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);
187 return (*p)(env, cookie,
188 LUSTRE_MDD_NAME"-object@%p(open_count=%d, valid=%x, cltime=%llu, flags=%lx)",
189 mdd, mdd->mod_count, mdd->mod_valid,
190 mdd->mod_cltime, mdd->mod_flags);
193 static const struct lu_object_operations mdd_lu_obj_ops = {
194 .loo_object_init = mdd_object_init,
195 .loo_object_start = mdd_object_start,
196 .loo_object_free = mdd_object_free,
197 .loo_object_print = mdd_object_print,
200 struct mdd_object *mdd_object_find(const struct lu_env *env,
201 struct mdd_device *d,
202 const struct lu_fid *f)
204 return md2mdd_obj(md_object_find_slice(env, &d->mdd_md_dev, f));
208 * No permission check is needed.
210 int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
213 struct mdd_object *mdd_obj = md2mdd_obj(obj);
218 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr);
219 if ((ma->ma_need & MA_INODE) != 0 && mdd_is_dead_obj(mdd_obj))
220 ma->ma_attr.la_nlink = 0;
226 * No permission check is needed.
228 static int mdd_xattr_get(const struct lu_env *env,
229 struct md_object *obj, struct lu_buf *buf,
232 struct mdd_object *mdd_obj = md2mdd_obj(obj);
237 if (mdd_object_exists(mdd_obj) == 0) {
238 CERROR("%s: object "DFID" not found: rc = -2\n",
239 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
243 /* If the object has been destroyed, then do not get LMVEA, because
244 * it needs to load stripes from the iteration of the master object,
245 * and it will cause problem if master object has been destroyed, see
247 if (unlikely((mdd_obj->mod_flags & DEAD_OBJ) &&
248 !(mdd_obj->mod_flags & ORPHAN_OBJ) &&
249 strcmp(name, XATTR_NAME_LMV) == 0))
252 /* If the object has been delete from the namespace, then
253 * get linkEA should return -ENOENT as well */
254 if (unlikely((mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ)) &&
255 strcmp(name, XATTR_NAME_LINK) == 0))
258 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
259 rc = mdo_xattr_get(env, mdd_obj, buf, name);
260 mdd_read_unlock(env, mdd_obj);
266 * Permission check is done when open,
267 * no need check again.
269 int mdd_readlink(const struct lu_env *env, struct md_object *obj,
272 struct mdd_object *mdd_obj = md2mdd_obj(obj);
273 struct dt_object *next;
278 if (mdd_object_exists(mdd_obj) == 0) {
279 CERROR("%s: object "DFID" not found: rc = -2\n",
280 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
284 next = mdd_object_child(mdd_obj);
285 LASSERT(next != NULL);
286 LASSERT(next->do_body_ops != NULL);
287 LASSERT(next->do_body_ops->dbo_read != NULL);
288 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
289 rc = dt_read(env, next, buf, &pos);
290 mdd_read_unlock(env, mdd_obj);
295 * No permission check is needed.
297 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
300 struct mdd_object *mdd_obj = md2mdd_obj(obj);
305 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
306 rc = mdo_xattr_list(env, mdd_obj, buf);
307 mdd_read_unlock(env, mdd_obj);
313 * Filter out XATTR_NAME_LINK if this is an orphan object. See
316 if (unlikely(mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ))) {
317 char *end = (char *)buf->lb_buf + rc;
318 char *p = buf->lb_buf;
321 char *next = p + strlen(p) + 1;
323 if (strcmp(p, XATTR_NAME_LINK) == 0) {
325 memmove(p, next, end - next);
327 CDEBUG(D_INFO, "Filtered out "XATTR_NAME_LINK
328 " of orphan "DFID"\n",
329 PFID(mdd_object_fid(mdd_obj)));
340 int mdd_invalidate(const struct lu_env *env, struct md_object *obj)
342 return mdo_invalidate(env, md2mdd_obj(obj));
345 int mdd_declare_object_create_internal(const struct lu_env *env,
346 struct mdd_object *p,
347 struct mdd_object *c,
348 struct lu_attr *attr,
349 struct thandle *handle,
350 const struct md_op_spec *spec,
351 struct dt_allocation_hint *hint)
353 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
354 const struct dt_index_features *feat = spec->sp_feat;
358 if (feat != &dt_directory_features && feat != NULL) {
359 dof->dof_type = DFT_INDEX;
360 dof->u.dof_idx.di_feat = feat;
363 dof->dof_type = dt_mode_to_dft(attr->la_mode);
364 if (dof->dof_type == DFT_REGULAR) {
365 dof->u.dof_reg.striped =
366 md_should_create(spec->sp_cr_flags);
367 if (spec->sp_cr_flags & MDS_OPEN_HAS_EA)
368 dof->u.dof_reg.striped = 0;
369 /* is this replay? */
371 dof->u.dof_reg.striped = 0;
375 rc = mdo_declare_create_obj(env, c, attr, hint, dof, handle);
380 int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p,
381 struct mdd_object *c, struct lu_attr *attr,
382 struct thandle *handle,
383 const struct md_op_spec *spec,
384 struct dt_allocation_hint *hint)
386 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
390 LASSERT(!mdd_object_exists(c));
392 rc = mdo_create_obj(env, c, attr, hint, dof, handle);
397 int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj,
398 const struct lu_attr *attr, struct thandle *handle,
404 rc = mdo_attr_set(env, obj, attr, handle);
405 #ifdef CONFIG_FS_POSIX_ACL
406 if (!rc && (attr->la_valid & LA_MODE) && needacl)
407 rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
412 int mdd_update_time(const struct lu_env *env, struct mdd_object *obj,
413 const struct lu_attr *oattr, struct lu_attr *attr,
414 struct thandle *handle)
419 LASSERT(attr->la_valid & LA_CTIME);
420 LASSERT(oattr != NULL);
422 /* Make sure the ctime is increased only, however, it's not strictly
423 * reliable at here because there is not guarantee to hold lock on
424 * object, so we just bypass some unnecessary cmtime setting first
425 * and OSD has to check it again. */
426 if (attr->la_ctime < oattr->la_ctime)
427 attr->la_valid &= ~(LA_MTIME | LA_CTIME);
428 else if (attr->la_valid == LA_CTIME &&
429 attr->la_ctime == oattr->la_ctime)
430 attr->la_valid &= ~LA_CTIME;
432 if (attr->la_valid != 0)
433 rc = mdd_attr_set_internal(env, obj, attr, handle, 0);
438 * This gives the same functionality as the code between
439 * sys_chmod and inode_setattr
440 * chown_common and inode_setattr
441 * utimes and inode_setattr
442 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
444 static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
445 const struct lu_attr *oattr, struct lu_attr *la,
446 const unsigned long flags)
455 /* Do not permit change file type */
456 if (la->la_valid & LA_TYPE)
459 /* They should not be processed by setattr */
460 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
463 LASSERT(oattr != NULL);
465 uc = lu_ucred_check(env);
469 if (la->la_valid == LA_CTIME) {
470 if (!(flags & MDS_PERM_BYPASS))
471 /* This is only for set ctime when rename's source is
473 rc = mdd_may_delete(env, NULL, NULL, obj, oattr, NULL,
475 if (rc == 0 && la->la_ctime <= oattr->la_ctime)
476 la->la_valid &= ~LA_CTIME;
480 if (la->la_valid == LA_ATIME) {
481 /* This is an atime-only attribute update for close RPCs. */
482 if (la->la_atime < (oattr->la_atime +
483 mdd_obj2mdd_dev(obj)->mdd_atime_diff))
484 la->la_valid &= ~LA_ATIME;
488 /* Check if flags change. */
489 if (la->la_valid & LA_FLAGS) {
490 unsigned int oldflags = oattr->la_flags &
491 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
492 unsigned int newflags = la->la_flags &
493 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
495 if ((uc->uc_fsuid != oattr->la_uid) &&
496 !md_capable(uc, CFS_CAP_FOWNER))
499 /* The IMMUTABLE and APPEND_ONLY flags can
500 * only be changed by the relevant capability. */
501 if ((oldflags ^ newflags) &&
502 !md_capable(uc, CFS_CAP_LINUX_IMMUTABLE))
505 if (!S_ISDIR(oattr->la_mode))
506 la->la_flags &= ~(LUSTRE_DIRSYNC_FL | LUSTRE_TOPDIR_FL);
509 if (oattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL) &&
510 (la->la_valid & ~LA_FLAGS) &&
511 !(flags & MDS_PERM_BYPASS))
514 /* Check for setting the obj time. */
515 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
516 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
517 if ((uc->uc_fsuid != oattr->la_uid) &&
518 !md_capable(uc, CFS_CAP_FOWNER)) {
519 rc = mdd_permission_internal(env, obj, oattr,
526 if (la->la_valid & LA_KILL_SUID) {
527 la->la_valid &= ~LA_KILL_SUID;
528 if ((oattr->la_mode & S_ISUID) &&
529 !(la->la_valid & LA_MODE)) {
530 la->la_mode = oattr->la_mode;
531 la->la_valid |= LA_MODE;
533 la->la_mode &= ~S_ISUID;
536 if (la->la_valid & LA_KILL_SGID) {
537 la->la_valid &= ~LA_KILL_SGID;
538 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
539 (S_ISGID | S_IXGRP)) &&
540 !(la->la_valid & LA_MODE)) {
541 la->la_mode = oattr->la_mode;
542 la->la_valid |= LA_MODE;
544 la->la_mode &= ~S_ISGID;
547 /* Make sure a caller can chmod. */
548 if (la->la_valid & LA_MODE) {
549 if (!(flags & MDS_PERM_BYPASS) &&
550 (uc->uc_fsuid != oattr->la_uid) &&
551 !md_capable(uc, CFS_CAP_FOWNER))
554 if (la->la_mode == (umode_t) -1)
555 la->la_mode = oattr->la_mode;
557 la->la_mode = (la->la_mode & S_IALLUGO) |
558 (oattr->la_mode & ~S_IALLUGO);
560 /* Also check the setgid bit! */
561 if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ?
562 la->la_gid : oattr->la_gid) &&
563 !md_capable(uc, CFS_CAP_FSETID))
564 la->la_mode &= ~S_ISGID;
566 la->la_mode = oattr->la_mode;
569 /* Make sure a caller can chown. */
570 if (la->la_valid & LA_UID) {
571 if (la->la_uid == (uid_t) -1)
572 la->la_uid = oattr->la_uid;
573 if (((uc->uc_fsuid != oattr->la_uid) ||
574 (la->la_uid != oattr->la_uid)) &&
575 !md_capable(uc, CFS_CAP_CHOWN))
578 /* If the user or group of a non-directory has been
579 * changed by a non-root user, remove the setuid bit.
580 * 19981026 David C Niemi <niemi@tux.org>
582 * Changed this to apply to all users, including root,
583 * to avoid some races. This is the behavior we had in
584 * 2.0. The check for non-root was definitely wrong
585 * for 2.2 anyway, as it should have been using
586 * CAP_FSETID rather than fsuid -- 19990830 SD. */
587 if (((oattr->la_mode & S_ISUID) == S_ISUID) &&
588 !S_ISDIR(oattr->la_mode)) {
589 la->la_mode &= ~S_ISUID;
590 la->la_valid |= LA_MODE;
594 /* Make sure caller can chgrp. */
595 if (la->la_valid & LA_GID) {
596 if (la->la_gid == (gid_t) -1)
597 la->la_gid = oattr->la_gid;
598 if (((uc->uc_fsuid != oattr->la_uid) ||
599 ((la->la_gid != oattr->la_gid) &&
600 !lustre_in_group_p(uc, la->la_gid))) &&
601 !md_capable(uc, CFS_CAP_CHOWN))
604 /* Likewise, if the user or group of a non-directory
605 * has been changed by a non-root user, remove the
606 * setgid bit UNLESS there is no group execute bit
607 * (this would be a file marked for mandatory
608 * locking). 19981026 David C Niemi <niemi@tux.org>
610 * Removed the fsuid check (see the comment above) --
612 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
613 (S_ISGID | S_IXGRP)) && !S_ISDIR(oattr->la_mode)) {
614 la->la_mode &= ~S_ISGID;
615 la->la_valid |= LA_MODE;
619 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
620 if (!((flags & MDS_OWNEROVERRIDE) &&
621 (uc->uc_fsuid == oattr->la_uid)) &&
622 !(flags & MDS_PERM_BYPASS)) {
623 rc = mdd_permission_internal(env, obj, oattr,
630 if (la->la_valid & LA_CTIME) {
631 /* The pure setattr, it has the priority over what is
632 * already set, do not drop it if ctime is equal. */
633 if (la->la_ctime < oattr->la_ctime)
634 la->la_valid &= ~(LA_ATIME | LA_MTIME | LA_CTIME);
640 /** Store a data change changelog record
641 * If this fails, we must fail the whole transaction; we don't
642 * want the change to commit without the log entry.
643 * \param mdd_obj - mdd_object of change
644 * \param handle - transaction handle
646 int mdd_changelog_data_store(const struct lu_env *env, struct mdd_device *mdd,
647 enum changelog_rec_type type, int flags,
648 struct mdd_object *mdd_obj, struct thandle *handle)
650 const struct lu_ucred *uc = lu_ucred(env);
651 const struct lu_fid *tfid;
652 struct llog_changelog_rec *rec;
658 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
660 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
663 LASSERT(mdd_obj != NULL);
664 LASSERT(handle != NULL);
666 tfid = mdo2fid(mdd_obj);
668 if ((type >= CL_MTIME) && (type <= CL_ATIME) &&
669 cfs_time_before_64(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) {
670 /* Don't need multiple updates in this log */
671 /* Don't check under lock - no big deal if we get an extra
676 flags = (flags & CLF_FLAGMASK) | CLF_VERSION;
677 if (uc != NULL && uc->uc_jobid[0] != '\0')
680 reclen = llog_data_len(changelog_rec_offset(flags & CLF_SUPPORTED));
681 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
682 if (buf->lb_buf == NULL)
686 rec->cr.cr_flags = flags;
687 rec->cr.cr_type = (__u32)type;
688 rec->cr.cr_tfid = *tfid;
689 rec->cr.cr_namelen = 0;
690 mdd_obj->mod_cltime = cfs_time_current_64();
692 if (flags & CLF_JOBID)
693 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
695 rc = mdd_changelog_store(env, mdd, rec, handle);
700 static int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type,
701 int flags, struct md_object *obj)
703 struct thandle *handle;
704 struct mdd_object *mdd_obj = md2mdd_obj(obj);
705 struct mdd_device *mdd = mdo2mdd(obj);
709 handle = mdd_trans_create(env, mdd);
711 RETURN(PTR_ERR(handle));
713 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
717 rc = mdd_trans_start(env, mdd, handle);
721 rc = mdd_changelog_data_store(env, mdd, type, flags, mdd_obj,
725 rc = mdd_trans_stop(env, mdd, rc, handle);
731 * Save LMA extended attributes with data from \a ma.
733 * HSM and Size-On-MDS data will be extracted from \ma if they are valid, if
734 * not, LMA EA will be first read from disk, modified and write back.
737 /* Precedence for choosing record type when multiple
738 * attributes change: setattr > mtime > ctime > atime
739 * (ctime changes when mtime does, plus chmod/chown.
740 * atime and ctime are independent.) */
741 static int mdd_attr_set_changelog(const struct lu_env *env,
742 struct md_object *obj, struct thandle *handle,
745 struct mdd_device *mdd = mdo2mdd(obj);
748 bits = (valid & LA_SIZE) ? 1 << CL_TRUNC : 0;
749 bits |= (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0;
750 bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0;
751 bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0;
752 bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0;
753 bits = bits & mdd->mdd_cl.mc_mask;
754 /* This is an implementation limit rather than a protocol limit */
755 CLASSERT(CL_LAST <= sizeof(int) * 8);
759 /* The record type is the lowest non-masked set bit */
762 /* FYI we only store the first CLF_FLAGMASK bits of la_valid */
763 return mdd_changelog_data_store(env, mdd, type, (int)valid,
764 md2mdd_obj(obj), handle);
767 static int mdd_declare_attr_set(const struct lu_env *env,
768 struct mdd_device *mdd,
769 struct mdd_object *obj,
770 const struct lu_attr *attr,
771 struct thandle *handle)
775 rc = mdo_declare_attr_set(env, obj, attr, handle);
779 #ifdef CONFIG_FS_POSIX_ACL
780 if (attr->la_valid & LA_MODE) {
781 mdd_read_lock(env, obj, MOR_TGT_CHILD);
782 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL,
783 XATTR_NAME_ACL_ACCESS);
784 mdd_read_unlock(env, obj);
785 if (rc == -EOPNOTSUPP || rc == -ENODATA)
791 struct lu_buf *buf = mdd_buf_get(env, NULL, rc);
792 rc = mdo_declare_xattr_set(env, obj, buf,
793 XATTR_NAME_ACL_ACCESS, 0,
801 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
808 * permission changes may require sync operation, to mitigate performance
809 * impact, only do this for dir and when permission is reduced.
811 * For regular files, version is updated with permission change (see VBR), async
812 * permission won't cause any issue, while missing permission change on
813 * directory may affect accessibility of other objects after recovery.
815 static inline bool permission_needs_sync(const struct lu_attr *old,
816 const struct lu_attr *new)
818 if (!S_ISDIR(old->la_mode))
821 if (new->la_valid & (LA_UID | LA_GID))
824 if (new->la_valid & LA_MODE &&
825 new->la_mode & (S_ISUID | S_ISGID | S_ISVTX))
828 if ((new->la_valid & LA_MODE) &&
829 ((new->la_mode & old->la_mode) & S_IRWXUGO) !=
830 (old->la_mode & S_IRWXUGO))
836 /* set attr and LOV EA at once, return updated attr */
837 int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
838 const struct md_attr *ma)
840 struct mdd_object *mdd_obj = md2mdd_obj(obj);
841 struct mdd_device *mdd = mdo2mdd(obj);
842 struct thandle *handle;
843 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
844 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
845 const struct lu_attr *la = &ma->ma_attr;
849 /* we do not use ->attr_set() for LOV/HSM EA any more */
850 LASSERT((ma->ma_valid & MA_LOV) == 0);
851 LASSERT((ma->ma_valid & MA_HSM) == 0);
853 rc = mdd_la_get(env, mdd_obj, attr);
857 *la_copy = ma->ma_attr;
858 rc = mdd_fix_attr(env, mdd_obj, attr, la_copy, ma->ma_attr_flags);
862 /* no need to setattr anymore */
863 if (la_copy->la_valid == 0) {
864 CDEBUG(D_INODE, "%s: no valid attribute on "DFID", previous"
865 "valid is %#llx\n", mdd2obd_dev(mdd)->obd_name,
866 PFID(mdo2fid(mdd_obj)), la->la_valid);
871 handle = mdd_trans_create(env, mdd);
873 RETURN(PTR_ERR(handle));
875 rc = mdd_declare_attr_set(env, mdd, mdd_obj, la_copy, handle);
879 rc = mdd_trans_start(env, mdd, handle);
883 if (mdd->mdd_sync_permission && permission_needs_sync(attr, la))
886 if (la->la_valid & (LA_MTIME | LA_CTIME))
887 CDEBUG(D_INODE, "setting mtime %llu, ctime %llu\n",
888 la->la_mtime, la->la_ctime);
890 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
891 if (la_copy->la_valid & LA_FLAGS)
892 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
893 else if (la_copy->la_valid) /* setattr */
894 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
895 mdd_write_unlock(env, mdd_obj);
898 rc = mdd_attr_set_changelog(env, obj, handle,
904 rc = mdd_trans_stop(env, mdd, rc, handle);
909 static int mdd_xattr_sanity_check(const struct lu_env *env,
910 struct mdd_object *obj,
911 const struct lu_attr *attr,
914 struct lu_ucred *uc = lu_ucred_assert(env);
917 if (attr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
920 if (strncmp(XATTR_USER_PREFIX, name,
921 sizeof(XATTR_USER_PREFIX) - 1) == 0) {
922 /* For sticky directories, only the owner and privileged user
923 * can write attributes. */
924 if (S_ISDIR(attr->la_mode) && (attr->la_mode & S_ISVTX) &&
925 (uc->uc_fsuid != attr->la_uid) &&
926 !md_capable(uc, CFS_CAP_FOWNER))
929 if ((uc->uc_fsuid != attr->la_uid) &&
930 !md_capable(uc, CFS_CAP_FOWNER))
937 static int mdd_declare_xattr_set(const struct lu_env *env,
938 struct mdd_device *mdd,
939 struct mdd_object *obj,
940 const struct lu_buf *buf,
942 int fl, struct thandle *handle)
946 rc = mdo_declare_xattr_set(env, obj, buf, name, fl, handle);
950 /* Only record user and layout xattr changes */
951 if (strncmp(XATTR_USER_PREFIX, name,
952 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
953 strcmp(XATTR_NAME_LOV, name) == 0) {
954 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
959 /* If HSM data is modified, this could add a changelog */
960 if (strcmp(XATTR_NAME_HSM, name) == 0) {
961 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
966 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
971 * Compare current and future data of HSM EA and add a changelog if needed.
973 * Caller should have write-locked \param obj.
975 * \param buf - Future HSM EA content.
976 * \retval 0 if no changelog is needed or changelog was added properly.
977 * \retval -ve errno if there was a problem
979 static int mdd_hsm_update_locked(const struct lu_env *env,
980 struct md_object *obj,
981 const struct lu_buf *buf,
982 struct thandle *handle)
984 struct mdd_thread_info *info = mdd_env_info(env);
985 struct mdd_device *mdd = mdo2mdd(obj);
986 struct mdd_object *mdd_obj = md2mdd_obj(obj);
987 struct lu_buf *current_buf;
988 struct md_hsm *current_mh;
989 struct md_hsm *new_mh;
993 OBD_ALLOC_PTR(current_mh);
994 if (current_mh == NULL)
997 /* Read HSM attrs from disk */
998 CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
999 current_buf = mdd_buf_get(env, info->mti_xattr_buf,
1000 sizeof(info->mti_xattr_buf));
1001 rc = mdo_xattr_get(env, mdd_obj, current_buf, XATTR_NAME_HSM);
1002 rc = lustre_buf2hsm(current_buf->lb_buf, rc, current_mh);
1003 if (rc < 0 && rc != -ENODATA)
1005 else if (rc == -ENODATA)
1006 current_mh->mh_flags = 0;
1008 /* Map future HSM xattr */
1009 OBD_ALLOC_PTR(new_mh);
1011 GOTO(free, rc = -ENOMEM);
1012 lustre_buf2hsm(buf->lb_buf, buf->lb_len, new_mh);
1014 /* If HSM flags are different, add a changelog */
1016 if (current_mh->mh_flags != new_mh->mh_flags) {
1018 hsm_set_cl_event(&flags, HE_STATE);
1019 if (new_mh->mh_flags & HS_DIRTY)
1020 hsm_set_cl_flags(&flags, CLF_HSM_DIRTY);
1022 rc = mdd_changelog_data_store(env, mdd, CL_HSM, flags, mdd_obj,
1026 OBD_FREE_PTR(new_mh);
1028 OBD_FREE_PTR(current_mh);
1032 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1036 * The caller should guarantee to update the object ctime
1037 * after xattr_set if needed.
1039 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
1040 const struct lu_buf *buf, const char *name,
1043 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1044 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1045 struct mdd_device *mdd = mdo2mdd(obj);
1046 struct thandle *handle;
1050 rc = mdd_la_get(env, mdd_obj, attr);
1054 rc = mdd_xattr_sanity_check(env, mdd_obj, attr, name);
1058 if (strcmp(name, XATTR_NAME_ACL_ACCESS) == 0 ||
1059 strcmp(name, XATTR_NAME_ACL_DEFAULT) == 0) {
1060 struct posix_acl *acl;
1062 /* user may set empty ACL, which should be treated as removing
1064 acl = posix_acl_from_xattr(&init_user_ns, buf->lb_buf,
1067 RETURN(PTR_ERR(acl));
1069 rc = mdd_xattr_del(env, obj, name);
1072 posix_acl_release(acl);
1075 if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) {
1076 rc = mdd_acl_set(env, mdd_obj, attr, buf, fl);
1080 handle = mdd_trans_create(env, mdd);
1082 RETURN(PTR_ERR(handle));
1084 rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, fl, handle);
1088 rc = mdd_trans_start(env, mdd, handle);
1092 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1094 if (strcmp(XATTR_NAME_HSM, name) == 0) {
1095 rc = mdd_hsm_update_locked(env, obj, buf, handle);
1097 mdd_write_unlock(env, mdd_obj);
1102 rc = mdo_xattr_set(env, mdd_obj, buf, name, fl, handle);
1103 mdd_write_unlock(env, mdd_obj);
1107 if (strcmp(XATTR_NAME_LOV, name) == 0)
1108 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, mdd_obj,
1110 else if (strncmp(XATTR_USER_PREFIX, name,
1111 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1112 strncmp(XATTR_NAME_POSIX_ACL_ACCESS, name,
1113 sizeof(XATTR_NAME_POSIX_ACL_ACCESS) - 1) == 0 ||
1114 strncmp(XATTR_NAME_POSIX_ACL_DEFAULT, name,
1115 sizeof(XATTR_NAME_POSIX_ACL_DEFAULT) - 1) == 0)
1116 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1120 rc = mdd_trans_stop(env, mdd, rc, handle);
1125 static int mdd_declare_xattr_del(const struct lu_env *env,
1126 struct mdd_device *mdd,
1127 struct mdd_object *obj,
1129 struct thandle *handle)
1133 rc = mdo_declare_xattr_del(env, obj, name, handle);
1137 /* Only record system & user xattr changes */
1138 if (strncmp(XATTR_USER_PREFIX, name,
1139 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1140 strncmp(XATTR_NAME_POSIX_ACL_ACCESS, name,
1141 sizeof(XATTR_NAME_POSIX_ACL_ACCESS) - 1) == 0 ||
1142 strncmp(XATTR_NAME_POSIX_ACL_DEFAULT, name,
1143 sizeof(XATTR_NAME_POSIX_ACL_DEFAULT) - 1) == 0)
1144 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1150 * The caller should guarantee to update the object ctime
1151 * after xattr_set if needed.
1153 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1156 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1157 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1158 struct mdd_device *mdd = mdo2mdd(obj);
1159 struct thandle *handle;
1163 rc = mdd_la_get(env, mdd_obj, attr);
1167 rc = mdd_xattr_sanity_check(env, mdd_obj, attr, name);
1171 handle = mdd_trans_create(env, mdd);
1173 RETURN(PTR_ERR(handle));
1175 rc = mdd_declare_xattr_del(env, mdd, mdd_obj, name, handle);
1179 rc = mdd_trans_start(env, mdd, handle);
1183 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1184 rc = mdo_xattr_del(env, mdd_obj, name, handle);
1185 mdd_write_unlock(env, mdd_obj);
1189 /* Only record system & user xattr changes */
1190 if (strncmp(XATTR_USER_PREFIX, name,
1191 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1192 strncmp(XATTR_NAME_POSIX_ACL_ACCESS, name,
1193 sizeof(XATTR_NAME_POSIX_ACL_ACCESS) - 1) == 0 ||
1194 strncmp(XATTR_NAME_POSIX_ACL_DEFAULT, name,
1195 sizeof(XATTR_NAME_POSIX_ACL_DEFAULT) - 1) == 0)
1196 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1200 rc = mdd_trans_stop(env, mdd, rc, handle);
1206 * read lov EA of an object
1207 * return the lov EA in an allocated lu_buf
1209 int mdd_get_lov_ea(const struct lu_env *env, struct mdd_object *obj,
1210 struct lu_buf *lmm_buf)
1212 struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf;
1217 rc = mdo_xattr_get(env, obj, buf, XATTR_NAME_LOV);
1219 if (rc == -ERANGE) {
1220 /* mti_big_buf is allocated but is too small
1221 * we need to increase it */
1222 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1224 if (buf->lb_buf == NULL)
1225 GOTO(out, rc = -ENOMEM);
1236 if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) {
1237 /* mti_big_buf was not allocated, so we have to
1238 * allocate it based on the ea size */
1239 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1241 if (buf->lb_buf == NULL)
1242 GOTO(out, rc = -ENOMEM);
1246 lu_buf_alloc(lmm_buf, bufsize);
1247 if (lmm_buf->lb_buf == NULL)
1248 GOTO(out, rc = -ENOMEM);
1250 memcpy(lmm_buf->lb_buf, buf->lb_buf, bufsize);
1256 lu_buf_free(lmm_buf);
1260 static int mdd_xattr_hsm_replace(const struct lu_env *env,
1261 struct mdd_object *o, struct lu_buf *buf,
1262 struct thandle *handle)
1264 struct hsm_attrs *attrs;
1270 rc = mdo_xattr_set(env, o, buf, XATTR_NAME_HSM, LU_XATTR_REPLACE,
1275 attrs = buf->lb_buf;
1276 hsm_flags = le32_to_cpu(attrs->hsm_flags);
1277 if (!(hsm_flags & HS_RELEASED) || mdd_is_dead_obj(o))
1280 /* Add a changelog record for release. */
1281 hsm_set_cl_event(&flags, HE_RELEASE);
1282 rc = mdd_changelog_data_store(env, mdo2mdd(&o->mod_obj), CL_HSM,
1288 * check if layout swapping between 2 objects is allowed
1290 * - only normal FIDs or non-system IGIFs
1291 * - same type of objects
1292 * - same owner/group (so quotas are still valid)
1294 static int mdd_layout_swap_allowed(const struct lu_env *env,
1295 struct mdd_object *o1,
1296 const struct lu_attr *attr1,
1297 struct mdd_object *o2,
1298 const struct lu_attr *attr2)
1300 const struct lu_fid *fid1, *fid2;
1306 if (!fid_is_norm(fid1) &&
1307 (!fid_is_igif(fid1) || IS_ERR(mdd_links_get(env, o1))))
1310 if (!fid_is_norm(fid2) &&
1311 (!fid_is_igif(fid2) || IS_ERR(mdd_links_get(env, o2))))
1314 if (mdd_object_type(o1) != mdd_object_type(o2)) {
1315 if (S_ISDIR(mdd_object_type(o1)))
1317 if (S_ISREG(mdd_object_type(o1)))
1322 if ((attr1->la_uid != attr2->la_uid) ||
1323 (attr1->la_gid != attr2->la_gid))
1330 * swap layouts between 2 lustre objects
1332 static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1,
1333 struct md_object *obj2, __u64 flags)
1335 struct mdd_thread_info *info = mdd_env_info(env);
1336 struct mdd_object *fst_o = md2mdd_obj(obj1);
1337 struct mdd_object *snd_o = md2mdd_obj(obj2);
1338 struct lu_attr *fst_la = MDD_ENV_VAR(env, cattr);
1339 struct lu_attr *snd_la = MDD_ENV_VAR(env, tattr);
1340 struct mdd_device *mdd = mdo2mdd(obj1);
1341 struct lov_mds_md *fst_lmm, *snd_lmm;
1342 struct lu_buf *fst_buf = &info->mti_buf[0];
1343 struct lu_buf *snd_buf = &info->mti_buf[1];
1344 struct lu_buf *fst_hsm_buf = &info->mti_buf[2];
1345 struct lu_buf *snd_hsm_buf = &info->mti_buf[3];
1346 struct ost_id *saved_oi = NULL;
1347 struct thandle *handle;
1348 __u16 fst_gen, snd_gen;
1354 CLASSERT(ARRAY_SIZE(info->mti_buf) >= 4);
1355 memset(info->mti_buf, 0, sizeof(info->mti_buf));
1357 /* we have to sort the 2 obj, so locking will always
1358 * be in the same order, even in case of 2 concurrent swaps */
1359 rc = lu_fid_cmp(mdo2fid(fst_o), mdo2fid(snd_o));
1360 if (rc == 0) /* same fid ? */
1366 rc = mdd_la_get(env, fst_o, fst_la);
1370 rc = mdd_la_get(env, snd_o, snd_la);
1374 /* check if layout swapping is allowed */
1375 rc = mdd_layout_swap_allowed(env, fst_o, fst_la, snd_o, snd_la);
1379 handle = mdd_trans_create(env, mdd);
1381 RETURN(PTR_ERR(handle));
1383 /* objects are already sorted */
1384 mdd_write_lock(env, fst_o, MOR_TGT_CHILD);
1385 mdd_write_lock(env, snd_o, MOR_TGT_CHILD);
1387 rc = mdd_get_lov_ea(env, fst_o, fst_buf);
1388 if (rc < 0 && rc != -ENODATA)
1391 rc = mdd_get_lov_ea(env, snd_o, snd_buf);
1392 if (rc < 0 && rc != -ENODATA)
1395 /* swapping 2 non existant layouts is a success */
1396 if (fst_buf->lb_buf == NULL && snd_buf->lb_buf == NULL)
1399 /* to help inode migration between MDT, it is better to
1400 * start by the no layout file (if one), so we order the swap */
1401 if (snd_buf->lb_buf == NULL) {
1403 swap(fst_buf, snd_buf);
1406 /* lmm and generation layout initialization */
1407 if (fst_buf->lb_buf != NULL) {
1408 fst_lmm = fst_buf->lb_buf;
1409 fst_gen = le16_to_cpu(fst_lmm->lmm_layout_gen);
1410 fst_fl = LU_XATTR_REPLACE;
1414 fst_fl = LU_XATTR_CREATE;
1417 snd_lmm = snd_buf->lb_buf;
1418 snd_gen = le16_to_cpu(snd_lmm->lmm_layout_gen);
1420 /* increase the generation layout numbers */
1424 /* set the file specific informations in lmm */
1425 if (fst_lmm != NULL) {
1426 saved_oi = &info->mti_oa.o_oi;
1428 *saved_oi = fst_lmm->lmm_oi;
1429 fst_lmm->lmm_layout_gen = cpu_to_le16(snd_gen);
1430 fst_lmm->lmm_oi = snd_lmm->lmm_oi;
1431 snd_lmm->lmm_oi = *saved_oi;
1433 if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1))
1434 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEF);
1435 else if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))
1436 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V3_DEF);
1438 GOTO(stop, rc = -EPROTO);
1440 snd_lmm->lmm_layout_gen = cpu_to_le16(fst_gen);
1442 /* Prepare HSM attribute if it's required */
1443 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1444 const int buflen = sizeof(struct hsm_attrs);
1446 lu_buf_alloc(fst_hsm_buf, buflen);
1447 lu_buf_alloc(snd_hsm_buf, buflen);
1448 if (fst_hsm_buf->lb_buf == NULL || snd_hsm_buf->lb_buf == NULL)
1449 GOTO(stop, rc = -ENOMEM);
1451 /* Read HSM attribute */
1452 rc = mdo_xattr_get(env, fst_o, fst_hsm_buf, XATTR_NAME_HSM);
1456 rc = mdo_xattr_get(env, snd_o, snd_hsm_buf, XATTR_NAME_HSM);
1460 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_hsm_buf,
1461 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1466 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_hsm_buf,
1467 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1473 /* prepare transaction */
1474 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_buf, XATTR_NAME_LOV,
1479 if (fst_buf->lb_buf != NULL)
1480 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_buf,
1481 XATTR_NAME_LOV, LU_XATTR_REPLACE,
1484 rc = mdd_declare_xattr_del(env, mdd, snd_o, XATTR_NAME_LOV,
1489 rc = mdd_trans_start(env, mdd, handle);
1493 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1494 rc = mdd_xattr_hsm_replace(env, fst_o, snd_hsm_buf, handle);
1498 rc = mdd_xattr_hsm_replace(env, snd_o, fst_hsm_buf, handle);
1500 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf,
1503 CERROR("%s: restore "DFID" HSM error: %d/%d\n",
1504 mdd_obj_dev_name(fst_o),
1505 PFID(mdo2fid(fst_o)), rc, rc2);
1510 rc = mdo_xattr_set(env, fst_o, snd_buf, XATTR_NAME_LOV, fst_fl, handle);
1514 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_MDS_HSM_SWAP_LAYOUTS))) {
1517 if (fst_buf->lb_buf != NULL)
1518 rc = mdo_xattr_set(env, snd_o, fst_buf, XATTR_NAME_LOV,
1519 LU_XATTR_REPLACE, handle);
1521 rc = mdo_xattr_del(env, snd_o, XATTR_NAME_LOV, handle);
1527 /* failure on second file, but first was done, so we have
1528 * to roll back first. */
1529 if (fst_buf->lb_buf != NULL) {
1530 fst_lmm->lmm_oi = *saved_oi;
1531 fst_lmm->lmm_layout_gen = cpu_to_le16(fst_gen - 1);
1532 rc2 = mdo_xattr_set(env, fst_o, fst_buf, XATTR_NAME_LOV,
1533 LU_XATTR_REPLACE, handle);
1535 rc2 = mdo_xattr_del(env, fst_o, XATTR_NAME_LOV, handle);
1541 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, handle);
1546 rc2 = mdd_xattr_hsm_replace(env, snd_o, snd_hsm_buf, handle);
1551 CERROR("%s: unable to roll back layout swap. FIDs: "
1552 DFID" and "DFID "error: %d/%d, steps: %d\n",
1553 mdd_obj_dev_name(fst_o),
1554 PFID(mdo2fid(snd_o)), PFID(mdo2fid(fst_o)),
1556 /* a solution to avoid journal commit is to panic,
1557 * but it has strong consequences so we use LBUG to
1558 * allow sysdamin to choose to panic or not
1565 /* Issue one changelog record per file */
1566 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle);
1570 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle);
1576 rc = mdd_trans_stop(env, mdd, rc, handle);
1578 mdd_write_unlock(env, snd_o);
1579 mdd_write_unlock(env, fst_o);
1581 lu_buf_free(fst_buf);
1582 lu_buf_free(snd_buf);
1583 lu_buf_free(fst_hsm_buf);
1584 lu_buf_free(snd_hsm_buf);
1588 void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent,
1589 struct mdd_object *child, const struct lu_attr *attr,
1590 const struct md_op_spec *spec,
1591 struct dt_allocation_hint *hint)
1593 struct dt_object *np = parent ? mdd_object_child(parent) : NULL;
1594 struct dt_object *nc = mdd_object_child(child);
1596 memset(hint, 0, sizeof(*hint));
1598 /* For striped directory, give striping EA to lod_ah_init, which will
1599 * decide the stripe_offset and stripe count by it. */
1600 if (S_ISDIR(attr->la_mode) &&
1601 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1602 hint->dah_eadata = spec->u.sp_ea.eadata;
1603 hint->dah_eadata_len = spec->u.sp_ea.eadatalen;
1605 hint->dah_eadata = NULL;
1606 hint->dah_eadata_len = 0;
1609 CDEBUG(D_INFO, DFID" eadata %p len %d\n", PFID(mdd_object_fid(child)),
1610 hint->dah_eadata, hint->dah_eadata_len);
1611 /* @hint will be initialized by underlying device. */
1612 nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT);
1616 * do NOT or the MAY_*'s, you'll get the weakest
1618 int accmode(const struct lu_env *env, const struct lu_attr *la, int flags)
1622 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1623 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1624 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1625 * owner can write to a file even if it is marked readonly to hide
1626 * its brokenness. (bug 5781) */
1627 if (flags & MDS_OPEN_OWNEROVERRIDE) {
1628 struct lu_ucred *uc = lu_ucred_check(env);
1630 if ((uc == NULL) || (la->la_uid == uc->uc_fsuid))
1634 if (flags & FMODE_READ)
1636 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1638 if (flags & MDS_FMODE_EXEC)
1643 static int mdd_open_sanity_check(const struct lu_env *env,
1644 struct mdd_object *obj,
1645 const struct lu_attr *attr, int flag)
1651 if (mdd_is_dead_obj(obj))
1654 if (S_ISLNK(attr->la_mode))
1657 mode = accmode(env, attr, flag);
1659 if (S_ISDIR(attr->la_mode) && (mode & MAY_WRITE))
1662 if (!(flag & MDS_OPEN_CREATED)) {
1663 rc = mdd_permission_internal(env, obj, attr, mode);
1668 if (S_ISFIFO(attr->la_mode) || S_ISSOCK(attr->la_mode) ||
1669 S_ISBLK(attr->la_mode) || S_ISCHR(attr->la_mode))
1670 flag &= ~MDS_OPEN_TRUNC;
1672 /* For writing append-only file must open it with append mode. */
1673 if (attr->la_flags & LUSTRE_APPEND_FL) {
1674 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1676 if (flag & MDS_OPEN_TRUNC)
1683 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1686 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1687 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1690 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1692 rc = mdd_la_get(env, mdd_obj, attr);
1696 rc = mdd_open_sanity_check(env, mdd_obj, attr, flags);
1700 mdd_obj->mod_count++;
1703 mdd_write_unlock(env, mdd_obj);
1707 static int mdd_declare_close(const struct lu_env *env,
1708 struct mdd_object *obj,
1710 struct thandle *handle)
1714 rc = orph_declare_index_delete(env, obj, handle);
1718 return mdo_declare_destroy(env, obj, handle);
1722 * No permission check is needed.
1724 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1725 struct md_attr *ma, int mode)
1727 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1728 struct mdd_device *mdd = mdo2mdd(obj);
1729 struct thandle *handle = NULL;
1734 if (ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_KEEP_ORPHAN) {
1735 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1736 mdd_obj->mod_count--;
1737 mdd_write_unlock(env, mdd_obj);
1739 if (mdd_obj->mod_flags & ORPHAN_OBJ && !mdd_obj->mod_count)
1740 CDEBUG(D_HA, "Object "DFID" is retained in orphan "
1741 "list\n", PFID(mdd_object_fid(mdd_obj)));
1745 /* mdd_finish_unlink() will always set orphan object as DEAD_OBJ, but
1746 * it might fail to add the object to orphan list (w/o ORPHAN_OBJ). */
1747 /* check without any lock */
1748 is_orphan = mdd_obj->mod_count == 1 &&
1749 (mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0;
1753 handle = mdd_trans_create(env, mdo2mdd(obj));
1755 RETURN(PTR_ERR(handle));
1757 rc = mdd_declare_close(env, mdd_obj, ma, handle);
1761 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1765 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1770 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1771 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr);
1773 CERROR("Failed to get lu_attr of "DFID": %d\n",
1774 PFID(mdd_object_fid(mdd_obj)), rc);
1778 /* check again with lock */
1779 is_orphan = (mdd_obj->mod_count == 1) &&
1780 ((mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0 ||
1781 ma->ma_attr.la_nlink == 0);
1783 if (is_orphan && handle == NULL) {
1784 mdd_write_unlock(env, mdd_obj);
1788 mdd_obj->mod_count--; /*release open count */
1794 /* NB: Object maybe not in orphan list originally, it is rare case for
1795 * mdd_finish_unlink() failure, in that case, the object doesn't have
1796 * ORPHAN_OBJ flag */
1797 if ((mdd_obj->mod_flags & ORPHAN_OBJ) != 0) {
1798 /* remove link to object from orphan index */
1799 LASSERT(handle != NULL);
1800 rc = __mdd_orphan_del(env, mdd_obj, handle);
1802 CERROR("%s: unable to delete "DFID" from orphan list: "
1803 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1804 PFID(mdd_object_fid(mdd_obj)), rc);
1805 /* If object was not deleted from orphan list, do not
1806 * destroy OSS objects, which will be done when next
1811 CDEBUG(D_HA, "Object "DFID" is deleted from orphan "
1812 "list, OSS objects to be destroyed.\n",
1813 PFID(mdd_object_fid(mdd_obj)));
1816 rc = mdo_destroy(env, mdd_obj, handle);
1819 CERROR("%s: unable to delete "DFID" from orphan list: "
1820 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1821 PFID(mdd_object_fid(mdd_obj)), rc);
1826 mdd_write_unlock(env, mdd_obj);
1829 (mode & (FMODE_WRITE | MDS_OPEN_APPEND | MDS_OPEN_TRUNC)) &&
1830 !(ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_RECOV_OPEN)) {
1831 if (handle == NULL) {
1832 handle = mdd_trans_create(env, mdo2mdd(obj));
1834 GOTO(stop, rc = PTR_ERR(handle));
1836 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL,
1841 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1846 mdd_changelog_data_store(env, mdd, CL_CLOSE, mode,
1851 if (handle != NULL && !IS_ERR(handle))
1852 rc = mdd_trans_stop(env, mdd, rc, handle);
1858 * Permission check is done when open,
1859 * no need check again.
1861 static int mdd_readpage_sanity_check(const struct lu_env *env,
1862 struct mdd_object *obj)
1864 struct dt_object *next = mdd_object_child(obj);
1868 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
1876 static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp,
1877 size_t nob, const struct dt_it_ops *iops,
1878 struct dt_it *it, __u32 attr, void *arg)
1880 struct lu_dirpage *dp = &lp->lp_dir;
1884 struct lu_dirent *ent;
1885 struct lu_dirent *last = NULL;
1889 if (nob < sizeof(*dp))
1892 memset(area, 0, sizeof (*dp));
1893 area += sizeof (*dp);
1894 nob -= sizeof (*dp);
1901 len = iops->key_size(env, it);
1903 /* IAM iterator can return record with zero len. */
1907 hash = iops->store(env, it);
1908 if (unlikely(first)) {
1910 dp->ldp_hash_start = cpu_to_le64(hash);
1913 /* calculate max space required for lu_dirent */
1914 recsize = lu_dirent_calc_size(len, attr);
1916 if (nob >= recsize) {
1917 result = iops->rec(env, it, (struct dt_rec *)ent, attr);
1918 if (result == -ESTALE)
1923 /* osd might not able to pack all attributes,
1924 * so recheck rec length */
1925 recsize = le16_to_cpu(ent->lde_reclen);
1927 if (le32_to_cpu(ent->lde_attrs) & LUDA_FID) {
1928 fid_le_to_cpu(&fid, &ent->lde_fid);
1929 if (fid_is_dot_lustre(&fid))
1933 result = (last != NULL) ? 0 :-EINVAL;
1937 ent = (void *)ent + recsize;
1941 result = iops->next(env, it);
1942 if (result == -ESTALE)
1944 } while (result == 0);
1947 dp->ldp_hash_end = cpu_to_le64(hash);
1949 if (last->lde_hash == dp->ldp_hash_end)
1950 dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE);
1951 last->lde_reclen = 0; /* end mark */
1954 /* end of directory */
1955 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1956 else if (result < 0)
1957 CWARN("build page failed: %d!\n", result);
1961 int mdd_readpage(const struct lu_env *env, struct md_object *obj,
1962 const struct lu_rdpg *rdpg)
1964 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1968 if (mdd_object_exists(mdd_obj) == 0) {
1969 CERROR("%s: object "DFID" not found: rc = -2\n",
1970 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
1974 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
1975 rc = mdd_readpage_sanity_check(env, mdd_obj);
1977 GOTO(out_unlock, rc);
1979 if (mdd_is_dead_obj(mdd_obj)) {
1981 struct lu_dirpage *dp;
1984 * According to POSIX, please do not return any entry to client:
1985 * even dot and dotdot should not be returned.
1987 CDEBUG(D_INODE, "readdir from dead object: "DFID"\n",
1988 PFID(mdd_object_fid(mdd_obj)));
1990 if (rdpg->rp_count <= 0)
1991 GOTO(out_unlock, rc = -EFAULT);
1992 LASSERT(rdpg->rp_pages != NULL);
1994 pg = rdpg->rp_pages[0];
1995 dp = (struct lu_dirpage *)kmap(pg);
1996 memset(dp, 0 , sizeof(struct lu_dirpage));
1997 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
1998 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1999 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
2001 GOTO(out_unlock, rc = LU_PAGE_SIZE);
2004 rc = dt_index_walk(env, mdd_object_child(mdd_obj), rdpg,
2005 mdd_dir_page_build, NULL);
2007 struct lu_dirpage *dp;
2009 dp = kmap(rdpg->rp_pages[0]);
2010 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
2013 * No pages were processed, mark this for first page
2016 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
2017 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
2018 rc = min_t(unsigned int, LU_PAGE_SIZE, rdpg->rp_count);
2020 kunmap(rdpg->rp_pages[0]);
2023 GOTO(out_unlock, rc);
2025 mdd_read_unlock(env, mdd_obj);
2029 static int mdd_object_sync(const struct lu_env *env, struct md_object *obj)
2031 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2033 if (mdd_object_exists(mdd_obj) == 0) {
2036 CERROR("%s: object "DFID" not found: rc = %d\n",
2037 mdd_obj_dev_name(mdd_obj),
2038 PFID(mdd_object_fid(mdd_obj)), rc);
2041 return dt_object_sync(env, mdd_object_child(mdd_obj),
2045 static int mdd_object_lock(const struct lu_env *env,
2046 struct md_object *obj,
2047 struct lustre_handle *lh,
2048 struct ldlm_enqueue_info *einfo,
2049 union ldlm_policy_data *policy)
2051 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2052 return dt_object_lock(env, mdd_object_child(mdd_obj), lh,
2056 static int mdd_object_unlock(const struct lu_env *env,
2057 struct md_object *obj,
2058 struct ldlm_enqueue_info *einfo,
2059 union ldlm_policy_data *policy)
2061 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2062 return dt_object_unlock(env, mdd_object_child(mdd_obj), einfo, policy);
2065 const struct md_object_operations mdd_obj_ops = {
2066 .moo_permission = mdd_permission,
2067 .moo_attr_get = mdd_attr_get,
2068 .moo_attr_set = mdd_attr_set,
2069 .moo_xattr_get = mdd_xattr_get,
2070 .moo_xattr_set = mdd_xattr_set,
2071 .moo_xattr_list = mdd_xattr_list,
2072 .moo_invalidate = mdd_invalidate,
2073 .moo_xattr_del = mdd_xattr_del,
2074 .moo_swap_layouts = mdd_swap_layouts,
2075 .moo_open = mdd_open,
2076 .moo_close = mdd_close,
2077 .moo_readpage = mdd_readpage,
2078 .moo_readlink = mdd_readlink,
2079 .moo_changelog = mdd_changelog,
2080 .moo_object_sync = mdd_object_sync,
2081 .moo_object_lock = mdd_object_lock,
2082 .moo_object_unlock = mdd_object_unlock,