */
1172 rc = dt_sync(env, mdd->mdd_child);
1177 handle = mdd_trans_create(env, mdd);
1178 if (IS_ERR(handle)) {
1179 rc = PTR_ERR(handle);
1185 rc = mdd_declare_attr_set(env, mdd, mdd_obj, la_copy, handle);
1189 rc = mdd_trans_start(env, mdd, handle);
1193 if (mdd->mdd_sync_permission && permission_needs_sync(attr, la))
1194 handle->th_sync = 1;
1196 if (la->la_valid & (LA_MTIME | LA_CTIME))
1197 CDEBUG(D_INODE, "setting mtime %llu, ctime %llu\n",
1198 la->la_mtime, la->la_ctime);
1200 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1201 if (la_copy->la_valid) {
1202 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
1204 if (rc == -EDQUOT && la_copy->la_flags & LUSTRE_SET_SYNC_FL) {
1205 /* rollback to the original gid */
1206 la_copy->la_flags &= ~LUSTRE_SET_SYNC_FL;
1207 la_copy->la_gid = attr->la_gid;
1208 mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
1211 mdd_write_unlock(env, mdd_obj);
1215 rc = mdd_attr_set_changelog(env, obj, handle,
1219 rc = mdd_trans_stop(env, mdd, rc, handle);
1224 static int mdd_xattr_sanity_check(const struct lu_env *env,
1225 struct mdd_object *obj,
1226 const struct lu_attr *attr,
1229 struct lu_ucred *uc = lu_ucred_assert(env);
1232 if (attr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
1235 if (strncmp(XATTR_USER_PREFIX, name,
1236 sizeof(XATTR_USER_PREFIX) - 1) == 0) {
1237 /* For sticky directories, only the owner and privileged user
1238 * can write attributes. */
1239 if (S_ISDIR(attr->la_mode) && (attr->la_mode & S_ISVTX) &&
1240 (uc->uc_fsuid != attr->la_uid) &&
1241 !md_capable(uc, CFS_CAP_FOWNER))
1243 } else if (strcmp(name, XATTR_NAME_SOM) != 0 &&
1244 (uc->uc_fsuid != attr->la_uid) &&
1245 !md_capable(uc, CFS_CAP_FOWNER)) {
1253 * Check if a string begins with a given prefix.
1255 * \param str String to check
1256 * \param prefix Substring to check at the beginning of \a str
1257 * \return true/false whether the condition is verified.
1259 static inline bool has_prefix(const char *str, const char *prefix)
1261 return strncmp(prefix, str, strlen(prefix)) == 0;
1265 * Indicate the kind of changelog to store (if any) for a xattr set/del.
1267 * \param[in] xattr_name Full extended attribute name.
1269 * \return The type of changelog to use, or -1 if no changelog is to be emitted.
1271 static enum changelog_rec_type
1272 mdd_xattr_changelog_type(const struct lu_env *env, struct mdd_device *mdd,
1273 const char *xattr_name)
1275 /* Layout changes systematically recorded */
1276 if (strcmp(XATTR_NAME_LOV, xattr_name) == 0 ||
1277 strncmp(XATTR_LUSTRE_LOV, xattr_name,
1278 strlen(XATTR_LUSTRE_LOV)) == 0)
1281 /* HSM information changes systematically recorded */
1282 if (strcmp(XATTR_NAME_HSM, xattr_name) == 0)
1285 if (has_prefix(xattr_name, XATTR_USER_PREFIX) ||
1286 has_prefix(xattr_name, XATTR_NAME_POSIX_ACL_ACCESS) ||
1287 has_prefix(xattr_name, XATTR_NAME_POSIX_ACL_DEFAULT) ||
1288 has_prefix(xattr_name, XATTR_TRUSTED_PREFIX) ||
1289 has_prefix(xattr_name, XATTR_SECURITY_PREFIX))
1295 static int mdd_declare_xattr_set(const struct lu_env *env,
1296 struct mdd_device *mdd,
1297 struct mdd_object *obj,
1298 const struct lu_buf *buf,
1300 int fl, struct thandle *handle)
1302 enum changelog_rec_type type;
1305 rc = mdo_declare_xattr_set(env, obj, buf, name, fl, handle);
1309 type = mdd_xattr_changelog_type(env, mdd, name);
1311 return 0; /* no changelog to store */
1313 return mdd_declare_changelog_store(env, mdd, type, NULL, NULL, handle);
1317 * Compare current and future data of HSM EA and add a changelog if needed.
1319 * Caller should have write-locked \param obj.
1321 * \param buf - Future HSM EA content.
1322 * \retval 0 if no changelog is needed or changelog was added properly.
1323 * \retval -ve errno if there was a problem
1325 static int mdd_hsm_update_locked(const struct lu_env *env,
1326 struct md_object *obj,
1327 const struct lu_buf *buf,
1328 struct thandle *handle, int *cl_flags)
1330 struct mdd_thread_info *info = mdd_env_info(env);
1331 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1332 struct lu_buf *current_buf;
1333 struct md_hsm *current_mh;
1334 struct md_hsm *new_mh;
1338 OBD_ALLOC_PTR(current_mh);
1339 if (current_mh == NULL)
1342 /* Read HSM attrs from disk */
1343 current_buf = lu_buf_check_and_alloc(&info->mti_xattr_buf,
1344 mdo2mdd(obj)->mdd_dt_conf.ddp_max_ea_size);
1345 rc = mdo_xattr_get(env, mdd_obj, current_buf, XATTR_NAME_HSM);
1346 rc = lustre_buf2hsm(current_buf->lb_buf, rc, current_mh);
1347 if (rc < 0 && rc != -ENODATA)
1349 else if (rc == -ENODATA)
1350 current_mh->mh_flags = 0;
1352 /* Map future HSM xattr */
1353 OBD_ALLOC_PTR(new_mh);
1355 GOTO(free, rc = -ENOMEM);
1356 lustre_buf2hsm(buf->lb_buf, buf->lb_len, new_mh);
1360 /* Flags differ, set flags for the changelog that will be added */
1361 if (current_mh->mh_flags != new_mh->mh_flags) {
1362 hsm_set_cl_event(cl_flags, HE_STATE);
1363 if (new_mh->mh_flags & HS_DIRTY)
1364 hsm_set_cl_flags(cl_flags, CLF_HSM_DIRTY);
1367 OBD_FREE_PTR(new_mh);
1370 OBD_FREE_PTR(current_mh);
1374 static int mdd_object_pfid_replace(const struct lu_env *env,
1375 struct mdd_object *o)
1377 struct mdd_device *mdd = mdo2mdd(&o->mod_obj);
1378 struct thandle *handle;
1381 handle = mdd_trans_create(env, mdd);
1383 RETURN(PTR_ERR(handle));
1385 handle->th_complex = 1;
1387 /* it doesn't need to track the PFID update via llog, because LFSCK
1388 * will repair it even it goes wrong */
1389 rc = mdd_declare_xattr_set(env, mdd, o, NULL, XATTR_NAME_FID,
1394 rc = mdd_trans_start(env, mdd, handle);
1398 rc = mdo_xattr_set(env, o, NULL, XATTR_NAME_FID, 0, handle);
1403 mdd_trans_stop(env, mdd, rc, handle);
1408 static int mdd_declare_xattr_del(const struct lu_env *env,
1409 struct mdd_device *mdd,
1410 struct mdd_object *obj,
1412 struct thandle *handle);
1414 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1417 static int mdd_xattr_merge(const struct lu_env *env, struct md_object *md_obj,
1418 struct md_object *md_vic)
1420 struct mdd_device *mdd = mdo2mdd(md_obj);
1421 struct mdd_object *obj = md2mdd_obj(md_obj);
1422 struct mdd_object *vic = md2mdd_obj(md_vic);
1423 struct lu_buf *buf = &mdd_env_info(env)->mti_buf[0];
1424 struct lu_buf *buf_vic = &mdd_env_info(env)->mti_buf[1];
1425 struct lov_mds_md *lmm;
1426 struct thandle *handle;
1430 rc = lu_fid_cmp(mdo2fid(obj), mdo2fid(vic));
1431 if (rc == 0) /* same fid */
1434 handle = mdd_trans_create(env, mdd);
1436 RETURN(PTR_ERR(handle));
1439 mdd_write_lock(env, obj, MOR_TGT_CHILD);
1440 mdd_write_lock(env, vic, MOR_TGT_CHILD);
1442 mdd_write_lock(env, vic, MOR_TGT_CHILD);
1443 mdd_write_lock(env, obj, MOR_TGT_CHILD);
1446 /* get EA of victim file */
1447 memset(buf_vic, 0, sizeof(*buf_vic));
1448 rc = mdd_get_lov_ea(env, vic, buf_vic);
1455 /* parse the layout of victim file */
1456 lmm = buf_vic->lb_buf;
1457 if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
1458 GOTO(out, rc = -EINVAL);
1460 /* save EA of target file for restore */
1461 memset(buf, 0, sizeof(*buf));
1462 rc = mdd_get_lov_ea(env, obj, buf);
1466 /* Get rid of the layout from victim object */
1467 rc = mdd_declare_xattr_del(env, mdd, vic, XATTR_NAME_LOV, handle);
1471 rc = mdd_declare_xattr_set(env, mdd, obj, buf_vic, XATTR_LUSTRE_LOV,
1472 LU_XATTR_MERGE, handle);
1476 rc = mdd_trans_start(env, mdd, handle);
1480 rc = mdo_xattr_set(env, obj, buf_vic, XATTR_LUSTRE_LOV, LU_XATTR_MERGE,
1485 rc = mdo_xattr_del(env, vic, XATTR_NAME_LOV, handle);
1487 GOTO(out_restore, rc);
1489 (void)mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, obj, handle);
1490 (void)mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, vic, handle);
1495 int rc2 = mdo_xattr_set(env, obj, buf, XATTR_NAME_LOV,
1496 LU_XATTR_REPLACE, handle);
1498 CERROR("%s: failed to rollback of layout of: "DFID
1499 ": %d, file state unknown\n",
1500 mdd_obj_dev_name(obj), PFID(mdo2fid(obj)), rc2);
1504 mdd_trans_stop(env, mdd, rc, handle);
1505 mdd_write_unlock(env, obj);
1506 mdd_write_unlock(env, vic);
1508 lu_buf_free(buf_vic);
1511 (void) mdd_object_pfid_replace(env, obj);
1517 * Extract the mirror with specified mirror id, and store the splitted
1518 * mirror layout to @buf.
1520 * \param[in] comp_v1 mirrored layout
1521 * \param[in] mirror_id the mirror with mirror_id to be extracted
1522 * \param[out] buf store the layout excluding the extracted mirror,
1523 * caller free the buffer we allocated in this function
1524 * \param[out] buf_vic store the extracted layout, caller free the buffer
1525 * we allocated in this function
1527 * \retval 0 on success; < 0 if error happens
1529 static int mdd_split_ea(struct lov_comp_md_v1 *comp_v1, __u16 mirror_id,
1530 struct lu_buf *buf, struct lu_buf *buf_vic)
1532 struct lov_comp_md_v1 *comp_rem;
1533 struct lov_comp_md_v1 *comp_vic;
1534 struct lov_comp_md_entry_v1 *entry;
1535 struct lov_comp_md_entry_v1 *entry_rem;
1536 struct lov_comp_md_entry_v1 *entry_vic;
1538 __u16 comp_cnt, count = 0;
1539 int lmm_size, lmm_size_vic = 0;
1541 int offset, offset_rem, offset_vic;
1543 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1544 /* comp_v1 should contains more than 1 mirror */
1545 if (mirror_cnt <= 1)
1547 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1548 lmm_size = le32_to_cpu(comp_v1->lcm_size);
1550 for (i = 0; i < comp_cnt; i++) {
1551 entry = &comp_v1->lcm_entries[i];
1552 if (mirror_id_of(le32_to_cpu(entry->lcme_id)) == mirror_id) {
1554 lmm_size_vic += sizeof(*entry);
1555 lmm_size_vic += le32_to_cpu(entry->lcme_size);
1556 } else if (count > 0) {
1557 /* find the specified mirror */
1565 lu_buf_alloc(buf, lmm_size - lmm_size_vic);
1569 lu_buf_alloc(buf_vic, sizeof(*comp_vic) + lmm_size_vic);
1570 if (!buf_vic->lb_buf) {
1575 comp_rem = (struct lov_comp_md_v1 *)buf->lb_buf;
1576 comp_vic = (struct lov_comp_md_v1 *)buf_vic->lb_buf;
1578 memcpy(comp_rem, comp_v1, sizeof(*comp_v1));
1579 comp_rem->lcm_mirror_count = cpu_to_le16(mirror_cnt - 2);
1580 comp_rem->lcm_entry_count = cpu_to_le32(comp_cnt - count);
1581 comp_rem->lcm_size = cpu_to_le32(lmm_size - lmm_size_vic);
1582 if (!comp_rem->lcm_mirror_count)
1583 comp_rem->lcm_flags = cpu_to_le16(LCM_FL_NONE);
1585 memset(comp_vic, 0, sizeof(*comp_v1));
1586 comp_vic->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1587 comp_vic->lcm_mirror_count = 0;
1588 comp_vic->lcm_entry_count = cpu_to_le32(count);
1589 comp_vic->lcm_size = cpu_to_le32(lmm_size_vic + sizeof(*comp_vic));
1590 comp_vic->lcm_flags = cpu_to_le16(LCM_FL_NONE);
1591 comp_vic->lcm_layout_gen = 0;
1593 offset = sizeof(*comp_v1) + sizeof(*entry) * comp_cnt;
1594 offset_rem = sizeof(*comp_rem) +
1595 sizeof(*entry_rem) * (comp_cnt - count);
1596 offset_vic = sizeof(*comp_vic) + sizeof(*entry_vic) * count;
1597 for (i = j = k = 0; i < comp_cnt; i++) {
1598 struct lov_mds_md *lmm, *lmm_dst;
1601 entry = &comp_v1->lcm_entries[i];
1602 entry_vic = &comp_vic->lcm_entries[j];
1603 entry_rem = &comp_rem->lcm_entries[k];
1605 if (mirror_id_of(le32_to_cpu(entry->lcme_id)) == mirror_id)
1608 /* copy component entry */
1610 memcpy(entry_vic, entry, sizeof(*entry));
1611 entry_vic->lcme_flags &= cpu_to_le32(LCME_FL_INIT);
1612 entry_vic->lcme_offset = cpu_to_le32(offset_vic);
1615 memcpy(entry_rem, entry, sizeof(*entry));
1616 entry_rem->lcme_offset = cpu_to_le32(offset_rem);
1620 lmm = (struct lov_mds_md *)((char *)comp_v1 + offset);
1622 lmm_dst = (struct lov_mds_md *)
1623 ((char *)comp_vic + offset_vic);
1625 lmm_dst = (struct lov_mds_md *)
1626 ((char *)comp_rem + offset_rem);
1628 /* copy component entry blob */
1629 memcpy(lmm_dst, lmm, le32_to_cpu(entry->lcme_size));
1631 /* blob offset advance */
1632 offset += le32_to_cpu(entry->lcme_size);
1634 offset_vic += le32_to_cpu(entry->lcme_size);
1636 offset_rem += le32_to_cpu(entry->lcme_size);
1642 static int mdd_xattr_split(const struct lu_env *env, struct md_object *md_obj,
1643 struct md_rejig_data *mrd)
1645 struct mdd_device *mdd = mdo2mdd(md_obj);
1646 struct mdd_object *obj = md2mdd_obj(md_obj);
1647 struct mdd_object *vic = md2mdd_obj(mrd->mrd_obj);
1648 struct lu_buf *buf = &mdd_env_info(env)->mti_buf[0];
1649 struct lu_buf *buf_save = &mdd_env_info(env)->mti_buf[1];
1650 struct lu_buf *buf_vic = &mdd_env_info(env)->mti_buf[2];
1651 struct lov_comp_md_v1 *lcm;
1652 struct thandle *handle;
1656 rc = lu_fid_cmp(mdo2fid(obj), mdo2fid(vic));
1657 if (rc == 0) /* same fid */
1660 handle = mdd_trans_create(env, mdd);
1662 RETURN(PTR_ERR(handle));
1665 mdd_write_lock(env, obj, MOR_TGT_CHILD);
1666 mdd_write_lock(env, vic, MOR_TGT_CHILD);
1668 mdd_write_lock(env, vic, MOR_TGT_CHILD);
1669 mdd_write_lock(env, obj, MOR_TGT_CHILD);
1672 /* get EA of mirrored file */
1673 memset(buf_save, 0, sizeof(*buf));
1674 rc = mdd_get_lov_ea(env, obj, buf_save);
1678 lcm = buf_save->lb_buf;
1679 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
1680 GOTO(out, rc = -EINVAL);
1683 * Extract the mirror with specified mirror id, and store the splitted
1684 * mirror layout to the victim file.
1686 memset(buf, 0, sizeof(*buf));
1687 memset(buf_vic, 0, sizeof(*buf_vic));
1688 rc = mdd_split_ea(lcm, mrd->mrd_mirror_id, buf, buf_vic);
1692 rc = mdd_declare_xattr_set(env, mdd, obj, buf, XATTR_NAME_LOV,
1693 LU_XATTR_SPLIT, handle);
1696 rc = mdd_declare_xattr_set(env, mdd, vic, buf_vic, XATTR_NAME_LOV,
1697 LU_XATTR_SPLIT, handle);
1701 rc = mdd_trans_start(env, mdd, handle);
1705 rc = mdo_xattr_set(env, obj, buf, XATTR_NAME_LOV, LU_XATTR_REPLACE,
1710 rc = mdo_xattr_set(env, vic, buf_vic, XATTR_NAME_LOV, LU_XATTR_CREATE,
1713 GOTO(out_restore, rc);
1715 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, obj, handle);
1719 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, vic, handle);
1726 /* restore obj's layout */
1727 int rc2 = mdo_xattr_set(env, obj, buf_save, XATTR_NAME_LOV,
1728 LU_XATTR_REPLACE, handle);
1730 CERROR("%s: failed to rollback of layout of: "DFID
1731 ": %d, file state unkonwn.\n",
1732 mdd_obj_dev_name(obj), PFID(mdo2fid(obj)), rc2);
1735 mdd_trans_stop(env, mdd, rc, handle);
1736 mdd_write_unlock(env, obj);
1737 mdd_write_unlock(env, vic);
1738 lu_buf_free(buf_save);
1740 lu_buf_free(buf_vic);
1743 (void) mdd_object_pfid_replace(env, obj);
1748 static int mdd_layout_merge_allowed(const struct lu_env *env,
1749 struct md_object *target,
1750 struct md_object *victim)
1752 struct mdd_object *o1 = md2mdd_obj(target);
1754 /* cannot extend directory's LOVEA */
1755 if (S_ISDIR(mdd_object_type(o1))) {
1756 CERROR("%s: Don't extend directory's LOVEA, just set it.\n",
1757 mdd_obj_dev_name(o1));
1765 * The caller should guarantee to update the object ctime
1766 * after xattr_set if needed.
1768 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
1769 const struct lu_buf *buf, const char *name,
1772 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1773 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1774 struct mdd_device *mdd = mdo2mdd(obj);
1775 struct thandle *handle;
1776 enum changelog_rec_type cl_type;
1781 rc = mdd_la_get(env, mdd_obj, attr);
1785 rc = mdd_xattr_sanity_check(env, mdd_obj, attr, name);
1789 if (strcmp(name, XATTR_LUSTRE_LOV) == 0 &&
1790 (fl == LU_XATTR_MERGE || fl == LU_XATTR_SPLIT)) {
1791 struct md_rejig_data *mrd = buf->lb_buf;
1792 struct md_object *victim = mrd->mrd_obj;
1794 if (buf->lb_len != sizeof(*mrd))
1797 rc = mdd_layout_merge_allowed(env, obj, victim);
1801 if (fl == LU_XATTR_MERGE)
1802 /* merge layout of victim as a mirror of obj's. */
1803 rc = mdd_xattr_merge(env, obj, victim);
1805 rc = mdd_xattr_split(env, obj, mrd);
1809 if (strcmp(name, XATTR_NAME_ACL_ACCESS) == 0 ||
1810 strcmp(name, XATTR_NAME_ACL_DEFAULT) == 0) {
1811 struct posix_acl *acl;
1813 /* user may set empty ACL, which should be treated as removing
1815 acl = posix_acl_from_xattr(&init_user_ns, buf->lb_buf,
1818 RETURN(PTR_ERR(acl));
1820 rc = mdd_xattr_del(env, obj, name);
1823 posix_acl_release(acl);
1826 if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) {
1827 rc = mdd_acl_set(env, mdd_obj, attr, buf, fl);
1831 handle = mdd_trans_create(env, mdd);
1833 RETURN(PTR_ERR(handle));
1835 rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, fl, handle);
1839 rc = mdd_trans_start(env, mdd, handle);
1843 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1845 if (strcmp(XATTR_NAME_HSM, name) == 0) {
1846 rc = mdd_hsm_update_locked(env, obj, buf, handle, &cl_flags);
1848 mdd_write_unlock(env, mdd_obj);
1853 rc = mdo_xattr_set(env, mdd_obj, buf, name, fl, handle);
1854 mdd_write_unlock(env, mdd_obj);
1858 cl_type = mdd_xattr_changelog_type(env, mdd, name);
1862 rc = mdd_changelog_data_store_xattr(env, mdd, cl_type, cl_flags,
1863 mdd_obj, name, handle);
1867 return mdd_trans_stop(env, mdd, rc, handle);
1870 static int mdd_declare_xattr_del(const struct lu_env *env,
1871 struct mdd_device *mdd,
1872 struct mdd_object *obj,
1874 struct thandle *handle)
1876 enum changelog_rec_type type;
1879 rc = mdo_declare_xattr_del(env, obj, name, handle);
1883 type = mdd_xattr_changelog_type(env, mdd, name);
1885 return 0; /* no changelog to store */
1887 return mdd_declare_changelog_store(env, mdd, type, NULL, NULL, handle);
1891 * The caller should guarantee to update the object ctime
1892 * after xattr_set if needed.
1894 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1897 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1898 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1899 struct mdd_device *mdd = mdo2mdd(obj);
1900 struct thandle *handle;
1904 rc = mdd_la_get(env, mdd_obj, attr);
1908 rc = mdd_xattr_sanity_check(env, mdd_obj, attr, name);
1912 handle = mdd_trans_create(env, mdd);
1914 RETURN(PTR_ERR(handle));
1916 rc = mdd_declare_xattr_del(env, mdd, mdd_obj, name, handle);
1920 rc = mdd_trans_start(env, mdd, handle);
1924 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1925 rc = mdo_xattr_del(env, mdd_obj, name, handle);
1926 mdd_write_unlock(env, mdd_obj);
1930 if (mdd_xattr_changelog_type(env, mdd, name) < 0)
1933 rc = mdd_changelog_data_store_xattr(env, mdd, CL_SETXATTR, 0, mdd_obj,
1938 return mdd_trans_stop(env, mdd, rc, handle);
1942 * read lov EA of an object
1943 * return the lov EA in an allocated lu_buf
1945 int mdd_get_lov_ea(const struct lu_env *env, struct mdd_object *obj,
1946 struct lu_buf *lmm_buf)
1948 struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf;
1953 rc = mdo_xattr_get(env, obj, buf, XATTR_NAME_LOV);
1955 if (rc == -ERANGE) {
1956 /* mti_big_buf is allocated but is too small
1957 * we need to increase it */
1958 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1960 if (buf->lb_buf == NULL)
1961 GOTO(out, rc = -ENOMEM);
1972 if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) {
1973 /* mti_big_buf was not allocated, so we have to
1974 * allocate it based on the ea size */
1975 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1977 if (buf->lb_buf == NULL)
1978 GOTO(out, rc = -ENOMEM);
1982 lu_buf_alloc(lmm_buf, bufsize);
1983 if (lmm_buf->lb_buf == NULL)
1984 GOTO(out, rc = -ENOMEM);
1986 memcpy(lmm_buf->lb_buf, buf->lb_buf, bufsize);
1992 lu_buf_free(lmm_buf);
1996 static int mdd_xattr_hsm_replace(const struct lu_env *env,
1997 struct mdd_object *o, struct lu_buf *buf,
1998 struct thandle *handle)
2000 struct hsm_attrs *attrs;
2006 rc = mdo_xattr_set(env, o, buf, XATTR_NAME_HSM, LU_XATTR_REPLACE,
2011 attrs = buf->lb_buf;
2012 hsm_flags = le32_to_cpu(attrs->hsm_flags);
2013 if (!(hsm_flags & HS_RELEASED) || mdd_is_dead_obj(o))
2016 /* Add a changelog record for release. */
2017 hsm_set_cl_event(&flags, HE_RELEASE);
2018 rc = mdd_changelog_data_store(env, mdo2mdd(&o->mod_obj), CL_HSM,
2024 * check if layout swapping between 2 objects is allowed
2026 * - only normal FIDs or non-system IGIFs
2027 * - same type of objects
2028 * - same owner/group (so quotas are still valid) unless this is from HSM
2031 static int mdd_layout_swap_allowed(const struct lu_env *env,
2032 struct mdd_object *o1,
2033 const struct lu_attr *attr1,
2034 struct mdd_object *o2,
2035 const struct lu_attr *attr2,
2038 const struct lu_fid *fid1, *fid2;
2044 if (!fid_is_norm(fid1) &&
2045 (!fid_is_igif(fid1) || IS_ERR(mdd_links_get(env, o1))))
2048 if (!fid_is_norm(fid2) &&
2049 (!fid_is_igif(fid2) || IS_ERR(mdd_links_get(env, o2))))
2052 if (mdd_object_type(o1) != mdd_object_type(o2)) {
2053 if (S_ISDIR(mdd_object_type(o1)))
2055 if (S_ISREG(mdd_object_type(o1)))
2060 if (flags & SWAP_LAYOUTS_MDS_HSM)
2063 if ((attr1->la_uid != attr2->la_uid) ||
2064 (attr1->la_gid != attr2->la_gid))
2070 /* XXX To set the proper lmm_oi & lmm_layout_gen when swap layouts, we have to
2071 * look into the layout in MDD layer. */
2072 static int mdd_lmm_oi(struct lov_mds_md *lmm, struct ost_id *oi, bool get)
2074 struct lov_comp_md_v1 *comp_v1;
2075 struct lov_mds_md *v1;
2079 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1) {
2080 comp_v1 = (struct lov_comp_md_v1 *)lmm;
2081 ent_count = le16_to_cpu(comp_v1->lcm_entry_count);
2087 off = le32_to_cpu(comp_v1->lcm_entries[0].lcme_offset);
2088 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
2091 for (i = 0; i < le32_to_cpu(ent_count); i++) {
2092 off = le32_to_cpu(comp_v1->lcm_entries[i].
2094 v1 = (struct lov_mds_md *)((char *)comp_v1 +
2099 } else if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1 ||
2100 le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V3) {
2111 static inline int mdd_get_lmm_oi(struct lov_mds_md *lmm, struct ost_id *oi)
2113 return mdd_lmm_oi(lmm, oi, true);
2116 static inline int mdd_set_lmm_oi(struct lov_mds_md *lmm, struct ost_id *oi)
2118 return mdd_lmm_oi(lmm, oi, false);
2121 static int mdd_lmm_gen(struct lov_mds_md *lmm, __u32 *gen, bool get)
2123 struct lov_comp_md_v1 *comp_v1;
2125 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1) {
2126 comp_v1 = (struct lov_comp_md_v1 *)lmm;
2128 *gen = le32_to_cpu(comp_v1->lcm_layout_gen);
2130 comp_v1->lcm_layout_gen = cpu_to_le32(*gen);
2131 } else if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1 ||
2132 le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V3) {
2133 __u16 tmp_gen = *gen;
2135 *gen = le16_to_cpu(lmm->lmm_layout_gen);
2137 lmm->lmm_layout_gen = cpu_to_le16(tmp_gen);
2144 static inline int mdd_get_lmm_gen(struct lov_mds_md *lmm, __u32 *gen)
2146 return mdd_lmm_gen(lmm, gen, true);
2149 static inline int mdd_set_lmm_gen(struct lov_mds_md *lmm, __u32 *gen)
2151 return mdd_lmm_gen(lmm, gen, false);
2154 static inline __u64 mdd_lmm_dom_size(void *buf)
2156 struct lov_mds_md *lmm = buf;
2157 struct lov_comp_md_v1 *comp_v1;
2158 struct lov_mds_md *v1;
2164 if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
2167 comp_v1 = (struct lov_comp_md_v1 *)lmm;
2168 off = le32_to_cpu(comp_v1->lcm_entries[0].lcme_offset);
2169 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
2171 /* DoM entry is the first entry always */
2172 if (lov_pattern(le32_to_cpu(v1->lmm_pattern)) == LOV_PATTERN_MDT)
2173 return le64_to_cpu(comp_v1->lcm_entries[0].lcme_extent.e_end);
2179 * swap layouts between 2 lustre objects
2181 static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1,
2182 struct md_object *obj2, __u64 flags)
2184 struct mdd_thread_info *info = mdd_env_info(env);
2185 struct mdd_object *fst_o = md2mdd_obj(obj1);
2186 struct mdd_object *snd_o = md2mdd_obj(obj2);
2187 struct lu_attr *fst_la = MDD_ENV_VAR(env, cattr);
2188 struct lu_attr *snd_la = MDD_ENV_VAR(env, tattr);
2189 struct mdd_device *mdd = mdo2mdd(obj1);
2190 struct lov_mds_md *fst_lmm, *snd_lmm;
2191 struct lu_buf *fst_buf = &info->mti_buf[0];
2192 struct lu_buf *snd_buf = &info->mti_buf[1];
2193 struct lu_buf *fst_hsm_buf = &info->mti_buf[2];
2194 struct lu_buf *snd_hsm_buf = &info->mti_buf[3];
2195 struct ost_id *saved_oi = NULL;
2196 struct thandle *handle;
2197 __u32 fst_gen, snd_gen, saved_gen;
2203 CLASSERT(ARRAY_SIZE(info->mti_buf) >= 4);
2204 memset(info->mti_buf, 0, sizeof(info->mti_buf));
2206 /* we have to sort the 2 obj, so locking will always
2207 * be in the same order, even in case of 2 concurrent swaps */
2208 rc = lu_fid_cmp(mdo2fid(fst_o), mdo2fid(snd_o));
2209 if (rc == 0) /* same fid ? */
2215 rc = mdd_la_get(env, fst_o, fst_la);
2219 rc = mdd_la_get(env, snd_o, snd_la);
2223 /* check if layout swapping is allowed */
2224 rc = mdd_layout_swap_allowed(env, fst_o, fst_la, snd_o, snd_la, flags);
2228 handle = mdd_trans_create(env, mdd);
2230 RETURN(PTR_ERR(handle));
2232 /* objects are already sorted */
2233 mdd_write_lock(env, fst_o, MOR_TGT_CHILD);
2234 mdd_write_lock(env, snd_o, MOR_TGT_CHILD);
2236 rc = mdd_get_lov_ea(env, fst_o, fst_buf);
2237 if (rc < 0 && rc != -ENODATA)
2240 rc = mdd_get_lov_ea(env, snd_o, snd_buf);
2241 if (rc < 0 && rc != -ENODATA)
2244 /* check if file is DoM, it can be migrated only to another DoM layout
2245 * with the same DoM component size
2247 if (mdd_lmm_dom_size(fst_buf->lb_buf) !=
2248 mdd_lmm_dom_size(snd_buf->lb_buf)) {
2250 CDEBUG(D_LAYOUT, "cannot swap layout for "DFID": new layout "
2251 "must have the same DoM component: rc = %d\n",
2252 PFID(mdo2fid(fst_o)), rc);
2256 /* swapping 2 non existant layouts is a success */
2257 if (fst_buf->lb_buf == NULL && snd_buf->lb_buf == NULL)
2260 /* to help inode migration between MDT, it is better to
2261 * start by the no layout file (if one), so we order the swap */
2262 if (snd_buf->lb_buf == NULL) {
2264 swap(fst_buf, snd_buf);
2267 fst_gen = snd_gen = 0;
2268 /* lmm and generation layout initialization */
2269 if (fst_buf->lb_buf != NULL) {
2270 fst_lmm = fst_buf->lb_buf;
2271 mdd_get_lmm_gen(fst_lmm, &fst_gen);
2272 fst_fl = LU_XATTR_REPLACE;
2276 fst_fl = LU_XATTR_CREATE;
2279 snd_lmm = snd_buf->lb_buf;
2280 mdd_get_lmm_gen(snd_lmm, &snd_gen);
2282 saved_gen = fst_gen;
2283 /* increase the generation layout numbers */
2288 * XXX The layout generation is used to generate component IDs for
2289 * the composite file, we have to do some special tweaks to make
2290 * sure the layout generation is always adequate for that job.
2293 /* Skip invalid generation number for composite layout */
2294 if ((snd_gen & LCME_ID_MASK) == 0)
2296 if ((fst_gen & LCME_ID_MASK) == 0)
2298 /* Make sure the generation is greater than all the component IDs */
2299 if (fst_gen < snd_gen)
2301 else if (fst_gen > snd_gen)
2304 /* set the file specific informations in lmm */
2305 if (fst_lmm != NULL) {
2306 saved_oi = &info->mti_oa.o_oi;
2307 mdd_get_lmm_oi(fst_lmm, saved_oi);
2308 mdd_set_lmm_gen(fst_lmm, &snd_gen);
2309 mdd_set_lmm_oi(fst_lmm, &snd_lmm->lmm_oi);
2310 mdd_set_lmm_oi(snd_lmm, saved_oi);
2312 if ((snd_lmm->lmm_magic & cpu_to_le32(LOV_MAGIC_MASK)) ==
2313 cpu_to_le32(LOV_MAGIC_MAGIC))
2314 snd_lmm->lmm_magic |= cpu_to_le32(LOV_MAGIC_DEFINED);
2316 GOTO(stop, rc = -EPROTO);
2318 mdd_set_lmm_gen(snd_lmm, &fst_gen);
2320 /* Prepare HSM attribute if it's required */
2321 if (flags & SWAP_LAYOUTS_MDS_HSM) {
2322 const int buflen = sizeof(struct hsm_attrs);
2324 lu_buf_alloc(fst_hsm_buf, buflen);
2325 lu_buf_alloc(snd_hsm_buf, buflen);
2326 if (fst_hsm_buf->lb_buf == NULL || snd_hsm_buf->lb_buf == NULL)
2327 GOTO(stop, rc = -ENOMEM);
2329 /* Read HSM attribute */
2330 rc = mdo_xattr_get(env, fst_o, fst_hsm_buf, XATTR_NAME_HSM);
2334 rc = mdo_xattr_get(env, snd_o, snd_hsm_buf, XATTR_NAME_HSM);
2338 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_hsm_buf,
2339 XATTR_NAME_HSM, LU_XATTR_REPLACE,
2344 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_hsm_buf,
2345 XATTR_NAME_HSM, LU_XATTR_REPLACE,
2351 /* prepare transaction */
2352 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_buf, XATTR_NAME_LOV,
2357 if (fst_buf->lb_buf != NULL)
2358 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_buf,
2359 XATTR_NAME_LOV, LU_XATTR_REPLACE,
2362 rc = mdd_declare_xattr_del(env, mdd, snd_o, XATTR_NAME_LOV,
2367 rc = mdd_trans_start(env, mdd, handle);
2371 if (flags & SWAP_LAYOUTS_MDS_HSM) {
2372 rc = mdd_xattr_hsm_replace(env, fst_o, snd_hsm_buf, handle);
2376 rc = mdd_xattr_hsm_replace(env, snd_o, fst_hsm_buf, handle);
2378 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf,
2381 CERROR("%s: restore "DFID" HSM error: %d/%d\n",
2382 mdd_obj_dev_name(fst_o),
2383 PFID(mdo2fid(fst_o)), rc, rc2);
2388 rc = mdo_xattr_set(env, fst_o, snd_buf, XATTR_NAME_LOV, fst_fl, handle);
2392 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_MDS_HSM_SWAP_LAYOUTS))) {
2395 if (fst_buf->lb_buf != NULL)
2396 rc = mdo_xattr_set(env, snd_o, fst_buf, XATTR_NAME_LOV,
2397 LU_XATTR_REPLACE, handle);
2399 rc = mdo_xattr_del(env, snd_o, XATTR_NAME_LOV, handle);
2402 GOTO(out_restore, rc);
2404 /* Issue one changelog record per file */
2405 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle);
2409 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle);
2418 /* failure on second file, but first was done, so we have
2419 * to roll back first. */
2420 if (fst_buf->lb_buf != NULL) {
2421 mdd_set_lmm_oi(fst_lmm, saved_oi);
2422 mdd_set_lmm_gen(fst_lmm, &saved_gen);
2423 rc2 = mdo_xattr_set(env, fst_o, fst_buf, XATTR_NAME_LOV,
2424 LU_XATTR_REPLACE, handle);
2426 rc2 = mdo_xattr_del(env, fst_o, XATTR_NAME_LOV, handle);
2432 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, handle);
2437 rc2 = mdd_xattr_hsm_replace(env, snd_o, snd_hsm_buf, handle);
2442 CERROR("%s: unable to roll back layout swap. FIDs: "
2443 DFID" and "DFID "error: %d/%d, steps: %d\n",
2444 mdd_obj_dev_name(fst_o),
2445 PFID(mdo2fid(snd_o)), PFID(mdo2fid(fst_o)),
2447 /* a solution to avoid journal commit is to panic,
2448 * but it has strong consequences so we use LBUG to
2449 * allow sysdamin to choose to panic or not
2456 rc = mdd_trans_stop(env, mdd, rc, handle);
2458 mdd_write_unlock(env, snd_o);
2459 mdd_write_unlock(env, fst_o);
2461 lu_buf_free(fst_buf);
2462 lu_buf_free(snd_buf);
2463 lu_buf_free(fst_hsm_buf);
2464 lu_buf_free(snd_hsm_buf);
2467 (void) mdd_object_pfid_replace(env, fst_o);
2468 (void) mdd_object_pfid_replace(env, snd_o);
2473 static int mdd_declare_layout_change(const struct lu_env *env,
2474 struct mdd_device *mdd,
2475 struct mdd_object *obj,
2476 struct md_layout_change *mlc,
2477 struct thandle *handle)
2481 rc = mdo_declare_layout_change(env, obj, mlc, handle);
2485 return mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
2489 /* For PFL, this is used to instantiate necessary component objects. */
2491 mdd_layout_instantiate_component(const struct lu_env *env,
2492 struct mdd_object *obj, struct md_layout_change *mlc,
2493 struct thandle *handle)
2495 struct mdd_device *mdd = mdd_obj2mdd_dev(obj);
2499 if (mlc->mlc_opc != MD_LAYOUT_WRITE)
2502 rc = mdd_declare_layout_change(env, mdd, obj, mlc, handle);
2504 * It's possible that another layout write intent has already
2505 * instantiated our objects, so a -EALREADY returned, and we need to
2509 RETURN(rc == -EALREADY ? 0 : rc);
2511 rc = mdd_trans_start(env, mdd, handle);
2515 mdd_write_lock(env, obj, MOR_TGT_CHILD);
2516 rc = mdo_layout_change(env, obj, mlc, handle);
2517 mdd_write_unlock(env, obj);
2521 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, obj, handle);
2526 * Change the FLR layout from RDONLY to WRITE_PENDING.
2528 * It picks the primary mirror, and bumps the layout version, and set
2529 * layout version xattr to OST objects in a sync tx. In order to facilitate
2530 * the handling of phantom writers from evicted clients, the clients carry
2531 * layout version of the file with write RPC, so that the OSTs can verify
2532 * if the write RPCs are legitimate, meaning not from evicted clients.
2535 mdd_layout_update_rdonly(const struct lu_env *env, struct mdd_object *obj,
2536 struct md_layout_change *mlc, struct thandle *handle)
2538 struct mdd_device *mdd = mdd_obj2mdd_dev(obj);
2539 struct lu_buf *som_buf = &mdd_env_info(env)->mti_buf[1];
2540 struct lustre_som_attrs *som = &mlc->mlc_som;
2545 /* Verify acceptable operations */
2546 switch (mlc->mlc_opc) {
2547 case MD_LAYOUT_WRITE:
2548 case MD_LAYOUT_RESYNC:
2549 /* these are legal operations - this represents the case that
2550 * a few mirrors were missed in the last resync. */
2552 case MD_LAYOUT_RESYNC_DONE:
2557 som_buf->lb_buf = som;
2558 som_buf->lb_len = sizeof(*som);
2559 rc = mdo_xattr_get(env, obj, som_buf, XATTR_NAME_SOM);
2560 if (rc < 0 && rc != -ENODATA)
2564 lustre_som_swab(som);
2565 if (som->lsa_valid & SOM_FL_STRICT)
2566 fl = LU_XATTR_REPLACE;
2569 rc = mdd_declare_layout_change(env, mdd, obj, mlc, handle);
2574 rc = mdd_declare_xattr_set(env, mdd, obj, som_buf,
2575 XATTR_NAME_SOM, fl, handle);
2580 /* record a changelog for data mover to consume */
2581 rc = mdd_declare_changelog_store(env, mdd, CL_FLRW, NULL, NULL, handle);
2585 rc = mdd_trans_start(env, mdd, handle);
2589 /* it needs a sync tx to make FLR to work properly */
2590 handle->th_sync = 1;
2592 mdd_write_lock(env, obj, MOR_TGT_CHILD);
2593 rc = mdo_layout_change(env, obj, mlc, handle);
2595 /* SOM state transition from STRICT to STALE */
2596 som->lsa_valid = SOM_FL_STALE;
2597 lustre_som_swab(som);
2598 rc = mdo_xattr_set(env, obj, som_buf, XATTR_NAME_SOM,
2601 mdd_write_unlock(env, obj);
2605 rc = mdd_changelog_data_store(env, mdd, CL_FLRW, 0, obj, handle);
2616 * Handle mirrored file state transition when it's in WRITE_PENDING.
2618 * Only MD_LAYOUT_RESYNC, which represents start of resync, is allowed when
2619 * the file is in WRITE_PENDING state. If everything goes fine, the file's
2620 * layout version will be increased, and the file's state will be changed to
2624 mdd_layout_update_write_pending(const struct lu_env *env,
2625 struct mdd_object *obj, struct md_layout_change *mlc,
2626 struct thandle *handle)
2628 struct mdd_device *mdd = mdd_obj2mdd_dev(obj);
2632 switch (mlc->mlc_opc) {
2633 case MD_LAYOUT_RESYNC:
2634 /* Upon receiving the resync request, it should
2635 * instantiate all stale components right away to get ready
2636 * for mirror copy. In order to avoid layout version change,
2637 * client should avoid sending LAYOUT_WRITE request at the
2640 case MD_LAYOUT_WRITE:
2641 /* legal race for concurrent write, the file state has been
2642 * changed by another client. */
2648 rc = mdd_declare_layout_change(env, mdd, obj, mlc, handle);
2652 rc = mdd_trans_start(env, mdd, handle);
2656 /* it needs a sync tx to make FLR to work properly */
2657 handle->th_sync = 1;
2659 mdd_write_lock(env, obj, MOR_TGT_CHILD);
2660 rc = mdo_layout_change(env, obj, mlc, handle);
2661 mdd_write_unlock(env, obj);
2672 * Handle the requests when a FLR file's state is in SYNC_PENDING.
2674 * Only concurrent write and sync complete requests are possible when the
2675 * file is in SYNC_PENDING. For the latter request, it will pass in the
2676 * mirrors that have been synchronized, then the stale bit will be cleared
2677 * to make the file's state turn into RDONLY.
2678 * For concurrent write reqeust, it just needs to change the file's state
2679 * to WRITE_PENDING in a sync tx. It doesn't have to change the layout
2680 * version because the version will be increased in the transition to
2681 * SYNC_PENDING later so that it can deny the write request from potential
2682 * evicted SYNC clients. */
2684 mdd_object_update_sync_pending(const struct lu_env *env, struct mdd_object *obj,
2685 struct md_layout_change *mlc, struct thandle *handle)
2687 struct mdd_device *mdd = mdd_obj2mdd_dev(obj);
2688 struct lu_buf *som_buf = &mdd_env_info(env)->mti_buf[1];
2693 /* operation validation */
2694 switch (mlc->mlc_opc) {
2695 case MD_LAYOUT_RESYNC_DONE:
2696 /* resync complete. */
2697 case MD_LAYOUT_WRITE:
2698 /* concurrent write. */
2700 case MD_LAYOUT_RESYNC:
2701 /* resync again, most likely the previous run failed.
2702 * no-op if it's already in SYNC_PENDING state */
2708 if (mlc->mlc_som.lsa_valid & SOM_FL_STRICT) {
2709 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_SOM);
2710 if (rc < 0 && rc != -ENODATA)
2713 fl = rc == -ENODATA ? LU_XATTR_CREATE : LU_XATTR_REPLACE;
2714 lustre_som_swab(&mlc->mlc_som);
2715 som_buf->lb_buf = &mlc->mlc_som;
2716 som_buf->lb_len = sizeof(mlc->mlc_som);
2719 rc = mdd_declare_layout_change(env, mdd, obj, mlc, handle);
2723 /* record a changelog for the completion of resync */
2724 rc = mdd_declare_changelog_store(env, mdd, CL_RESYNC, NULL, NULL,
2729 /* RESYNC_DONE has piggybacked size and blocks */
2731 rc = mdd_declare_xattr_set(env, mdd, obj, som_buf,
2732 XATTR_NAME_SOM, fl, handle);
2737 rc = mdd_trans_start(env, mdd, handle);
2741 /* it needs a sync tx to make FLR to work properly */
2742 handle->th_sync = 1;
2744 rc = mdo_layout_change(env, obj, mlc, handle);
2749 rc = mdo_xattr_set(env, obj, som_buf, XATTR_NAME_SOM,
2755 rc = mdd_changelog_data_store(env, mdd, CL_RESYNC, 0, obj, handle);
2764 * Layout change callback for object.
2766 * This is only used by FLR for now. In the future, it can be exteneded to
2767 * handle all layout change.
2770 mdd_layout_change(const struct lu_env *env, struct md_object *o,
2771 struct md_layout_change *mlc)
2773 struct mdd_object *obj = md2mdd_obj(o);
2774 struct mdd_device *mdd = mdd_obj2mdd_dev(obj);
2775 struct lu_buf *buf = mdd_buf_get(env, NULL, 0);
2776 struct lov_comp_md_v1 *lcm;
2777 struct thandle *handle;
2782 /* Verify acceptable operations */
2783 switch (mlc->mlc_opc) {
2784 case MD_LAYOUT_WRITE:
2785 case MD_LAYOUT_RESYNC:
2786 case MD_LAYOUT_RESYNC_DONE:
2792 handle = mdd_trans_create(env, mdd);
2794 RETURN(PTR_ERR(handle));
2796 rc = mdd_get_lov_ea(env, obj, buf);
2803 /* analyze the layout to make sure it's a FLR file */
2805 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
2806 GOTO(out, rc = -EINVAL);
2808 flr_state = le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK;
2810 /* please refer to HLD of FLR for state transition */
2811 switch (flr_state) {
2813 rc = mdd_layout_instantiate_component(env, obj, mlc, handle);
2815 case LCM_FL_WRITE_PENDING:
2816 rc = mdd_layout_update_write_pending(env, obj, mlc, handle);
2819 rc = mdd_layout_update_rdonly(env, obj, mlc, handle);
2821 case LCM_FL_SYNC_PENDING:
2822 rc = mdd_object_update_sync_pending(env, obj, mlc, handle);
2831 mdd_trans_stop(env, mdd, rc, handle);
2836 void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent,
2837 struct mdd_object *child, const struct lu_attr *attr,
2838 const struct md_op_spec *spec,
2839 struct dt_allocation_hint *hint)
2841 struct dt_object *np = parent ? mdd_object_child(parent) : NULL;
2842 struct dt_object *nc = mdd_object_child(child);
2844 memset(hint, 0, sizeof(*hint));
2846 /* For striped directory, give striping EA to lod_ah_init, which will
2847 * decide the stripe_offset and stripe count by it. */
2848 if (S_ISDIR(attr->la_mode) &&
2849 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
2850 hint->dah_eadata = spec->u.sp_ea.eadata;
2851 hint->dah_eadata_len = spec->u.sp_ea.eadatalen;
2853 hint->dah_eadata = NULL;
2854 hint->dah_eadata_len = 0;
2857 CDEBUG(D_INFO, DFID" eadata %p len %d\n", PFID(mdd_object_fid(child)),
2858 hint->dah_eadata, hint->dah_eadata_len);
2859 /* @hint will be initialized by underlying device. */
2860 nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT);
2864 * do NOT or the MAY_*'s, you'll get the weakest
2866 int accmode(const struct lu_env *env, const struct lu_attr *la, int flags)
2870 /* Sadly, NFSD reopens a file repeatedly during operation, so the
2871 * "acc_mode = 0" allowance for newly-created files isn't honoured.
2872 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
2873 * owner can write to a file even if it is marked readonly to hide
2874 * its brokenness. (bug 5781) */
2875 if (flags & MDS_OPEN_OWNEROVERRIDE) {
2876 struct lu_ucred *uc = lu_ucred_check(env);
2878 if ((uc == NULL) || (la->la_uid == uc->uc_fsuid))
2882 if (flags & MDS_FMODE_READ)
2884 if (flags & (MDS_FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
2886 if (flags & MDS_FMODE_EXEC)
2891 static int mdd_open_sanity_check(const struct lu_env *env,
2892 struct mdd_object *obj,
2893 const struct lu_attr *attr, int flag)
2899 if (mdd_is_dead_obj(obj))
2902 if (S_ISLNK(attr->la_mode))
2905 mode = accmode(env, attr, flag);
2907 if (S_ISDIR(attr->la_mode) && (mode & MAY_WRITE))
2910 if (!(flag & MDS_OPEN_CREATED)) {
2911 rc = mdd_permission_internal(env, obj, attr, mode);
2916 if (S_ISFIFO(attr->la_mode) || S_ISSOCK(attr->la_mode) ||
2917 S_ISBLK(attr->la_mode) || S_ISCHR(attr->la_mode))
2918 flag &= ~MDS_OPEN_TRUNC;
2920 /* For writing append-only file must open it with append mode. */
2921 if (attr->la_flags & LUSTRE_APPEND_FL) {
2922 if ((flag & MDS_FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
2924 if (flag & MDS_OPEN_TRUNC)
2931 static int mdd_open(const struct lu_env *env, struct md_object *obj,
2934 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2935 struct md_device *md_dev = lu2md_dev(mdd2lu_dev(mdo2mdd(obj)));
2936 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
2937 struct mdd_object_user *mou = NULL;
2938 const struct lu_ucred *uc = lu_ucred(env);
2939 struct mdd_device *mdd = mdo2mdd(obj);
2940 enum changelog_rec_type type = CL_OPEN;
2943 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
2945 rc = mdd_la_get(env, mdd_obj, attr);
2949 rc = mdd_open_sanity_check(env, mdd_obj, attr, flags);
2950 if ((rc == -EACCES) && (mdd->mdd_cl.mc_mask & (1 << CL_DN_OPEN)))
2955 mdd_obj->mod_count++;
2957 if (!mdd_changelog_enabled(env, mdd, type))
2961 /* look for existing opener in list under mdd_write_lock */
2962 mou = mdd_obj_user_find(mdd_obj, uc->uc_uid, uc->uc_gid, flags);
2967 /* add user to list */
2968 mou = mdd_obj_user_alloc(flags, uc->uc_uid, uc->uc_gid);
2974 rc2 = mdd_obj_user_add(mdd_obj, mou, type == CL_DN_OPEN);
2976 mdd_obj_user_free(mou);
2981 if (type == CL_DN_OPEN) {
2982 if (ktime_before(ktime_get(), mou->mou_deniednext))
2983 /* same user denied again same access within
2984 * time interval: do not record
2988 /* this user already denied, but some time ago:
2989 * update denied time
2991 mou->mou_deniednext =
2992 ktime_add(ktime_get(),
2993 ktime_set(mdd->mdd_cl.mc_deniednext,
2996 mou->mou_opencount++;
2997 /* same user opening file again with same flags:
3004 mdd_changelog(env, type, flags, md_dev, mdo2fid(mdd_obj));
3008 mdd_write_unlock(env, mdd_obj);
3012 static int mdd_declare_close(const struct lu_env *env, struct mdd_object *obj,
3013 struct md_attr *ma, struct thandle *handle)
3017 rc = mdd_orphan_declare_delete(env, obj, handle);
3021 return mdo_declare_destroy(env, obj, handle);
3025 * No permission check is needed.
3027 static int mdd_close(const struct lu_env *env, struct md_object *obj,
3028 struct md_attr *ma, int mode)
3030 struct mdd_object *mdd_obj = md2mdd_obj(obj);
3031 struct mdd_device *mdd = mdo2mdd(obj);
3032 struct thandle *handle = NULL;
3035 bool blocked = false;
3036 int last_close_by_uid = 0;
3037 const struct lu_ucred *uc = lu_ucred(env);
3040 if (ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_KEEP_ORPHAN) {
3041 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
3042 mdd_obj->mod_count--;
3043 mdd_write_unlock(env, mdd_obj);
3045 if (mdd_obj->mod_flags & ORPHAN_OBJ && !mdd_obj->mod_count)
3046 CDEBUG(D_HA, "Object "DFID" is retained in orphan "
3047 "list\n", PFID(mdd_object_fid(mdd_obj)));
3051 /* mdd_finish_unlink() will always set orphan object as DEAD_OBJ, but
3052 * it might fail to add the object to orphan list (w/o ORPHAN_OBJ). */
3053 /* check without any lock */
3054 is_orphan = mdd_obj->mod_count == 1 &&
3055 (mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0;
3059 /* mdd_trans_create() maybe failed because of barrier_entry(),
3060 * under such case, the orphan MDT-object will be left in the
3061 * orphan list, and when the MDT remount next time, the unused
3062 * orphans will be destroyed automatically.
3064 * One exception: the former mdd_finish_unlink may failed to
3065 * add the orphan MDT-object to the orphan list, then if the
3066 * mdd_trans_create() failed because of barrier_entry(), the
3067 * MDT-object will become real orphan that is neither in the
3068 * namespace nor in the orphan list. Such bad case should be
3069 * very rare and will be handled by e2fsck/lfsck. */
3070 handle = mdd_trans_create(env, mdo2mdd(obj));
3071 if (IS_ERR(handle)) {
3072 rc = PTR_ERR(handle);
3073 if (rc != -EINPROGRESS)
3081 rc = mdd_declare_close(env, mdd_obj, ma, handle);
3085 rc = mdd_declare_changelog_store(env, mdd, CL_CLOSE, NULL, NULL,
3090 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
3096 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
3097 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr);
3099 CERROR("%s: failed to get lu_attr of "DFID": rc = %d\n",
3100 lu_dev_name(mdd2lu_dev(mdd)),
3101 PFID(mdd_object_fid(mdd_obj)), rc);
3105 /* check again with lock */
3106 is_orphan = (mdd_obj->mod_count == 1) &&
3107 ((mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0 ||
3108 ma->ma_attr.la_nlink == 0);
3110 if (is_orphan && !handle && !blocked) {
3111 mdd_write_unlock(env, mdd_obj);
3115 mdd_obj->mod_count--; /*release open count */
3117 /* under mdd write lock */
3118 /* If recording, see if we need to remove UID from list */
3119 if (mdd_changelog_enabled(env, mdd, CL_OPEN)) {
3120 struct mdd_object_user *mou;
3122 /* look for UID in list */
3123 /* If mou is NULL, it probably means logging was enabled after
3124 * the user had the file open. So the corresponding close
3125 * will not be logged.
3127 mou = mdd_obj_user_find(mdd_obj, uc->uc_uid, uc->uc_gid, mode);
3129 mou->mou_opencount--;
3130 if (mou->mou_opencount == 0) {
3131 mdd_obj_user_remove(mdd_obj, mou);
3132 last_close_by_uid = 1;
3137 if (!is_orphan || blocked)
3141 /* NB: Object maybe not in orphan list originally, it is rare case for
3142 * mdd_finish_unlink() failure, in that case, the object doesn't have
3143 * ORPHAN_OBJ flag */
3144 if ((mdd_obj->mod_flags & ORPHAN_OBJ) != 0) {
3145 /* remove link to object from orphan index */
3146 LASSERT(handle != NULL);
3147 rc = mdd_orphan_delete(env, mdd_obj, handle);
3149 CERROR("%s: unable to delete "DFID" from orphan list: "
3150 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
3151 PFID(mdd_object_fid(mdd_obj)), rc);
3152 /* If object was not deleted from orphan list, do not
3153 * destroy OSS objects, which will be done when next
3158 CDEBUG(D_HA, "Object "DFID" is deleted from orphan "
3159 "list, OSS objects to be destroyed.\n",
3160 PFID(mdd_object_fid(mdd_obj)));
3163 rc = mdo_destroy(env, mdd_obj, handle);
3166 CERROR("%s: unable to delete "DFID" from orphan list: "
3167 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
3168 PFID(mdd_object_fid(mdd_obj)), rc);
3173 mdd_write_unlock(env, mdd_obj);
3175 if (rc != 0 || blocked ||
3176 !mdd_changelog_enabled(env, mdd, CL_CLOSE))
3179 /* Record CL_CLOSE in changelog only if file was opened in write mode,
3180 * or if CL_OPEN was recorded and it's last close by user.
3181 * Changelogs mask may change between open and close operations, but
3182 * this is not a big deal if we have a CL_CLOSE entry with no matching
3183 * CL_OPEN. Plus Changelogs mask may not change often.
3185 if (((!(mdd->mdd_cl.mc_mask & (1 << CL_OPEN)) &&
3186 (mode & (MDS_FMODE_WRITE | MDS_OPEN_APPEND | MDS_OPEN_TRUNC))) ||
3187 ((mdd->mdd_cl.mc_mask & (1 << CL_OPEN)) && last_close_by_uid)) &&
3188 !(ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_RECOV_OPEN)) {
3189 if (handle == NULL) {
3190 handle = mdd_trans_create(env, mdo2mdd(obj));
3192 GOTO(stop, rc = PTR_ERR(handle));
3194 rc = mdd_declare_changelog_store(env, mdd, CL_CLOSE,
3195 NULL, NULL, handle);
3199 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
3204 mdd_changelog_data_store(env, mdd, CL_CLOSE, mode,
3209 if (handle != NULL && !IS_ERR(handle))
3210 rc = mdd_trans_stop(env, mdd, rc, handle);
3216 * Permission check is done when open,
3217 * no need check again.
3219 static int mdd_readpage_sanity_check(const struct lu_env *env,
3220 struct mdd_object *obj)
3222 struct dt_object *next = mdd_object_child(obj);
3226 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
3234 static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp,
3235 size_t nob, const struct dt_it_ops *iops,
3236 struct dt_it *it, __u32 attr, void *arg)
3238 struct lu_dirpage *dp = &lp->lp_dir;
3242 struct lu_dirent *ent;
3243 struct lu_dirent *last = NULL;
3247 if (nob < sizeof(*dp))
3250 memset(area, 0, sizeof (*dp));
3251 area += sizeof (*dp);
3252 nob -= sizeof (*dp);
3259 len = iops->key_size(env, it);
3261 /* IAM iterator can return record with zero len. */
3265 hash = iops->store(env, it);
3266 if (unlikely(first)) {
3268 dp->ldp_hash_start = cpu_to_le64(hash);
3271 /* calculate max space required for lu_dirent */
3272 recsize = lu_dirent_calc_size(len, attr);
3274 if (nob >= recsize) {
3275 result = iops->rec(env, it, (struct dt_rec *)ent, attr);
3276 if (result == -ESTALE)
3281 /* osd might not able to pack all attributes,
3282 * so recheck rec length */
3283 recsize = le16_to_cpu(ent->lde_reclen);
3285 if (le32_to_cpu(ent->lde_attrs) & LUDA_FID) {
3286 fid_le_to_cpu(&fid, &ent->lde_fid);
3287 if (fid_is_dot_lustre(&fid))
3291 result = (last != NULL) ? 0 :-EINVAL;
3295 ent = (void *)ent + recsize;
3299 result = iops->next(env, it);
3300 if (result == -ESTALE)
3302 } while (result == 0);
3305 dp->ldp_hash_end = cpu_to_le64(hash);
3307 if (last->lde_hash == dp->ldp_hash_end)
3308 dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE);
3309 last->lde_reclen = 0; /* end mark */
3312 /* end of directory */
3313 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
3314 else if (result < 0)
3315 CWARN("build page failed: %d!\n", result);
3319 int mdd_readpage(const struct lu_env *env, struct md_object *obj,
3320 const struct lu_rdpg *rdpg)
3322 struct mdd_object *mdd_obj = md2mdd_obj(obj);
3326 if (mdd_object_exists(mdd_obj) == 0) {
3327 CERROR("%s: object "DFID" not found: rc = -2\n",
3328 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
3332 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
3333 rc = mdd_readpage_sanity_check(env, mdd_obj);
3335 GOTO(out_unlock, rc);
3337 if (mdd_is_dead_obj(mdd_obj)) {
3339 struct lu_dirpage *dp;
3342 * According to POSIX, please do not return any entry to client:
3343 * even dot and dotdot should not be returned.
3345 CDEBUG(D_INODE, "readdir from dead object: "DFID"\n",
3346 PFID(mdd_object_fid(mdd_obj)));
3348 if (rdpg->rp_count <= 0)
3349 GOTO(out_unlock, rc = -EFAULT);
3350 LASSERT(rdpg->rp_pages != NULL);
3352 pg = rdpg->rp_pages[0];
3353 dp = (struct lu_dirpage *)kmap(pg);
3354 memset(dp, 0 , sizeof(struct lu_dirpage));
3355 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
3356 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
3357 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
3359 GOTO(out_unlock, rc = LU_PAGE_SIZE);
3362 rc = dt_index_walk(env, mdd_object_child(mdd_obj), rdpg,
3363 mdd_dir_page_build, NULL);
3365 struct lu_dirpage *dp;
3367 dp = kmap(rdpg->rp_pages[0]);
3368 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
3371 * No pages were processed, mark this for first page
3374 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
3375 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
3376 rc = min_t(unsigned int, LU_PAGE_SIZE, rdpg->rp_count);
3378 kunmap(rdpg->rp_pages[0]);
3381 GOTO(out_unlock, rc);
3383 mdd_read_unlock(env, mdd_obj);
3387 static int mdd_object_sync(const struct lu_env *env, struct md_object *obj)
3389 struct mdd_object *mdd_obj = md2mdd_obj(obj);
3391 if (mdd_object_exists(mdd_obj) == 0) {
3394 CERROR("%s: object "DFID" not found: rc = %d\n",
3395 mdd_obj_dev_name(mdd_obj),
3396 PFID(mdd_object_fid(mdd_obj)), rc);
3399 return dt_object_sync(env, mdd_object_child(mdd_obj),
3403 static int mdd_object_lock(const struct lu_env *env,
3404 struct md_object *obj,
3405 struct lustre_handle *lh,
3406 struct ldlm_enqueue_info *einfo,
3407 union ldlm_policy_data *policy)
3409 struct mdd_object *mdd_obj = md2mdd_obj(obj);
3410 return dt_object_lock(env, mdd_object_child(mdd_obj), lh,
3414 static int mdd_object_unlock(const struct lu_env *env,
3415 struct md_object *obj,
3416 struct ldlm_enqueue_info *einfo,
3417 union ldlm_policy_data *policy)
3419 struct mdd_object *mdd_obj = md2mdd_obj(obj);
3420 return dt_object_unlock(env, mdd_object_child(mdd_obj), einfo, policy);
3423 const struct md_object_operations mdd_obj_ops = {
3424 .moo_permission = mdd_permission,
3425 .moo_attr_get = mdd_attr_get,
3426 .moo_attr_set = mdd_attr_set,
3427 .moo_xattr_get = mdd_xattr_get,
3428 .moo_xattr_set = mdd_xattr_set,
3429 .moo_xattr_list = mdd_xattr_list,
3430 .moo_invalidate = mdd_invalidate,
3431 .moo_xattr_del = mdd_xattr_del,
3432 .moo_swap_layouts = mdd_swap_layouts,
3433 .moo_open = mdd_open,
3434 .moo_close = mdd_close,
3435 .moo_readpage = mdd_readpage,
3436 .moo_readlink = mdd_readlink,
3437 .moo_changelog = mdd_changelog,
3438 .moo_object_sync = mdd_object_sync,
3439 .moo_object_lock = mdd_object_lock,
3440 .moo_object_unlock = mdd_object_unlock,
3441 .moo_layout_change = mdd_layout_change,