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, 2016, 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 <lustre_param.h>
50 #include <lustre_mds.h>
51 #include <lustre/lustre_idl.h>
53 #include "mdd_internal.h"
55 static const struct lu_object_operations mdd_lu_obj_ops;
57 static int mdd_xattr_get(const struct lu_env *env,
58 struct md_object *obj, struct lu_buf *buf,
61 int mdd_la_get(const struct lu_env *env, struct mdd_object *obj,
66 if (mdd_object_exists(obj) == 0)
69 rc = mdo_attr_get(env, obj, la);
70 if (unlikely(rc != 0)) {
72 obj->mod_flags |= DEAD_OBJ;
76 if (la->la_valid & LA_FLAGS &&
77 la->la_flags & LUSTRE_ORPHAN_FL)
78 obj->mod_flags |= ORPHAN_OBJ | DEAD_OBJ;
83 struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
85 struct mdd_thread_info *info;
87 lu_env_refill((struct lu_env *)env);
88 info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
89 LASSERT(info != NULL);
93 struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
97 buf = &mdd_env_info(env)->mti_buf[0];
103 const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
104 const void *area, ssize_t len)
108 buf = &mdd_env_info(env)->mti_buf[0];
109 buf->lb_buf = (void *)area;
114 struct lu_object *mdd_object_alloc(const struct lu_env *env,
115 const struct lu_object_header *hdr,
118 struct mdd_object *mdd_obj;
120 OBD_SLAB_ALLOC_PTR_GFP(mdd_obj, mdd_object_kmem, GFP_NOFS);
121 if (mdd_obj != NULL) {
124 o = mdd2lu_obj(mdd_obj);
125 lu_object_init(o, NULL, d);
126 mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
127 mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
128 mdd_obj->mod_count = 0;
129 o->lo_ops = &mdd_lu_obj_ops;
136 static int mdd_object_init(const struct lu_env *env, struct lu_object *o,
137 const struct lu_object_conf *unused)
139 struct mdd_device *d = lu2mdd_dev(o->lo_dev);
140 struct mdd_object *mdd_obj = lu2mdd_obj(o);
141 struct lu_object *below;
142 struct lu_device *under;
145 mdd_obj->mod_cltime = 0;
146 under = &d->mdd_child->dd_lu_dev;
147 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
149 RETURN(PTR_ERR(below));
151 lu_object_add(o, below);
156 static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
160 if (lu_object_exists(o)) {
161 struct mdd_object *mdd_obj = lu2mdd_obj(o);
162 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
164 rc = mdd_la_get(env, mdd_obj, attr);
170 static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
172 struct mdd_object *mdd = lu2mdd_obj(o);
175 OBD_SLAB_FREE_PTR(mdd, mdd_object_kmem);
178 static int mdd_object_print(const struct lu_env *env, void *cookie,
179 lu_printer_t p, const struct lu_object *o)
181 struct mdd_object *mdd = lu2mdd_obj((struct lu_object *)o);
183 return (*p)(env, cookie,
184 LUSTRE_MDD_NAME"-object@%p(open_count=%d, valid=%x, cltime=%llu, flags=%lx)",
185 mdd, mdd->mod_count, mdd->mod_valid,
186 mdd->mod_cltime, mdd->mod_flags);
189 static const struct lu_object_operations mdd_lu_obj_ops = {
190 .loo_object_init = mdd_object_init,
191 .loo_object_start = mdd_object_start,
192 .loo_object_free = mdd_object_free,
193 .loo_object_print = mdd_object_print,
196 struct mdd_object *mdd_object_find(const struct lu_env *env,
197 struct mdd_device *d,
198 const struct lu_fid *f)
200 return md2mdd_obj(md_object_find_slice(env, &d->mdd_md_dev, f));
204 * No permission check is needed.
206 int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
209 struct mdd_object *mdd_obj = md2mdd_obj(obj);
214 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr);
215 if ((ma->ma_need & MA_INODE) != 0 && mdd_is_dead_obj(mdd_obj))
216 ma->ma_attr.la_nlink = 0;
222 * No permission check is needed.
224 static int mdd_xattr_get(const struct lu_env *env,
225 struct md_object *obj, struct lu_buf *buf,
228 struct mdd_object *mdd_obj = md2mdd_obj(obj);
233 if (mdd_object_exists(mdd_obj) == 0) {
234 CERROR("%s: object "DFID" not found: rc = -2\n",
235 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
239 /* If the object has been destroyed, then do not get LMVEA, because
240 * it needs to load stripes from the iteration of the master object,
241 * and it will cause problem if master object has been destroyed, see
243 if (unlikely((mdd_obj->mod_flags & DEAD_OBJ) &&
244 !(mdd_obj->mod_flags & ORPHAN_OBJ) &&
245 strcmp(name, XATTR_NAME_LMV) == 0))
248 /* If the object has been delete from the namespace, then
249 * get linkEA should return -ENOENT as well */
250 if (unlikely((mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ)) &&
251 strcmp(name, XATTR_NAME_LINK) == 0))
254 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
255 rc = mdo_xattr_get(env, mdd_obj, buf, name);
256 mdd_read_unlock(env, mdd_obj);
262 * Permission check is done when open,
263 * no need check again.
265 int mdd_readlink(const struct lu_env *env, struct md_object *obj,
268 struct mdd_object *mdd_obj = md2mdd_obj(obj);
269 struct dt_object *next;
274 if (mdd_object_exists(mdd_obj) == 0) {
275 CERROR("%s: object "DFID" not found: rc = -2\n",
276 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
280 next = mdd_object_child(mdd_obj);
281 LASSERT(next != NULL);
282 LASSERT(next->do_body_ops != NULL);
283 LASSERT(next->do_body_ops->dbo_read != NULL);
284 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
285 rc = dt_read(env, next, buf, &pos);
286 mdd_read_unlock(env, mdd_obj);
291 * No permission check is needed.
293 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
296 struct mdd_object *mdd_obj = md2mdd_obj(obj);
301 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
302 rc = mdo_xattr_list(env, mdd_obj, buf);
303 mdd_read_unlock(env, mdd_obj);
309 * Filter out XATTR_NAME_LINK if this is an orphan object. See
312 if (unlikely(mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ))) {
313 char *end = (char *)buf->lb_buf + rc;
314 char *p = buf->lb_buf;
317 char *next = p + strlen(p) + 1;
319 if (strcmp(p, XATTR_NAME_LINK) == 0) {
321 memmove(p, next, end - next);
323 CDEBUG(D_INFO, "Filtered out "XATTR_NAME_LINK
324 " of orphan "DFID"\n",
325 PFID(mdd_object_fid(mdd_obj)));
336 int mdd_invalidate(const struct lu_env *env, struct md_object *obj)
338 return mdo_invalidate(env, md2mdd_obj(obj));
341 int mdd_declare_object_create_internal(const struct lu_env *env,
342 struct mdd_object *p,
343 struct mdd_object *c,
344 struct lu_attr *attr,
345 struct thandle *handle,
346 const struct md_op_spec *spec,
347 struct dt_allocation_hint *hint)
349 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
350 const struct dt_index_features *feat = spec->sp_feat;
354 if (feat != &dt_directory_features && feat != NULL) {
355 dof->dof_type = DFT_INDEX;
356 dof->u.dof_idx.di_feat = feat;
359 dof->dof_type = dt_mode_to_dft(attr->la_mode);
360 if (dof->dof_type == DFT_REGULAR) {
361 dof->u.dof_reg.striped =
362 md_should_create(spec->sp_cr_flags);
363 if (spec->sp_cr_flags & MDS_OPEN_HAS_EA)
364 dof->u.dof_reg.striped = 0;
365 /* is this replay? */
367 dof->u.dof_reg.striped = 0;
371 rc = mdo_declare_create_obj(env, c, attr, hint, dof, handle);
376 int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p,
377 struct mdd_object *c, struct lu_attr *attr,
378 struct thandle *handle,
379 const struct md_op_spec *spec,
380 struct dt_allocation_hint *hint)
382 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
386 LASSERT(!mdd_object_exists(c));
388 rc = mdo_create_obj(env, c, attr, hint, dof, handle);
393 int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj,
394 const struct lu_attr *attr, struct thandle *handle,
400 rc = mdo_attr_set(env, obj, attr, handle);
401 #ifdef CONFIG_FS_POSIX_ACL
402 if (!rc && (attr->la_valid & LA_MODE) && needacl)
403 rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
408 int mdd_update_time(const struct lu_env *env, struct mdd_object *obj,
409 const struct lu_attr *oattr, struct lu_attr *attr,
410 struct thandle *handle)
415 LASSERT(attr->la_valid & LA_CTIME);
416 LASSERT(oattr != NULL);
418 /* Make sure the ctime is increased only, however, it's not strictly
419 * reliable at here because there is not guarantee to hold lock on
420 * object, so we just bypass some unnecessary cmtime setting first
421 * and OSD has to check it again. */
422 if (attr->la_ctime < oattr->la_ctime)
423 attr->la_valid &= ~(LA_MTIME | LA_CTIME);
424 else if (attr->la_valid == LA_CTIME &&
425 attr->la_ctime == oattr->la_ctime)
426 attr->la_valid &= ~LA_CTIME;
428 if (attr->la_valid != 0)
429 rc = mdd_attr_set_internal(env, obj, attr, handle, 0);
434 * This gives the same functionality as the code between
435 * sys_chmod and inode_setattr
436 * chown_common and inode_setattr
437 * utimes and inode_setattr
438 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
440 static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
441 const struct lu_attr *oattr, struct lu_attr *la,
442 const unsigned long flags)
451 /* Do not permit change file type */
452 if (la->la_valid & LA_TYPE)
455 /* They should not be processed by setattr */
456 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
459 LASSERT(oattr != NULL);
461 uc = lu_ucred_check(env);
465 if (la->la_valid == LA_CTIME) {
466 if (!(flags & MDS_PERM_BYPASS))
467 /* This is only for set ctime when rename's source is
469 rc = mdd_may_delete(env, NULL, NULL, obj, oattr, NULL,
471 if (rc == 0 && la->la_ctime <= oattr->la_ctime)
472 la->la_valid &= ~LA_CTIME;
476 if (la->la_valid == LA_ATIME) {
477 /* This is an atime-only attribute update for close RPCs. */
478 if (la->la_atime < (oattr->la_atime +
479 mdd_obj2mdd_dev(obj)->mdd_atime_diff))
480 la->la_valid &= ~LA_ATIME;
484 /* Check if flags change. */
485 if (la->la_valid & LA_FLAGS) {
486 unsigned int oldflags = oattr->la_flags &
487 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
488 unsigned int newflags = la->la_flags &
489 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
491 if ((uc->uc_fsuid != oattr->la_uid) &&
492 !md_capable(uc, CFS_CAP_FOWNER))
495 /* The IMMUTABLE and APPEND_ONLY flags can
496 * only be changed by the relevant capability. */
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 | LUSTRE_TOPDIR_FL);
505 if (oattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL) &&
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 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
616 if (!((flags & MDS_OWNEROVERRIDE) &&
617 (uc->uc_fsuid == oattr->la_uid)) &&
618 !(flags & MDS_PERM_BYPASS)) {
619 rc = mdd_permission_internal(env, obj, oattr,
626 if (la->la_valid & LA_CTIME) {
627 /* The pure setattr, it has the priority over what is
628 * already set, do not drop it if ctime is equal. */
629 if (la->la_ctime < oattr->la_ctime)
630 la->la_valid &= ~(LA_ATIME | LA_MTIME | LA_CTIME);
636 static int mdd_changelog_data_store_by_fid(const struct lu_env *env,
637 struct mdd_device *mdd,
638 enum changelog_rec_type type, int flags,
639 const struct lu_fid *fid,
640 struct thandle *handle)
642 const struct lu_ucred *uc = lu_ucred(env);
643 struct llog_changelog_rec *rec;
648 flags = (flags & CLF_FLAGMASK) | CLF_VERSION;
649 if (uc != NULL && uc->uc_jobid[0] != '\0')
652 reclen = llog_data_len(changelog_rec_offset(flags & CLF_SUPPORTED));
653 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
654 if (buf->lb_buf == NULL)
658 rec->cr.cr_flags = flags;
659 rec->cr.cr_type = (__u32)type;
660 rec->cr.cr_tfid = *fid;
661 rec->cr.cr_namelen = 0;
663 if (flags & CLF_JOBID)
664 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
666 rc = mdd_changelog_store(env, mdd, rec, handle);
672 /** Store a data change changelog record
673 * If this fails, we must fail the whole transaction; we don't
674 * want the change to commit without the log entry.
675 * \param mdd_obj - mdd_object of change
676 * \param handle - transaction handle
678 int mdd_changelog_data_store(const struct lu_env *env, struct mdd_device *mdd,
679 enum changelog_rec_type type, int flags,
680 struct mdd_object *mdd_obj, struct thandle *handle)
685 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
687 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
690 LASSERT(mdd_obj != NULL);
691 LASSERT(handle != NULL);
693 if ((type >= CL_MTIME) && (type <= CL_ATIME) &&
694 cfs_time_before_64(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) {
695 /* Don't need multiple updates in this log */
696 /* Don't check under lock - no big deal if we get an extra
701 rc = mdd_changelog_data_store_by_fid(env, mdd, type, flags,
702 mdo2fid(mdd_obj), handle);
704 mdd_obj->mod_cltime = cfs_time_current_64();
709 static int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type,
710 int flags, struct md_device *m, const struct lu_fid *fid)
712 struct thandle *handle;
713 struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
718 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
720 if (!(mdd->mdd_cl.mc_mask & (1 << type)))
723 LASSERT(fid != NULL);
725 handle = mdd_trans_create(env, mdd);
727 RETURN(PTR_ERR(handle));
729 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
733 rc = mdd_trans_start(env, mdd, handle);
737 rc = mdd_changelog_data_store_by_fid(env, mdd, type, flags,
741 rc = mdd_trans_stop(env, mdd, rc, handle);
747 * Save LMA extended attributes with data from \a ma.
749 * HSM and Size-On-MDS data will be extracted from \ma if they are valid, if
750 * not, LMA EA will be first read from disk, modified and write back.
753 /* Precedence for choosing record type when multiple
754 * attributes change: setattr > mtime > ctime > atime
755 * (ctime changes when mtime does, plus chmod/chown.
756 * atime and ctime are independent.) */
757 static int mdd_attr_set_changelog(const struct lu_env *env,
758 struct md_object *obj, struct thandle *handle,
761 struct mdd_device *mdd = mdo2mdd(obj);
764 bits = (valid & LA_SIZE) ? 1 << CL_TRUNC : 0;
765 bits |= (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0;
766 bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0;
767 bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0;
768 bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0;
769 bits = bits & mdd->mdd_cl.mc_mask;
770 /* This is an implementation limit rather than a protocol limit */
771 CLASSERT(CL_LAST <= sizeof(int) * 8);
775 /* The record type is the lowest non-masked set bit */
778 /* FYI we only store the first CLF_FLAGMASK bits of la_valid */
779 return mdd_changelog_data_store(env, mdd, type, (int)valid,
780 md2mdd_obj(obj), handle);
783 static int mdd_declare_attr_set(const struct lu_env *env,
784 struct mdd_device *mdd,
785 struct mdd_object *obj,
786 const struct lu_attr *attr,
787 struct thandle *handle)
791 rc = mdo_declare_attr_set(env, obj, attr, handle);
795 #ifdef CONFIG_FS_POSIX_ACL
796 if (attr->la_valid & LA_MODE) {
797 mdd_read_lock(env, obj, MOR_TGT_CHILD);
798 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL,
799 XATTR_NAME_ACL_ACCESS);
800 mdd_read_unlock(env, obj);
801 if (rc == -EOPNOTSUPP || rc == -ENODATA)
807 struct lu_buf *buf = mdd_buf_get(env, NULL, rc);
808 rc = mdo_declare_xattr_set(env, obj, buf,
809 XATTR_NAME_ACL_ACCESS, 0,
817 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
824 * permission changes may require sync operation, to mitigate performance
825 * impact, only do this for dir and when permission is reduced.
827 * For regular files, version is updated with permission change (see VBR), async
828 * permission won't cause any issue, while missing permission change on
829 * directory may affect accessibility of other objects after recovery.
831 static inline bool permission_needs_sync(const struct lu_attr *old,
832 const struct lu_attr *new)
834 if (!S_ISDIR(old->la_mode))
837 if (new->la_valid & (LA_UID | LA_GID))
840 if (new->la_valid & LA_MODE &&
841 new->la_mode & (S_ISUID | S_ISGID | S_ISVTX))
844 if ((new->la_valid & LA_MODE) &&
845 ((new->la_mode & old->la_mode) & S_IRWXUGO) !=
846 (old->la_mode & S_IRWXUGO))
852 /* set attr and LOV EA at once, return updated attr */
853 int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
854 const struct md_attr *ma)
856 struct mdd_object *mdd_obj = md2mdd_obj(obj);
857 struct mdd_device *mdd = mdo2mdd(obj);
858 struct thandle *handle;
859 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
860 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
861 const struct lu_attr *la = &ma->ma_attr;
865 /* we do not use ->attr_set() for LOV/HSM EA any more */
866 LASSERT((ma->ma_valid & MA_LOV) == 0);
867 LASSERT((ma->ma_valid & MA_HSM) == 0);
869 rc = mdd_la_get(env, mdd_obj, attr);
873 *la_copy = ma->ma_attr;
874 rc = mdd_fix_attr(env, mdd_obj, attr, la_copy, ma->ma_attr_flags);
878 /* no need to setattr anymore */
879 if (la_copy->la_valid == 0) {
880 CDEBUG(D_INODE, "%s: no valid attribute on "DFID", previous"
881 "valid is %#llx\n", mdd2obd_dev(mdd)->obd_name,
882 PFID(mdo2fid(mdd_obj)), la->la_valid);
887 handle = mdd_trans_create(env, mdd);
889 RETURN(PTR_ERR(handle));
891 rc = mdd_declare_attr_set(env, mdd, mdd_obj, la_copy, handle);
895 rc = mdd_trans_start(env, mdd, handle);
899 if (mdd->mdd_sync_permission && permission_needs_sync(attr, la))
902 if (la->la_valid & (LA_MTIME | LA_CTIME))
903 CDEBUG(D_INODE, "setting mtime %llu, ctime %llu\n",
904 la->la_mtime, la->la_ctime);
906 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
907 if (la_copy->la_valid & LA_FLAGS)
908 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
909 else if (la_copy->la_valid) /* setattr */
910 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
911 mdd_write_unlock(env, mdd_obj);
914 rc = mdd_attr_set_changelog(env, obj, handle,
920 rc = mdd_trans_stop(env, mdd, rc, handle);
925 static int mdd_xattr_sanity_check(const struct lu_env *env,
926 struct mdd_object *obj,
927 const struct lu_attr *attr,
930 struct lu_ucred *uc = lu_ucred_assert(env);
933 if (attr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
936 if (strncmp(XATTR_USER_PREFIX, name,
937 sizeof(XATTR_USER_PREFIX) - 1) == 0) {
938 /* For sticky directories, only the owner and privileged user
939 * can write attributes. */
940 if (S_ISDIR(attr->la_mode) && (attr->la_mode & S_ISVTX) &&
941 (uc->uc_fsuid != attr->la_uid) &&
942 !md_capable(uc, CFS_CAP_FOWNER))
945 if ((uc->uc_fsuid != attr->la_uid) &&
946 !md_capable(uc, CFS_CAP_FOWNER))
954 * Check if a string begins with a given prefix.
956 * \param str String to check
957 * \param prefix Substring to check at the beginning of \a str
958 * \return true/false whether the condition is verified.
960 static inline bool has_prefix(const char *str, const char *prefix)
962 return strncmp(prefix, str, strlen(prefix)) == 0;
966 * Indicate the kind of changelog to store (if any) for a xattr set/del.
968 * \param[in] xattr_name Full extended attribute name.
970 * \return The type of changelog to use, or -1 if no changelog is to be emitted.
972 static enum changelog_rec_type
973 mdd_xattr_changelog_type(const struct lu_env *env, struct mdd_device *mdd,
974 const char *xattr_name)
976 /* Layout changes systematically recorded */
977 if (strcmp(XATTR_NAME_LOV, xattr_name) == 0)
980 /* HSM information changes systematically recorded */
981 if (strcmp(XATTR_NAME_HSM, xattr_name) == 0)
984 if (has_prefix(xattr_name, XATTR_USER_PREFIX) ||
985 has_prefix(xattr_name, XATTR_NAME_POSIX_ACL_ACCESS) ||
986 has_prefix(xattr_name, XATTR_NAME_POSIX_ACL_DEFAULT))
992 static int mdd_declare_xattr_set(const struct lu_env *env,
993 struct mdd_device *mdd,
994 struct mdd_object *obj,
995 const struct lu_buf *buf,
997 int fl, struct thandle *handle)
1001 rc = mdo_declare_xattr_set(env, obj, buf, name, fl, handle);
1005 if (mdd_xattr_changelog_type(env, mdd, name) < 0)
1006 return 0; /* no changelog to store */
1008 return mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1012 * Compare current and future data of HSM EA and add a changelog if needed.
1014 * Caller should have write-locked \param obj.
1016 * \param buf - Future HSM EA content.
1017 * \retval 0 if no changelog is needed or changelog was added properly.
1018 * \retval -ve errno if there was a problem
1020 static int mdd_hsm_update_locked(const struct lu_env *env,
1021 struct md_object *obj,
1022 const struct lu_buf *buf,
1023 struct thandle *handle, int *cl_flags)
1025 struct mdd_thread_info *info = mdd_env_info(env);
1026 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1027 struct lu_buf *current_buf;
1028 struct md_hsm *current_mh;
1029 struct md_hsm *new_mh;
1033 OBD_ALLOC_PTR(current_mh);
1034 if (current_mh == NULL)
1037 /* Read HSM attrs from disk */
1038 CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
1039 current_buf = mdd_buf_get(env, info->mti_xattr_buf,
1040 sizeof(info->mti_xattr_buf));
1041 rc = mdo_xattr_get(env, mdd_obj, current_buf, XATTR_NAME_HSM);
1042 rc = lustre_buf2hsm(current_buf->lb_buf, rc, current_mh);
1043 if (rc < 0 && rc != -ENODATA)
1045 else if (rc == -ENODATA)
1046 current_mh->mh_flags = 0;
1048 /* Map future HSM xattr */
1049 OBD_ALLOC_PTR(new_mh);
1051 GOTO(free, rc = -ENOMEM);
1052 lustre_buf2hsm(buf->lb_buf, buf->lb_len, new_mh);
1056 /* Flags differ, set flags for the changelog that will be added */
1057 if (current_mh->mh_flags != new_mh->mh_flags) {
1058 hsm_set_cl_event(cl_flags, HE_STATE);
1059 if (new_mh->mh_flags & HS_DIRTY)
1060 hsm_set_cl_flags(cl_flags, CLF_HSM_DIRTY);
1063 OBD_FREE_PTR(new_mh);
1066 OBD_FREE_PTR(current_mh);
1070 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1074 * The caller should guarantee to update the object ctime
1075 * after xattr_set if needed.
1077 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
1078 const struct lu_buf *buf, const char *name,
1081 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1082 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1083 struct mdd_device *mdd = mdo2mdd(obj);
1084 struct thandle *handle;
1085 enum changelog_rec_type cl_type;
1090 rc = mdd_la_get(env, mdd_obj, attr);
1094 rc = mdd_xattr_sanity_check(env, mdd_obj, attr, name);
1098 if (strcmp(name, XATTR_NAME_ACL_ACCESS) == 0 ||
1099 strcmp(name, XATTR_NAME_ACL_DEFAULT) == 0) {
1100 struct posix_acl *acl;
1102 /* user may set empty ACL, which should be treated as removing
1104 acl = posix_acl_from_xattr(&init_user_ns, buf->lb_buf,
1107 RETURN(PTR_ERR(acl));
1109 rc = mdd_xattr_del(env, obj, name);
1112 posix_acl_release(acl);
1115 if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) {
1116 rc = mdd_acl_set(env, mdd_obj, attr, buf, fl);
1120 handle = mdd_trans_create(env, mdd);
1122 RETURN(PTR_ERR(handle));
1124 rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, fl, handle);
1128 rc = mdd_trans_start(env, mdd, handle);
1132 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1134 if (strcmp(XATTR_NAME_HSM, name) == 0) {
1135 rc = mdd_hsm_update_locked(env, obj, buf, handle, &cl_flags);
1137 mdd_write_unlock(env, mdd_obj);
1142 rc = mdo_xattr_set(env, mdd_obj, buf, name, fl, handle);
1143 mdd_write_unlock(env, mdd_obj);
1147 cl_type = mdd_xattr_changelog_type(env, mdd, name);
1151 rc = mdd_changelog_data_store(env, mdd, cl_type, cl_flags, mdd_obj,
1156 return mdd_trans_stop(env, mdd, rc, handle);
1159 static int mdd_declare_xattr_del(const struct lu_env *env,
1160 struct mdd_device *mdd,
1161 struct mdd_object *obj,
1163 struct thandle *handle)
1167 rc = mdo_declare_xattr_del(env, obj, name, handle);
1171 if (mdd_xattr_changelog_type(env, mdd, name) < 0)
1172 return 0; /* no changelog to store */
1174 return mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1178 * The caller should guarantee to update the object ctime
1179 * after xattr_set if needed.
1181 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1184 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1185 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1186 struct mdd_device *mdd = mdo2mdd(obj);
1187 struct thandle *handle;
1191 rc = mdd_la_get(env, mdd_obj, attr);
1195 rc = mdd_xattr_sanity_check(env, mdd_obj, attr, name);
1199 handle = mdd_trans_create(env, mdd);
1201 RETURN(PTR_ERR(handle));
1203 rc = mdd_declare_xattr_del(env, mdd, mdd_obj, name, handle);
1207 rc = mdd_trans_start(env, mdd, handle);
1211 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1212 rc = mdo_xattr_del(env, mdd_obj, name, handle);
1213 mdd_write_unlock(env, mdd_obj);
1217 if (mdd_xattr_changelog_type(env, mdd, name) < 0)
1220 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj, handle);
1224 return mdd_trans_stop(env, mdd, rc, handle);
1228 * read lov EA of an object
1229 * return the lov EA in an allocated lu_buf
1231 int mdd_get_lov_ea(const struct lu_env *env, struct mdd_object *obj,
1232 struct lu_buf *lmm_buf)
1234 struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf;
1239 rc = mdo_xattr_get(env, obj, buf, XATTR_NAME_LOV);
1241 if (rc == -ERANGE) {
1242 /* mti_big_buf is allocated but is too small
1243 * we need to increase it */
1244 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1246 if (buf->lb_buf == NULL)
1247 GOTO(out, rc = -ENOMEM);
1258 if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) {
1259 /* mti_big_buf was not allocated, so we have to
1260 * allocate it based on the ea size */
1261 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1263 if (buf->lb_buf == NULL)
1264 GOTO(out, rc = -ENOMEM);
1268 lu_buf_alloc(lmm_buf, bufsize);
1269 if (lmm_buf->lb_buf == NULL)
1270 GOTO(out, rc = -ENOMEM);
1272 memcpy(lmm_buf->lb_buf, buf->lb_buf, bufsize);
1278 lu_buf_free(lmm_buf);
1282 static int mdd_xattr_hsm_replace(const struct lu_env *env,
1283 struct mdd_object *o, struct lu_buf *buf,
1284 struct thandle *handle)
1286 struct hsm_attrs *attrs;
1292 rc = mdo_xattr_set(env, o, buf, XATTR_NAME_HSM, LU_XATTR_REPLACE,
1297 attrs = buf->lb_buf;
1298 hsm_flags = le32_to_cpu(attrs->hsm_flags);
1299 if (!(hsm_flags & HS_RELEASED) || mdd_is_dead_obj(o))
1302 /* Add a changelog record for release. */
1303 hsm_set_cl_event(&flags, HE_RELEASE);
1304 rc = mdd_changelog_data_store(env, mdo2mdd(&o->mod_obj), CL_HSM,
1310 * check if layout swapping between 2 objects is allowed
1312 * - only normal FIDs or non-system IGIFs
1313 * - same type of objects
1314 * - same owner/group (so quotas are still valid)
1316 static int mdd_layout_swap_allowed(const struct lu_env *env,
1317 struct mdd_object *o1,
1318 const struct lu_attr *attr1,
1319 struct mdd_object *o2,
1320 const struct lu_attr *attr2)
1322 const struct lu_fid *fid1, *fid2;
1328 if (!fid_is_norm(fid1) &&
1329 (!fid_is_igif(fid1) || IS_ERR(mdd_links_get(env, o1))))
1332 if (!fid_is_norm(fid2) &&
1333 (!fid_is_igif(fid2) || IS_ERR(mdd_links_get(env, o2))))
1336 if (mdd_object_type(o1) != mdd_object_type(o2)) {
1337 if (S_ISDIR(mdd_object_type(o1)))
1339 if (S_ISREG(mdd_object_type(o1)))
1344 if ((attr1->la_uid != attr2->la_uid) ||
1345 (attr1->la_gid != attr2->la_gid))
1352 * swap layouts between 2 lustre objects
1354 static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1,
1355 struct md_object *obj2, __u64 flags)
1357 struct mdd_thread_info *info = mdd_env_info(env);
1358 struct mdd_object *fst_o = md2mdd_obj(obj1);
1359 struct mdd_object *snd_o = md2mdd_obj(obj2);
1360 struct lu_attr *fst_la = MDD_ENV_VAR(env, cattr);
1361 struct lu_attr *snd_la = MDD_ENV_VAR(env, tattr);
1362 struct mdd_device *mdd = mdo2mdd(obj1);
1363 struct lov_mds_md *fst_lmm, *snd_lmm;
1364 struct lu_buf *fst_buf = &info->mti_buf[0];
1365 struct lu_buf *snd_buf = &info->mti_buf[1];
1366 struct lu_buf *fst_hsm_buf = &info->mti_buf[2];
1367 struct lu_buf *snd_hsm_buf = &info->mti_buf[3];
1368 struct ost_id *saved_oi = NULL;
1369 struct thandle *handle;
1370 __u16 fst_gen, snd_gen;
1376 CLASSERT(ARRAY_SIZE(info->mti_buf) >= 4);
1377 memset(info->mti_buf, 0, sizeof(info->mti_buf));
1379 /* we have to sort the 2 obj, so locking will always
1380 * be in the same order, even in case of 2 concurrent swaps */
1381 rc = lu_fid_cmp(mdo2fid(fst_o), mdo2fid(snd_o));
1382 if (rc == 0) /* same fid ? */
1388 rc = mdd_la_get(env, fst_o, fst_la);
1392 rc = mdd_la_get(env, snd_o, snd_la);
1396 /* check if layout swapping is allowed */
1397 rc = mdd_layout_swap_allowed(env, fst_o, fst_la, snd_o, snd_la);
1401 handle = mdd_trans_create(env, mdd);
1403 RETURN(PTR_ERR(handle));
1405 /* objects are already sorted */
1406 mdd_write_lock(env, fst_o, MOR_TGT_CHILD);
1407 mdd_write_lock(env, snd_o, MOR_TGT_CHILD);
1409 rc = mdd_get_lov_ea(env, fst_o, fst_buf);
1410 if (rc < 0 && rc != -ENODATA)
1413 rc = mdd_get_lov_ea(env, snd_o, snd_buf);
1414 if (rc < 0 && rc != -ENODATA)
1417 /* swapping 2 non existant layouts is a success */
1418 if (fst_buf->lb_buf == NULL && snd_buf->lb_buf == NULL)
1421 /* to help inode migration between MDT, it is better to
1422 * start by the no layout file (if one), so we order the swap */
1423 if (snd_buf->lb_buf == NULL) {
1425 swap(fst_buf, snd_buf);
1428 /* lmm and generation layout initialization */
1429 if (fst_buf->lb_buf != NULL) {
1430 fst_lmm = fst_buf->lb_buf;
1431 fst_gen = le16_to_cpu(fst_lmm->lmm_layout_gen);
1432 fst_fl = LU_XATTR_REPLACE;
1436 fst_fl = LU_XATTR_CREATE;
1439 snd_lmm = snd_buf->lb_buf;
1440 snd_gen = le16_to_cpu(snd_lmm->lmm_layout_gen);
1442 /* increase the generation layout numbers */
1446 /* set the file specific informations in lmm */
1447 if (fst_lmm != NULL) {
1448 saved_oi = &info->mti_oa.o_oi;
1450 *saved_oi = fst_lmm->lmm_oi;
1451 fst_lmm->lmm_layout_gen = cpu_to_le16(snd_gen);
1452 fst_lmm->lmm_oi = snd_lmm->lmm_oi;
1453 snd_lmm->lmm_oi = *saved_oi;
1455 if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1))
1456 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEF);
1457 else if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))
1458 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V3_DEF);
1460 GOTO(stop, rc = -EPROTO);
1462 snd_lmm->lmm_layout_gen = cpu_to_le16(fst_gen);
1464 /* Prepare HSM attribute if it's required */
1465 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1466 const int buflen = sizeof(struct hsm_attrs);
1468 lu_buf_alloc(fst_hsm_buf, buflen);
1469 lu_buf_alloc(snd_hsm_buf, buflen);
1470 if (fst_hsm_buf->lb_buf == NULL || snd_hsm_buf->lb_buf == NULL)
1471 GOTO(stop, rc = -ENOMEM);
1473 /* Read HSM attribute */
1474 rc = mdo_xattr_get(env, fst_o, fst_hsm_buf, XATTR_NAME_HSM);
1478 rc = mdo_xattr_get(env, snd_o, snd_hsm_buf, XATTR_NAME_HSM);
1482 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_hsm_buf,
1483 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1488 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_hsm_buf,
1489 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1495 /* prepare transaction */
1496 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_buf, XATTR_NAME_LOV,
1501 if (fst_buf->lb_buf != NULL)
1502 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_buf,
1503 XATTR_NAME_LOV, LU_XATTR_REPLACE,
1506 rc = mdd_declare_xattr_del(env, mdd, snd_o, XATTR_NAME_LOV,
1511 rc = mdd_trans_start(env, mdd, handle);
1515 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1516 rc = mdd_xattr_hsm_replace(env, fst_o, snd_hsm_buf, handle);
1520 rc = mdd_xattr_hsm_replace(env, snd_o, fst_hsm_buf, handle);
1522 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf,
1525 CERROR("%s: restore "DFID" HSM error: %d/%d\n",
1526 mdd_obj_dev_name(fst_o),
1527 PFID(mdo2fid(fst_o)), rc, rc2);
1532 rc = mdo_xattr_set(env, fst_o, snd_buf, XATTR_NAME_LOV, fst_fl, handle);
1536 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_MDS_HSM_SWAP_LAYOUTS))) {
1539 if (fst_buf->lb_buf != NULL)
1540 rc = mdo_xattr_set(env, snd_o, fst_buf, XATTR_NAME_LOV,
1541 LU_XATTR_REPLACE, handle);
1543 rc = mdo_xattr_del(env, snd_o, XATTR_NAME_LOV, handle);
1549 /* failure on second file, but first was done, so we have
1550 * to roll back first. */
1551 if (fst_buf->lb_buf != NULL) {
1552 fst_lmm->lmm_oi = *saved_oi;
1553 fst_lmm->lmm_layout_gen = cpu_to_le16(fst_gen - 1);
1554 rc2 = mdo_xattr_set(env, fst_o, fst_buf, XATTR_NAME_LOV,
1555 LU_XATTR_REPLACE, handle);
1557 rc2 = mdo_xattr_del(env, fst_o, XATTR_NAME_LOV, handle);
1563 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, handle);
1568 rc2 = mdd_xattr_hsm_replace(env, snd_o, snd_hsm_buf, handle);
1573 CERROR("%s: unable to roll back layout swap. FIDs: "
1574 DFID" and "DFID "error: %d/%d, steps: %d\n",
1575 mdd_obj_dev_name(fst_o),
1576 PFID(mdo2fid(snd_o)), PFID(mdo2fid(fst_o)),
1578 /* a solution to avoid journal commit is to panic,
1579 * but it has strong consequences so we use LBUG to
1580 * allow sysdamin to choose to panic or not
1587 /* Issue one changelog record per file */
1588 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle);
1592 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle);
1598 rc = mdd_trans_stop(env, mdd, rc, handle);
1600 mdd_write_unlock(env, snd_o);
1601 mdd_write_unlock(env, fst_o);
1603 lu_buf_free(fst_buf);
1604 lu_buf_free(snd_buf);
1605 lu_buf_free(fst_hsm_buf);
1606 lu_buf_free(snd_hsm_buf);
1610 void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent,
1611 struct mdd_object *child, const struct lu_attr *attr,
1612 const struct md_op_spec *spec,
1613 struct dt_allocation_hint *hint)
1615 struct dt_object *np = parent ? mdd_object_child(parent) : NULL;
1616 struct dt_object *nc = mdd_object_child(child);
1618 memset(hint, 0, sizeof(*hint));
1620 /* For striped directory, give striping EA to lod_ah_init, which will
1621 * decide the stripe_offset and stripe count by it. */
1622 if (S_ISDIR(attr->la_mode) &&
1623 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1624 hint->dah_eadata = spec->u.sp_ea.eadata;
1625 hint->dah_eadata_len = spec->u.sp_ea.eadatalen;
1627 hint->dah_eadata = NULL;
1628 hint->dah_eadata_len = 0;
1631 CDEBUG(D_INFO, DFID" eadata %p len %d\n", PFID(mdd_object_fid(child)),
1632 hint->dah_eadata, hint->dah_eadata_len);
1633 /* @hint will be initialized by underlying device. */
1634 nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT);
1638 * do NOT or the MAY_*'s, you'll get the weakest
1640 int accmode(const struct lu_env *env, const struct lu_attr *la, int flags)
1644 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1645 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1646 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1647 * owner can write to a file even if it is marked readonly to hide
1648 * its brokenness. (bug 5781) */
1649 if (flags & MDS_OPEN_OWNEROVERRIDE) {
1650 struct lu_ucred *uc = lu_ucred_check(env);
1652 if ((uc == NULL) || (la->la_uid == uc->uc_fsuid))
1656 if (flags & FMODE_READ)
1658 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1660 if (flags & MDS_FMODE_EXEC)
1665 static int mdd_open_sanity_check(const struct lu_env *env,
1666 struct mdd_object *obj,
1667 const struct lu_attr *attr, int flag)
1673 if (mdd_is_dead_obj(obj))
1676 if (S_ISLNK(attr->la_mode))
1679 mode = accmode(env, attr, flag);
1681 if (S_ISDIR(attr->la_mode) && (mode & MAY_WRITE))
1684 if (!(flag & MDS_OPEN_CREATED)) {
1685 rc = mdd_permission_internal(env, obj, attr, mode);
1690 if (S_ISFIFO(attr->la_mode) || S_ISSOCK(attr->la_mode) ||
1691 S_ISBLK(attr->la_mode) || S_ISCHR(attr->la_mode))
1692 flag &= ~MDS_OPEN_TRUNC;
1694 /* For writing append-only file must open it with append mode. */
1695 if (attr->la_flags & LUSTRE_APPEND_FL) {
1696 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1698 if (flag & MDS_OPEN_TRUNC)
1705 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1708 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1709 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1712 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1714 rc = mdd_la_get(env, mdd_obj, attr);
1718 rc = mdd_open_sanity_check(env, mdd_obj, attr, flags);
1722 mdd_obj->mod_count++;
1725 mdd_write_unlock(env, mdd_obj);
1729 static int mdd_declare_close(const struct lu_env *env,
1730 struct mdd_object *obj,
1732 struct thandle *handle)
1736 rc = orph_declare_index_delete(env, obj, handle);
1740 return mdo_declare_destroy(env, obj, handle);
1744 * No permission check is needed.
1746 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1747 struct md_attr *ma, int mode)
1749 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1750 struct mdd_device *mdd = mdo2mdd(obj);
1751 struct thandle *handle = NULL;
1756 if (ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_KEEP_ORPHAN) {
1757 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1758 mdd_obj->mod_count--;
1759 mdd_write_unlock(env, mdd_obj);
1761 if (mdd_obj->mod_flags & ORPHAN_OBJ && !mdd_obj->mod_count)
1762 CDEBUG(D_HA, "Object "DFID" is retained in orphan "
1763 "list\n", PFID(mdd_object_fid(mdd_obj)));
1767 /* mdd_finish_unlink() will always set orphan object as DEAD_OBJ, but
1768 * it might fail to add the object to orphan list (w/o ORPHAN_OBJ). */
1769 /* check without any lock */
1770 is_orphan = mdd_obj->mod_count == 1 &&
1771 (mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0;
1775 handle = mdd_trans_create(env, mdo2mdd(obj));
1777 RETURN(PTR_ERR(handle));
1779 rc = mdd_declare_close(env, mdd_obj, ma, handle);
1783 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1787 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1792 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1793 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr);
1795 CERROR("Failed to get lu_attr of "DFID": %d\n",
1796 PFID(mdd_object_fid(mdd_obj)), rc);
1800 /* check again with lock */
1801 is_orphan = (mdd_obj->mod_count == 1) &&
1802 ((mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0 ||
1803 ma->ma_attr.la_nlink == 0);
1805 if (is_orphan && handle == NULL) {
1806 mdd_write_unlock(env, mdd_obj);
1810 mdd_obj->mod_count--; /*release open count */
1816 /* NB: Object maybe not in orphan list originally, it is rare case for
1817 * mdd_finish_unlink() failure, in that case, the object doesn't have
1818 * ORPHAN_OBJ flag */
1819 if ((mdd_obj->mod_flags & ORPHAN_OBJ) != 0) {
1820 /* remove link to object from orphan index */
1821 LASSERT(handle != NULL);
1822 rc = __mdd_orphan_del(env, mdd_obj, handle);
1824 CERROR("%s: unable to delete "DFID" from orphan list: "
1825 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1826 PFID(mdd_object_fid(mdd_obj)), rc);
1827 /* If object was not deleted from orphan list, do not
1828 * destroy OSS objects, which will be done when next
1833 CDEBUG(D_HA, "Object "DFID" is deleted from orphan "
1834 "list, OSS objects to be destroyed.\n",
1835 PFID(mdd_object_fid(mdd_obj)));
1838 rc = mdo_destroy(env, mdd_obj, handle);
1841 CERROR("%s: unable to delete "DFID" from orphan list: "
1842 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1843 PFID(mdd_object_fid(mdd_obj)), rc);
1848 mdd_write_unlock(env, mdd_obj);
1851 (mode & (FMODE_WRITE | MDS_OPEN_APPEND | MDS_OPEN_TRUNC)) &&
1852 !(ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_RECOV_OPEN)) {
1853 if (handle == NULL) {
1854 handle = mdd_trans_create(env, mdo2mdd(obj));
1856 GOTO(stop, rc = PTR_ERR(handle));
1858 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL,
1863 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1868 mdd_changelog_data_store(env, mdd, CL_CLOSE, mode,
1873 if (handle != NULL && !IS_ERR(handle))
1874 rc = mdd_trans_stop(env, mdd, rc, handle);
1880 * Permission check is done when open,
1881 * no need check again.
1883 static int mdd_readpage_sanity_check(const struct lu_env *env,
1884 struct mdd_object *obj)
1886 struct dt_object *next = mdd_object_child(obj);
1890 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
1898 static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp,
1899 size_t nob, const struct dt_it_ops *iops,
1900 struct dt_it *it, __u32 attr, void *arg)
1902 struct lu_dirpage *dp = &lp->lp_dir;
1906 struct lu_dirent *ent;
1907 struct lu_dirent *last = NULL;
1911 if (nob < sizeof(*dp))
1914 memset(area, 0, sizeof (*dp));
1915 area += sizeof (*dp);
1916 nob -= sizeof (*dp);
1923 len = iops->key_size(env, it);
1925 /* IAM iterator can return record with zero len. */
1929 hash = iops->store(env, it);
1930 if (unlikely(first)) {
1932 dp->ldp_hash_start = cpu_to_le64(hash);
1935 /* calculate max space required for lu_dirent */
1936 recsize = lu_dirent_calc_size(len, attr);
1938 if (nob >= recsize) {
1939 result = iops->rec(env, it, (struct dt_rec *)ent, attr);
1940 if (result == -ESTALE)
1945 /* osd might not able to pack all attributes,
1946 * so recheck rec length */
1947 recsize = le16_to_cpu(ent->lde_reclen);
1949 if (le32_to_cpu(ent->lde_attrs) & LUDA_FID) {
1950 fid_le_to_cpu(&fid, &ent->lde_fid);
1951 if (fid_is_dot_lustre(&fid))
1955 result = (last != NULL) ? 0 :-EINVAL;
1959 ent = (void *)ent + recsize;
1963 result = iops->next(env, it);
1964 if (result == -ESTALE)
1966 } while (result == 0);
1969 dp->ldp_hash_end = cpu_to_le64(hash);
1971 if (last->lde_hash == dp->ldp_hash_end)
1972 dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE);
1973 last->lde_reclen = 0; /* end mark */
1976 /* end of directory */
1977 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1978 else if (result < 0)
1979 CWARN("build page failed: %d!\n", result);
1983 int mdd_readpage(const struct lu_env *env, struct md_object *obj,
1984 const struct lu_rdpg *rdpg)
1986 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1990 if (mdd_object_exists(mdd_obj) == 0) {
1991 CERROR("%s: object "DFID" not found: rc = -2\n",
1992 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
1996 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
1997 rc = mdd_readpage_sanity_check(env, mdd_obj);
1999 GOTO(out_unlock, rc);
2001 if (mdd_is_dead_obj(mdd_obj)) {
2003 struct lu_dirpage *dp;
2006 * According to POSIX, please do not return any entry to client:
2007 * even dot and dotdot should not be returned.
2009 CDEBUG(D_INODE, "readdir from dead object: "DFID"\n",
2010 PFID(mdd_object_fid(mdd_obj)));
2012 if (rdpg->rp_count <= 0)
2013 GOTO(out_unlock, rc = -EFAULT);
2014 LASSERT(rdpg->rp_pages != NULL);
2016 pg = rdpg->rp_pages[0];
2017 dp = (struct lu_dirpage *)kmap(pg);
2018 memset(dp, 0 , sizeof(struct lu_dirpage));
2019 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
2020 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
2021 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
2023 GOTO(out_unlock, rc = LU_PAGE_SIZE);
2026 rc = dt_index_walk(env, mdd_object_child(mdd_obj), rdpg,
2027 mdd_dir_page_build, NULL);
2029 struct lu_dirpage *dp;
2031 dp = kmap(rdpg->rp_pages[0]);
2032 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
2035 * No pages were processed, mark this for first page
2038 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
2039 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
2040 rc = min_t(unsigned int, LU_PAGE_SIZE, rdpg->rp_count);
2042 kunmap(rdpg->rp_pages[0]);
2045 GOTO(out_unlock, rc);
2047 mdd_read_unlock(env, mdd_obj);
2051 static int mdd_object_sync(const struct lu_env *env, struct md_object *obj)
2053 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2055 if (mdd_object_exists(mdd_obj) == 0) {
2058 CERROR("%s: object "DFID" not found: rc = %d\n",
2059 mdd_obj_dev_name(mdd_obj),
2060 PFID(mdd_object_fid(mdd_obj)), rc);
2063 return dt_object_sync(env, mdd_object_child(mdd_obj),
2067 static int mdd_object_lock(const struct lu_env *env,
2068 struct md_object *obj,
2069 struct lustre_handle *lh,
2070 struct ldlm_enqueue_info *einfo,
2071 union ldlm_policy_data *policy)
2073 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2074 return dt_object_lock(env, mdd_object_child(mdd_obj), lh,
2078 static int mdd_object_unlock(const struct lu_env *env,
2079 struct md_object *obj,
2080 struct ldlm_enqueue_info *einfo,
2081 union ldlm_policy_data *policy)
2083 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2084 return dt_object_unlock(env, mdd_object_child(mdd_obj), einfo, policy);
2087 const struct md_object_operations mdd_obj_ops = {
2088 .moo_permission = mdd_permission,
2089 .moo_attr_get = mdd_attr_get,
2090 .moo_attr_set = mdd_attr_set,
2091 .moo_xattr_get = mdd_xattr_get,
2092 .moo_xattr_set = mdd_xattr_set,
2093 .moo_xattr_list = mdd_xattr_list,
2094 .moo_invalidate = mdd_invalidate,
2095 .moo_xattr_del = mdd_xattr_del,
2096 .moo_swap_layouts = mdd_swap_layouts,
2097 .moo_open = mdd_open,
2098 .moo_close = mdd_close,
2099 .moo_readpage = mdd_readpage,
2100 .moo_readlink = mdd_readlink,
2101 .moo_changelog = mdd_changelog,
2102 .moo_object_sync = mdd_object_sync,
2103 .moo_object_lock = mdd_object_lock,
2104 .moo_object_unlock = mdd_object_unlock,