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) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdd/mdd_object.c
38 * Lustre Metadata Server (mdd) routines
40 * Author: Wang Di <wangdi@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_MDS
45 #include <linux/module.h>
47 #include <obd_class.h>
48 #include <obd_support.h>
49 #include <lprocfs_status.h>
50 /* fid_be_cpu(), fid_cpu_to_be(). */
51 #include <lustre_fid.h>
53 #include <lustre_idmap.h>
54 #include <lustre_param.h>
55 #include <lustre_mds.h>
56 #include <lustre/lustre_idl.h>
58 #include "mdd_internal.h"
60 static const struct lu_object_operations mdd_lu_obj_ops;
61 extern struct kmem_cache *mdd_object_kmem;
63 static int mdd_xattr_get(const struct lu_env *env,
64 struct md_object *obj, struct lu_buf *buf,
67 int mdd_data_get(const struct lu_env *env, struct mdd_object *obj,
70 if (mdd_object_exists(obj) == 0) {
71 CERROR("%s: object "DFID" not found: rc = -2\n",
72 mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj)));
75 mdo_data_get(env, obj, data);
79 int mdd_la_get(const struct lu_env *env, struct mdd_object *obj,
80 struct lu_attr *la, struct lustre_capa *capa)
82 if (mdd_object_exists(obj) == 0) {
83 CERROR("%s: object "DFID" not found: rc = -2\n",
84 mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj)));
88 return mdo_attr_get(env, obj, la, capa);
91 void mdd_flags_xlate(struct mdd_object *obj, __u32 flags)
93 obj->mod_flags &= ~(APPEND_OBJ|IMMUTE_OBJ);
95 if (flags & LUSTRE_APPEND_FL)
96 obj->mod_flags |= APPEND_OBJ;
98 if (flags & LUSTRE_IMMUTABLE_FL)
99 obj->mod_flags |= IMMUTE_OBJ;
102 struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
104 struct mdd_thread_info *info;
106 lu_env_refill((struct lu_env *)env);
107 info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
108 LASSERT(info != NULL);
112 const struct lu_name *mdd_name_get_const(const struct lu_env *env,
113 const void *area, ssize_t len)
115 struct lu_name *lname;
117 lname = &mdd_env_info(env)->mti_name;
118 lname->ln_name = area;
119 lname->ln_namelen = len;
123 struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
127 buf = &mdd_env_info(env)->mti_buf[0];
133 const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
134 const void *area, ssize_t len)
138 buf = &mdd_env_info(env)->mti_buf[0];
139 buf->lb_buf = (void *)area;
144 struct lu_object *mdd_object_alloc(const struct lu_env *env,
145 const struct lu_object_header *hdr,
148 struct mdd_object *mdd_obj;
150 OBD_SLAB_ALLOC_PTR_GFP(mdd_obj, mdd_object_kmem, GFP_NOFS);
151 if (mdd_obj != NULL) {
154 o = mdd2lu_obj(mdd_obj);
155 lu_object_init(o, NULL, d);
156 mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
157 mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
158 mdd_obj->mod_count = 0;
159 o->lo_ops = &mdd_lu_obj_ops;
166 static int mdd_object_init(const struct lu_env *env, struct lu_object *o,
167 const struct lu_object_conf *unused)
169 struct mdd_device *d = lu2mdd_dev(o->lo_dev);
170 struct mdd_object *mdd_obj = lu2mdd_obj(o);
171 struct lu_object *below;
172 struct lu_device *under;
175 mdd_obj->mod_cltime = 0;
176 under = &d->mdd_child->dd_lu_dev;
177 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
179 RETURN(PTR_ERR(below));
181 lu_object_add(o, below);
186 static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
190 if (lu_object_exists(o)) {
191 struct mdd_object *mdd_obj = lu2mdd_obj(o);
192 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
194 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
196 mdd_flags_xlate(mdd_obj, attr->la_flags);
202 static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
204 struct mdd_object *mdd = lu2mdd_obj(o);
207 OBD_SLAB_FREE_PTR(mdd, mdd_object_kmem);
210 static int mdd_object_print(const struct lu_env *env, void *cookie,
211 lu_printer_t p, const struct lu_object *o)
213 struct mdd_object *mdd = lu2mdd_obj((struct lu_object *)o);
214 return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p(open_count=%d, "
215 "valid=%x, cltime="LPU64", flags=%lx)",
216 mdd, mdd->mod_count, mdd->mod_valid,
217 mdd->mod_cltime, mdd->mod_flags);
220 static const struct lu_object_operations mdd_lu_obj_ops = {
221 .loo_object_init = mdd_object_init,
222 .loo_object_start = mdd_object_start,
223 .loo_object_free = mdd_object_free,
224 .loo_object_print = mdd_object_print,
227 struct mdd_object *mdd_object_find(const struct lu_env *env,
228 struct mdd_device *d,
229 const struct lu_fid *f)
231 return md2mdd_obj(md_object_find_slice(env, &d->mdd_md_dev, f));
235 * No permission check is needed.
237 int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
240 struct mdd_object *mdd_obj = md2mdd_obj(obj);
245 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr,
246 mdd_object_capa(env, md2mdd_obj(obj)));
247 if ((ma->ma_need & MA_INODE) != 0 && mdd_is_dead_obj(mdd_obj))
248 ma->ma_attr.la_nlink = 0;
254 * No permission check is needed.
256 static int mdd_xattr_get(const struct lu_env *env,
257 struct md_object *obj, struct lu_buf *buf,
260 struct mdd_object *mdd_obj = md2mdd_obj(obj);
265 if (mdd_object_exists(mdd_obj) == 0) {
266 CERROR("%s: object "DFID" not found: rc = -2\n",
267 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
271 /* If the object has been delete from the namespace, then
272 * get linkEA should return -ENOENT as well */
273 if (unlikely((mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ)) &&
274 strcmp(name, XATTR_NAME_LINK) == 0))
277 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
278 rc = mdo_xattr_get(env, mdd_obj, buf, name,
279 mdd_object_capa(env, mdd_obj));
280 mdd_read_unlock(env, mdd_obj);
286 * Permission check is done when open,
287 * no need check again.
289 int mdd_readlink(const struct lu_env *env, struct md_object *obj,
292 struct mdd_object *mdd_obj = md2mdd_obj(obj);
293 struct dt_object *next;
298 if (mdd_object_exists(mdd_obj) == 0) {
299 CERROR("%s: object "DFID" not found: rc = -2\n",
300 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
304 next = mdd_object_child(mdd_obj);
305 LASSERT(next != NULL);
306 LASSERT(next->do_body_ops != NULL);
307 LASSERT(next->do_body_ops->dbo_read != NULL);
308 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
309 rc = next->do_body_ops->dbo_read(env, next, buf, &pos,
310 mdd_object_capa(env, mdd_obj));
311 mdd_read_unlock(env, mdd_obj);
316 * No permission check is needed.
318 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
321 struct mdd_object *mdd_obj = md2mdd_obj(obj);
326 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
327 rc = mdo_xattr_list(env, mdd_obj, buf, mdd_object_capa(env, mdd_obj));
328 mdd_read_unlock(env, mdd_obj);
333 int mdd_declare_object_create_internal(const struct lu_env *env,
334 struct mdd_object *p,
335 struct mdd_object *c,
336 struct lu_attr *attr,
337 struct thandle *handle,
338 const struct md_op_spec *spec,
339 struct dt_allocation_hint *hint)
341 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
342 const struct dt_index_features *feat = spec->sp_feat;
346 if (feat != &dt_directory_features && feat != NULL) {
347 dof->dof_type = DFT_INDEX;
348 dof->u.dof_idx.di_feat = feat;
351 dof->dof_type = dt_mode_to_dft(attr->la_mode);
352 if (dof->dof_type == DFT_REGULAR) {
353 dof->u.dof_reg.striped =
354 md_should_create(spec->sp_cr_flags);
355 if (spec->sp_cr_flags & MDS_OPEN_HAS_EA)
356 dof->u.dof_reg.striped = 0;
357 /* is this replay? */
359 dof->u.dof_reg.striped = 0;
363 rc = mdo_declare_create_obj(env, c, attr, hint, dof, handle);
368 int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p,
369 struct mdd_object *c, struct lu_attr *attr,
370 struct thandle *handle,
371 const struct md_op_spec *spec,
372 struct dt_allocation_hint *hint)
374 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
378 LASSERT(!mdd_object_exists(c));
380 rc = mdo_create_obj(env, c, attr, hint, dof, handle);
382 LASSERT(ergo(rc == 0, mdd_object_exists(c)));
387 int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj,
388 const struct lu_attr *attr, struct thandle *handle,
394 rc = mdo_attr_set(env, obj, attr, handle, mdd_object_capa(env, obj));
395 #ifdef CONFIG_FS_POSIX_ACL
396 if (!rc && (attr->la_valid & LA_MODE) && needacl)
397 rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
402 int mdd_update_time(const struct lu_env *env, struct mdd_object *obj,
403 const struct lu_attr *oattr, struct lu_attr *attr,
404 struct thandle *handle)
409 LASSERT(attr->la_valid & LA_CTIME);
410 LASSERT(oattr != NULL);
412 /* Make sure the ctime is increased only, however, it's not strictly
413 * reliable at here because there is not guarantee to hold lock on
414 * object, so we just bypass some unnecessary cmtime setting first
415 * and OSD has to check it again. */
416 if (attr->la_ctime < oattr->la_ctime)
417 attr->la_valid &= ~(LA_MTIME | LA_CTIME);
418 else if (attr->la_valid == LA_CTIME &&
419 attr->la_ctime == oattr->la_ctime)
420 attr->la_valid &= ~LA_CTIME;
422 if (attr->la_valid != 0)
423 rc = mdd_attr_set_internal(env, obj, attr, handle, 0);
428 * This gives the same functionality as the code between
429 * sys_chmod and inode_setattr
430 * chown_common and inode_setattr
431 * utimes and inode_setattr
432 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
434 static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
435 const struct lu_attr *oattr, struct lu_attr *la,
436 const unsigned long flags)
445 /* Do not permit change file type */
446 if (la->la_valid & LA_TYPE)
449 /* They should not be processed by setattr */
450 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
453 LASSERT(oattr != NULL);
455 /* export destroy does not have ->le_ses, but we may want
456 * to drop LUSTRE_SOM_FL. */
457 uc = lu_ucred_check(env);
461 if (la->la_valid == LA_CTIME) {
462 if (!(flags & MDS_PERM_BYPASS))
463 /* This is only for set ctime when rename's source is
465 rc = mdd_may_delete(env, NULL, NULL, obj, oattr, NULL,
467 if (rc == 0 && la->la_ctime <= oattr->la_ctime)
468 la->la_valid &= ~LA_CTIME;
472 if (la->la_valid == LA_ATIME) {
473 /* This is atime only set for read atime update on close. */
474 if (la->la_atime >= oattr->la_atime &&
475 la->la_atime < (oattr->la_atime +
476 mdd_obj2mdd_dev(obj)->mdd_atime_diff))
477 la->la_valid &= ~LA_ATIME;
481 /* Check if flags change. */
482 if (la->la_valid & LA_FLAGS) {
483 unsigned int oldflags = 0;
484 unsigned int newflags = la->la_flags &
485 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
487 if ((uc->uc_fsuid != oattr->la_uid) &&
488 !md_capable(uc, CFS_CAP_FOWNER))
491 /* XXX: the IMMUTABLE and APPEND_ONLY flags can
492 * only be changed by the relevant capability. */
493 if (mdd_is_immutable(obj))
494 oldflags |= LUSTRE_IMMUTABLE_FL;
495 if (mdd_is_append(obj))
496 oldflags |= LUSTRE_APPEND_FL;
497 if ((oldflags ^ newflags) &&
498 !md_capable(uc, CFS_CAP_LINUX_IMMUTABLE))
501 if (!S_ISDIR(oattr->la_mode))
502 la->la_flags &= ~LUSTRE_DIRSYNC_FL;
505 if ((mdd_is_immutable(obj) || mdd_is_append(obj)) &&
506 (la->la_valid & ~LA_FLAGS) &&
507 !(flags & MDS_PERM_BYPASS))
510 /* Check for setting the obj time. */
511 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
512 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
513 if ((uc->uc_fsuid != oattr->la_uid) &&
514 !md_capable(uc, CFS_CAP_FOWNER)) {
515 rc = mdd_permission_internal(env, obj, oattr,
522 if (la->la_valid & LA_KILL_SUID) {
523 la->la_valid &= ~LA_KILL_SUID;
524 if ((oattr->la_mode & S_ISUID) &&
525 !(la->la_valid & LA_MODE)) {
526 la->la_mode = oattr->la_mode;
527 la->la_valid |= LA_MODE;
529 la->la_mode &= ~S_ISUID;
532 if (la->la_valid & LA_KILL_SGID) {
533 la->la_valid &= ~LA_KILL_SGID;
534 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
535 (S_ISGID | S_IXGRP)) &&
536 !(la->la_valid & LA_MODE)) {
537 la->la_mode = oattr->la_mode;
538 la->la_valid |= LA_MODE;
540 la->la_mode &= ~S_ISGID;
543 /* Make sure a caller can chmod. */
544 if (la->la_valid & LA_MODE) {
545 if (!(flags & MDS_PERM_BYPASS) &&
546 (uc->uc_fsuid != oattr->la_uid) &&
547 !md_capable(uc, CFS_CAP_FOWNER))
550 if (la->la_mode == (umode_t) -1)
551 la->la_mode = oattr->la_mode;
553 la->la_mode = (la->la_mode & S_IALLUGO) |
554 (oattr->la_mode & ~S_IALLUGO);
556 /* Also check the setgid bit! */
557 if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ?
558 la->la_gid : oattr->la_gid) &&
559 !md_capable(uc, CFS_CAP_FSETID))
560 la->la_mode &= ~S_ISGID;
562 la->la_mode = oattr->la_mode;
565 /* Make sure a caller can chown. */
566 if (la->la_valid & LA_UID) {
567 if (la->la_uid == (uid_t) -1)
568 la->la_uid = oattr->la_uid;
569 if (((uc->uc_fsuid != oattr->la_uid) ||
570 (la->la_uid != oattr->la_uid)) &&
571 !md_capable(uc, CFS_CAP_CHOWN))
574 /* If the user or group of a non-directory has been
575 * changed by a non-root user, remove the setuid bit.
576 * 19981026 David C Niemi <niemi@tux.org>
578 * Changed this to apply to all users, including root,
579 * to avoid some races. This is the behavior we had in
580 * 2.0. The check for non-root was definitely wrong
581 * for 2.2 anyway, as it should have been using
582 * CAP_FSETID rather than fsuid -- 19990830 SD. */
583 if (((oattr->la_mode & S_ISUID) == S_ISUID) &&
584 !S_ISDIR(oattr->la_mode)) {
585 la->la_mode &= ~S_ISUID;
586 la->la_valid |= LA_MODE;
590 /* Make sure caller can chgrp. */
591 if (la->la_valid & LA_GID) {
592 if (la->la_gid == (gid_t) -1)
593 la->la_gid = oattr->la_gid;
594 if (((uc->uc_fsuid != oattr->la_uid) ||
595 ((la->la_gid != oattr->la_gid) &&
596 !lustre_in_group_p(uc, la->la_gid))) &&
597 !md_capable(uc, CFS_CAP_CHOWN))
600 /* Likewise, if the user or group of a non-directory
601 * has been changed by a non-root user, remove the
602 * setgid bit UNLESS there is no group execute bit
603 * (this would be a file marked for mandatory
604 * locking). 19981026 David C Niemi <niemi@tux.org>
606 * Removed the fsuid check (see the comment above) --
608 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
609 (S_ISGID | S_IXGRP)) && !S_ISDIR(oattr->la_mode)) {
610 la->la_mode &= ~S_ISGID;
611 la->la_valid |= LA_MODE;
615 /* For both Size-on-MDS case and truncate case,
616 * "la->la_valid & (LA_SIZE | LA_BLOCKS)" are ture.
617 * We distinguish them by "flags & MDS_SOM".
618 * For SOM case, it is true, the MAY_WRITE perm has been checked
619 * when open, no need check again. For truncate case, it is false,
620 * the MAY_WRITE perm should be checked here. */
621 if (flags & MDS_SOM) {
622 /* For the "Size-on-MDS" setattr update, merge coming
623 * attributes with the set in the inode. BUG 10641 */
624 if ((la->la_valid & LA_ATIME) &&
625 (la->la_atime <= oattr->la_atime))
626 la->la_valid &= ~LA_ATIME;
628 /* OST attributes do not have a priority over MDS attributes,
629 * so drop times if ctime is equal. */
630 if ((la->la_valid & LA_CTIME) &&
631 (la->la_ctime <= oattr->la_ctime))
632 la->la_valid &= ~(LA_MTIME | LA_CTIME);
634 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
635 if (!((flags & MDS_OWNEROVERRIDE) &&
636 (uc->uc_fsuid == oattr->la_uid)) &&
637 !(flags & MDS_PERM_BYPASS)) {
638 rc = mdd_permission_internal(env, obj,
644 if (la->la_valid & LA_CTIME) {
645 /* The pure setattr, it has the priority over what is
646 * already set, do not drop it if ctime is equal. */
647 if (la->la_ctime < oattr->la_ctime)
648 la->la_valid &= ~(LA_ATIME | LA_MTIME |
656 /** Store a data change changelog record
657 * If this fails, we must fail the whole transaction; we don't
658 * want the change to commit without the log entry.
659 * \param mdd_obj - mdd_object of change
660 * \param handle - transacion handle
662 int mdd_changelog_data_store(const struct lu_env *env, struct mdd_device *mdd,
663 enum changelog_rec_type type, int flags,
664 struct mdd_object *mdd_obj, struct thandle *handle)
666 const struct lu_fid *tfid;
667 struct llog_changelog_rec *rec;
673 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
675 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
678 LASSERT(mdd_obj != NULL);
679 LASSERT(handle != NULL);
681 tfid = mdo2fid(mdd_obj);
683 if ((type >= CL_MTIME) && (type <= CL_ATIME) &&
684 cfs_time_before_64(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) {
685 /* Don't need multiple updates in this log */
686 /* Don't check under lock - no big deal if we get an extra
691 reclen = llog_data_len(sizeof(*rec));
692 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
693 if (buf->lb_buf == NULL)
697 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
698 rec->cr.cr_type = (__u32)type;
699 rec->cr.cr_tfid = *tfid;
700 rec->cr.cr_namelen = 0;
701 mdd_obj->mod_cltime = cfs_time_current_64();
703 rc = mdd_changelog_store(env, mdd, rec, handle);
708 int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type,
709 int flags, struct md_object *obj)
711 struct thandle *handle;
712 struct mdd_object *mdd_obj = md2mdd_obj(obj);
713 struct mdd_device *mdd = mdo2mdd(obj);
717 handle = mdd_trans_create(env, mdd);
719 RETURN(PTR_ERR(handle));
721 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
725 rc = mdd_trans_start(env, mdd, handle);
729 rc = mdd_changelog_data_store(env, mdd, type, flags, mdd_obj,
733 mdd_trans_stop(env, mdd, rc, handle);
739 * Save LMA extended attributes with data from \a ma.
741 * HSM and Size-On-MDS data will be extracted from \ma if they are valid, if
742 * not, LMA EA will be first read from disk, modified and write back.
745 /* Precedence for choosing record type when multiple
746 * attributes change: setattr > mtime > ctime > atime
747 * (ctime changes when mtime does, plus chmod/chown.
748 * atime and ctime are independent.) */
749 static int mdd_attr_set_changelog(const struct lu_env *env,
750 struct md_object *obj, struct thandle *handle,
753 struct mdd_device *mdd = mdo2mdd(obj);
756 bits = (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0;
757 bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0;
758 bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0;
759 bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0;
760 bits = bits & mdd->mdd_cl.mc_mask;
764 /* The record type is the lowest non-masked set bit */
765 while (bits && ((bits & 1) == 0)) {
770 /* FYI we only store the first CLF_FLAGMASK bits of la_valid */
771 return mdd_changelog_data_store(env, mdd, type, (int)valid,
772 md2mdd_obj(obj), handle);
775 static int mdd_declare_attr_set(const struct lu_env *env,
776 struct mdd_device *mdd,
777 struct mdd_object *obj,
778 const struct lu_attr *attr,
779 struct thandle *handle)
783 rc = mdo_declare_attr_set(env, obj, attr, handle);
787 #ifdef CONFIG_FS_POSIX_ACL
788 if (attr->la_valid & LA_MODE) {
789 mdd_read_lock(env, obj, MOR_TGT_CHILD);
790 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL,
791 XATTR_NAME_ACL_ACCESS, BYPASS_CAPA);
792 mdd_read_unlock(env, obj);
793 if (rc == -EOPNOTSUPP || rc == -ENODATA)
799 struct lu_buf *buf = mdd_buf_get(env, NULL, rc);
800 rc = mdo_declare_xattr_set(env, obj, buf,
801 XATTR_NAME_ACL_ACCESS, 0,
809 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
816 * permission changes may require sync operation, to mitigate performance
817 * impact, only do this for dir and when permission is reduced.
819 * For regular files, version is updated with permission change (see VBR), async
820 * permission won't cause any issue, while missing permission change on
821 * directory may affect accessibility of other objects after recovery.
823 static inline bool permission_needs_sync(const struct lu_attr *old,
824 const struct lu_attr *new)
826 if (!S_ISDIR(old->la_mode))
829 if (new->la_valid & (LA_UID | LA_GID))
832 if (new->la_valid & LA_MODE &&
833 new->la_mode & (S_ISUID | S_ISGID | S_ISVTX))
836 if ((new->la_valid & LA_MODE) &&
837 ((new->la_mode & old->la_mode) & S_IRWXUGO) !=
838 (old->la_mode & S_IRWXUGO))
844 /* set attr and LOV EA at once, return updated attr */
845 int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
846 const struct md_attr *ma)
848 struct mdd_object *mdd_obj = md2mdd_obj(obj);
849 struct mdd_device *mdd = mdo2mdd(obj);
850 struct thandle *handle;
851 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
852 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
853 const struct lu_attr *la = &ma->ma_attr;
857 /* we do not use ->attr_set() for LOV/SOM/HSM EA any more */
858 LASSERT((ma->ma_valid & MA_LOV) == 0);
859 LASSERT((ma->ma_valid & MA_HSM) == 0);
860 LASSERT((ma->ma_valid & MA_SOM) == 0);
862 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
866 *la_copy = ma->ma_attr;
867 rc = mdd_fix_attr(env, mdd_obj, attr, la_copy, ma->ma_attr_flags);
871 /* setattr on "close" only change atime, or do nothing */
872 if (la->la_valid == LA_ATIME && la_copy->la_valid == 0)
875 handle = mdd_trans_create(env, mdd);
877 RETURN(PTR_ERR(handle));
879 rc = mdd_declare_attr_set(env, mdd, mdd_obj, la, handle);
883 rc = mdd_trans_start(env, mdd, handle);
887 if (mdd->mdd_sync_permission && permission_needs_sync(attr, la))
890 if (la->la_valid & (LA_MTIME | LA_CTIME))
891 CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n",
892 la->la_mtime, la->la_ctime);
894 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
895 if (la_copy->la_valid & LA_FLAGS) {
896 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
898 mdd_flags_xlate(mdd_obj, la_copy->la_flags);
899 } else if (la_copy->la_valid) { /* setattr */
900 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
902 mdd_write_unlock(env, mdd_obj);
905 rc = mdd_attr_set_changelog(env, obj, handle, la->la_valid);
910 mdd_trans_stop(env, mdd, rc, handle);
914 static int mdd_xattr_sanity_check(const struct lu_env *env,
915 struct mdd_object *obj,
916 const struct lu_attr *attr)
918 struct lu_ucred *uc = lu_ucred_assert(env);
921 if (mdd_is_immutable(obj) || mdd_is_append(obj))
924 if ((uc->uc_fsuid != attr->la_uid) && !md_capable(uc, CFS_CAP_FOWNER))
930 static int mdd_declare_xattr_set(const struct lu_env *env,
931 struct mdd_device *mdd,
932 struct mdd_object *obj,
933 const struct lu_buf *buf,
935 int fl, struct thandle *handle)
939 rc = mdo_declare_xattr_set(env, obj, buf, name, fl, handle);
943 /* Only record user and layout xattr changes */
944 if (strncmp(XATTR_USER_PREFIX, name,
945 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
946 strcmp(XATTR_NAME_LOV, name) == 0) {
947 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
952 /* If HSM data is modified, this could add a changelog */
953 if (strcmp(XATTR_NAME_HSM, name) == 0) {
954 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
959 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
964 * Compare current and future data of HSM EA and add a changelog if needed.
966 * Caller should have write-locked \param obj.
968 * \param buf - Future HSM EA content.
969 * \retval 0 if no changelog is needed or changelog was added properly.
970 * \retval -ve errno if there was a problem
972 static int mdd_hsm_update_locked(const struct lu_env *env,
973 struct md_object *obj,
974 const struct lu_buf *buf,
975 struct thandle *handle)
977 struct mdd_thread_info *info = mdd_env_info(env);
978 struct mdd_device *mdd = mdo2mdd(obj);
979 struct mdd_object *mdd_obj = md2mdd_obj(obj);
980 struct lu_buf *current_buf;
981 struct md_hsm *current_mh;
982 struct md_hsm *new_mh;
986 OBD_ALLOC_PTR(current_mh);
987 if (current_mh == NULL)
990 /* Read HSM attrs from disk */
991 CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
992 current_buf = mdd_buf_get(env, info->mti_xattr_buf,
993 sizeof(info->mti_xattr_buf));
994 rc = mdo_xattr_get(env, mdd_obj, current_buf, XATTR_NAME_HSM,
995 mdd_object_capa(env, mdd_obj));
996 rc = lustre_buf2hsm(current_buf->lb_buf, rc, current_mh);
997 if (rc < 0 && rc != -ENODATA)
999 else if (rc == -ENODATA)
1000 current_mh->mh_flags = 0;
1002 /* Map future HSM xattr */
1003 OBD_ALLOC_PTR(new_mh);
1005 GOTO(free, rc = -ENOMEM);
1006 lustre_buf2hsm(buf->lb_buf, buf->lb_len, new_mh);
1008 /* If HSM flags are different, add a changelog */
1010 if (current_mh->mh_flags != new_mh->mh_flags) {
1012 hsm_set_cl_event(&flags, HE_STATE);
1013 if (new_mh->mh_flags & HS_DIRTY)
1014 hsm_set_cl_flags(&flags, CLF_HSM_DIRTY);
1016 rc = mdd_changelog_data_store(env, mdd, CL_HSM, flags, mdd_obj,
1020 OBD_FREE_PTR(new_mh);
1022 OBD_FREE_PTR(current_mh);
1027 * The caller should guarantee to update the object ctime
1028 * after xattr_set if needed.
1030 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
1031 const struct lu_buf *buf, const char *name,
1034 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1035 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1036 struct mdd_device *mdd = mdo2mdd(obj);
1037 struct thandle *handle;
1041 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
1045 if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) {
1046 rc = mdd_acl_set(env, mdd_obj, attr, buf, fl);
1050 rc = mdd_xattr_sanity_check(env, mdd_obj, attr);
1054 handle = mdd_trans_create(env, mdd);
1056 RETURN(PTR_ERR(handle));
1058 rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, fl, handle);
1062 rc = mdd_trans_start(env, mdd, handle);
1066 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1068 if (strcmp(XATTR_NAME_HSM, name) == 0) {
1069 rc = mdd_hsm_update_locked(env, obj, buf, handle);
1071 mdd_write_unlock(env, mdd_obj);
1076 rc = mdo_xattr_set(env, mdd_obj, buf, name, fl, handle,
1077 mdd_object_capa(env, mdd_obj));
1078 mdd_write_unlock(env, mdd_obj);
1082 if (strcmp(XATTR_NAME_LOV, name) == 0)
1083 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, mdd_obj,
1085 else if (strncmp(XATTR_USER_PREFIX, name,
1086 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1087 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1088 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1089 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1090 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1091 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1095 mdd_trans_stop(env, mdd, rc, handle);
1100 static int mdd_declare_xattr_del(const struct lu_env *env,
1101 struct mdd_device *mdd,
1102 struct mdd_object *obj,
1104 struct thandle *handle)
1108 rc = mdo_declare_xattr_del(env, obj, name, handle);
1112 /* Only record user xattr changes */
1113 if ((strncmp(XATTR_USER_PREFIX, name,
1114 sizeof(XATTR_USER_PREFIX) - 1) == 0))
1115 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1121 * The caller should guarantee to update the object ctime
1122 * after xattr_set if needed.
1124 int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1127 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1128 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1129 struct mdd_device *mdd = mdo2mdd(obj);
1130 struct thandle *handle;
1134 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
1138 rc = mdd_xattr_sanity_check(env, mdd_obj, attr);
1142 handle = mdd_trans_create(env, mdd);
1144 RETURN(PTR_ERR(handle));
1146 rc = mdd_declare_xattr_del(env, mdd, mdd_obj, name, handle);
1150 rc = mdd_trans_start(env, mdd, handle);
1154 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1155 rc = mdo_xattr_del(env, mdd_obj, name, handle,
1156 mdd_object_capa(env, mdd_obj));
1157 mdd_write_unlock(env, mdd_obj);
1161 /* Only record system & user xattr changes */
1162 if (strncmp(XATTR_USER_PREFIX, name,
1163 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1164 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1165 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1166 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1167 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1168 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1172 mdd_trans_stop(env, mdd, rc, handle);
1178 * read lov EA of an object
1179 * return the lov EA in an allocated lu_buf
1181 int mdd_get_lov_ea(const struct lu_env *env, struct mdd_object *obj,
1182 struct lu_buf *lmm_buf)
1184 struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf;
1189 rc = mdo_xattr_get(env, obj, buf, XATTR_NAME_LOV,
1190 mdd_object_capa(env, obj));
1192 if (rc == -ERANGE) {
1193 /* mti_big_buf is allocated but is too small
1194 * we need to increase it */
1195 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1197 if (buf->lb_buf == NULL)
1198 GOTO(out, rc = -ENOMEM);
1209 if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) {
1210 /* mti_big_buf was not allocated, so we have to
1211 * allocate it based on the ea size */
1212 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1214 if (buf->lb_buf == NULL)
1215 GOTO(out, rc = -ENOMEM);
1219 lu_buf_alloc(lmm_buf, bufsize);
1220 if (lmm_buf->lb_buf == NULL)
1221 GOTO(out, rc = -ENOMEM);
1223 memcpy(lmm_buf->lb_buf, buf->lb_buf, bufsize);
1229 lu_buf_free(lmm_buf);
1233 static int mdd_xattr_hsm_replace(const struct lu_env *env,
1234 struct mdd_object *o, struct lu_buf *buf,
1235 struct thandle *handle)
1237 struct hsm_attrs *attrs;
1243 rc = mdo_xattr_set(env, o, buf, XATTR_NAME_HSM, LU_XATTR_REPLACE,
1244 handle, mdd_object_capa(env, o));
1248 attrs = buf->lb_buf;
1249 hsm_flags = le32_to_cpu(attrs->hsm_flags);
1250 if (!(hsm_flags & HS_RELEASED) || mdd_is_dead_obj(o))
1253 /* Add a changelog record for release. */
1254 hsm_set_cl_event(&flags, HE_RELEASE);
1255 rc = mdd_changelog_data_store(env, mdo2mdd(&o->mod_obj), CL_HSM,
1261 * check if layout swapping between 2 objects is allowed
1263 * - only normal FIDs or non-system IGIFs
1264 * - same type of objects
1265 * - same owner/group (so quotas are still valid)
1267 static int mdd_layout_swap_allowed(const struct lu_env *env,
1268 struct mdd_object *o1,
1269 const struct lu_attr *attr1,
1270 struct mdd_object *o2,
1271 const struct lu_attr *attr2)
1273 const struct lu_fid *fid1, *fid2;
1279 if (!fid_is_norm(fid1) &&
1280 (!fid_is_igif(fid1) || IS_ERR(mdd_links_get(env, o1))))
1283 if (!fid_is_norm(fid2) &&
1284 (!fid_is_igif(fid2) || IS_ERR(mdd_links_get(env, o2))))
1287 if (mdd_object_type(o1) != mdd_object_type(o2)) {
1288 if (S_ISDIR(mdd_object_type(o1)))
1290 if (S_ISREG(mdd_object_type(o1)))
1295 if ((attr1->la_uid != attr2->la_uid) ||
1296 (attr1->la_gid != attr2->la_gid))
1303 * swap layouts between 2 lustre objects
1305 static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1,
1306 struct md_object *obj2, __u64 flags)
1308 struct mdd_thread_info *info = mdd_env_info(env);
1309 struct mdd_object *fst_o = md2mdd_obj(obj1);
1310 struct mdd_object *snd_o = md2mdd_obj(obj2);
1311 struct lu_attr *fst_la = MDD_ENV_VAR(env, cattr);
1312 struct lu_attr *snd_la = MDD_ENV_VAR(env, tattr);
1313 struct mdd_device *mdd = mdo2mdd(obj1);
1314 struct lov_mds_md *fst_lmm, *snd_lmm;
1315 struct lu_buf *fst_buf = &info->mti_buf[0];
1316 struct lu_buf *snd_buf = &info->mti_buf[1];
1317 struct lu_buf *fst_hsm_buf = &info->mti_buf[2];
1318 struct lu_buf *snd_hsm_buf = &info->mti_buf[3];
1319 struct ost_id *saved_oi = NULL;
1320 struct thandle *handle;
1321 __u16 fst_gen, snd_gen;
1327 CLASSERT(ARRAY_SIZE(info->mti_buf) >= 4);
1328 memset(info->mti_buf, 0, sizeof(info->mti_buf));
1330 /* we have to sort the 2 obj, so locking will always
1331 * be in the same order, even in case of 2 concurrent swaps */
1332 rc = lu_fid_cmp(mdo2fid(fst_o), mdo2fid(snd_o));
1333 if (rc == 0) /* same fid ? */
1339 rc = mdd_la_get(env, fst_o, fst_la, BYPASS_CAPA);
1343 rc = mdd_la_get(env, snd_o, snd_la, BYPASS_CAPA);
1347 /* check if layout swapping is allowed */
1348 rc = mdd_layout_swap_allowed(env, fst_o, fst_la, snd_o, snd_la);
1352 handle = mdd_trans_create(env, mdd);
1354 RETURN(PTR_ERR(handle));
1356 /* objects are already sorted */
1357 mdd_write_lock(env, fst_o, MOR_TGT_CHILD);
1358 mdd_write_lock(env, snd_o, MOR_TGT_CHILD);
1360 rc = mdd_get_lov_ea(env, fst_o, fst_buf);
1361 if (rc < 0 && rc != -ENODATA)
1364 rc = mdd_get_lov_ea(env, snd_o, snd_buf);
1365 if (rc < 0 && rc != -ENODATA)
1368 /* swapping 2 non existant layouts is a success */
1369 if (fst_buf->lb_buf == NULL && snd_buf->lb_buf == NULL)
1372 /* to help inode migration between MDT, it is better to
1373 * start by the no layout file (if one), so we order the swap */
1374 if (snd_buf->lb_buf == NULL) {
1376 swap(fst_buf, snd_buf);
1379 /* lmm and generation layout initialization */
1380 if (fst_buf->lb_buf != NULL) {
1381 fst_lmm = fst_buf->lb_buf;
1382 fst_gen = le16_to_cpu(fst_lmm->lmm_layout_gen);
1383 fst_fl = LU_XATTR_REPLACE;
1387 fst_fl = LU_XATTR_CREATE;
1390 snd_lmm = snd_buf->lb_buf;
1391 snd_gen = le16_to_cpu(snd_lmm->lmm_layout_gen);
1393 /* increase the generation layout numbers */
1397 /* set the file specific informations in lmm */
1398 if (fst_lmm != NULL) {
1399 saved_oi = &info->mti_oa.o_oi;
1401 *saved_oi = fst_lmm->lmm_oi;
1402 fst_lmm->lmm_layout_gen = cpu_to_le16(snd_gen);
1403 fst_lmm->lmm_oi = snd_lmm->lmm_oi;
1404 snd_lmm->lmm_oi = *saved_oi;
1406 if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1))
1407 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEF);
1408 else if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))
1409 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V3_DEF);
1411 GOTO(stop, rc = -EPROTO);
1413 snd_lmm->lmm_layout_gen = cpu_to_le16(fst_gen);
1415 /* Prepare HSM attribute if it's required */
1416 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1417 const int buflen = sizeof(struct hsm_attrs);
1419 lu_buf_alloc(fst_hsm_buf, buflen);
1420 lu_buf_alloc(snd_hsm_buf, buflen);
1421 if (fst_hsm_buf->lb_buf == NULL || snd_hsm_buf->lb_buf == NULL)
1422 GOTO(stop, rc = -ENOMEM);
1424 /* Read HSM attribute */
1425 rc = mdo_xattr_get(env, fst_o, fst_hsm_buf, XATTR_NAME_HSM,
1430 rc = mdo_xattr_get(env, snd_o, snd_hsm_buf, XATTR_NAME_HSM,
1435 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_hsm_buf,
1436 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1441 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_hsm_buf,
1442 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1448 /* prepare transaction */
1449 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_buf, XATTR_NAME_LOV,
1454 if (fst_buf->lb_buf != NULL)
1455 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_buf,
1456 XATTR_NAME_LOV, LU_XATTR_REPLACE,
1459 rc = mdd_declare_xattr_del(env, mdd, snd_o, XATTR_NAME_LOV,
1464 rc = mdd_trans_start(env, mdd, handle);
1468 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1469 rc = mdd_xattr_hsm_replace(env, fst_o, snd_hsm_buf, handle);
1473 rc = mdd_xattr_hsm_replace(env, snd_o, fst_hsm_buf, handle);
1475 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf,
1478 CERROR("%s: restore "DFID" HSM error: %d/%d\n",
1479 mdd_obj_dev_name(fst_o),
1480 PFID(mdo2fid(fst_o)), rc, rc2);
1485 rc = mdo_xattr_set(env, fst_o, snd_buf, XATTR_NAME_LOV, fst_fl, handle,
1486 mdd_object_capa(env, fst_o));
1490 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_MDS_HSM_SWAP_LAYOUTS))) {
1493 if (fst_buf->lb_buf != NULL)
1494 rc = mdo_xattr_set(env, snd_o, fst_buf, XATTR_NAME_LOV,
1495 LU_XATTR_REPLACE, handle,
1496 mdd_object_capa(env, snd_o));
1498 rc = mdo_xattr_del(env, snd_o, XATTR_NAME_LOV, handle,
1499 mdd_object_capa(env, snd_o));
1505 /* failure on second file, but first was done, so we have
1506 * to roll back first. */
1507 if (fst_buf->lb_buf != NULL) {
1508 fst_lmm->lmm_oi = *saved_oi;
1509 fst_lmm->lmm_layout_gen = cpu_to_le16(fst_gen - 1);
1510 rc2 = mdo_xattr_set(env, fst_o, fst_buf, XATTR_NAME_LOV,
1511 LU_XATTR_REPLACE, handle,
1512 mdd_object_capa(env, fst_o));
1514 rc2 = mdo_xattr_del(env, fst_o, XATTR_NAME_LOV, handle,
1515 mdd_object_capa(env, fst_o));
1521 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, handle);
1526 rc2 = mdd_xattr_hsm_replace(env, snd_o, snd_hsm_buf, handle);
1531 CERROR("%s: unable to roll back layout swap. FIDs: "
1532 DFID" and "DFID "error: %d/%d, steps: %d\n",
1533 mdd_obj_dev_name(fst_o),
1534 PFID(mdo2fid(snd_o)), PFID(mdo2fid(fst_o)),
1536 /* a solution to avoid journal commit is to panic,
1537 * but it has strong consequences so we use LBUG to
1538 * allow sysdamin to choose to panic or not
1545 /* Issue one changelog record per file */
1546 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle);
1550 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle);
1556 mdd_trans_stop(env, mdd, rc, handle);
1557 mdd_write_unlock(env, snd_o);
1558 mdd_write_unlock(env, fst_o);
1560 lu_buf_free(fst_buf);
1561 lu_buf_free(snd_buf);
1562 lu_buf_free(fst_hsm_buf);
1563 lu_buf_free(snd_hsm_buf);
1567 void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent,
1568 struct mdd_object *child, const struct lu_attr *attr,
1569 const struct md_op_spec *spec,
1570 struct dt_allocation_hint *hint)
1572 struct dt_object *np = parent ? mdd_object_child(parent) : NULL;
1573 struct dt_object *nc = mdd_object_child(child);
1575 memset(hint, 0, sizeof(*hint));
1577 /* For striped directory, give striping EA to lod_ah_init, which will
1578 * decide the stripe_offset and stripe count by it. */
1579 if (S_ISDIR(attr->la_mode) &&
1580 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1581 hint->dah_eadata = spec->u.sp_ea.eadata;
1582 hint->dah_eadata_len = spec->u.sp_ea.eadatalen;
1584 hint->dah_eadata = NULL;
1585 hint->dah_eadata_len = 0;
1588 CDEBUG(D_INFO, DFID" eadata %p len %d\n", PFID(mdd_object_fid(child)),
1589 hint->dah_eadata, hint->dah_eadata_len);
1590 /* @hint will be initialized by underlying device. */
1591 nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT);
1595 * do NOT or the MAY_*'s, you'll get the weakest
1597 int accmode(const struct lu_env *env, const struct lu_attr *la, int flags)
1601 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1602 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1603 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1604 * owner can write to a file even if it is marked readonly to hide
1605 * its brokenness. (bug 5781) */
1606 if (flags & MDS_OPEN_OWNEROVERRIDE) {
1607 struct lu_ucred *uc = lu_ucred_check(env);
1609 if ((uc == NULL) || (la->la_uid == uc->uc_fsuid))
1613 if (flags & FMODE_READ)
1615 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1617 if (flags & MDS_FMODE_EXEC)
1622 static int mdd_open_sanity_check(const struct lu_env *env,
1623 struct mdd_object *obj,
1624 const struct lu_attr *attr, int flag)
1630 if (mdd_is_dead_obj(obj))
1633 if (S_ISLNK(attr->la_mode))
1636 mode = accmode(env, attr, flag);
1638 if (S_ISDIR(attr->la_mode) && (mode & MAY_WRITE))
1641 if (!(flag & MDS_OPEN_CREATED)) {
1642 rc = mdd_permission_internal(env, obj, attr, mode);
1647 if (S_ISFIFO(attr->la_mode) || S_ISSOCK(attr->la_mode) ||
1648 S_ISBLK(attr->la_mode) || S_ISCHR(attr->la_mode))
1649 flag &= ~MDS_OPEN_TRUNC;
1651 /* For writing append-only file must open it with append mode. */
1652 if (mdd_is_append(obj)) {
1653 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1655 if (flag & MDS_OPEN_TRUNC)
1661 * Now, flag -- O_NOATIME does not be packed by client.
1663 if (flag & O_NOATIME) {
1664 struct lu_ucred *uc = lu_ucred(env);
1666 if (uc && ((uc->uc_valid == UCRED_OLD) ||
1667 (uc->uc_valid == UCRED_NEW)) &&
1668 (uc->uc_fsuid != attr->la_uid) &&
1669 !md_capable(uc, CFS_CAP_FOWNER))
1677 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1680 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1681 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1684 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1686 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
1690 rc = mdd_open_sanity_check(env, mdd_obj, attr, flags);
1692 mdd_obj->mod_count++;
1694 mdd_write_unlock(env, mdd_obj);
1698 int mdd_declare_object_kill(const struct lu_env *env, struct mdd_object *obj,
1699 struct md_attr *ma, struct thandle *handle)
1701 return mdo_declare_destroy(env, obj, handle);
1704 /* return md_attr back,
1705 * if it is last unlink then return lov ea + llog cookie*/
1706 int mdd_object_kill(const struct lu_env *env, struct mdd_object *obj,
1707 struct md_attr *ma, struct thandle *handle)
1712 rc = mdo_destroy(env, obj, handle);
1717 static int mdd_declare_close(const struct lu_env *env,
1718 struct mdd_object *obj,
1720 struct thandle *handle)
1724 rc = orph_declare_index_delete(env, obj, handle);
1728 return mdo_declare_destroy(env, obj, handle);
1732 * No permission check is needed.
1734 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1735 struct md_attr *ma, int mode)
1737 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1738 struct mdd_device *mdd = mdo2mdd(obj);
1739 struct thandle *handle = NULL;
1740 int rc, is_orphan = 0;
1743 if (ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_KEEP_ORPHAN) {
1744 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1745 mdd_obj->mod_count--;
1746 mdd_write_unlock(env, mdd_obj);
1748 if (mdd_obj->mod_flags & ORPHAN_OBJ && !mdd_obj->mod_count)
1749 CDEBUG(D_HA, "Object "DFID" is retained in orphan "
1750 "list\n", PFID(mdd_object_fid(mdd_obj)));
1754 /* mdd_finish_unlink() will always set orphan object as DEAD_OBJ, but
1755 * it might fail to add the object to orphan list (w/o ORPHAN_OBJ). */
1756 /* check without any lock */
1757 is_orphan = mdd_obj->mod_count == 1 &&
1758 (mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0;
1762 handle = mdd_trans_create(env, mdo2mdd(obj));
1764 RETURN(PTR_ERR(handle));
1766 rc = mdd_declare_close(env, mdd_obj, ma, handle);
1770 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1774 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1779 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1780 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr,
1781 mdd_object_capa(env, mdd_obj));
1783 CERROR("Failed to get lu_attr of "DFID": %d\n",
1784 PFID(mdd_object_fid(mdd_obj)), rc);
1788 /* check again with lock */
1789 is_orphan = (mdd_obj->mod_count == 1) &&
1790 ((mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0 ||
1791 ma->ma_attr.la_nlink == 0);
1793 if (is_orphan && handle == NULL) {
1794 mdd_write_unlock(env, mdd_obj);
1798 mdd_obj->mod_count--; /*release open count */
1804 /* NB: Object maybe not in orphan list originally, it is rare case for
1805 * mdd_finish_unlink() failure, in that case, the object doesn't have
1806 * ORPHAN_OBJ flag */
1807 if ((mdd_obj->mod_flags & ORPHAN_OBJ) != 0) {
1808 /* remove link to object from orphan index */
1809 LASSERT(handle != NULL);
1810 rc = __mdd_orphan_del(env, mdd_obj, handle);
1812 CERROR("%s: unable to delete "DFID" from orphan list: "
1813 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1814 PFID(mdd_object_fid(mdd_obj)), rc);
1815 /* If object was not deleted from orphan list, do not
1816 * destroy OSS objects, which will be done when next
1821 CDEBUG(D_HA, "Object "DFID" is deleted from orphan "
1822 "list, OSS objects to be destroyed.\n",
1823 PFID(mdd_object_fid(mdd_obj)));
1826 rc = mdo_destroy(env, mdd_obj, handle);
1829 CERROR("%s: unable to delete "DFID" from orphan list: "
1830 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1831 PFID(mdd_object_fid(mdd_obj)), rc);
1836 mdd_write_unlock(env, mdd_obj);
1839 (mode & (FMODE_WRITE | MDS_OPEN_APPEND | MDS_OPEN_TRUNC)) &&
1840 !(ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_RECOV_OPEN)) {
1841 if (handle == NULL) {
1842 handle = mdd_trans_create(env, mdo2mdd(obj));
1844 GOTO(stop, rc = PTR_ERR(handle));
1846 rc = mdd_declare_changelog_store(env, mdd, NULL,
1851 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1856 mdd_changelog_data_store(env, mdd, CL_CLOSE, mode,
1861 if (handle != NULL && !IS_ERR(handle))
1862 mdd_trans_stop(env, mdd, rc, handle);
1868 * Permission check is done when open,
1869 * no need check again.
1871 static int mdd_readpage_sanity_check(const struct lu_env *env,
1872 struct mdd_object *obj)
1874 struct dt_object *next = mdd_object_child(obj);
1878 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
1886 static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp,
1887 int nob, const struct dt_it_ops *iops,
1888 struct dt_it *it, __u32 attr, void *arg)
1890 struct lu_dirpage *dp = &lp->lp_dir;
1894 struct lu_dirent *ent;
1895 struct lu_dirent *last = NULL;
1898 memset(area, 0, sizeof (*dp));
1899 area += sizeof (*dp);
1900 nob -= sizeof (*dp);
1907 len = iops->key_size(env, it);
1909 /* IAM iterator can return record with zero len. */
1913 hash = iops->store(env, it);
1914 if (unlikely(first)) {
1916 dp->ldp_hash_start = cpu_to_le64(hash);
1919 /* calculate max space required for lu_dirent */
1920 recsize = lu_dirent_calc_size(len, attr);
1922 if (nob >= recsize) {
1923 result = iops->rec(env, it, (struct dt_rec *)ent, attr);
1924 if (result == -ESTALE)
1929 /* osd might not able to pack all attributes,
1930 * so recheck rec length */
1931 recsize = le16_to_cpu(ent->lde_reclen);
1933 result = (last != NULL) ? 0 :-EINVAL;
1937 ent = (void *)ent + recsize;
1941 result = iops->next(env, it);
1942 if (result == -ESTALE)
1944 } while (result == 0);
1947 dp->ldp_hash_end = cpu_to_le64(hash);
1949 if (last->lde_hash == dp->ldp_hash_end)
1950 dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE);
1951 last->lde_reclen = 0; /* end mark */
1954 /* end of directory */
1955 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1956 else if (result < 0)
1957 CWARN("build page failed: %d!\n", result);
1961 int mdd_readpage(const struct lu_env *env, struct md_object *obj,
1962 const struct lu_rdpg *rdpg)
1964 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1968 if (mdd_object_exists(mdd_obj) == 0) {
1969 CERROR("%s: object "DFID" not found: rc = -2\n",
1970 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
1974 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
1975 rc = mdd_readpage_sanity_check(env, mdd_obj);
1977 GOTO(out_unlock, rc);
1979 if (mdd_is_dead_obj(mdd_obj)) {
1981 struct lu_dirpage *dp;
1984 * According to POSIX, please do not return any entry to client:
1985 * even dot and dotdot should not be returned.
1987 CDEBUG(D_INODE, "readdir from dead object: "DFID"\n",
1988 PFID(mdd_object_fid(mdd_obj)));
1990 if (rdpg->rp_count <= 0)
1991 GOTO(out_unlock, rc = -EFAULT);
1992 LASSERT(rdpg->rp_pages != NULL);
1994 pg = rdpg->rp_pages[0];
1995 dp = (struct lu_dirpage *)kmap(pg);
1996 memset(dp, 0 , sizeof(struct lu_dirpage));
1997 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
1998 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1999 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
2001 GOTO(out_unlock, rc = LU_PAGE_SIZE);
2004 rc = dt_index_walk(env, mdd_object_child(mdd_obj), rdpg,
2005 mdd_dir_page_build, NULL);
2007 struct lu_dirpage *dp;
2009 dp = kmap(rdpg->rp_pages[0]);
2010 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
2013 * No pages were processed, mark this for first page
2016 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
2017 rc = min_t(unsigned int, LU_PAGE_SIZE, rdpg->rp_count);
2019 kunmap(rdpg->rp_pages[0]);
2022 GOTO(out_unlock, rc);
2024 mdd_read_unlock(env, mdd_obj);
2028 static int mdd_object_sync(const struct lu_env *env, struct md_object *obj)
2030 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2032 if (mdd_object_exists(mdd_obj) == 0) {
2033 CERROR("%s: object "DFID" not found: rc = -2\n",
2034 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
2037 return dt_object_sync(env, mdd_object_child(mdd_obj));
2040 static int mdd_object_lock(const struct lu_env *env,
2041 struct md_object *obj,
2042 struct lustre_handle *lh,
2043 struct ldlm_enqueue_info *einfo,
2044 ldlm_policy_data_t *policy)
2046 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2047 return dt_object_lock(env, mdd_object_child(mdd_obj), lh,
2051 static int mdd_object_unlock(const struct lu_env *env,
2052 struct md_object *obj,
2053 struct ldlm_enqueue_info *einfo,
2054 ldlm_policy_data_t *policy)
2056 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2057 return dt_object_unlock(env, mdd_object_child(mdd_obj), einfo, policy);
2060 const struct md_object_operations mdd_obj_ops = {
2061 .moo_permission = mdd_permission,
2062 .moo_attr_get = mdd_attr_get,
2063 .moo_attr_set = mdd_attr_set,
2064 .moo_xattr_get = mdd_xattr_get,
2065 .moo_xattr_set = mdd_xattr_set,
2066 .moo_xattr_list = mdd_xattr_list,
2067 .moo_xattr_del = mdd_xattr_del,
2068 .moo_swap_layouts = mdd_swap_layouts,
2069 .moo_open = mdd_open,
2070 .moo_close = mdd_close,
2071 .moo_readpage = mdd_readpage,
2072 .moo_readlink = mdd_readlink,
2073 .moo_changelog = mdd_changelog,
2074 .moo_capa_get = mdd_capa_get,
2075 .moo_object_sync = mdd_object_sync,
2076 .moo_object_lock = mdd_object_lock,
2077 .moo_object_unlock = mdd_object_unlock,