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, 2015, 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>
52 #include <lustre_idmap.h>
53 #include <lustre_param.h>
54 #include <lustre_mds.h>
55 #include <lustre/lustre_idl.h>
57 #include "mdd_internal.h"
59 static const struct lu_object_operations mdd_lu_obj_ops;
61 static int mdd_xattr_get(const struct lu_env *env,
62 struct md_object *obj, struct lu_buf *buf,
65 int mdd_la_get(const struct lu_env *env, struct mdd_object *obj,
70 if (mdd_object_exists(obj) == 0)
73 rc = mdo_attr_get(env, obj, la);
74 if (unlikely(rc != 0)) {
76 obj->mod_flags |= DEAD_OBJ;
80 if (la->la_valid & LA_FLAGS &&
81 la->la_flags & LUSTRE_ORPHAN_FL)
82 obj->mod_flags |= ORPHAN_OBJ | DEAD_OBJ;
87 struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
89 struct mdd_thread_info *info;
91 lu_env_refill((struct lu_env *)env);
92 info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
93 LASSERT(info != NULL);
97 struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
101 buf = &mdd_env_info(env)->mti_buf[0];
107 const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
108 const void *area, ssize_t len)
112 buf = &mdd_env_info(env)->mti_buf[0];
113 buf->lb_buf = (void *)area;
118 struct lu_object *mdd_object_alloc(const struct lu_env *env,
119 const struct lu_object_header *hdr,
122 struct mdd_object *mdd_obj;
124 OBD_SLAB_ALLOC_PTR_GFP(mdd_obj, mdd_object_kmem, GFP_NOFS);
125 if (mdd_obj != NULL) {
128 o = mdd2lu_obj(mdd_obj);
129 lu_object_init(o, NULL, d);
130 mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
131 mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
132 mdd_obj->mod_count = 0;
133 o->lo_ops = &mdd_lu_obj_ops;
140 static int mdd_object_init(const struct lu_env *env, struct lu_object *o,
141 const struct lu_object_conf *unused)
143 struct mdd_device *d = lu2mdd_dev(o->lo_dev);
144 struct mdd_object *mdd_obj = lu2mdd_obj(o);
145 struct lu_object *below;
146 struct lu_device *under;
149 mdd_obj->mod_cltime = 0;
150 under = &d->mdd_child->dd_lu_dev;
151 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
153 RETURN(PTR_ERR(below));
155 lu_object_add(o, below);
160 static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
164 if (lu_object_exists(o)) {
165 struct mdd_object *mdd_obj = lu2mdd_obj(o);
166 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
168 rc = mdd_la_get(env, mdd_obj, attr);
174 static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
176 struct mdd_object *mdd = lu2mdd_obj(o);
179 OBD_SLAB_FREE_PTR(mdd, mdd_object_kmem);
182 static int mdd_object_print(const struct lu_env *env, void *cookie,
183 lu_printer_t p, const struct lu_object *o)
185 struct mdd_object *mdd = lu2mdd_obj((struct lu_object *)o);
186 return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p(open_count=%d, "
187 "valid=%x, cltime="LPU64", flags=%lx)",
188 mdd, mdd->mod_count, mdd->mod_valid,
189 mdd->mod_cltime, mdd->mod_flags);
192 static const struct lu_object_operations mdd_lu_obj_ops = {
193 .loo_object_init = mdd_object_init,
194 .loo_object_start = mdd_object_start,
195 .loo_object_free = mdd_object_free,
196 .loo_object_print = mdd_object_print,
199 struct mdd_object *mdd_object_find(const struct lu_env *env,
200 struct mdd_device *d,
201 const struct lu_fid *f)
203 return md2mdd_obj(md_object_find_slice(env, &d->mdd_md_dev, f));
207 * No permission check is needed.
209 int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
212 struct mdd_object *mdd_obj = md2mdd_obj(obj);
217 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr);
218 if ((ma->ma_need & MA_INODE) != 0 && mdd_is_dead_obj(mdd_obj))
219 ma->ma_attr.la_nlink = 0;
225 * No permission check is needed.
227 static int mdd_xattr_get(const struct lu_env *env,
228 struct md_object *obj, struct lu_buf *buf,
231 struct mdd_object *mdd_obj = md2mdd_obj(obj);
236 if (mdd_object_exists(mdd_obj) == 0) {
237 CERROR("%s: object "DFID" not found: rc = -2\n",
238 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
242 /* If the object has been destroyed, then do not get LMVEA, because
243 * it needs to load stripes from the iteration of the master object,
244 * and it will cause problem if master object has been destroyed, see
246 if (unlikely((mdd_obj->mod_flags & DEAD_OBJ) &&
247 !(mdd_obj->mod_flags & ORPHAN_OBJ) &&
248 strcmp(name, XATTR_NAME_LMV) == 0))
251 /* If the object has been delete from the namespace, then
252 * get linkEA should return -ENOENT as well */
253 if (unlikely((mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ)) &&
254 strcmp(name, XATTR_NAME_LINK) == 0))
257 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
258 rc = mdo_xattr_get(env, mdd_obj, buf, name);
259 mdd_read_unlock(env, mdd_obj);
265 * Permission check is done when open,
266 * no need check again.
268 int mdd_readlink(const struct lu_env *env, struct md_object *obj,
271 struct mdd_object *mdd_obj = md2mdd_obj(obj);
272 struct dt_object *next;
277 if (mdd_object_exists(mdd_obj) == 0) {
278 CERROR("%s: object "DFID" not found: rc = -2\n",
279 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
283 next = mdd_object_child(mdd_obj);
284 LASSERT(next != NULL);
285 LASSERT(next->do_body_ops != NULL);
286 LASSERT(next->do_body_ops->dbo_read != NULL);
287 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
288 rc = dt_read(env, next, buf, &pos);
289 mdd_read_unlock(env, mdd_obj);
294 * No permission check is needed.
296 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
299 struct mdd_object *mdd_obj = md2mdd_obj(obj);
304 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
305 rc = mdo_xattr_list(env, mdd_obj, buf);
306 mdd_read_unlock(env, mdd_obj);
312 * Filter out XATTR_NAME_LINK if this is an orphan object. See
315 if (unlikely(mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ))) {
316 char *end = (char *)buf->lb_buf + rc;
317 char *p = buf->lb_buf;
320 char *next = p + strlen(p) + 1;
322 if (strcmp(p, XATTR_NAME_LINK) == 0) {
324 memmove(p, next, end - next);
326 CDEBUG(D_INFO, "Filtered out "XATTR_NAME_LINK
327 " of orphan "DFID"\n",
328 PFID(mdd_object_fid(mdd_obj)));
339 int mdd_invalidate(const struct lu_env *env, struct md_object *obj)
341 return mdo_invalidate(env, md2mdd_obj(obj));
344 int mdd_declare_object_create_internal(const struct lu_env *env,
345 struct mdd_object *p,
346 struct mdd_object *c,
347 struct lu_attr *attr,
348 struct thandle *handle,
349 const struct md_op_spec *spec,
350 struct dt_allocation_hint *hint)
352 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
353 const struct dt_index_features *feat = spec->sp_feat;
357 if (feat != &dt_directory_features && feat != NULL) {
358 dof->dof_type = DFT_INDEX;
359 dof->u.dof_idx.di_feat = feat;
362 dof->dof_type = dt_mode_to_dft(attr->la_mode);
363 if (dof->dof_type == DFT_REGULAR) {
364 dof->u.dof_reg.striped =
365 md_should_create(spec->sp_cr_flags);
366 if (spec->sp_cr_flags & MDS_OPEN_HAS_EA)
367 dof->u.dof_reg.striped = 0;
368 /* is this replay? */
370 dof->u.dof_reg.striped = 0;
374 rc = mdo_declare_create_obj(env, c, attr, hint, dof, handle);
379 int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p,
380 struct mdd_object *c, struct lu_attr *attr,
381 struct thandle *handle,
382 const struct md_op_spec *spec,
383 struct dt_allocation_hint *hint)
385 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
389 LASSERT(!mdd_object_exists(c));
391 rc = mdo_create_obj(env, c, attr, hint, dof, handle);
396 int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj,
397 const struct lu_attr *attr, struct thandle *handle,
403 rc = mdo_attr_set(env, obj, attr, handle);
404 #ifdef CONFIG_FS_POSIX_ACL
405 if (!rc && (attr->la_valid & LA_MODE) && needacl)
406 rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
411 int mdd_update_time(const struct lu_env *env, struct mdd_object *obj,
412 const struct lu_attr *oattr, struct lu_attr *attr,
413 struct thandle *handle)
418 LASSERT(attr->la_valid & LA_CTIME);
419 LASSERT(oattr != NULL);
421 /* Make sure the ctime is increased only, however, it's not strictly
422 * reliable at here because there is not guarantee to hold lock on
423 * object, so we just bypass some unnecessary cmtime setting first
424 * and OSD has to check it again. */
425 if (attr->la_ctime < oattr->la_ctime)
426 attr->la_valid &= ~(LA_MTIME | LA_CTIME);
427 else if (attr->la_valid == LA_CTIME &&
428 attr->la_ctime == oattr->la_ctime)
429 attr->la_valid &= ~LA_CTIME;
431 if (attr->la_valid != 0)
432 rc = mdd_attr_set_internal(env, obj, attr, handle, 0);
437 * This gives the same functionality as the code between
438 * sys_chmod and inode_setattr
439 * chown_common and inode_setattr
440 * utimes and inode_setattr
441 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
443 static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
444 const struct lu_attr *oattr, struct lu_attr *la,
445 const unsigned long flags)
454 /* Do not permit change file type */
455 if (la->la_valid & LA_TYPE)
458 /* They should not be processed by setattr */
459 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
462 LASSERT(oattr != NULL);
464 uc = lu_ucred_check(env);
468 if (la->la_valid == LA_CTIME) {
469 if (!(flags & MDS_PERM_BYPASS))
470 /* This is only for set ctime when rename's source is
472 rc = mdd_may_delete(env, NULL, NULL, obj, oattr, NULL,
474 if (rc == 0 && la->la_ctime <= oattr->la_ctime)
475 la->la_valid &= ~LA_CTIME;
479 if (la->la_valid == LA_ATIME) {
480 /* This is an atime-only attribute update for close RPCs. */
481 if (la->la_atime < (oattr->la_atime +
482 mdd_obj2mdd_dev(obj)->mdd_atime_diff))
483 la->la_valid &= ~LA_ATIME;
487 /* Check if flags change. */
488 if (la->la_valid & LA_FLAGS) {
489 unsigned int oldflags = oattr->la_flags &
490 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
491 unsigned int newflags = la->la_flags &
492 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
494 if ((uc->uc_fsuid != oattr->la_uid) &&
495 !md_capable(uc, CFS_CAP_FOWNER))
498 /* The IMMUTABLE and APPEND_ONLY flags can
499 * only be changed by the relevant capability. */
500 if ((oldflags ^ newflags) &&
501 !md_capable(uc, CFS_CAP_LINUX_IMMUTABLE))
504 if (!S_ISDIR(oattr->la_mode))
505 la->la_flags &= ~(LUSTRE_DIRSYNC_FL | LUSTRE_TOPDIR_FL);
508 if (oattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL) &&
509 (la->la_valid & ~LA_FLAGS) &&
510 !(flags & MDS_PERM_BYPASS))
513 /* Check for setting the obj time. */
514 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
515 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
516 if ((uc->uc_fsuid != oattr->la_uid) &&
517 !md_capable(uc, CFS_CAP_FOWNER)) {
518 rc = mdd_permission_internal(env, obj, oattr,
525 if (la->la_valid & LA_KILL_SUID) {
526 la->la_valid &= ~LA_KILL_SUID;
527 if ((oattr->la_mode & S_ISUID) &&
528 !(la->la_valid & LA_MODE)) {
529 la->la_mode = oattr->la_mode;
530 la->la_valid |= LA_MODE;
532 la->la_mode &= ~S_ISUID;
535 if (la->la_valid & LA_KILL_SGID) {
536 la->la_valid &= ~LA_KILL_SGID;
537 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
538 (S_ISGID | S_IXGRP)) &&
539 !(la->la_valid & LA_MODE)) {
540 la->la_mode = oattr->la_mode;
541 la->la_valid |= LA_MODE;
543 la->la_mode &= ~S_ISGID;
546 /* Make sure a caller can chmod. */
547 if (la->la_valid & LA_MODE) {
548 if (!(flags & MDS_PERM_BYPASS) &&
549 (uc->uc_fsuid != oattr->la_uid) &&
550 !md_capable(uc, CFS_CAP_FOWNER))
553 if (la->la_mode == (umode_t) -1)
554 la->la_mode = oattr->la_mode;
556 la->la_mode = (la->la_mode & S_IALLUGO) |
557 (oattr->la_mode & ~S_IALLUGO);
559 /* Also check the setgid bit! */
560 if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ?
561 la->la_gid : oattr->la_gid) &&
562 !md_capable(uc, CFS_CAP_FSETID))
563 la->la_mode &= ~S_ISGID;
565 la->la_mode = oattr->la_mode;
568 /* Make sure a caller can chown. */
569 if (la->la_valid & LA_UID) {
570 if (la->la_uid == (uid_t) -1)
571 la->la_uid = oattr->la_uid;
572 if (((uc->uc_fsuid != oattr->la_uid) ||
573 (la->la_uid != oattr->la_uid)) &&
574 !md_capable(uc, CFS_CAP_CHOWN))
577 /* If the user or group of a non-directory has been
578 * changed by a non-root user, remove the setuid bit.
579 * 19981026 David C Niemi <niemi@tux.org>
581 * Changed this to apply to all users, including root,
582 * to avoid some races. This is the behavior we had in
583 * 2.0. The check for non-root was definitely wrong
584 * for 2.2 anyway, as it should have been using
585 * CAP_FSETID rather than fsuid -- 19990830 SD. */
586 if (((oattr->la_mode & S_ISUID) == S_ISUID) &&
587 !S_ISDIR(oattr->la_mode)) {
588 la->la_mode &= ~S_ISUID;
589 la->la_valid |= LA_MODE;
593 /* Make sure caller can chgrp. */
594 if (la->la_valid & LA_GID) {
595 if (la->la_gid == (gid_t) -1)
596 la->la_gid = oattr->la_gid;
597 if (((uc->uc_fsuid != oattr->la_uid) ||
598 ((la->la_gid != oattr->la_gid) &&
599 !lustre_in_group_p(uc, la->la_gid))) &&
600 !md_capable(uc, CFS_CAP_CHOWN))
603 /* Likewise, if the user or group of a non-directory
604 * has been changed by a non-root user, remove the
605 * setgid bit UNLESS there is no group execute bit
606 * (this would be a file marked for mandatory
607 * locking). 19981026 David C Niemi <niemi@tux.org>
609 * Removed the fsuid check (see the comment above) --
611 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
612 (S_ISGID | S_IXGRP)) && !S_ISDIR(oattr->la_mode)) {
613 la->la_mode &= ~S_ISGID;
614 la->la_valid |= LA_MODE;
618 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
619 if (!((flags & MDS_OWNEROVERRIDE) &&
620 (uc->uc_fsuid == oattr->la_uid)) &&
621 !(flags & MDS_PERM_BYPASS)) {
622 rc = mdd_permission_internal(env, obj, oattr,
629 if (la->la_valid & LA_CTIME) {
630 /* The pure setattr, it has the priority over what is
631 * already set, do not drop it if ctime is equal. */
632 if (la->la_ctime < oattr->la_ctime)
633 la->la_valid &= ~(LA_ATIME | LA_MTIME | LA_CTIME);
639 /** Store a data change changelog record
640 * If this fails, we must fail the whole transaction; we don't
641 * want the change to commit without the log entry.
642 * \param mdd_obj - mdd_object of change
643 * \param handle - transaction handle
645 int mdd_changelog_data_store(const struct lu_env *env, struct mdd_device *mdd,
646 enum changelog_rec_type type, int flags,
647 struct mdd_object *mdd_obj, struct thandle *handle)
649 const struct lu_ucred *uc = lu_ucred(env);
650 const struct lu_fid *tfid;
651 struct llog_changelog_rec *rec;
657 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
659 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
662 LASSERT(mdd_obj != NULL);
663 LASSERT(handle != NULL);
665 tfid = mdo2fid(mdd_obj);
667 if ((type >= CL_MTIME) && (type <= CL_ATIME) &&
668 cfs_time_before_64(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) {
669 /* Don't need multiple updates in this log */
670 /* Don't check under lock - no big deal if we get an extra
675 flags = (flags & CLF_FLAGMASK) | CLF_VERSION;
676 if (uc != NULL && uc->uc_jobid[0] != '\0')
679 reclen = llog_data_len(changelog_rec_offset(flags & CLF_SUPPORTED));
680 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
681 if (buf->lb_buf == NULL)
685 rec->cr.cr_flags = flags;
686 rec->cr.cr_type = (__u32)type;
687 rec->cr.cr_tfid = *tfid;
688 rec->cr.cr_namelen = 0;
689 mdd_obj->mod_cltime = cfs_time_current_64();
691 if (flags & CLF_JOBID)
692 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
694 rc = mdd_changelog_store(env, mdd, rec, handle);
699 static int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type,
700 int flags, struct md_object *obj)
702 struct thandle *handle;
703 struct mdd_object *mdd_obj = md2mdd_obj(obj);
704 struct mdd_device *mdd = mdo2mdd(obj);
708 handle = mdd_trans_create(env, mdd);
710 RETURN(PTR_ERR(handle));
712 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
716 rc = mdd_trans_start(env, mdd, handle);
720 rc = mdd_changelog_data_store(env, mdd, type, flags, mdd_obj,
724 rc = mdd_trans_stop(env, mdd, rc, handle);
730 * Save LMA extended attributes with data from \a ma.
732 * HSM and Size-On-MDS data will be extracted from \ma if they are valid, if
733 * not, LMA EA will be first read from disk, modified and write back.
736 /* Precedence for choosing record type when multiple
737 * attributes change: setattr > mtime > ctime > atime
738 * (ctime changes when mtime does, plus chmod/chown.
739 * atime and ctime are independent.) */
740 static int mdd_attr_set_changelog(const struct lu_env *env,
741 struct md_object *obj, struct thandle *handle,
744 struct mdd_device *mdd = mdo2mdd(obj);
747 bits = (valid & LA_SIZE) ? 1 << CL_TRUNC : 0;
748 bits |= (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0;
749 bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0;
750 bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0;
751 bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0;
752 bits = bits & mdd->mdd_cl.mc_mask;
753 /* This is an implementation limit rather than a protocol limit */
754 CLASSERT(CL_LAST <= sizeof(int) * 8);
758 /* The record type is the lowest non-masked set bit */
761 /* FYI we only store the first CLF_FLAGMASK bits of la_valid */
762 return mdd_changelog_data_store(env, mdd, type, (int)valid,
763 md2mdd_obj(obj), handle);
766 static int mdd_declare_attr_set(const struct lu_env *env,
767 struct mdd_device *mdd,
768 struct mdd_object *obj,
769 const struct lu_attr *attr,
770 struct thandle *handle)
774 rc = mdo_declare_attr_set(env, obj, attr, handle);
778 #ifdef CONFIG_FS_POSIX_ACL
779 if (attr->la_valid & LA_MODE) {
780 mdd_read_lock(env, obj, MOR_TGT_CHILD);
781 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL,
782 XATTR_NAME_ACL_ACCESS);
783 mdd_read_unlock(env, obj);
784 if (rc == -EOPNOTSUPP || rc == -ENODATA)
790 struct lu_buf *buf = mdd_buf_get(env, NULL, rc);
791 rc = mdo_declare_xattr_set(env, obj, buf,
792 XATTR_NAME_ACL_ACCESS, 0,
800 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
807 * permission changes may require sync operation, to mitigate performance
808 * impact, only do this for dir and when permission is reduced.
810 * For regular files, version is updated with permission change (see VBR), async
811 * permission won't cause any issue, while missing permission change on
812 * directory may affect accessibility of other objects after recovery.
814 static inline bool permission_needs_sync(const struct lu_attr *old,
815 const struct lu_attr *new)
817 if (!S_ISDIR(old->la_mode))
820 if (new->la_valid & (LA_UID | LA_GID))
823 if (new->la_valid & LA_MODE &&
824 new->la_mode & (S_ISUID | S_ISGID | S_ISVTX))
827 if ((new->la_valid & LA_MODE) &&
828 ((new->la_mode & old->la_mode) & S_IRWXUGO) !=
829 (old->la_mode & S_IRWXUGO))
835 /* set attr and LOV EA at once, return updated attr */
836 int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
837 const struct md_attr *ma)
839 struct mdd_object *mdd_obj = md2mdd_obj(obj);
840 struct mdd_device *mdd = mdo2mdd(obj);
841 struct thandle *handle;
842 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
843 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
844 const struct lu_attr *la = &ma->ma_attr;
848 /* we do not use ->attr_set() for LOV/HSM EA any more */
849 LASSERT((ma->ma_valid & MA_LOV) == 0);
850 LASSERT((ma->ma_valid & MA_HSM) == 0);
852 rc = mdd_la_get(env, mdd_obj, attr);
856 *la_copy = ma->ma_attr;
857 rc = mdd_fix_attr(env, mdd_obj, attr, la_copy, ma->ma_attr_flags);
861 /* no need to setattr anymore */
862 if (la_copy->la_valid == 0) {
863 CDEBUG(D_INODE, "%s: no valid attribute on "DFID", previous"
864 "valid is "LPX64"\n", mdd2obd_dev(mdd)->obd_name,
865 PFID(mdo2fid(mdd_obj)), la->la_valid);
870 handle = mdd_trans_create(env, mdd);
872 RETURN(PTR_ERR(handle));
874 rc = mdd_declare_attr_set(env, mdd, mdd_obj, la_copy, handle);
878 rc = mdd_trans_start(env, mdd, handle);
882 if (mdd->mdd_sync_permission && permission_needs_sync(attr, la))
885 if (la->la_valid & (LA_MTIME | LA_CTIME))
886 CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n",
887 la->la_mtime, la->la_ctime);
889 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
890 if (la_copy->la_valid & LA_FLAGS)
891 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
892 else if (la_copy->la_valid) /* setattr */
893 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
894 mdd_write_unlock(env, mdd_obj);
897 rc = mdd_attr_set_changelog(env, obj, handle,
903 rc = mdd_trans_stop(env, mdd, rc, handle);
908 static int mdd_xattr_sanity_check(const struct lu_env *env,
909 struct mdd_object *obj,
910 const struct lu_attr *attr,
913 struct lu_ucred *uc = lu_ucred_assert(env);
916 if (attr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
919 if (strncmp(XATTR_USER_PREFIX, name,
920 sizeof(XATTR_USER_PREFIX) - 1) == 0) {
921 /* For sticky directories, only the owner and privileged user
922 * can write attributes. */
923 if (S_ISDIR(attr->la_mode) && (attr->la_mode & S_ISVTX) &&
924 (uc->uc_fsuid != attr->la_uid) &&
925 !md_capable(uc, CFS_CAP_FOWNER))
928 if ((uc->uc_fsuid != attr->la_uid) &&
929 !md_capable(uc, CFS_CAP_FOWNER))
936 static int mdd_declare_xattr_set(const struct lu_env *env,
937 struct mdd_device *mdd,
938 struct mdd_object *obj,
939 const struct lu_buf *buf,
941 int fl, struct thandle *handle)
945 rc = mdo_declare_xattr_set(env, obj, buf, name, fl, handle);
949 /* Only record user and layout xattr changes */
950 if (strncmp(XATTR_USER_PREFIX, name,
951 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
952 strcmp(XATTR_NAME_LOV, name) == 0) {
953 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
958 /* If HSM data is modified, this could add a changelog */
959 if (strcmp(XATTR_NAME_HSM, name) == 0) {
960 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
965 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
970 * Compare current and future data of HSM EA and add a changelog if needed.
972 * Caller should have write-locked \param obj.
974 * \param buf - Future HSM EA content.
975 * \retval 0 if no changelog is needed or changelog was added properly.
976 * \retval -ve errno if there was a problem
978 static int mdd_hsm_update_locked(const struct lu_env *env,
979 struct md_object *obj,
980 const struct lu_buf *buf,
981 struct thandle *handle)
983 struct mdd_thread_info *info = mdd_env_info(env);
984 struct mdd_device *mdd = mdo2mdd(obj);
985 struct mdd_object *mdd_obj = md2mdd_obj(obj);
986 struct lu_buf *current_buf;
987 struct md_hsm *current_mh;
988 struct md_hsm *new_mh;
992 OBD_ALLOC_PTR(current_mh);
993 if (current_mh == NULL)
996 /* Read HSM attrs from disk */
997 CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
998 current_buf = mdd_buf_get(env, info->mti_xattr_buf,
999 sizeof(info->mti_xattr_buf));
1000 rc = mdo_xattr_get(env, mdd_obj, current_buf, XATTR_NAME_HSM);
1001 rc = lustre_buf2hsm(current_buf->lb_buf, rc, current_mh);
1002 if (rc < 0 && rc != -ENODATA)
1004 else if (rc == -ENODATA)
1005 current_mh->mh_flags = 0;
1007 /* Map future HSM xattr */
1008 OBD_ALLOC_PTR(new_mh);
1010 GOTO(free, rc = -ENOMEM);
1011 lustre_buf2hsm(buf->lb_buf, buf->lb_len, new_mh);
1013 /* If HSM flags are different, add a changelog */
1015 if (current_mh->mh_flags != new_mh->mh_flags) {
1017 hsm_set_cl_event(&flags, HE_STATE);
1018 if (new_mh->mh_flags & HS_DIRTY)
1019 hsm_set_cl_flags(&flags, CLF_HSM_DIRTY);
1021 rc = mdd_changelog_data_store(env, mdd, CL_HSM, flags, mdd_obj,
1025 OBD_FREE_PTR(new_mh);
1027 OBD_FREE_PTR(current_mh);
1031 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1035 * The caller should guarantee to update the object ctime
1036 * after xattr_set if needed.
1038 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
1039 const struct lu_buf *buf, const char *name,
1042 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1043 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1044 struct mdd_device *mdd = mdo2mdd(obj);
1045 struct thandle *handle;
1049 rc = mdd_la_get(env, mdd_obj, attr);
1053 rc = mdd_xattr_sanity_check(env, mdd_obj, attr, name);
1057 if (strcmp(name, XATTR_NAME_ACL_ACCESS) == 0 ||
1058 strcmp(name, XATTR_NAME_ACL_DEFAULT) == 0) {
1059 struct posix_acl *acl;
1061 /* user may set empty ACL, which should be treated as removing
1063 acl = posix_acl_from_xattr(&init_user_ns, buf->lb_buf,
1066 RETURN(PTR_ERR(acl));
1068 rc = mdd_xattr_del(env, obj, name);
1071 posix_acl_release(acl);
1074 if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) {
1075 rc = mdd_acl_set(env, mdd_obj, attr, buf, fl);
1079 handle = mdd_trans_create(env, mdd);
1081 RETURN(PTR_ERR(handle));
1083 rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, fl, handle);
1087 rc = mdd_trans_start(env, mdd, handle);
1091 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1093 if (strcmp(XATTR_NAME_HSM, name) == 0) {
1094 rc = mdd_hsm_update_locked(env, obj, buf, handle);
1096 mdd_write_unlock(env, mdd_obj);
1101 rc = mdo_xattr_set(env, mdd_obj, buf, name, fl, handle);
1102 mdd_write_unlock(env, mdd_obj);
1106 if (strcmp(XATTR_NAME_LOV, name) == 0)
1107 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, mdd_obj,
1109 else if (strncmp(XATTR_USER_PREFIX, name,
1110 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1111 strncmp(XATTR_NAME_POSIX_ACL_ACCESS, name,
1112 sizeof(XATTR_NAME_POSIX_ACL_ACCESS) - 1) == 0 ||
1113 strncmp(XATTR_NAME_POSIX_ACL_DEFAULT, name,
1114 sizeof(XATTR_NAME_POSIX_ACL_DEFAULT) - 1) == 0)
1115 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1119 rc = mdd_trans_stop(env, mdd, rc, handle);
1124 static int mdd_declare_xattr_del(const struct lu_env *env,
1125 struct mdd_device *mdd,
1126 struct mdd_object *obj,
1128 struct thandle *handle)
1132 rc = mdo_declare_xattr_del(env, obj, name, handle);
1136 /* Only record system & user xattr changes */
1137 if (strncmp(XATTR_USER_PREFIX, name,
1138 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1139 strncmp(XATTR_NAME_POSIX_ACL_ACCESS, name,
1140 sizeof(XATTR_NAME_POSIX_ACL_ACCESS) - 1) == 0 ||
1141 strncmp(XATTR_NAME_POSIX_ACL_DEFAULT, name,
1142 sizeof(XATTR_NAME_POSIX_ACL_DEFAULT) - 1) == 0)
1143 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1149 * The caller should guarantee to update the object ctime
1150 * after xattr_set if needed.
1152 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1155 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1156 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1157 struct mdd_device *mdd = mdo2mdd(obj);
1158 struct thandle *handle;
1162 rc = mdd_la_get(env, mdd_obj, attr);
1166 rc = mdd_xattr_sanity_check(env, mdd_obj, attr, name);
1170 handle = mdd_trans_create(env, mdd);
1172 RETURN(PTR_ERR(handle));
1174 rc = mdd_declare_xattr_del(env, mdd, mdd_obj, name, handle);
1178 rc = mdd_trans_start(env, mdd, handle);
1182 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1183 rc = mdo_xattr_del(env, mdd_obj, name, handle);
1184 mdd_write_unlock(env, mdd_obj);
1188 /* Only record system & user xattr changes */
1189 if (strncmp(XATTR_USER_PREFIX, name,
1190 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1191 strncmp(XATTR_NAME_POSIX_ACL_ACCESS, name,
1192 sizeof(XATTR_NAME_POSIX_ACL_ACCESS) - 1) == 0 ||
1193 strncmp(XATTR_NAME_POSIX_ACL_DEFAULT, name,
1194 sizeof(XATTR_NAME_POSIX_ACL_DEFAULT) - 1) == 0)
1195 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1199 rc = mdd_trans_stop(env, mdd, rc, handle);
1205 * read lov EA of an object
1206 * return the lov EA in an allocated lu_buf
1208 int mdd_get_lov_ea(const struct lu_env *env, struct mdd_object *obj,
1209 struct lu_buf *lmm_buf)
1211 struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf;
1216 rc = mdo_xattr_get(env, obj, buf, XATTR_NAME_LOV);
1218 if (rc == -ERANGE) {
1219 /* mti_big_buf is allocated but is too small
1220 * we need to increase it */
1221 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1223 if (buf->lb_buf == NULL)
1224 GOTO(out, rc = -ENOMEM);
1235 if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) {
1236 /* mti_big_buf was not allocated, so we have to
1237 * allocate it based on the ea size */
1238 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1240 if (buf->lb_buf == NULL)
1241 GOTO(out, rc = -ENOMEM);
1245 lu_buf_alloc(lmm_buf, bufsize);
1246 if (lmm_buf->lb_buf == NULL)
1247 GOTO(out, rc = -ENOMEM);
1249 memcpy(lmm_buf->lb_buf, buf->lb_buf, bufsize);
1255 lu_buf_free(lmm_buf);
1259 static int mdd_xattr_hsm_replace(const struct lu_env *env,
1260 struct mdd_object *o, struct lu_buf *buf,
1261 struct thandle *handle)
1263 struct hsm_attrs *attrs;
1269 rc = mdo_xattr_set(env, o, buf, XATTR_NAME_HSM, LU_XATTR_REPLACE,
1274 attrs = buf->lb_buf;
1275 hsm_flags = le32_to_cpu(attrs->hsm_flags);
1276 if (!(hsm_flags & HS_RELEASED) || mdd_is_dead_obj(o))
1279 /* Add a changelog record for release. */
1280 hsm_set_cl_event(&flags, HE_RELEASE);
1281 rc = mdd_changelog_data_store(env, mdo2mdd(&o->mod_obj), CL_HSM,
1287 * check if layout swapping between 2 objects is allowed
1289 * - only normal FIDs or non-system IGIFs
1290 * - same type of objects
1291 * - same owner/group (so quotas are still valid)
1293 static int mdd_layout_swap_allowed(const struct lu_env *env,
1294 struct mdd_object *o1,
1295 const struct lu_attr *attr1,
1296 struct mdd_object *o2,
1297 const struct lu_attr *attr2)
1299 const struct lu_fid *fid1, *fid2;
1305 if (!fid_is_norm(fid1) &&
1306 (!fid_is_igif(fid1) || IS_ERR(mdd_links_get(env, o1))))
1309 if (!fid_is_norm(fid2) &&
1310 (!fid_is_igif(fid2) || IS_ERR(mdd_links_get(env, o2))))
1313 if (mdd_object_type(o1) != mdd_object_type(o2)) {
1314 if (S_ISDIR(mdd_object_type(o1)))
1316 if (S_ISREG(mdd_object_type(o1)))
1321 if ((attr1->la_uid != attr2->la_uid) ||
1322 (attr1->la_gid != attr2->la_gid))
1329 * swap layouts between 2 lustre objects
1331 static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1,
1332 struct md_object *obj2, __u64 flags)
1334 struct mdd_thread_info *info = mdd_env_info(env);
1335 struct mdd_object *fst_o = md2mdd_obj(obj1);
1336 struct mdd_object *snd_o = md2mdd_obj(obj2);
1337 struct lu_attr *fst_la = MDD_ENV_VAR(env, cattr);
1338 struct lu_attr *snd_la = MDD_ENV_VAR(env, tattr);
1339 struct mdd_device *mdd = mdo2mdd(obj1);
1340 struct lov_mds_md *fst_lmm, *snd_lmm;
1341 struct lu_buf *fst_buf = &info->mti_buf[0];
1342 struct lu_buf *snd_buf = &info->mti_buf[1];
1343 struct lu_buf *fst_hsm_buf = &info->mti_buf[2];
1344 struct lu_buf *snd_hsm_buf = &info->mti_buf[3];
1345 struct ost_id *saved_oi = NULL;
1346 struct thandle *handle;
1347 __u16 fst_gen, snd_gen;
1353 CLASSERT(ARRAY_SIZE(info->mti_buf) >= 4);
1354 memset(info->mti_buf, 0, sizeof(info->mti_buf));
1356 /* we have to sort the 2 obj, so locking will always
1357 * be in the same order, even in case of 2 concurrent swaps */
1358 rc = lu_fid_cmp(mdo2fid(fst_o), mdo2fid(snd_o));
1359 if (rc == 0) /* same fid ? */
1365 rc = mdd_la_get(env, fst_o, fst_la);
1369 rc = mdd_la_get(env, snd_o, snd_la);
1373 /* check if layout swapping is allowed */
1374 rc = mdd_layout_swap_allowed(env, fst_o, fst_la, snd_o, snd_la);
1378 handle = mdd_trans_create(env, mdd);
1380 RETURN(PTR_ERR(handle));
1382 /* objects are already sorted */
1383 mdd_write_lock(env, fst_o, MOR_TGT_CHILD);
1384 mdd_write_lock(env, snd_o, MOR_TGT_CHILD);
1386 rc = mdd_get_lov_ea(env, fst_o, fst_buf);
1387 if (rc < 0 && rc != -ENODATA)
1390 rc = mdd_get_lov_ea(env, snd_o, snd_buf);
1391 if (rc < 0 && rc != -ENODATA)
1394 /* swapping 2 non existant layouts is a success */
1395 if (fst_buf->lb_buf == NULL && snd_buf->lb_buf == NULL)
1398 /* to help inode migration between MDT, it is better to
1399 * start by the no layout file (if one), so we order the swap */
1400 if (snd_buf->lb_buf == NULL) {
1402 swap(fst_buf, snd_buf);
1405 /* lmm and generation layout initialization */
1406 if (fst_buf->lb_buf != NULL) {
1407 fst_lmm = fst_buf->lb_buf;
1408 fst_gen = le16_to_cpu(fst_lmm->lmm_layout_gen);
1409 fst_fl = LU_XATTR_REPLACE;
1413 fst_fl = LU_XATTR_CREATE;
1416 snd_lmm = snd_buf->lb_buf;
1417 snd_gen = le16_to_cpu(snd_lmm->lmm_layout_gen);
1419 /* increase the generation layout numbers */
1423 /* set the file specific informations in lmm */
1424 if (fst_lmm != NULL) {
1425 saved_oi = &info->mti_oa.o_oi;
1427 *saved_oi = fst_lmm->lmm_oi;
1428 fst_lmm->lmm_layout_gen = cpu_to_le16(snd_gen);
1429 fst_lmm->lmm_oi = snd_lmm->lmm_oi;
1430 snd_lmm->lmm_oi = *saved_oi;
1432 if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1))
1433 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEF);
1434 else if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))
1435 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V3_DEF);
1437 GOTO(stop, rc = -EPROTO);
1439 snd_lmm->lmm_layout_gen = cpu_to_le16(fst_gen);
1441 /* Prepare HSM attribute if it's required */
1442 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1443 const int buflen = sizeof(struct hsm_attrs);
1445 lu_buf_alloc(fst_hsm_buf, buflen);
1446 lu_buf_alloc(snd_hsm_buf, buflen);
1447 if (fst_hsm_buf->lb_buf == NULL || snd_hsm_buf->lb_buf == NULL)
1448 GOTO(stop, rc = -ENOMEM);
1450 /* Read HSM attribute */
1451 rc = mdo_xattr_get(env, fst_o, fst_hsm_buf, XATTR_NAME_HSM);
1455 rc = mdo_xattr_get(env, snd_o, snd_hsm_buf, XATTR_NAME_HSM);
1459 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_hsm_buf,
1460 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1465 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_hsm_buf,
1466 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1472 /* prepare transaction */
1473 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_buf, XATTR_NAME_LOV,
1478 if (fst_buf->lb_buf != NULL)
1479 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_buf,
1480 XATTR_NAME_LOV, LU_XATTR_REPLACE,
1483 rc = mdd_declare_xattr_del(env, mdd, snd_o, XATTR_NAME_LOV,
1488 rc = mdd_trans_start(env, mdd, handle);
1492 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1493 rc = mdd_xattr_hsm_replace(env, fst_o, snd_hsm_buf, handle);
1497 rc = mdd_xattr_hsm_replace(env, snd_o, fst_hsm_buf, handle);
1499 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf,
1502 CERROR("%s: restore "DFID" HSM error: %d/%d\n",
1503 mdd_obj_dev_name(fst_o),
1504 PFID(mdo2fid(fst_o)), rc, rc2);
1509 rc = mdo_xattr_set(env, fst_o, snd_buf, XATTR_NAME_LOV, fst_fl, handle);
1513 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_MDS_HSM_SWAP_LAYOUTS))) {
1516 if (fst_buf->lb_buf != NULL)
1517 rc = mdo_xattr_set(env, snd_o, fst_buf, XATTR_NAME_LOV,
1518 LU_XATTR_REPLACE, handle);
1520 rc = mdo_xattr_del(env, snd_o, XATTR_NAME_LOV, handle);
1526 /* failure on second file, but first was done, so we have
1527 * to roll back first. */
1528 if (fst_buf->lb_buf != NULL) {
1529 fst_lmm->lmm_oi = *saved_oi;
1530 fst_lmm->lmm_layout_gen = cpu_to_le16(fst_gen - 1);
1531 rc2 = mdo_xattr_set(env, fst_o, fst_buf, XATTR_NAME_LOV,
1532 LU_XATTR_REPLACE, handle);
1534 rc2 = mdo_xattr_del(env, fst_o, XATTR_NAME_LOV, handle);
1540 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, handle);
1545 rc2 = mdd_xattr_hsm_replace(env, snd_o, snd_hsm_buf, handle);
1550 CERROR("%s: unable to roll back layout swap. FIDs: "
1551 DFID" and "DFID "error: %d/%d, steps: %d\n",
1552 mdd_obj_dev_name(fst_o),
1553 PFID(mdo2fid(snd_o)), PFID(mdo2fid(fst_o)),
1555 /* a solution to avoid journal commit is to panic,
1556 * but it has strong consequences so we use LBUG to
1557 * allow sysdamin to choose to panic or not
1564 /* Issue one changelog record per file */
1565 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle);
1569 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle);
1575 rc = mdd_trans_stop(env, mdd, rc, handle);
1577 mdd_write_unlock(env, snd_o);
1578 mdd_write_unlock(env, fst_o);
1580 lu_buf_free(fst_buf);
1581 lu_buf_free(snd_buf);
1582 lu_buf_free(fst_hsm_buf);
1583 lu_buf_free(snd_hsm_buf);
1587 void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent,
1588 struct mdd_object *child, const struct lu_attr *attr,
1589 const struct md_op_spec *spec,
1590 struct dt_allocation_hint *hint)
1592 struct dt_object *np = parent ? mdd_object_child(parent) : NULL;
1593 struct dt_object *nc = mdd_object_child(child);
1595 memset(hint, 0, sizeof(*hint));
1597 /* For striped directory, give striping EA to lod_ah_init, which will
1598 * decide the stripe_offset and stripe count by it. */
1599 if (S_ISDIR(attr->la_mode) &&
1600 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1601 hint->dah_eadata = spec->u.sp_ea.eadata;
1602 hint->dah_eadata_len = spec->u.sp_ea.eadatalen;
1604 hint->dah_eadata = NULL;
1605 hint->dah_eadata_len = 0;
1608 CDEBUG(D_INFO, DFID" eadata %p len %d\n", PFID(mdd_object_fid(child)),
1609 hint->dah_eadata, hint->dah_eadata_len);
1610 /* @hint will be initialized by underlying device. */
1611 nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT);
1615 * do NOT or the MAY_*'s, you'll get the weakest
1617 int accmode(const struct lu_env *env, const struct lu_attr *la, int flags)
1621 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1622 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1623 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1624 * owner can write to a file even if it is marked readonly to hide
1625 * its brokenness. (bug 5781) */
1626 if (flags & MDS_OPEN_OWNEROVERRIDE) {
1627 struct lu_ucred *uc = lu_ucred_check(env);
1629 if ((uc == NULL) || (la->la_uid == uc->uc_fsuid))
1633 if (flags & FMODE_READ)
1635 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1637 if (flags & MDS_FMODE_EXEC)
1642 static int mdd_open_sanity_check(const struct lu_env *env,
1643 struct mdd_object *obj,
1644 const struct lu_attr *attr, int flag)
1650 if (mdd_is_dead_obj(obj))
1653 if (S_ISLNK(attr->la_mode))
1656 mode = accmode(env, attr, flag);
1658 if (S_ISDIR(attr->la_mode) && (mode & MAY_WRITE))
1661 if (!(flag & MDS_OPEN_CREATED)) {
1662 rc = mdd_permission_internal(env, obj, attr, mode);
1667 if (S_ISFIFO(attr->la_mode) || S_ISSOCK(attr->la_mode) ||
1668 S_ISBLK(attr->la_mode) || S_ISCHR(attr->la_mode))
1669 flag &= ~MDS_OPEN_TRUNC;
1671 /* For writing append-only file must open it with append mode. */
1672 if (attr->la_flags & LUSTRE_APPEND_FL) {
1673 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1675 if (flag & MDS_OPEN_TRUNC)
1682 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1685 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1686 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1689 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1691 rc = mdd_la_get(env, mdd_obj, attr);
1695 rc = mdd_open_sanity_check(env, mdd_obj, attr, flags);
1699 mdd_obj->mod_count++;
1702 mdd_write_unlock(env, mdd_obj);
1706 static int mdd_declare_close(const struct lu_env *env,
1707 struct mdd_object *obj,
1709 struct thandle *handle)
1713 rc = orph_declare_index_delete(env, obj, handle);
1717 return mdo_declare_destroy(env, obj, handle);
1721 * No permission check is needed.
1723 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1724 struct md_attr *ma, int mode)
1726 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1727 struct mdd_device *mdd = mdo2mdd(obj);
1728 struct thandle *handle = NULL;
1733 if (ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_KEEP_ORPHAN) {
1734 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1735 mdd_obj->mod_count--;
1736 mdd_write_unlock(env, mdd_obj);
1738 if (mdd_obj->mod_flags & ORPHAN_OBJ && !mdd_obj->mod_count)
1739 CDEBUG(D_HA, "Object "DFID" is retained in orphan "
1740 "list\n", PFID(mdd_object_fid(mdd_obj)));
1744 /* mdd_finish_unlink() will always set orphan object as DEAD_OBJ, but
1745 * it might fail to add the object to orphan list (w/o ORPHAN_OBJ). */
1746 /* check without any lock */
1747 is_orphan = mdd_obj->mod_count == 1 &&
1748 (mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0;
1752 handle = mdd_trans_create(env, mdo2mdd(obj));
1754 RETURN(PTR_ERR(handle));
1756 rc = mdd_declare_close(env, mdd_obj, ma, handle);
1760 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1764 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1769 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1770 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr);
1772 CERROR("Failed to get lu_attr of "DFID": %d\n",
1773 PFID(mdd_object_fid(mdd_obj)), rc);
1777 /* check again with lock */
1778 is_orphan = (mdd_obj->mod_count == 1) &&
1779 ((mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0 ||
1780 ma->ma_attr.la_nlink == 0);
1782 if (is_orphan && handle == NULL) {
1783 mdd_write_unlock(env, mdd_obj);
1787 mdd_obj->mod_count--; /*release open count */
1793 /* NB: Object maybe not in orphan list originally, it is rare case for
1794 * mdd_finish_unlink() failure, in that case, the object doesn't have
1795 * ORPHAN_OBJ flag */
1796 if ((mdd_obj->mod_flags & ORPHAN_OBJ) != 0) {
1797 /* remove link to object from orphan index */
1798 LASSERT(handle != NULL);
1799 rc = __mdd_orphan_del(env, mdd_obj, handle);
1801 CERROR("%s: unable to delete "DFID" from orphan list: "
1802 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1803 PFID(mdd_object_fid(mdd_obj)), rc);
1804 /* If object was not deleted from orphan list, do not
1805 * destroy OSS objects, which will be done when next
1810 CDEBUG(D_HA, "Object "DFID" is deleted from orphan "
1811 "list, OSS objects to be destroyed.\n",
1812 PFID(mdd_object_fid(mdd_obj)));
1815 rc = mdo_destroy(env, mdd_obj, handle);
1818 CERROR("%s: unable to delete "DFID" from orphan list: "
1819 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1820 PFID(mdd_object_fid(mdd_obj)), rc);
1825 mdd_write_unlock(env, mdd_obj);
1828 (mode & (FMODE_WRITE | MDS_OPEN_APPEND | MDS_OPEN_TRUNC)) &&
1829 !(ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_RECOV_OPEN)) {
1830 if (handle == NULL) {
1831 handle = mdd_trans_create(env, mdo2mdd(obj));
1833 GOTO(stop, rc = PTR_ERR(handle));
1835 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL,
1840 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1845 mdd_changelog_data_store(env, mdd, CL_CLOSE, mode,
1850 if (handle != NULL && !IS_ERR(handle))
1851 rc = mdd_trans_stop(env, mdd, rc, handle);
1857 * Permission check is done when open,
1858 * no need check again.
1860 static int mdd_readpage_sanity_check(const struct lu_env *env,
1861 struct mdd_object *obj)
1863 struct dt_object *next = mdd_object_child(obj);
1867 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
1875 static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp,
1876 size_t nob, const struct dt_it_ops *iops,
1877 struct dt_it *it, __u32 attr, void *arg)
1879 struct lu_dirpage *dp = &lp->lp_dir;
1883 struct lu_dirent *ent;
1884 struct lu_dirent *last = NULL;
1888 if (nob < sizeof(*dp))
1891 memset(area, 0, sizeof (*dp));
1892 area += sizeof (*dp);
1893 nob -= sizeof (*dp);
1900 len = iops->key_size(env, it);
1902 /* IAM iterator can return record with zero len. */
1906 hash = iops->store(env, it);
1907 if (unlikely(first)) {
1909 dp->ldp_hash_start = cpu_to_le64(hash);
1912 /* calculate max space required for lu_dirent */
1913 recsize = lu_dirent_calc_size(len, attr);
1915 if (nob >= recsize) {
1916 result = iops->rec(env, it, (struct dt_rec *)ent, attr);
1917 if (result == -ESTALE)
1922 /* osd might not able to pack all attributes,
1923 * so recheck rec length */
1924 recsize = le16_to_cpu(ent->lde_reclen);
1926 if (le32_to_cpu(ent->lde_attrs) & LUDA_FID) {
1927 fid_le_to_cpu(&fid, &ent->lde_fid);
1928 if (fid_is_dot_lustre(&fid))
1932 result = (last != NULL) ? 0 :-EINVAL;
1936 ent = (void *)ent + recsize;
1940 result = iops->next(env, it);
1941 if (result == -ESTALE)
1943 } while (result == 0);
1946 dp->ldp_hash_end = cpu_to_le64(hash);
1948 if (last->lde_hash == dp->ldp_hash_end)
1949 dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE);
1950 last->lde_reclen = 0; /* end mark */
1953 /* end of directory */
1954 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1955 else if (result < 0)
1956 CWARN("build page failed: %d!\n", result);
1960 int mdd_readpage(const struct lu_env *env, struct md_object *obj,
1961 const struct lu_rdpg *rdpg)
1963 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1967 if (mdd_object_exists(mdd_obj) == 0) {
1968 CERROR("%s: object "DFID" not found: rc = -2\n",
1969 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
1973 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
1974 rc = mdd_readpage_sanity_check(env, mdd_obj);
1976 GOTO(out_unlock, rc);
1978 if (mdd_is_dead_obj(mdd_obj)) {
1980 struct lu_dirpage *dp;
1983 * According to POSIX, please do not return any entry to client:
1984 * even dot and dotdot should not be returned.
1986 CDEBUG(D_INODE, "readdir from dead object: "DFID"\n",
1987 PFID(mdd_object_fid(mdd_obj)));
1989 if (rdpg->rp_count <= 0)
1990 GOTO(out_unlock, rc = -EFAULT);
1991 LASSERT(rdpg->rp_pages != NULL);
1993 pg = rdpg->rp_pages[0];
1994 dp = (struct lu_dirpage *)kmap(pg);
1995 memset(dp, 0 , sizeof(struct lu_dirpage));
1996 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
1997 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1998 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
2000 GOTO(out_unlock, rc = LU_PAGE_SIZE);
2003 rc = dt_index_walk(env, mdd_object_child(mdd_obj), rdpg,
2004 mdd_dir_page_build, NULL);
2006 struct lu_dirpage *dp;
2008 dp = kmap(rdpg->rp_pages[0]);
2009 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
2012 * No pages were processed, mark this for first page
2015 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
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) {
2035 CERROR("%s: object "DFID" not found: rc = %d\n",
2036 mdd_obj_dev_name(mdd_obj),
2037 PFID(mdd_object_fid(mdd_obj)), rc);
2040 return dt_object_sync(env, mdd_object_child(mdd_obj),
2044 static int mdd_object_lock(const struct lu_env *env,
2045 struct md_object *obj,
2046 struct lustre_handle *lh,
2047 struct ldlm_enqueue_info *einfo,
2048 union ldlm_policy_data *policy)
2050 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2051 return dt_object_lock(env, mdd_object_child(mdd_obj), lh,
2055 static int mdd_object_unlock(const struct lu_env *env,
2056 struct md_object *obj,
2057 struct ldlm_enqueue_info *einfo,
2058 union ldlm_policy_data *policy)
2060 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2061 return dt_object_unlock(env, mdd_object_child(mdd_obj), einfo, policy);
2064 const struct md_object_operations mdd_obj_ops = {
2065 .moo_permission = mdd_permission,
2066 .moo_attr_get = mdd_attr_get,
2067 .moo_attr_set = mdd_attr_set,
2068 .moo_xattr_get = mdd_xattr_get,
2069 .moo_xattr_set = mdd_xattr_set,
2070 .moo_xattr_list = mdd_xattr_list,
2071 .moo_invalidate = mdd_invalidate,
2072 .moo_xattr_del = mdd_xattr_del,
2073 .moo_swap_layouts = mdd_swap_layouts,
2074 .moo_open = mdd_open,
2075 .moo_close = mdd_close,
2076 .moo_readpage = mdd_readpage,
2077 .moo_readlink = mdd_readlink,
2078 .moo_changelog = mdd_changelog,
2079 .moo_object_sync = mdd_object_sync,
2080 .moo_object_lock = mdd_object_lock,
2081 .moo_object_unlock = mdd_object_unlock,