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) {
71 CERROR("%s: object "DFID" not found: rc = -2\n",
72 mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj)));
76 rc = mdo_attr_get(env, obj, la);
77 if (unlikely(rc != 0)) {
79 obj->mod_flags |= DEAD_OBJ;
83 if (la->la_valid & LA_FLAGS &&
84 la->la_flags & LUSTRE_ORPHAN_FL)
85 obj->mod_flags |= ORPHAN_OBJ | DEAD_OBJ;
90 struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
92 struct mdd_thread_info *info;
94 lu_env_refill((struct lu_env *)env);
95 info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
96 LASSERT(info != NULL);
100 struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
104 buf = &mdd_env_info(env)->mti_buf[0];
110 const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
111 const void *area, ssize_t len)
115 buf = &mdd_env_info(env)->mti_buf[0];
116 buf->lb_buf = (void *)area;
121 struct lu_object *mdd_object_alloc(const struct lu_env *env,
122 const struct lu_object_header *hdr,
125 struct mdd_object *mdd_obj;
127 OBD_SLAB_ALLOC_PTR_GFP(mdd_obj, mdd_object_kmem, GFP_NOFS);
128 if (mdd_obj != NULL) {
131 o = mdd2lu_obj(mdd_obj);
132 lu_object_init(o, NULL, d);
133 mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
134 mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
135 mdd_obj->mod_count = 0;
136 o->lo_ops = &mdd_lu_obj_ops;
143 static int mdd_object_init(const struct lu_env *env, struct lu_object *o,
144 const struct lu_object_conf *unused)
146 struct mdd_device *d = lu2mdd_dev(o->lo_dev);
147 struct mdd_object *mdd_obj = lu2mdd_obj(o);
148 struct lu_object *below;
149 struct lu_device *under;
152 mdd_obj->mod_cltime = 0;
153 under = &d->mdd_child->dd_lu_dev;
154 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
156 RETURN(PTR_ERR(below));
158 lu_object_add(o, below);
163 static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
167 if (lu_object_exists(o)) {
168 struct mdd_object *mdd_obj = lu2mdd_obj(o);
169 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
171 rc = mdd_la_get(env, mdd_obj, attr);
177 static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
179 struct mdd_object *mdd = lu2mdd_obj(o);
182 OBD_SLAB_FREE_PTR(mdd, mdd_object_kmem);
185 static int mdd_object_print(const struct lu_env *env, void *cookie,
186 lu_printer_t p, const struct lu_object *o)
188 struct mdd_object *mdd = lu2mdd_obj((struct lu_object *)o);
189 return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p(open_count=%d, "
190 "valid=%x, cltime="LPU64", flags=%lx)",
191 mdd, mdd->mod_count, mdd->mod_valid,
192 mdd->mod_cltime, mdd->mod_flags);
195 static const struct lu_object_operations mdd_lu_obj_ops = {
196 .loo_object_init = mdd_object_init,
197 .loo_object_start = mdd_object_start,
198 .loo_object_free = mdd_object_free,
199 .loo_object_print = mdd_object_print,
202 struct mdd_object *mdd_object_find(const struct lu_env *env,
203 struct mdd_device *d,
204 const struct lu_fid *f)
206 return md2mdd_obj(md_object_find_slice(env, &d->mdd_md_dev, f));
210 * No permission check is needed.
212 int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
215 struct mdd_object *mdd_obj = md2mdd_obj(obj);
220 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr);
221 if ((ma->ma_need & MA_INODE) != 0 && mdd_is_dead_obj(mdd_obj))
222 ma->ma_attr.la_nlink = 0;
228 * No permission check is needed.
230 static int mdd_xattr_get(const struct lu_env *env,
231 struct md_object *obj, struct lu_buf *buf,
234 struct mdd_object *mdd_obj = md2mdd_obj(obj);
239 if (mdd_object_exists(mdd_obj) == 0) {
240 CERROR("%s: object "DFID" not found: rc = -2\n",
241 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
245 /* If the object has been destroyed, then do not get LMVEA, because
246 * it needs to load stripes from the iteration of the master object,
247 * and it will cause problem if master object has been destroyed, see
249 if (unlikely((mdd_obj->mod_flags & DEAD_OBJ) &&
250 !(mdd_obj->mod_flags & ORPHAN_OBJ) &&
251 strcmp(name, XATTR_NAME_LMV) == 0))
254 /* If the object has been delete from the namespace, then
255 * get linkEA should return -ENOENT as well */
256 if (unlikely((mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ)) &&
257 strcmp(name, XATTR_NAME_LINK) == 0))
260 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
261 rc = mdo_xattr_get(env, mdd_obj, buf, name);
262 mdd_read_unlock(env, mdd_obj);
268 * Permission check is done when open,
269 * no need check again.
271 int mdd_readlink(const struct lu_env *env, struct md_object *obj,
274 struct mdd_object *mdd_obj = md2mdd_obj(obj);
275 struct dt_object *next;
280 if (mdd_object_exists(mdd_obj) == 0) {
281 CERROR("%s: object "DFID" not found: rc = -2\n",
282 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
286 next = mdd_object_child(mdd_obj);
287 LASSERT(next != NULL);
288 LASSERT(next->do_body_ops != NULL);
289 LASSERT(next->do_body_ops->dbo_read != NULL);
290 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
291 rc = dt_read(env, next, buf, &pos);
292 mdd_read_unlock(env, mdd_obj);
297 * No permission check is needed.
299 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
302 struct mdd_object *mdd_obj = md2mdd_obj(obj);
307 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
308 rc = mdo_xattr_list(env, mdd_obj, buf);
309 mdd_read_unlock(env, mdd_obj);
315 * Filter out XATTR_NAME_LINK if this is an orphan object. See
318 if (unlikely(mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ))) {
319 char *end = (char *)buf->lb_buf + rc;
320 char *p = buf->lb_buf;
323 char *next = p + strlen(p) + 1;
325 if (strcmp(p, XATTR_NAME_LINK) == 0) {
327 memmove(p, next, end - next);
329 CDEBUG(D_INFO, "Filtered out "XATTR_NAME_LINK
330 " of orphan "DFID"\n",
331 PFID(mdd_object_fid(mdd_obj)));
342 int mdd_invalidate(const struct lu_env *env, struct md_object *obj)
344 return mdo_invalidate(env, md2mdd_obj(obj));
347 int mdd_declare_object_create_internal(const struct lu_env *env,
348 struct mdd_object *p,
349 struct mdd_object *c,
350 struct lu_attr *attr,
351 struct thandle *handle,
352 const struct md_op_spec *spec,
353 struct dt_allocation_hint *hint)
355 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
356 const struct dt_index_features *feat = spec->sp_feat;
360 if (feat != &dt_directory_features && feat != NULL) {
361 dof->dof_type = DFT_INDEX;
362 dof->u.dof_idx.di_feat = feat;
365 dof->dof_type = dt_mode_to_dft(attr->la_mode);
366 if (dof->dof_type == DFT_REGULAR) {
367 dof->u.dof_reg.striped =
368 md_should_create(spec->sp_cr_flags);
369 if (spec->sp_cr_flags & MDS_OPEN_HAS_EA)
370 dof->u.dof_reg.striped = 0;
371 /* is this replay? */
373 dof->u.dof_reg.striped = 0;
377 rc = mdo_declare_create_obj(env, c, attr, hint, dof, handle);
382 int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p,
383 struct mdd_object *c, struct lu_attr *attr,
384 struct thandle *handle,
385 const struct md_op_spec *spec,
386 struct dt_allocation_hint *hint)
388 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
392 LASSERT(!mdd_object_exists(c));
394 rc = mdo_create_obj(env, c, attr, hint, dof, handle);
399 int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj,
400 const struct lu_attr *attr, struct thandle *handle,
406 rc = mdo_attr_set(env, obj, attr, handle);
407 #ifdef CONFIG_FS_POSIX_ACL
408 if (!rc && (attr->la_valid & LA_MODE) && needacl)
409 rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
414 int mdd_update_time(const struct lu_env *env, struct mdd_object *obj,
415 const struct lu_attr *oattr, struct lu_attr *attr,
416 struct thandle *handle)
421 LASSERT(attr->la_valid & LA_CTIME);
422 LASSERT(oattr != NULL);
424 /* Make sure the ctime is increased only, however, it's not strictly
425 * reliable at here because there is not guarantee to hold lock on
426 * object, so we just bypass some unnecessary cmtime setting first
427 * and OSD has to check it again. */
428 if (attr->la_ctime < oattr->la_ctime)
429 attr->la_valid &= ~(LA_MTIME | LA_CTIME);
430 else if (attr->la_valid == LA_CTIME &&
431 attr->la_ctime == oattr->la_ctime)
432 attr->la_valid &= ~LA_CTIME;
434 if (attr->la_valid != 0)
435 rc = mdd_attr_set_internal(env, obj, attr, handle, 0);
440 * This gives the same functionality as the code between
441 * sys_chmod and inode_setattr
442 * chown_common and inode_setattr
443 * utimes and inode_setattr
444 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
446 static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
447 const struct lu_attr *oattr, struct lu_attr *la,
448 const unsigned long flags)
457 /* Do not permit change file type */
458 if (la->la_valid & LA_TYPE)
461 /* They should not be processed by setattr */
462 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
465 LASSERT(oattr != NULL);
467 uc = lu_ucred_check(env);
471 if (la->la_valid == LA_CTIME) {
472 if (!(flags & MDS_PERM_BYPASS))
473 /* This is only for set ctime when rename's source is
475 rc = mdd_may_delete(env, NULL, NULL, obj, oattr, NULL,
477 if (rc == 0 && la->la_ctime <= oattr->la_ctime)
478 la->la_valid &= ~LA_CTIME;
482 if (la->la_valid == LA_ATIME) {
483 /* This is an atime-only attribute update for close RPCs. */
484 if (la->la_atime < (oattr->la_atime +
485 mdd_obj2mdd_dev(obj)->mdd_atime_diff))
486 la->la_valid &= ~LA_ATIME;
490 /* Check if flags change. */
491 if (la->la_valid & LA_FLAGS) {
492 unsigned int oldflags = oattr->la_flags &
493 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
494 unsigned int newflags = la->la_flags &
495 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
497 if ((uc->uc_fsuid != oattr->la_uid) &&
498 !md_capable(uc, CFS_CAP_FOWNER))
501 /* The IMMUTABLE and APPEND_ONLY flags can
502 * only be changed by the relevant capability. */
503 if ((oldflags ^ newflags) &&
504 !md_capable(uc, CFS_CAP_LINUX_IMMUTABLE))
507 if (!S_ISDIR(oattr->la_mode))
508 la->la_flags &= ~(LUSTRE_DIRSYNC_FL | LUSTRE_TOPDIR_FL);
511 if (oattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL) &&
512 (la->la_valid & ~LA_FLAGS) &&
513 !(flags & MDS_PERM_BYPASS))
516 /* Check for setting the obj time. */
517 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
518 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
519 if ((uc->uc_fsuid != oattr->la_uid) &&
520 !md_capable(uc, CFS_CAP_FOWNER)) {
521 rc = mdd_permission_internal(env, obj, oattr,
528 if (la->la_valid & LA_KILL_SUID) {
529 la->la_valid &= ~LA_KILL_SUID;
530 if ((oattr->la_mode & S_ISUID) &&
531 !(la->la_valid & LA_MODE)) {
532 la->la_mode = oattr->la_mode;
533 la->la_valid |= LA_MODE;
535 la->la_mode &= ~S_ISUID;
538 if (la->la_valid & LA_KILL_SGID) {
539 la->la_valid &= ~LA_KILL_SGID;
540 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
541 (S_ISGID | S_IXGRP)) &&
542 !(la->la_valid & LA_MODE)) {
543 la->la_mode = oattr->la_mode;
544 la->la_valid |= LA_MODE;
546 la->la_mode &= ~S_ISGID;
549 /* Make sure a caller can chmod. */
550 if (la->la_valid & LA_MODE) {
551 if (!(flags & MDS_PERM_BYPASS) &&
552 (uc->uc_fsuid != oattr->la_uid) &&
553 !md_capable(uc, CFS_CAP_FOWNER))
556 if (la->la_mode == (umode_t) -1)
557 la->la_mode = oattr->la_mode;
559 la->la_mode = (la->la_mode & S_IALLUGO) |
560 (oattr->la_mode & ~S_IALLUGO);
562 /* Also check the setgid bit! */
563 if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ?
564 la->la_gid : oattr->la_gid) &&
565 !md_capable(uc, CFS_CAP_FSETID))
566 la->la_mode &= ~S_ISGID;
568 la->la_mode = oattr->la_mode;
571 /* Make sure a caller can chown. */
572 if (la->la_valid & LA_UID) {
573 if (la->la_uid == (uid_t) -1)
574 la->la_uid = oattr->la_uid;
575 if (((uc->uc_fsuid != oattr->la_uid) ||
576 (la->la_uid != oattr->la_uid)) &&
577 !md_capable(uc, CFS_CAP_CHOWN))
580 /* If the user or group of a non-directory has been
581 * changed by a non-root user, remove the setuid bit.
582 * 19981026 David C Niemi <niemi@tux.org>
584 * Changed this to apply to all users, including root,
585 * to avoid some races. This is the behavior we had in
586 * 2.0. The check for non-root was definitely wrong
587 * for 2.2 anyway, as it should have been using
588 * CAP_FSETID rather than fsuid -- 19990830 SD. */
589 if (((oattr->la_mode & S_ISUID) == S_ISUID) &&
590 !S_ISDIR(oattr->la_mode)) {
591 la->la_mode &= ~S_ISUID;
592 la->la_valid |= LA_MODE;
596 /* Make sure caller can chgrp. */
597 if (la->la_valid & LA_GID) {
598 if (la->la_gid == (gid_t) -1)
599 la->la_gid = oattr->la_gid;
600 if (((uc->uc_fsuid != oattr->la_uid) ||
601 ((la->la_gid != oattr->la_gid) &&
602 !lustre_in_group_p(uc, la->la_gid))) &&
603 !md_capable(uc, CFS_CAP_CHOWN))
606 /* Likewise, if the user or group of a non-directory
607 * has been changed by a non-root user, remove the
608 * setgid bit UNLESS there is no group execute bit
609 * (this would be a file marked for mandatory
610 * locking). 19981026 David C Niemi <niemi@tux.org>
612 * Removed the fsuid check (see the comment above) --
614 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
615 (S_ISGID | S_IXGRP)) && !S_ISDIR(oattr->la_mode)) {
616 la->la_mode &= ~S_ISGID;
617 la->la_valid |= LA_MODE;
621 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
622 if (!((flags & MDS_OWNEROVERRIDE) &&
623 (uc->uc_fsuid == oattr->la_uid)) &&
624 !(flags & MDS_PERM_BYPASS)) {
625 rc = mdd_permission_internal(env, obj, oattr,
632 if (la->la_valid & LA_CTIME) {
633 /* The pure setattr, it has the priority over what is
634 * already set, do not drop it if ctime is equal. */
635 if (la->la_ctime < oattr->la_ctime)
636 la->la_valid &= ~(LA_ATIME | LA_MTIME | LA_CTIME);
642 /** Store a data change changelog record
643 * If this fails, we must fail the whole transaction; we don't
644 * want the change to commit without the log entry.
645 * \param mdd_obj - mdd_object of change
646 * \param handle - transaction handle
648 int mdd_changelog_data_store(const struct lu_env *env, struct mdd_device *mdd,
649 enum changelog_rec_type type, int flags,
650 struct mdd_object *mdd_obj, struct thandle *handle)
652 const struct lu_ucred *uc = lu_ucred(env);
653 const struct lu_fid *tfid;
654 struct llog_changelog_rec *rec;
660 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
662 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
665 LASSERT(mdd_obj != NULL);
666 LASSERT(handle != NULL);
668 tfid = mdo2fid(mdd_obj);
670 if ((type >= CL_MTIME) && (type <= CL_ATIME) &&
671 cfs_time_before_64(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) {
672 /* Don't need multiple updates in this log */
673 /* Don't check under lock - no big deal if we get an extra
678 flags = (flags & CLF_FLAGMASK) | CLF_VERSION;
679 if (uc != NULL && uc->uc_jobid[0] != '\0')
682 reclen = llog_data_len(changelog_rec_offset(flags & CLF_SUPPORTED));
683 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
684 if (buf->lb_buf == NULL)
688 rec->cr.cr_flags = flags;
689 rec->cr.cr_type = (__u32)type;
690 rec->cr.cr_tfid = *tfid;
691 rec->cr.cr_namelen = 0;
692 mdd_obj->mod_cltime = cfs_time_current_64();
694 if (flags & CLF_JOBID)
695 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
697 rc = mdd_changelog_store(env, mdd, rec, handle);
702 static int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type,
703 int flags, struct md_object *obj)
705 struct thandle *handle;
706 struct mdd_object *mdd_obj = md2mdd_obj(obj);
707 struct mdd_device *mdd = mdo2mdd(obj);
711 handle = mdd_trans_create(env, mdd);
713 RETURN(PTR_ERR(handle));
715 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
719 rc = mdd_trans_start(env, mdd, handle);
723 rc = mdd_changelog_data_store(env, mdd, type, flags, mdd_obj,
727 mdd_trans_stop(env, mdd, rc, handle);
733 * Save LMA extended attributes with data from \a ma.
735 * HSM and Size-On-MDS data will be extracted from \ma if they are valid, if
736 * not, LMA EA will be first read from disk, modified and write back.
739 /* Precedence for choosing record type when multiple
740 * attributes change: setattr > mtime > ctime > atime
741 * (ctime changes when mtime does, plus chmod/chown.
742 * atime and ctime are independent.) */
743 static int mdd_attr_set_changelog(const struct lu_env *env,
744 struct md_object *obj, struct thandle *handle,
747 struct mdd_device *mdd = mdo2mdd(obj);
750 bits = (valid & LA_SIZE) ? 1 << CL_TRUNC : 0;
751 bits |= (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0;
752 bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0;
753 bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0;
754 bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0;
755 bits = bits & mdd->mdd_cl.mc_mask;
756 /* This is an implementation limit rather than a protocol limit */
757 CLASSERT(CL_LAST <= sizeof(int) * 8);
761 /* The record type is the lowest non-masked set bit */
764 /* FYI we only store the first CLF_FLAGMASK bits of la_valid */
765 return mdd_changelog_data_store(env, mdd, type, (int)valid,
766 md2mdd_obj(obj), handle);
769 static int mdd_declare_attr_set(const struct lu_env *env,
770 struct mdd_device *mdd,
771 struct mdd_object *obj,
772 const struct lu_attr *attr,
773 struct thandle *handle)
777 rc = mdo_declare_attr_set(env, obj, attr, handle);
781 #ifdef CONFIG_FS_POSIX_ACL
782 if (attr->la_valid & LA_MODE) {
783 mdd_read_lock(env, obj, MOR_TGT_CHILD);
784 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL,
785 XATTR_NAME_ACL_ACCESS);
786 mdd_read_unlock(env, obj);
787 if (rc == -EOPNOTSUPP || rc == -ENODATA)
793 struct lu_buf *buf = mdd_buf_get(env, NULL, rc);
794 rc = mdo_declare_xattr_set(env, obj, buf,
795 XATTR_NAME_ACL_ACCESS, 0,
803 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
810 * permission changes may require sync operation, to mitigate performance
811 * impact, only do this for dir and when permission is reduced.
813 * For regular files, version is updated with permission change (see VBR), async
814 * permission won't cause any issue, while missing permission change on
815 * directory may affect accessibility of other objects after recovery.
817 static inline bool permission_needs_sync(const struct lu_attr *old,
818 const struct lu_attr *new)
820 if (!S_ISDIR(old->la_mode))
823 if (new->la_valid & (LA_UID | LA_GID))
826 if (new->la_valid & LA_MODE &&
827 new->la_mode & (S_ISUID | S_ISGID | S_ISVTX))
830 if ((new->la_valid & LA_MODE) &&
831 ((new->la_mode & old->la_mode) & S_IRWXUGO) !=
832 (old->la_mode & S_IRWXUGO))
838 /* set attr and LOV EA at once, return updated attr */
839 int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
840 const struct md_attr *ma)
842 struct mdd_object *mdd_obj = md2mdd_obj(obj);
843 struct mdd_device *mdd = mdo2mdd(obj);
844 struct thandle *handle;
845 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
846 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
847 const struct lu_attr *la = &ma->ma_attr;
851 /* we do not use ->attr_set() for LOV/HSM EA any more */
852 LASSERT((ma->ma_valid & MA_LOV) == 0);
853 LASSERT((ma->ma_valid & MA_HSM) == 0);
855 rc = mdd_la_get(env, mdd_obj, attr);
859 *la_copy = ma->ma_attr;
860 rc = mdd_fix_attr(env, mdd_obj, attr, la_copy, ma->ma_attr_flags);
864 /* no need to setattr anymore */
865 if (la_copy->la_valid == 0) {
866 CDEBUG(D_INODE, "%s: no valid attribute on "DFID", previous"
867 "valid is "LPX64"\n", mdd2obd_dev(mdd)->obd_name,
868 PFID(mdo2fid(mdd_obj)), la->la_valid);
873 handle = mdd_trans_create(env, mdd);
875 RETURN(PTR_ERR(handle));
877 rc = mdd_declare_attr_set(env, mdd, mdd_obj, la_copy, handle);
881 rc = mdd_trans_start(env, mdd, handle);
885 if (mdd->mdd_sync_permission && permission_needs_sync(attr, la))
888 if (la->la_valid & (LA_MTIME | LA_CTIME))
889 CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n",
890 la->la_mtime, la->la_ctime);
892 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
893 if (la_copy->la_valid & LA_FLAGS)
894 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
895 else if (la_copy->la_valid) /* setattr */
896 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
897 mdd_write_unlock(env, mdd_obj);
900 rc = mdd_attr_set_changelog(env, obj, handle,
906 mdd_trans_stop(env, mdd, rc, handle);
910 static int mdd_xattr_sanity_check(const struct lu_env *env,
911 struct mdd_object *obj,
912 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 ((uc->uc_fsuid != attr->la_uid) && !md_capable(uc, CFS_CAP_FOWNER))
926 static int mdd_declare_xattr_set(const struct lu_env *env,
927 struct mdd_device *mdd,
928 struct mdd_object *obj,
929 const struct lu_buf *buf,
931 int fl, struct thandle *handle)
935 rc = mdo_declare_xattr_set(env, obj, buf, name, fl, handle);
939 /* Only record user and layout xattr changes */
940 if (strncmp(XATTR_USER_PREFIX, name,
941 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
942 strcmp(XATTR_NAME_LOV, name) == 0) {
943 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
948 /* If HSM data is modified, this could add a changelog */
949 if (strcmp(XATTR_NAME_HSM, name) == 0) {
950 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
955 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
960 * Compare current and future data of HSM EA and add a changelog if needed.
962 * Caller should have write-locked \param obj.
964 * \param buf - Future HSM EA content.
965 * \retval 0 if no changelog is needed or changelog was added properly.
966 * \retval -ve errno if there was a problem
968 static int mdd_hsm_update_locked(const struct lu_env *env,
969 struct md_object *obj,
970 const struct lu_buf *buf,
971 struct thandle *handle)
973 struct mdd_thread_info *info = mdd_env_info(env);
974 struct mdd_device *mdd = mdo2mdd(obj);
975 struct mdd_object *mdd_obj = md2mdd_obj(obj);
976 struct lu_buf *current_buf;
977 struct md_hsm *current_mh;
978 struct md_hsm *new_mh;
982 OBD_ALLOC_PTR(current_mh);
983 if (current_mh == NULL)
986 /* Read HSM attrs from disk */
987 CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
988 current_buf = mdd_buf_get(env, info->mti_xattr_buf,
989 sizeof(info->mti_xattr_buf));
990 rc = mdo_xattr_get(env, mdd_obj, current_buf, XATTR_NAME_HSM);
991 rc = lustre_buf2hsm(current_buf->lb_buf, rc, current_mh);
992 if (rc < 0 && rc != -ENODATA)
994 else if (rc == -ENODATA)
995 current_mh->mh_flags = 0;
997 /* Map future HSM xattr */
998 OBD_ALLOC_PTR(new_mh);
1000 GOTO(free, rc = -ENOMEM);
1001 lustre_buf2hsm(buf->lb_buf, buf->lb_len, new_mh);
1003 /* If HSM flags are different, add a changelog */
1005 if (current_mh->mh_flags != new_mh->mh_flags) {
1007 hsm_set_cl_event(&flags, HE_STATE);
1008 if (new_mh->mh_flags & HS_DIRTY)
1009 hsm_set_cl_flags(&flags, CLF_HSM_DIRTY);
1011 rc = mdd_changelog_data_store(env, mdd, CL_HSM, flags, mdd_obj,
1015 OBD_FREE_PTR(new_mh);
1017 OBD_FREE_PTR(current_mh);
1021 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1025 * The caller should guarantee to update the object ctime
1026 * after xattr_set if needed.
1028 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
1029 const struct lu_buf *buf, const char *name,
1032 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1033 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1034 struct mdd_device *mdd = mdo2mdd(obj);
1035 struct thandle *handle;
1039 rc = mdd_la_get(env, mdd_obj, attr);
1043 rc = mdd_xattr_sanity_check(env, mdd_obj, attr);
1047 if (strcmp(name, XATTR_NAME_ACL_ACCESS) == 0 ||
1048 strcmp(name, XATTR_NAME_ACL_DEFAULT) == 0) {
1049 struct posix_acl *acl;
1051 /* user may set empty ACL, which should be treated as removing
1053 acl = posix_acl_from_xattr(&init_user_ns, buf->lb_buf,
1056 RETURN(PTR_ERR(acl));
1058 rc = mdd_xattr_del(env, obj, name);
1061 posix_acl_release(acl);
1064 if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) {
1065 rc = mdd_acl_set(env, mdd_obj, attr, buf, fl);
1069 handle = mdd_trans_create(env, mdd);
1071 RETURN(PTR_ERR(handle));
1073 rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, fl, handle);
1077 rc = mdd_trans_start(env, mdd, handle);
1081 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1083 if (strcmp(XATTR_NAME_HSM, name) == 0) {
1084 rc = mdd_hsm_update_locked(env, obj, buf, handle);
1086 mdd_write_unlock(env, mdd_obj);
1091 rc = mdo_xattr_set(env, mdd_obj, buf, name, fl, handle);
1092 mdd_write_unlock(env, mdd_obj);
1096 if (strcmp(XATTR_NAME_LOV, name) == 0)
1097 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, mdd_obj,
1099 else if (strncmp(XATTR_USER_PREFIX, name,
1100 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1101 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1102 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1103 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1104 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1105 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1109 mdd_trans_stop(env, mdd, rc, handle);
1114 static int mdd_declare_xattr_del(const struct lu_env *env,
1115 struct mdd_device *mdd,
1116 struct mdd_object *obj,
1118 struct thandle *handle)
1122 rc = mdo_declare_xattr_del(env, obj, name, handle);
1126 /* Only record system & user xattr changes */
1127 if (strncmp(XATTR_USER_PREFIX, name,
1128 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1129 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1130 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1131 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1132 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1133 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1139 * The caller should guarantee to update the object ctime
1140 * after xattr_set if needed.
1142 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1145 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1146 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1147 struct mdd_device *mdd = mdo2mdd(obj);
1148 struct thandle *handle;
1152 rc = mdd_la_get(env, mdd_obj, attr);
1156 rc = mdd_xattr_sanity_check(env, mdd_obj, attr);
1160 handle = mdd_trans_create(env, mdd);
1162 RETURN(PTR_ERR(handle));
1164 rc = mdd_declare_xattr_del(env, mdd, mdd_obj, name, handle);
1168 rc = mdd_trans_start(env, mdd, handle);
1172 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1173 rc = mdo_xattr_del(env, mdd_obj, name, handle);
1174 mdd_write_unlock(env, mdd_obj);
1178 /* Only record system & user xattr changes */
1179 if (strncmp(XATTR_USER_PREFIX, name,
1180 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1181 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1182 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1183 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1184 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1185 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1189 mdd_trans_stop(env, mdd, rc, handle);
1195 * read lov EA of an object
1196 * return the lov EA in an allocated lu_buf
1198 int mdd_get_lov_ea(const struct lu_env *env, struct mdd_object *obj,
1199 struct lu_buf *lmm_buf)
1201 struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf;
1206 rc = mdo_xattr_get(env, obj, buf, XATTR_NAME_LOV);
1208 if (rc == -ERANGE) {
1209 /* mti_big_buf is allocated but is too small
1210 * we need to increase it */
1211 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1213 if (buf->lb_buf == NULL)
1214 GOTO(out, rc = -ENOMEM);
1225 if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) {
1226 /* mti_big_buf was not allocated, so we have to
1227 * allocate it based on the ea size */
1228 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1230 if (buf->lb_buf == NULL)
1231 GOTO(out, rc = -ENOMEM);
1235 lu_buf_alloc(lmm_buf, bufsize);
1236 if (lmm_buf->lb_buf == NULL)
1237 GOTO(out, rc = -ENOMEM);
1239 memcpy(lmm_buf->lb_buf, buf->lb_buf, bufsize);
1245 lu_buf_free(lmm_buf);
1249 static int mdd_xattr_hsm_replace(const struct lu_env *env,
1250 struct mdd_object *o, struct lu_buf *buf,
1251 struct thandle *handle)
1253 struct hsm_attrs *attrs;
1259 rc = mdo_xattr_set(env, o, buf, XATTR_NAME_HSM, LU_XATTR_REPLACE,
1264 attrs = buf->lb_buf;
1265 hsm_flags = le32_to_cpu(attrs->hsm_flags);
1266 if (!(hsm_flags & HS_RELEASED) || mdd_is_dead_obj(o))
1269 /* Add a changelog record for release. */
1270 hsm_set_cl_event(&flags, HE_RELEASE);
1271 rc = mdd_changelog_data_store(env, mdo2mdd(&o->mod_obj), CL_HSM,
1277 * check if layout swapping between 2 objects is allowed
1279 * - only normal FIDs or non-system IGIFs
1280 * - same type of objects
1281 * - same owner/group (so quotas are still valid)
1283 static int mdd_layout_swap_allowed(const struct lu_env *env,
1284 struct mdd_object *o1,
1285 const struct lu_attr *attr1,
1286 struct mdd_object *o2,
1287 const struct lu_attr *attr2)
1289 const struct lu_fid *fid1, *fid2;
1295 if (!fid_is_norm(fid1) &&
1296 (!fid_is_igif(fid1) || IS_ERR(mdd_links_get(env, o1))))
1299 if (!fid_is_norm(fid2) &&
1300 (!fid_is_igif(fid2) || IS_ERR(mdd_links_get(env, o2))))
1303 if (mdd_object_type(o1) != mdd_object_type(o2)) {
1304 if (S_ISDIR(mdd_object_type(o1)))
1306 if (S_ISREG(mdd_object_type(o1)))
1311 if ((attr1->la_uid != attr2->la_uid) ||
1312 (attr1->la_gid != attr2->la_gid))
1319 * swap layouts between 2 lustre objects
1321 static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1,
1322 struct md_object *obj2, __u64 flags)
1324 struct mdd_thread_info *info = mdd_env_info(env);
1325 struct mdd_object *fst_o = md2mdd_obj(obj1);
1326 struct mdd_object *snd_o = md2mdd_obj(obj2);
1327 struct lu_attr *fst_la = MDD_ENV_VAR(env, cattr);
1328 struct lu_attr *snd_la = MDD_ENV_VAR(env, tattr);
1329 struct mdd_device *mdd = mdo2mdd(obj1);
1330 struct lov_mds_md *fst_lmm, *snd_lmm;
1331 struct lu_buf *fst_buf = &info->mti_buf[0];
1332 struct lu_buf *snd_buf = &info->mti_buf[1];
1333 struct lu_buf *fst_hsm_buf = &info->mti_buf[2];
1334 struct lu_buf *snd_hsm_buf = &info->mti_buf[3];
1335 struct ost_id *saved_oi = NULL;
1336 struct thandle *handle;
1337 __u16 fst_gen, snd_gen;
1343 CLASSERT(ARRAY_SIZE(info->mti_buf) >= 4);
1344 memset(info->mti_buf, 0, sizeof(info->mti_buf));
1346 /* we have to sort the 2 obj, so locking will always
1347 * be in the same order, even in case of 2 concurrent swaps */
1348 rc = lu_fid_cmp(mdo2fid(fst_o), mdo2fid(snd_o));
1349 if (rc == 0) /* same fid ? */
1355 rc = mdd_la_get(env, fst_o, fst_la);
1359 rc = mdd_la_get(env, snd_o, snd_la);
1363 /* check if layout swapping is allowed */
1364 rc = mdd_layout_swap_allowed(env, fst_o, fst_la, snd_o, snd_la);
1368 handle = mdd_trans_create(env, mdd);
1370 RETURN(PTR_ERR(handle));
1372 /* objects are already sorted */
1373 mdd_write_lock(env, fst_o, MOR_TGT_CHILD);
1374 mdd_write_lock(env, snd_o, MOR_TGT_CHILD);
1376 rc = mdd_get_lov_ea(env, fst_o, fst_buf);
1377 if (rc < 0 && rc != -ENODATA)
1380 rc = mdd_get_lov_ea(env, snd_o, snd_buf);
1381 if (rc < 0 && rc != -ENODATA)
1384 /* swapping 2 non existant layouts is a success */
1385 if (fst_buf->lb_buf == NULL && snd_buf->lb_buf == NULL)
1388 /* to help inode migration between MDT, it is better to
1389 * start by the no layout file (if one), so we order the swap */
1390 if (snd_buf->lb_buf == NULL) {
1392 swap(fst_buf, snd_buf);
1395 /* lmm and generation layout initialization */
1396 if (fst_buf->lb_buf != NULL) {
1397 fst_lmm = fst_buf->lb_buf;
1398 fst_gen = le16_to_cpu(fst_lmm->lmm_layout_gen);
1399 fst_fl = LU_XATTR_REPLACE;
1403 fst_fl = LU_XATTR_CREATE;
1406 snd_lmm = snd_buf->lb_buf;
1407 snd_gen = le16_to_cpu(snd_lmm->lmm_layout_gen);
1409 /* increase the generation layout numbers */
1413 /* set the file specific informations in lmm */
1414 if (fst_lmm != NULL) {
1415 saved_oi = &info->mti_oa.o_oi;
1417 *saved_oi = fst_lmm->lmm_oi;
1418 fst_lmm->lmm_layout_gen = cpu_to_le16(snd_gen);
1419 fst_lmm->lmm_oi = snd_lmm->lmm_oi;
1420 snd_lmm->lmm_oi = *saved_oi;
1422 if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1))
1423 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEF);
1424 else if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))
1425 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V3_DEF);
1427 GOTO(stop, rc = -EPROTO);
1429 snd_lmm->lmm_layout_gen = cpu_to_le16(fst_gen);
1431 /* Prepare HSM attribute if it's required */
1432 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1433 const int buflen = sizeof(struct hsm_attrs);
1435 lu_buf_alloc(fst_hsm_buf, buflen);
1436 lu_buf_alloc(snd_hsm_buf, buflen);
1437 if (fst_hsm_buf->lb_buf == NULL || snd_hsm_buf->lb_buf == NULL)
1438 GOTO(stop, rc = -ENOMEM);
1440 /* Read HSM attribute */
1441 rc = mdo_xattr_get(env, fst_o, fst_hsm_buf, XATTR_NAME_HSM);
1445 rc = mdo_xattr_get(env, snd_o, snd_hsm_buf, XATTR_NAME_HSM);
1449 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_hsm_buf,
1450 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1455 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_hsm_buf,
1456 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1462 /* prepare transaction */
1463 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_buf, XATTR_NAME_LOV,
1468 if (fst_buf->lb_buf != NULL)
1469 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_buf,
1470 XATTR_NAME_LOV, LU_XATTR_REPLACE,
1473 rc = mdd_declare_xattr_del(env, mdd, snd_o, XATTR_NAME_LOV,
1478 rc = mdd_trans_start(env, mdd, handle);
1482 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1483 rc = mdd_xattr_hsm_replace(env, fst_o, snd_hsm_buf, handle);
1487 rc = mdd_xattr_hsm_replace(env, snd_o, fst_hsm_buf, handle);
1489 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf,
1492 CERROR("%s: restore "DFID" HSM error: %d/%d\n",
1493 mdd_obj_dev_name(fst_o),
1494 PFID(mdo2fid(fst_o)), rc, rc2);
1499 rc = mdo_xattr_set(env, fst_o, snd_buf, XATTR_NAME_LOV, fst_fl, handle);
1503 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_MDS_HSM_SWAP_LAYOUTS))) {
1506 if (fst_buf->lb_buf != NULL)
1507 rc = mdo_xattr_set(env, snd_o, fst_buf, XATTR_NAME_LOV,
1508 LU_XATTR_REPLACE, handle);
1510 rc = mdo_xattr_del(env, snd_o, XATTR_NAME_LOV, handle);
1516 /* failure on second file, but first was done, so we have
1517 * to roll back first. */
1518 if (fst_buf->lb_buf != NULL) {
1519 fst_lmm->lmm_oi = *saved_oi;
1520 fst_lmm->lmm_layout_gen = cpu_to_le16(fst_gen - 1);
1521 rc2 = mdo_xattr_set(env, fst_o, fst_buf, XATTR_NAME_LOV,
1522 LU_XATTR_REPLACE, handle);
1524 rc2 = mdo_xattr_del(env, fst_o, XATTR_NAME_LOV, handle);
1530 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, handle);
1535 rc2 = mdd_xattr_hsm_replace(env, snd_o, snd_hsm_buf, handle);
1540 CERROR("%s: unable to roll back layout swap. FIDs: "
1541 DFID" and "DFID "error: %d/%d, steps: %d\n",
1542 mdd_obj_dev_name(fst_o),
1543 PFID(mdo2fid(snd_o)), PFID(mdo2fid(fst_o)),
1545 /* a solution to avoid journal commit is to panic,
1546 * but it has strong consequences so we use LBUG to
1547 * allow sysdamin to choose to panic or not
1554 /* Issue one changelog record per file */
1555 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle);
1559 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle);
1565 mdd_trans_stop(env, mdd, rc, handle);
1566 mdd_write_unlock(env, snd_o);
1567 mdd_write_unlock(env, fst_o);
1569 lu_buf_free(fst_buf);
1570 lu_buf_free(snd_buf);
1571 lu_buf_free(fst_hsm_buf);
1572 lu_buf_free(snd_hsm_buf);
1576 void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent,
1577 struct mdd_object *child, const struct lu_attr *attr,
1578 const struct md_op_spec *spec,
1579 struct dt_allocation_hint *hint)
1581 struct dt_object *np = parent ? mdd_object_child(parent) : NULL;
1582 struct dt_object *nc = mdd_object_child(child);
1584 memset(hint, 0, sizeof(*hint));
1586 /* For striped directory, give striping EA to lod_ah_init, which will
1587 * decide the stripe_offset and stripe count by it. */
1588 if (S_ISDIR(attr->la_mode) &&
1589 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1590 hint->dah_eadata = spec->u.sp_ea.eadata;
1591 hint->dah_eadata_len = spec->u.sp_ea.eadatalen;
1593 hint->dah_eadata = NULL;
1594 hint->dah_eadata_len = 0;
1597 CDEBUG(D_INFO, DFID" eadata %p len %d\n", PFID(mdd_object_fid(child)),
1598 hint->dah_eadata, hint->dah_eadata_len);
1599 /* @hint will be initialized by underlying device. */
1600 nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT);
1604 * do NOT or the MAY_*'s, you'll get the weakest
1606 int accmode(const struct lu_env *env, const struct lu_attr *la, int flags)
1610 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1611 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1612 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1613 * owner can write to a file even if it is marked readonly to hide
1614 * its brokenness. (bug 5781) */
1615 if (flags & MDS_OPEN_OWNEROVERRIDE) {
1616 struct lu_ucred *uc = lu_ucred_check(env);
1618 if ((uc == NULL) || (la->la_uid == uc->uc_fsuid))
1622 if (flags & FMODE_READ)
1624 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1626 if (flags & MDS_FMODE_EXEC)
1631 static int mdd_open_sanity_check(const struct lu_env *env,
1632 struct mdd_object *obj,
1633 const struct lu_attr *attr, int flag)
1639 if (mdd_is_dead_obj(obj))
1642 if (S_ISLNK(attr->la_mode))
1645 mode = accmode(env, attr, flag);
1647 if (S_ISDIR(attr->la_mode) && (mode & MAY_WRITE))
1650 if (!(flag & MDS_OPEN_CREATED)) {
1651 rc = mdd_permission_internal(env, obj, attr, mode);
1656 if (S_ISFIFO(attr->la_mode) || S_ISSOCK(attr->la_mode) ||
1657 S_ISBLK(attr->la_mode) || S_ISCHR(attr->la_mode))
1658 flag &= ~MDS_OPEN_TRUNC;
1660 /* For writing append-only file must open it with append mode. */
1661 if (attr->la_flags & LUSTRE_APPEND_FL) {
1662 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1664 if (flag & MDS_OPEN_TRUNC)
1670 * Now, flag -- O_NOATIME does not be packed by client.
1672 if (flag & O_NOATIME) {
1673 struct lu_ucred *uc = lu_ucred(env);
1675 if (uc && ((uc->uc_valid == UCRED_OLD) ||
1676 (uc->uc_valid == UCRED_NEW)) &&
1677 (uc->uc_fsuid != attr->la_uid) &&
1678 !md_capable(uc, CFS_CAP_FOWNER))
1686 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1689 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1690 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1693 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1695 rc = mdd_la_get(env, mdd_obj, attr);
1699 rc = mdd_open_sanity_check(env, mdd_obj, attr, flags);
1703 mdd_obj->mod_count++;
1706 mdd_write_unlock(env, mdd_obj);
1710 static int mdd_declare_close(const struct lu_env *env,
1711 struct mdd_object *obj,
1713 struct thandle *handle)
1717 rc = orph_declare_index_delete(env, obj, handle);
1721 return mdo_declare_destroy(env, obj, handle);
1725 * No permission check is needed.
1727 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1728 struct md_attr *ma, int mode)
1730 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1731 struct mdd_device *mdd = mdo2mdd(obj);
1732 struct thandle *handle = NULL;
1733 int rc, is_orphan = 0;
1736 if (ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_KEEP_ORPHAN) {
1737 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1738 mdd_obj->mod_count--;
1739 mdd_write_unlock(env, mdd_obj);
1741 if (mdd_obj->mod_flags & ORPHAN_OBJ && !mdd_obj->mod_count)
1742 CDEBUG(D_HA, "Object "DFID" is retained in orphan "
1743 "list\n", PFID(mdd_object_fid(mdd_obj)));
1747 /* mdd_finish_unlink() will always set orphan object as DEAD_OBJ, but
1748 * it might fail to add the object to orphan list (w/o ORPHAN_OBJ). */
1749 /* check without any lock */
1750 is_orphan = mdd_obj->mod_count == 1 &&
1751 (mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0;
1755 handle = mdd_trans_create(env, mdo2mdd(obj));
1757 RETURN(PTR_ERR(handle));
1759 rc = mdd_declare_close(env, mdd_obj, ma, handle);
1763 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1767 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1772 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1773 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr);
1775 CERROR("Failed to get lu_attr of "DFID": %d\n",
1776 PFID(mdd_object_fid(mdd_obj)), rc);
1780 /* check again with lock */
1781 is_orphan = (mdd_obj->mod_count == 1) &&
1782 ((mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0 ||
1783 ma->ma_attr.la_nlink == 0);
1785 if (is_orphan && handle == NULL) {
1786 mdd_write_unlock(env, mdd_obj);
1790 mdd_obj->mod_count--; /*release open count */
1796 /* NB: Object maybe not in orphan list originally, it is rare case for
1797 * mdd_finish_unlink() failure, in that case, the object doesn't have
1798 * ORPHAN_OBJ flag */
1799 if ((mdd_obj->mod_flags & ORPHAN_OBJ) != 0) {
1800 /* remove link to object from orphan index */
1801 LASSERT(handle != NULL);
1802 rc = __mdd_orphan_del(env, mdd_obj, handle);
1804 CERROR("%s: unable to delete "DFID" from orphan list: "
1805 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1806 PFID(mdd_object_fid(mdd_obj)), rc);
1807 /* If object was not deleted from orphan list, do not
1808 * destroy OSS objects, which will be done when next
1813 CDEBUG(D_HA, "Object "DFID" is deleted from orphan "
1814 "list, OSS objects to be destroyed.\n",
1815 PFID(mdd_object_fid(mdd_obj)));
1818 rc = mdo_destroy(env, mdd_obj, handle);
1821 CERROR("%s: unable to delete "DFID" from orphan list: "
1822 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1823 PFID(mdd_object_fid(mdd_obj)), rc);
1828 mdd_write_unlock(env, mdd_obj);
1831 (mode & (FMODE_WRITE | MDS_OPEN_APPEND | MDS_OPEN_TRUNC)) &&
1832 !(ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_RECOV_OPEN)) {
1833 if (handle == NULL) {
1834 handle = mdd_trans_create(env, mdo2mdd(obj));
1836 GOTO(stop, rc = PTR_ERR(handle));
1838 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL,
1843 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1848 mdd_changelog_data_store(env, mdd, CL_CLOSE, mode,
1853 if (handle != NULL && !IS_ERR(handle))
1854 mdd_trans_stop(env, mdd, rc, handle);
1860 * Permission check is done when open,
1861 * no need check again.
1863 static int mdd_readpage_sanity_check(const struct lu_env *env,
1864 struct mdd_object *obj)
1866 struct dt_object *next = mdd_object_child(obj);
1870 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
1878 static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp,
1879 size_t nob, const struct dt_it_ops *iops,
1880 struct dt_it *it, __u32 attr, void *arg)
1882 struct lu_dirpage *dp = &lp->lp_dir;
1886 struct lu_dirent *ent;
1887 struct lu_dirent *last = NULL;
1891 if (nob < sizeof(*dp))
1894 memset(area, 0, sizeof (*dp));
1895 area += sizeof (*dp);
1896 nob -= sizeof (*dp);
1903 len = iops->key_size(env, it);
1905 /* IAM iterator can return record with zero len. */
1909 hash = iops->store(env, it);
1910 if (unlikely(first)) {
1912 dp->ldp_hash_start = cpu_to_le64(hash);
1915 /* calculate max space required for lu_dirent */
1916 recsize = lu_dirent_calc_size(len, attr);
1918 if (nob >= recsize) {
1919 result = iops->rec(env, it, (struct dt_rec *)ent, attr);
1920 if (result == -ESTALE)
1925 /* osd might not able to pack all attributes,
1926 * so recheck rec length */
1927 recsize = le16_to_cpu(ent->lde_reclen);
1929 if (le32_to_cpu(ent->lde_attrs) & LUDA_FID) {
1930 fid_le_to_cpu(&fid, &ent->lde_fid);
1931 if (fid_is_dot_lustre(&fid))
1935 result = (last != NULL) ? 0 :-EINVAL;
1939 ent = (void *)ent + recsize;
1943 result = iops->next(env, it);
1944 if (result == -ESTALE)
1946 } while (result == 0);
1949 dp->ldp_hash_end = cpu_to_le64(hash);
1951 if (last->lde_hash == dp->ldp_hash_end)
1952 dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE);
1953 last->lde_reclen = 0; /* end mark */
1956 /* end of directory */
1957 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1958 else if (result < 0)
1959 CWARN("build page failed: %d!\n", result);
1963 int mdd_readpage(const struct lu_env *env, struct md_object *obj,
1964 const struct lu_rdpg *rdpg)
1966 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1970 if (mdd_object_exists(mdd_obj) == 0) {
1971 CERROR("%s: object "DFID" not found: rc = -2\n",
1972 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
1976 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
1977 rc = mdd_readpage_sanity_check(env, mdd_obj);
1979 GOTO(out_unlock, rc);
1981 if (mdd_is_dead_obj(mdd_obj)) {
1983 struct lu_dirpage *dp;
1986 * According to POSIX, please do not return any entry to client:
1987 * even dot and dotdot should not be returned.
1989 CDEBUG(D_INODE, "readdir from dead object: "DFID"\n",
1990 PFID(mdd_object_fid(mdd_obj)));
1992 if (rdpg->rp_count <= 0)
1993 GOTO(out_unlock, rc = -EFAULT);
1994 LASSERT(rdpg->rp_pages != NULL);
1996 pg = rdpg->rp_pages[0];
1997 dp = (struct lu_dirpage *)kmap(pg);
1998 memset(dp, 0 , sizeof(struct lu_dirpage));
1999 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
2000 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
2001 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
2003 GOTO(out_unlock, rc = LU_PAGE_SIZE);
2006 rc = dt_index_walk(env, mdd_object_child(mdd_obj), rdpg,
2007 mdd_dir_page_build, NULL);
2009 struct lu_dirpage *dp;
2011 dp = kmap(rdpg->rp_pages[0]);
2012 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
2015 * No pages were processed, mark this for first page
2018 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
2019 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
2020 rc = min_t(unsigned int, LU_PAGE_SIZE, rdpg->rp_count);
2022 kunmap(rdpg->rp_pages[0]);
2025 GOTO(out_unlock, rc);
2027 mdd_read_unlock(env, mdd_obj);
2031 static int mdd_object_sync(const struct lu_env *env, struct md_object *obj)
2033 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2035 if (mdd_object_exists(mdd_obj) == 0) {
2038 CERROR("%s: object "DFID" not found: rc = %d\n",
2039 mdd_obj_dev_name(mdd_obj),
2040 PFID(mdd_object_fid(mdd_obj)), rc);
2043 return dt_object_sync(env, mdd_object_child(mdd_obj),
2047 static int mdd_object_lock(const struct lu_env *env,
2048 struct md_object *obj,
2049 struct lustre_handle *lh,
2050 struct ldlm_enqueue_info *einfo,
2051 union ldlm_policy_data *policy)
2053 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2054 return dt_object_lock(env, mdd_object_child(mdd_obj), lh,
2058 static int mdd_object_unlock(const struct lu_env *env,
2059 struct md_object *obj,
2060 struct ldlm_enqueue_info *einfo,
2061 union ldlm_policy_data *policy)
2063 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2064 return dt_object_unlock(env, mdd_object_child(mdd_obj), einfo, policy);
2067 const struct md_object_operations mdd_obj_ops = {
2068 .moo_permission = mdd_permission,
2069 .moo_attr_get = mdd_attr_get,
2070 .moo_attr_set = mdd_attr_set,
2071 .moo_xattr_get = mdd_xattr_get,
2072 .moo_xattr_set = mdd_xattr_set,
2073 .moo_xattr_list = mdd_xattr_list,
2074 .moo_invalidate = mdd_invalidate,
2075 .moo_xattr_del = mdd_xattr_del,
2076 .moo_swap_layouts = mdd_swap_layouts,
2077 .moo_open = mdd_open,
2078 .moo_close = mdd_close,
2079 .moo_readpage = mdd_readpage,
2080 .moo_readlink = mdd_readlink,
2081 .moo_changelog = mdd_changelog,
2082 .moo_object_sync = mdd_object_sync,
2083 .moo_object_lock = mdd_object_lock,
2084 .moo_object_unlock = mdd_object_unlock,