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, 2014, 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,
68 if (mdd_object_exists(obj) == 0) {
69 CERROR("%s: object "DFID" not found: rc = -2\n",
70 mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj)));
74 return mdo_attr_get(env, obj, la);
77 struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
79 struct mdd_thread_info *info;
81 lu_env_refill((struct lu_env *)env);
82 info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
83 LASSERT(info != NULL);
87 struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
91 buf = &mdd_env_info(env)->mti_buf[0];
97 const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
98 const void *area, ssize_t len)
102 buf = &mdd_env_info(env)->mti_buf[0];
103 buf->lb_buf = (void *)area;
108 struct lu_object *mdd_object_alloc(const struct lu_env *env,
109 const struct lu_object_header *hdr,
112 struct mdd_object *mdd_obj;
114 OBD_SLAB_ALLOC_PTR_GFP(mdd_obj, mdd_object_kmem, GFP_NOFS);
115 if (mdd_obj != NULL) {
118 o = mdd2lu_obj(mdd_obj);
119 lu_object_init(o, NULL, d);
120 mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
121 mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
122 mdd_obj->mod_count = 0;
123 o->lo_ops = &mdd_lu_obj_ops;
130 static int mdd_object_init(const struct lu_env *env, struct lu_object *o,
131 const struct lu_object_conf *unused)
133 struct mdd_device *d = lu2mdd_dev(o->lo_dev);
134 struct mdd_object *mdd_obj = lu2mdd_obj(o);
135 struct lu_object *below;
136 struct lu_device *under;
139 mdd_obj->mod_cltime = 0;
140 under = &d->mdd_child->dd_lu_dev;
141 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
143 RETURN(PTR_ERR(below));
145 lu_object_add(o, below);
150 static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
154 if (lu_object_exists(o)) {
155 struct mdd_object *mdd_obj = lu2mdd_obj(o);
156 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
158 rc = mdd_la_get(env, mdd_obj, attr);
164 static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
166 struct mdd_object *mdd = lu2mdd_obj(o);
169 OBD_SLAB_FREE_PTR(mdd, mdd_object_kmem);
172 static int mdd_object_print(const struct lu_env *env, void *cookie,
173 lu_printer_t p, const struct lu_object *o)
175 struct mdd_object *mdd = lu2mdd_obj((struct lu_object *)o);
176 return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p(open_count=%d, "
177 "valid=%x, cltime="LPU64", flags=%lx)",
178 mdd, mdd->mod_count, mdd->mod_valid,
179 mdd->mod_cltime, mdd->mod_flags);
182 static const struct lu_object_operations mdd_lu_obj_ops = {
183 .loo_object_init = mdd_object_init,
184 .loo_object_start = mdd_object_start,
185 .loo_object_free = mdd_object_free,
186 .loo_object_print = mdd_object_print,
189 struct mdd_object *mdd_object_find(const struct lu_env *env,
190 struct mdd_device *d,
191 const struct lu_fid *f)
193 return md2mdd_obj(md_object_find_slice(env, &d->mdd_md_dev, f));
197 * No permission check is needed.
199 int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
202 struct mdd_object *mdd_obj = md2mdd_obj(obj);
207 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr);
208 if ((ma->ma_need & MA_INODE) != 0 && mdd_is_dead_obj(mdd_obj))
209 ma->ma_attr.la_nlink = 0;
215 * No permission check is needed.
217 static int mdd_xattr_get(const struct lu_env *env,
218 struct md_object *obj, struct lu_buf *buf,
221 struct mdd_object *mdd_obj = md2mdd_obj(obj);
226 if (mdd_object_exists(mdd_obj) == 0) {
227 CERROR("%s: object "DFID" not found: rc = -2\n",
228 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
232 /* If the object has been delete from the namespace, then
233 * get linkEA should return -ENOENT as well */
234 if (unlikely((mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ)) &&
235 strcmp(name, XATTR_NAME_LINK) == 0))
238 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
239 rc = mdo_xattr_get(env, mdd_obj, buf, name);
240 mdd_read_unlock(env, mdd_obj);
246 * Permission check is done when open,
247 * no need check again.
249 int mdd_readlink(const struct lu_env *env, struct md_object *obj,
252 struct mdd_object *mdd_obj = md2mdd_obj(obj);
253 struct dt_object *next;
258 if (mdd_object_exists(mdd_obj) == 0) {
259 CERROR("%s: object "DFID" not found: rc = -2\n",
260 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
264 next = mdd_object_child(mdd_obj);
265 LASSERT(next != NULL);
266 LASSERT(next->do_body_ops != NULL);
267 LASSERT(next->do_body_ops->dbo_read != NULL);
268 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
269 rc = dt_read(env, next, buf, &pos);
270 mdd_read_unlock(env, mdd_obj);
275 * No permission check is needed.
277 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
280 struct mdd_object *mdd_obj = md2mdd_obj(obj);
285 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
286 rc = mdo_xattr_list(env, mdd_obj, buf);
287 mdd_read_unlock(env, mdd_obj);
293 * Filter out XATTR_NAME_LINK if this is an orphan object. See
296 if (unlikely(mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ))) {
297 char *end = (char *)buf->lb_buf + rc;
298 char *p = buf->lb_buf;
301 char *next = p + strlen(p) + 1;
303 if (strcmp(p, XATTR_NAME_LINK) == 0) {
305 memmove(p, next, end - next);
307 CDEBUG(D_INFO, "Filtered out "XATTR_NAME_LINK
308 " of orphan "DFID"\n",
309 PFID(mdd_object_fid(mdd_obj)));
320 int mdd_declare_object_create_internal(const struct lu_env *env,
321 struct mdd_object *p,
322 struct mdd_object *c,
323 struct lu_attr *attr,
324 struct thandle *handle,
325 const struct md_op_spec *spec,
326 struct dt_allocation_hint *hint)
328 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
329 const struct dt_index_features *feat = spec->sp_feat;
333 if (feat != &dt_directory_features && feat != NULL) {
334 dof->dof_type = DFT_INDEX;
335 dof->u.dof_idx.di_feat = feat;
338 dof->dof_type = dt_mode_to_dft(attr->la_mode);
339 if (dof->dof_type == DFT_REGULAR) {
340 dof->u.dof_reg.striped =
341 md_should_create(spec->sp_cr_flags);
342 if (spec->sp_cr_flags & MDS_OPEN_HAS_EA)
343 dof->u.dof_reg.striped = 0;
344 /* is this replay? */
346 dof->u.dof_reg.striped = 0;
350 rc = mdo_declare_create_obj(env, c, attr, hint, dof, handle);
355 int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p,
356 struct mdd_object *c, struct lu_attr *attr,
357 struct thandle *handle,
358 const struct md_op_spec *spec,
359 struct dt_allocation_hint *hint)
361 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
365 LASSERT(!mdd_object_exists(c));
367 rc = mdo_create_obj(env, c, attr, hint, dof, handle);
372 int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj,
373 const struct lu_attr *attr, struct thandle *handle,
379 rc = mdo_attr_set(env, obj, attr, handle);
380 #ifdef CONFIG_FS_POSIX_ACL
381 if (!rc && (attr->la_valid & LA_MODE) && needacl)
382 rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
387 int mdd_update_time(const struct lu_env *env, struct mdd_object *obj,
388 const struct lu_attr *oattr, struct lu_attr *attr,
389 struct thandle *handle)
394 LASSERT(attr->la_valid & LA_CTIME);
395 LASSERT(oattr != NULL);
397 /* Make sure the ctime is increased only, however, it's not strictly
398 * reliable at here because there is not guarantee to hold lock on
399 * object, so we just bypass some unnecessary cmtime setting first
400 * and OSD has to check it again. */
401 if (attr->la_ctime < oattr->la_ctime)
402 attr->la_valid &= ~(LA_MTIME | LA_CTIME);
403 else if (attr->la_valid == LA_CTIME &&
404 attr->la_ctime == oattr->la_ctime)
405 attr->la_valid &= ~LA_CTIME;
407 if (attr->la_valid != 0)
408 rc = mdd_attr_set_internal(env, obj, attr, handle, 0);
413 * This gives the same functionality as the code between
414 * sys_chmod and inode_setattr
415 * chown_common and inode_setattr
416 * utimes and inode_setattr
417 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
419 static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
420 const struct lu_attr *oattr, struct lu_attr *la,
421 const unsigned long flags)
430 /* Do not permit change file type */
431 if (la->la_valid & LA_TYPE)
434 /* They should not be processed by setattr */
435 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
438 LASSERT(oattr != NULL);
440 /* export destroy does not have ->le_ses, but we may want
441 * to drop LUSTRE_SOM_FL. */
442 uc = lu_ucred_check(env);
446 if (la->la_valid == LA_CTIME) {
447 if (!(flags & MDS_PERM_BYPASS))
448 /* This is only for set ctime when rename's source is
450 rc = mdd_may_delete(env, NULL, NULL, obj, oattr, NULL,
452 if (rc == 0 && la->la_ctime <= oattr->la_ctime)
453 la->la_valid &= ~LA_CTIME;
457 if (la->la_valid == LA_ATIME) {
458 /* This is atime only set for read atime update on close. */
459 if (la->la_atime >= oattr->la_atime &&
460 la->la_atime < (oattr->la_atime +
461 mdd_obj2mdd_dev(obj)->mdd_atime_diff))
462 la->la_valid &= ~LA_ATIME;
466 /* Check if flags change. */
467 if (la->la_valid & LA_FLAGS) {
468 unsigned int oldflags = oattr->la_flags &
469 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
470 unsigned int newflags = la->la_flags &
471 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
473 if ((uc->uc_fsuid != oattr->la_uid) &&
474 !md_capable(uc, CFS_CAP_FOWNER))
477 /* The IMMUTABLE and APPEND_ONLY flags can
478 * only be changed by the relevant capability. */
479 if ((oldflags ^ newflags) &&
480 !md_capable(uc, CFS_CAP_LINUX_IMMUTABLE))
483 if (!S_ISDIR(oattr->la_mode))
484 la->la_flags &= ~LUSTRE_DIRSYNC_FL;
487 if (oattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL) &&
488 (la->la_valid & ~LA_FLAGS) &&
489 !(flags & MDS_PERM_BYPASS))
492 /* Check for setting the obj time. */
493 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
494 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
495 if ((uc->uc_fsuid != oattr->la_uid) &&
496 !md_capable(uc, CFS_CAP_FOWNER)) {
497 rc = mdd_permission_internal(env, obj, oattr,
504 if (la->la_valid & LA_KILL_SUID) {
505 la->la_valid &= ~LA_KILL_SUID;
506 if ((oattr->la_mode & S_ISUID) &&
507 !(la->la_valid & LA_MODE)) {
508 la->la_mode = oattr->la_mode;
509 la->la_valid |= LA_MODE;
511 la->la_mode &= ~S_ISUID;
514 if (la->la_valid & LA_KILL_SGID) {
515 la->la_valid &= ~LA_KILL_SGID;
516 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
517 (S_ISGID | S_IXGRP)) &&
518 !(la->la_valid & LA_MODE)) {
519 la->la_mode = oattr->la_mode;
520 la->la_valid |= LA_MODE;
522 la->la_mode &= ~S_ISGID;
525 /* Make sure a caller can chmod. */
526 if (la->la_valid & LA_MODE) {
527 if (!(flags & MDS_PERM_BYPASS) &&
528 (uc->uc_fsuid != oattr->la_uid) &&
529 !md_capable(uc, CFS_CAP_FOWNER))
532 if (la->la_mode == (umode_t) -1)
533 la->la_mode = oattr->la_mode;
535 la->la_mode = (la->la_mode & S_IALLUGO) |
536 (oattr->la_mode & ~S_IALLUGO);
538 /* Also check the setgid bit! */
539 if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ?
540 la->la_gid : oattr->la_gid) &&
541 !md_capable(uc, CFS_CAP_FSETID))
542 la->la_mode &= ~S_ISGID;
544 la->la_mode = oattr->la_mode;
547 /* Make sure a caller can chown. */
548 if (la->la_valid & LA_UID) {
549 if (la->la_uid == (uid_t) -1)
550 la->la_uid = oattr->la_uid;
551 if (((uc->uc_fsuid != oattr->la_uid) ||
552 (la->la_uid != oattr->la_uid)) &&
553 !md_capable(uc, CFS_CAP_CHOWN))
556 /* If the user or group of a non-directory has been
557 * changed by a non-root user, remove the setuid bit.
558 * 19981026 David C Niemi <niemi@tux.org>
560 * Changed this to apply to all users, including root,
561 * to avoid some races. This is the behavior we had in
562 * 2.0. The check for non-root was definitely wrong
563 * for 2.2 anyway, as it should have been using
564 * CAP_FSETID rather than fsuid -- 19990830 SD. */
565 if (((oattr->la_mode & S_ISUID) == S_ISUID) &&
566 !S_ISDIR(oattr->la_mode)) {
567 la->la_mode &= ~S_ISUID;
568 la->la_valid |= LA_MODE;
572 /* Make sure caller can chgrp. */
573 if (la->la_valid & LA_GID) {
574 if (la->la_gid == (gid_t) -1)
575 la->la_gid = oattr->la_gid;
576 if (((uc->uc_fsuid != oattr->la_uid) ||
577 ((la->la_gid != oattr->la_gid) &&
578 !lustre_in_group_p(uc, la->la_gid))) &&
579 !md_capable(uc, CFS_CAP_CHOWN))
582 /* Likewise, if the user or group of a non-directory
583 * has been changed by a non-root user, remove the
584 * setgid bit UNLESS there is no group execute bit
585 * (this would be a file marked for mandatory
586 * locking). 19981026 David C Niemi <niemi@tux.org>
588 * Removed the fsuid check (see the comment above) --
590 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
591 (S_ISGID | S_IXGRP)) && !S_ISDIR(oattr->la_mode)) {
592 la->la_mode &= ~S_ISGID;
593 la->la_valid |= LA_MODE;
597 /* For both Size-on-MDS case and truncate case,
598 * "la->la_valid & (LA_SIZE | LA_BLOCKS)" are ture.
599 * We distinguish them by "flags & MDS_SOM".
600 * For SOM case, it is true, the MAY_WRITE perm has been checked
601 * when open, no need check again. For truncate case, it is false,
602 * the MAY_WRITE perm should be checked here. */
603 if (flags & MDS_SOM) {
604 /* For the "Size-on-MDS" setattr update, merge coming
605 * attributes with the set in the inode. BUG 10641 */
606 if ((la->la_valid & LA_ATIME) &&
607 (la->la_atime <= oattr->la_atime))
608 la->la_valid &= ~LA_ATIME;
610 /* OST attributes do not have a priority over MDS attributes,
611 * so drop times if ctime is equal. */
612 if ((la->la_valid & LA_CTIME) &&
613 (la->la_ctime <= oattr->la_ctime))
614 la->la_valid &= ~(LA_MTIME | LA_CTIME);
616 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
617 if (!((flags & MDS_OWNEROVERRIDE) &&
618 (uc->uc_fsuid == oattr->la_uid)) &&
619 !(flags & MDS_PERM_BYPASS)) {
620 rc = mdd_permission_internal(env, obj,
626 if (la->la_valid & LA_CTIME) {
627 /* The pure setattr, it has the priority over what is
628 * already set, do not drop it if ctime is equal. */
629 if (la->la_ctime < oattr->la_ctime)
630 la->la_valid &= ~(LA_ATIME | LA_MTIME |
638 /** Store a data change changelog record
639 * If this fails, we must fail the whole transaction; we don't
640 * want the change to commit without the log entry.
641 * \param mdd_obj - mdd_object of change
642 * \param handle - transacion handle
644 int mdd_changelog_data_store(const struct lu_env *env, struct mdd_device *mdd,
645 enum changelog_rec_type type, int flags,
646 struct mdd_object *mdd_obj, struct thandle *handle)
648 const struct lu_ucred *uc = lu_ucred(env);
649 const struct lu_fid *tfid;
650 struct llog_changelog_rec *rec;
656 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
658 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
661 LASSERT(mdd_obj != NULL);
662 LASSERT(handle != NULL);
664 tfid = mdo2fid(mdd_obj);
666 if ((type >= CL_MTIME) && (type <= CL_ATIME) &&
667 cfs_time_before_64(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) {
668 /* Don't need multiple updates in this log */
669 /* Don't check under lock - no big deal if we get an extra
674 flags = (flags & CLF_FLAGMASK) | CLF_VERSION;
675 if (uc != NULL && uc->uc_jobid[0] != '\0')
678 reclen = llog_data_len(changelog_rec_offset(flags & CLF_SUPPORTED));
679 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
680 if (buf->lb_buf == NULL)
684 rec->cr.cr_flags = flags;
685 rec->cr.cr_type = (__u32)type;
686 rec->cr.cr_tfid = *tfid;
687 rec->cr.cr_namelen = 0;
688 mdd_obj->mod_cltime = cfs_time_current_64();
690 if (flags & CLF_JOBID)
691 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
693 rc = mdd_changelog_store(env, mdd, rec, handle);
698 static int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type,
699 int flags, struct md_object *obj)
701 struct thandle *handle;
702 struct mdd_object *mdd_obj = md2mdd_obj(obj);
703 struct mdd_device *mdd = mdo2mdd(obj);
707 handle = mdd_trans_create(env, mdd);
709 RETURN(PTR_ERR(handle));
711 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
715 rc = mdd_trans_start(env, mdd, handle);
719 rc = mdd_changelog_data_store(env, mdd, type, flags, mdd_obj,
723 mdd_trans_stop(env, mdd, rc, handle);
729 * Save LMA extended attributes with data from \a ma.
731 * HSM and Size-On-MDS data will be extracted from \ma if they are valid, if
732 * not, LMA EA will be first read from disk, modified and write back.
735 /* Precedence for choosing record type when multiple
736 * attributes change: setattr > mtime > ctime > atime
737 * (ctime changes when mtime does, plus chmod/chown.
738 * atime and ctime are independent.) */
739 static int mdd_attr_set_changelog(const struct lu_env *env,
740 struct md_object *obj, struct thandle *handle,
743 struct mdd_device *mdd = mdo2mdd(obj);
746 bits = (valid & LA_SIZE) ? 1 << CL_TRUNC : 0;
747 bits |= (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0;
748 bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0;
749 bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0;
750 bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0;
751 bits = bits & mdd->mdd_cl.mc_mask;
752 /* This is an implementation limit rather than a protocol limit */
753 CLASSERT(CL_LAST <= sizeof(int) * 8);
757 /* The record type is the lowest non-masked set bit */
760 /* FYI we only store the first CLF_FLAGMASK bits of la_valid */
761 return mdd_changelog_data_store(env, mdd, type, (int)valid,
762 md2mdd_obj(obj), handle);
765 static int mdd_declare_attr_set(const struct lu_env *env,
766 struct mdd_device *mdd,
767 struct mdd_object *obj,
768 const struct lu_attr *attr,
769 struct thandle *handle)
773 rc = mdo_declare_attr_set(env, obj, attr, handle);
777 #ifdef CONFIG_FS_POSIX_ACL
778 if (attr->la_valid & LA_MODE) {
779 mdd_read_lock(env, obj, MOR_TGT_CHILD);
780 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL,
781 XATTR_NAME_ACL_ACCESS);
782 mdd_read_unlock(env, obj);
783 if (rc == -EOPNOTSUPP || rc == -ENODATA)
789 struct lu_buf *buf = mdd_buf_get(env, NULL, rc);
790 rc = mdo_declare_xattr_set(env, obj, buf,
791 XATTR_NAME_ACL_ACCESS, 0,
799 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
806 * permission changes may require sync operation, to mitigate performance
807 * impact, only do this for dir and when permission is reduced.
809 * For regular files, version is updated with permission change (see VBR), async
810 * permission won't cause any issue, while missing permission change on
811 * directory may affect accessibility of other objects after recovery.
813 static inline bool permission_needs_sync(const struct lu_attr *old,
814 const struct lu_attr *new)
816 if (!S_ISDIR(old->la_mode))
819 if (new->la_valid & (LA_UID | LA_GID))
822 if (new->la_valid & LA_MODE &&
823 new->la_mode & (S_ISUID | S_ISGID | S_ISVTX))
826 if ((new->la_valid & LA_MODE) &&
827 ((new->la_mode & old->la_mode) & S_IRWXUGO) !=
828 (old->la_mode & S_IRWXUGO))
834 /* set attr and LOV EA at once, return updated attr */
835 int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
836 const struct md_attr *ma)
838 struct mdd_object *mdd_obj = md2mdd_obj(obj);
839 struct mdd_device *mdd = mdo2mdd(obj);
840 struct thandle *handle;
841 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
842 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
843 const struct lu_attr *la = &ma->ma_attr;
847 /* we do not use ->attr_set() for LOV/SOM/HSM EA any more */
848 LASSERT((ma->ma_valid & MA_LOV) == 0);
849 LASSERT((ma->ma_valid & MA_HSM) == 0);
850 LASSERT((ma->ma_valid & MA_SOM) == 0);
852 rc = mdd_la_get(env, mdd_obj, attr);
856 *la_copy = ma->ma_attr;
857 rc = mdd_fix_attr(env, mdd_obj, attr, la_copy, ma->ma_attr_flags);
861 /* setattr on "close" only change atime, or do nothing */
862 if (la->la_valid == LA_ATIME && la_copy->la_valid == 0)
865 handle = mdd_trans_create(env, mdd);
867 RETURN(PTR_ERR(handle));
869 rc = mdd_declare_attr_set(env, mdd, mdd_obj, la, handle);
873 rc = mdd_trans_start(env, mdd, handle);
877 if (mdd->mdd_sync_permission && permission_needs_sync(attr, la))
880 if (la->la_valid & (LA_MTIME | LA_CTIME))
881 CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n",
882 la->la_mtime, la->la_ctime);
884 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
885 if (la_copy->la_valid & LA_FLAGS)
886 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
887 else if (la_copy->la_valid) /* setattr */
888 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
889 mdd_write_unlock(env, mdd_obj);
892 rc = mdd_attr_set_changelog(env, obj, handle, la->la_valid);
897 mdd_trans_stop(env, mdd, rc, handle);
901 static int mdd_xattr_sanity_check(const struct lu_env *env,
902 struct mdd_object *obj,
903 const struct lu_attr *attr)
905 struct lu_ucred *uc = lu_ucred_assert(env);
908 if (attr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
911 if ((uc->uc_fsuid != attr->la_uid) && !md_capable(uc, CFS_CAP_FOWNER))
917 static int mdd_declare_xattr_set(const struct lu_env *env,
918 struct mdd_device *mdd,
919 struct mdd_object *obj,
920 const struct lu_buf *buf,
922 int fl, struct thandle *handle)
926 rc = mdo_declare_xattr_set(env, obj, buf, name, fl, handle);
930 /* Only record user and layout xattr changes */
931 if (strncmp(XATTR_USER_PREFIX, name,
932 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
933 strcmp(XATTR_NAME_LOV, name) == 0) {
934 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
939 /* If HSM data is modified, this could add a changelog */
940 if (strcmp(XATTR_NAME_HSM, name) == 0) {
941 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
946 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
951 * Compare current and future data of HSM EA and add a changelog if needed.
953 * Caller should have write-locked \param obj.
955 * \param buf - Future HSM EA content.
956 * \retval 0 if no changelog is needed or changelog was added properly.
957 * \retval -ve errno if there was a problem
959 static int mdd_hsm_update_locked(const struct lu_env *env,
960 struct md_object *obj,
961 const struct lu_buf *buf,
962 struct thandle *handle)
964 struct mdd_thread_info *info = mdd_env_info(env);
965 struct mdd_device *mdd = mdo2mdd(obj);
966 struct mdd_object *mdd_obj = md2mdd_obj(obj);
967 struct lu_buf *current_buf;
968 struct md_hsm *current_mh;
969 struct md_hsm *new_mh;
973 OBD_ALLOC_PTR(current_mh);
974 if (current_mh == NULL)
977 /* Read HSM attrs from disk */
978 CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
979 current_buf = mdd_buf_get(env, info->mti_xattr_buf,
980 sizeof(info->mti_xattr_buf));
981 rc = mdo_xattr_get(env, mdd_obj, current_buf, XATTR_NAME_HSM);
982 rc = lustre_buf2hsm(current_buf->lb_buf, rc, current_mh);
983 if (rc < 0 && rc != -ENODATA)
985 else if (rc == -ENODATA)
986 current_mh->mh_flags = 0;
988 /* Map future HSM xattr */
989 OBD_ALLOC_PTR(new_mh);
991 GOTO(free, rc = -ENOMEM);
992 lustre_buf2hsm(buf->lb_buf, buf->lb_len, new_mh);
994 /* If HSM flags are different, add a changelog */
996 if (current_mh->mh_flags != new_mh->mh_flags) {
998 hsm_set_cl_event(&flags, HE_STATE);
999 if (new_mh->mh_flags & HS_DIRTY)
1000 hsm_set_cl_flags(&flags, CLF_HSM_DIRTY);
1002 rc = mdd_changelog_data_store(env, mdd, CL_HSM, flags, mdd_obj,
1006 OBD_FREE_PTR(new_mh);
1008 OBD_FREE_PTR(current_mh);
1012 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1016 * The caller should guarantee to update the object ctime
1017 * after xattr_set if needed.
1019 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
1020 const struct lu_buf *buf, const char *name,
1023 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1024 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1025 struct mdd_device *mdd = mdo2mdd(obj);
1026 struct thandle *handle;
1030 rc = mdd_la_get(env, mdd_obj, attr);
1034 rc = mdd_xattr_sanity_check(env, mdd_obj, attr);
1038 if (strcmp(name, XATTR_NAME_ACL_ACCESS) == 0 ||
1039 strcmp(name, XATTR_NAME_ACL_DEFAULT) == 0) {
1040 struct posix_acl *acl;
1042 /* user may set empty ACL, which should be treated as removing
1044 acl = posix_acl_from_xattr(&init_user_ns, buf->lb_buf,
1047 rc = mdd_xattr_del(env, obj, name);
1050 posix_acl_release(acl);
1053 if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) {
1054 rc = mdd_acl_set(env, mdd_obj, attr, buf, fl);
1058 handle = mdd_trans_create(env, mdd);
1060 RETURN(PTR_ERR(handle));
1062 rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, fl, handle);
1066 rc = mdd_trans_start(env, mdd, handle);
1070 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1072 if (strcmp(XATTR_NAME_HSM, name) == 0) {
1073 rc = mdd_hsm_update_locked(env, obj, buf, handle);
1075 mdd_write_unlock(env, mdd_obj);
1080 rc = mdo_xattr_set(env, mdd_obj, buf, name, fl, handle);
1081 mdd_write_unlock(env, mdd_obj);
1085 if (strcmp(XATTR_NAME_LOV, name) == 0)
1086 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, mdd_obj,
1088 else if (strncmp(XATTR_USER_PREFIX, name,
1089 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1090 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1091 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1092 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1093 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1094 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1098 mdd_trans_stop(env, mdd, rc, handle);
1103 static int mdd_declare_xattr_del(const struct lu_env *env,
1104 struct mdd_device *mdd,
1105 struct mdd_object *obj,
1107 struct thandle *handle)
1111 rc = mdo_declare_xattr_del(env, obj, name, handle);
1115 /* Only record user xattr changes */
1116 if ((strncmp(XATTR_USER_PREFIX, name,
1117 sizeof(XATTR_USER_PREFIX) - 1) == 0))
1118 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1124 * The caller should guarantee to update the object ctime
1125 * after xattr_set if needed.
1127 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1130 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1131 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1132 struct mdd_device *mdd = mdo2mdd(obj);
1133 struct thandle *handle;
1137 rc = mdd_la_get(env, mdd_obj, attr);
1141 rc = mdd_xattr_sanity_check(env, mdd_obj, attr);
1145 handle = mdd_trans_create(env, mdd);
1147 RETURN(PTR_ERR(handle));
1149 rc = mdd_declare_xattr_del(env, mdd, mdd_obj, name, handle);
1153 rc = mdd_trans_start(env, mdd, handle);
1157 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1158 rc = mdo_xattr_del(env, mdd_obj, name, handle);
1159 mdd_write_unlock(env, mdd_obj);
1163 /* Only record system & user xattr changes */
1164 if (strncmp(XATTR_USER_PREFIX, name,
1165 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1166 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1167 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1168 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1169 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1170 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1174 mdd_trans_stop(env, mdd, rc, handle);
1180 * read lov EA of an object
1181 * return the lov EA in an allocated lu_buf
1183 int mdd_get_lov_ea(const struct lu_env *env, struct mdd_object *obj,
1184 struct lu_buf *lmm_buf)
1186 struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf;
1191 rc = mdo_xattr_get(env, obj, buf, XATTR_NAME_LOV);
1193 if (rc == -ERANGE) {
1194 /* mti_big_buf is allocated but is too small
1195 * we need to increase it */
1196 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1198 if (buf->lb_buf == NULL)
1199 GOTO(out, rc = -ENOMEM);
1210 if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) {
1211 /* mti_big_buf was not allocated, so we have to
1212 * allocate it based on the ea size */
1213 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1215 if (buf->lb_buf == NULL)
1216 GOTO(out, rc = -ENOMEM);
1220 lu_buf_alloc(lmm_buf, bufsize);
1221 if (lmm_buf->lb_buf == NULL)
1222 GOTO(out, rc = -ENOMEM);
1224 memcpy(lmm_buf->lb_buf, buf->lb_buf, bufsize);
1230 lu_buf_free(lmm_buf);
1234 static int mdd_xattr_hsm_replace(const struct lu_env *env,
1235 struct mdd_object *o, struct lu_buf *buf,
1236 struct thandle *handle)
1238 struct hsm_attrs *attrs;
1244 rc = mdo_xattr_set(env, o, buf, XATTR_NAME_HSM, LU_XATTR_REPLACE,
1249 attrs = buf->lb_buf;
1250 hsm_flags = le32_to_cpu(attrs->hsm_flags);
1251 if (!(hsm_flags & HS_RELEASED) || mdd_is_dead_obj(o))
1254 /* Add a changelog record for release. */
1255 hsm_set_cl_event(&flags, HE_RELEASE);
1256 rc = mdd_changelog_data_store(env, mdo2mdd(&o->mod_obj), CL_HSM,
1262 * check if layout swapping between 2 objects is allowed
1264 * - only normal FIDs or non-system IGIFs
1265 * - same type of objects
1266 * - same owner/group (so quotas are still valid)
1268 static int mdd_layout_swap_allowed(const struct lu_env *env,
1269 struct mdd_object *o1,
1270 const struct lu_attr *attr1,
1271 struct mdd_object *o2,
1272 const struct lu_attr *attr2)
1274 const struct lu_fid *fid1, *fid2;
1280 if (!fid_is_norm(fid1) &&
1281 (!fid_is_igif(fid1) || IS_ERR(mdd_links_get(env, o1))))
1284 if (!fid_is_norm(fid2) &&
1285 (!fid_is_igif(fid2) || IS_ERR(mdd_links_get(env, o2))))
1288 if (mdd_object_type(o1) != mdd_object_type(o2)) {
1289 if (S_ISDIR(mdd_object_type(o1)))
1291 if (S_ISREG(mdd_object_type(o1)))
1296 if ((attr1->la_uid != attr2->la_uid) ||
1297 (attr1->la_gid != attr2->la_gid))
1304 * swap layouts between 2 lustre objects
1306 static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1,
1307 struct md_object *obj2, __u64 flags)
1309 struct mdd_thread_info *info = mdd_env_info(env);
1310 struct mdd_object *fst_o = md2mdd_obj(obj1);
1311 struct mdd_object *snd_o = md2mdd_obj(obj2);
1312 struct lu_attr *fst_la = MDD_ENV_VAR(env, cattr);
1313 struct lu_attr *snd_la = MDD_ENV_VAR(env, tattr);
1314 struct mdd_device *mdd = mdo2mdd(obj1);
1315 struct lov_mds_md *fst_lmm, *snd_lmm;
1316 struct lu_buf *fst_buf = &info->mti_buf[0];
1317 struct lu_buf *snd_buf = &info->mti_buf[1];
1318 struct lu_buf *fst_hsm_buf = &info->mti_buf[2];
1319 struct lu_buf *snd_hsm_buf = &info->mti_buf[3];
1320 struct ost_id *saved_oi = NULL;
1321 struct thandle *handle;
1322 __u16 fst_gen, snd_gen;
1328 CLASSERT(ARRAY_SIZE(info->mti_buf) >= 4);
1329 memset(info->mti_buf, 0, sizeof(info->mti_buf));
1331 /* we have to sort the 2 obj, so locking will always
1332 * be in the same order, even in case of 2 concurrent swaps */
1333 rc = lu_fid_cmp(mdo2fid(fst_o), mdo2fid(snd_o));
1334 if (rc == 0) /* same fid ? */
1340 rc = mdd_la_get(env, fst_o, fst_la);
1344 rc = mdd_la_get(env, snd_o, snd_la);
1348 /* check if layout swapping is allowed */
1349 rc = mdd_layout_swap_allowed(env, fst_o, fst_la, snd_o, snd_la);
1353 handle = mdd_trans_create(env, mdd);
1355 RETURN(PTR_ERR(handle));
1357 /* objects are already sorted */
1358 mdd_write_lock(env, fst_o, MOR_TGT_CHILD);
1359 mdd_write_lock(env, snd_o, MOR_TGT_CHILD);
1361 rc = mdd_get_lov_ea(env, fst_o, fst_buf);
1362 if (rc < 0 && rc != -ENODATA)
1365 rc = mdd_get_lov_ea(env, snd_o, snd_buf);
1366 if (rc < 0 && rc != -ENODATA)
1369 /* swapping 2 non existant layouts is a success */
1370 if (fst_buf->lb_buf == NULL && snd_buf->lb_buf == NULL)
1373 /* to help inode migration between MDT, it is better to
1374 * start by the no layout file (if one), so we order the swap */
1375 if (snd_buf->lb_buf == NULL) {
1377 swap(fst_buf, snd_buf);
1380 /* lmm and generation layout initialization */
1381 if (fst_buf->lb_buf != NULL) {
1382 fst_lmm = fst_buf->lb_buf;
1383 fst_gen = le16_to_cpu(fst_lmm->lmm_layout_gen);
1384 fst_fl = LU_XATTR_REPLACE;
1388 fst_fl = LU_XATTR_CREATE;
1391 snd_lmm = snd_buf->lb_buf;
1392 snd_gen = le16_to_cpu(snd_lmm->lmm_layout_gen);
1394 /* increase the generation layout numbers */
1398 /* set the file specific informations in lmm */
1399 if (fst_lmm != NULL) {
1400 saved_oi = &info->mti_oa.o_oi;
1402 *saved_oi = fst_lmm->lmm_oi;
1403 fst_lmm->lmm_layout_gen = cpu_to_le16(snd_gen);
1404 fst_lmm->lmm_oi = snd_lmm->lmm_oi;
1405 snd_lmm->lmm_oi = *saved_oi;
1407 if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1))
1408 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEF);
1409 else if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))
1410 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V3_DEF);
1412 GOTO(stop, rc = -EPROTO);
1414 snd_lmm->lmm_layout_gen = cpu_to_le16(fst_gen);
1416 /* Prepare HSM attribute if it's required */
1417 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1418 const int buflen = sizeof(struct hsm_attrs);
1420 lu_buf_alloc(fst_hsm_buf, buflen);
1421 lu_buf_alloc(snd_hsm_buf, buflen);
1422 if (fst_hsm_buf->lb_buf == NULL || snd_hsm_buf->lb_buf == NULL)
1423 GOTO(stop, rc = -ENOMEM);
1425 /* Read HSM attribute */
1426 rc = mdo_xattr_get(env, fst_o, fst_hsm_buf, XATTR_NAME_HSM);
1430 rc = mdo_xattr_get(env, snd_o, snd_hsm_buf, XATTR_NAME_HSM);
1434 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_hsm_buf,
1435 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1440 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_hsm_buf,
1441 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1447 /* prepare transaction */
1448 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_buf, XATTR_NAME_LOV,
1453 if (fst_buf->lb_buf != NULL)
1454 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_buf,
1455 XATTR_NAME_LOV, LU_XATTR_REPLACE,
1458 rc = mdd_declare_xattr_del(env, mdd, snd_o, XATTR_NAME_LOV,
1463 rc = mdd_trans_start(env, mdd, handle);
1467 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1468 rc = mdd_xattr_hsm_replace(env, fst_o, snd_hsm_buf, handle);
1472 rc = mdd_xattr_hsm_replace(env, snd_o, fst_hsm_buf, handle);
1474 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf,
1477 CERROR("%s: restore "DFID" HSM error: %d/%d\n",
1478 mdd_obj_dev_name(fst_o),
1479 PFID(mdo2fid(fst_o)), rc, rc2);
1484 rc = mdo_xattr_set(env, fst_o, snd_buf, XATTR_NAME_LOV, fst_fl, handle);
1488 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_MDS_HSM_SWAP_LAYOUTS))) {
1491 if (fst_buf->lb_buf != NULL)
1492 rc = mdo_xattr_set(env, snd_o, fst_buf, XATTR_NAME_LOV,
1493 LU_XATTR_REPLACE, handle);
1495 rc = mdo_xattr_del(env, snd_o, XATTR_NAME_LOV, handle);
1501 /* failure on second file, but first was done, so we have
1502 * to roll back first. */
1503 if (fst_buf->lb_buf != NULL) {
1504 fst_lmm->lmm_oi = *saved_oi;
1505 fst_lmm->lmm_layout_gen = cpu_to_le16(fst_gen - 1);
1506 rc2 = mdo_xattr_set(env, fst_o, fst_buf, XATTR_NAME_LOV,
1507 LU_XATTR_REPLACE, handle);
1509 rc2 = mdo_xattr_del(env, fst_o, XATTR_NAME_LOV, handle);
1515 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, handle);
1520 rc2 = mdd_xattr_hsm_replace(env, snd_o, snd_hsm_buf, handle);
1525 CERROR("%s: unable to roll back layout swap. FIDs: "
1526 DFID" and "DFID "error: %d/%d, steps: %d\n",
1527 mdd_obj_dev_name(fst_o),
1528 PFID(mdo2fid(snd_o)), PFID(mdo2fid(fst_o)),
1530 /* a solution to avoid journal commit is to panic,
1531 * but it has strong consequences so we use LBUG to
1532 * allow sysdamin to choose to panic or not
1539 /* Issue one changelog record per file */
1540 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle);
1544 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle);
1550 mdd_trans_stop(env, mdd, rc, handle);
1551 mdd_write_unlock(env, snd_o);
1552 mdd_write_unlock(env, fst_o);
1554 lu_buf_free(fst_buf);
1555 lu_buf_free(snd_buf);
1556 lu_buf_free(fst_hsm_buf);
1557 lu_buf_free(snd_hsm_buf);
1561 void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent,
1562 struct mdd_object *child, const struct lu_attr *attr,
1563 const struct md_op_spec *spec,
1564 struct dt_allocation_hint *hint)
1566 struct dt_object *np = parent ? mdd_object_child(parent) : NULL;
1567 struct dt_object *nc = mdd_object_child(child);
1569 memset(hint, 0, sizeof(*hint));
1571 /* For striped directory, give striping EA to lod_ah_init, which will
1572 * decide the stripe_offset and stripe count by it. */
1573 if (S_ISDIR(attr->la_mode) &&
1574 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1575 hint->dah_eadata = spec->u.sp_ea.eadata;
1576 hint->dah_eadata_len = spec->u.sp_ea.eadatalen;
1578 hint->dah_eadata = NULL;
1579 hint->dah_eadata_len = 0;
1582 CDEBUG(D_INFO, DFID" eadata %p len %d\n", PFID(mdd_object_fid(child)),
1583 hint->dah_eadata, hint->dah_eadata_len);
1584 /* @hint will be initialized by underlying device. */
1585 nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT);
1589 * do NOT or the MAY_*'s, you'll get the weakest
1591 int accmode(const struct lu_env *env, const struct lu_attr *la, int flags)
1595 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1596 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1597 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1598 * owner can write to a file even if it is marked readonly to hide
1599 * its brokenness. (bug 5781) */
1600 if (flags & MDS_OPEN_OWNEROVERRIDE) {
1601 struct lu_ucred *uc = lu_ucred_check(env);
1603 if ((uc == NULL) || (la->la_uid == uc->uc_fsuid))
1607 if (flags & FMODE_READ)
1609 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1611 if (flags & MDS_FMODE_EXEC)
1616 static int mdd_open_sanity_check(const struct lu_env *env,
1617 struct mdd_object *obj,
1618 const struct lu_attr *attr, int flag)
1624 if (mdd_is_dead_obj(obj))
1627 if (S_ISLNK(attr->la_mode))
1630 mode = accmode(env, attr, flag);
1632 if (S_ISDIR(attr->la_mode) && (mode & MAY_WRITE))
1635 if (!(flag & MDS_OPEN_CREATED)) {
1636 rc = mdd_permission_internal(env, obj, attr, mode);
1641 if (S_ISFIFO(attr->la_mode) || S_ISSOCK(attr->la_mode) ||
1642 S_ISBLK(attr->la_mode) || S_ISCHR(attr->la_mode))
1643 flag &= ~MDS_OPEN_TRUNC;
1645 /* For writing append-only file must open it with append mode. */
1646 if (attr->la_flags & LUSTRE_APPEND_FL) {
1647 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1649 if (flag & MDS_OPEN_TRUNC)
1655 * Now, flag -- O_NOATIME does not be packed by client.
1657 if (flag & O_NOATIME) {
1658 struct lu_ucred *uc = lu_ucred(env);
1660 if (uc && ((uc->uc_valid == UCRED_OLD) ||
1661 (uc->uc_valid == UCRED_NEW)) &&
1662 (uc->uc_fsuid != attr->la_uid) &&
1663 !md_capable(uc, CFS_CAP_FOWNER))
1671 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1674 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1675 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1678 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1680 rc = mdd_la_get(env, mdd_obj, attr);
1684 rc = mdd_open_sanity_check(env, mdd_obj, attr, flags);
1688 mdd_obj->mod_count++;
1691 mdd_write_unlock(env, mdd_obj);
1695 static int mdd_declare_close(const struct lu_env *env,
1696 struct mdd_object *obj,
1698 struct thandle *handle)
1702 rc = orph_declare_index_delete(env, obj, handle);
1706 return mdo_declare_destroy(env, obj, handle);
1710 * No permission check is needed.
1712 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1713 struct md_attr *ma, int mode)
1715 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1716 struct mdd_device *mdd = mdo2mdd(obj);
1717 struct thandle *handle = NULL;
1718 int rc, is_orphan = 0;
1721 if (ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_KEEP_ORPHAN) {
1722 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1723 mdd_obj->mod_count--;
1724 mdd_write_unlock(env, mdd_obj);
1726 if (mdd_obj->mod_flags & ORPHAN_OBJ && !mdd_obj->mod_count)
1727 CDEBUG(D_HA, "Object "DFID" is retained in orphan "
1728 "list\n", PFID(mdd_object_fid(mdd_obj)));
1732 /* mdd_finish_unlink() will always set orphan object as DEAD_OBJ, but
1733 * it might fail to add the object to orphan list (w/o ORPHAN_OBJ). */
1734 /* check without any lock */
1735 is_orphan = mdd_obj->mod_count == 1 &&
1736 (mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0;
1740 handle = mdd_trans_create(env, mdo2mdd(obj));
1742 RETURN(PTR_ERR(handle));
1744 rc = mdd_declare_close(env, mdd_obj, ma, handle);
1748 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1752 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1757 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1758 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr);
1760 CERROR("Failed to get lu_attr of "DFID": %d\n",
1761 PFID(mdd_object_fid(mdd_obj)), rc);
1765 /* check again with lock */
1766 is_orphan = (mdd_obj->mod_count == 1) &&
1767 ((mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0 ||
1768 ma->ma_attr.la_nlink == 0);
1770 if (is_orphan && handle == NULL) {
1771 mdd_write_unlock(env, mdd_obj);
1775 mdd_obj->mod_count--; /*release open count */
1781 /* NB: Object maybe not in orphan list originally, it is rare case for
1782 * mdd_finish_unlink() failure, in that case, the object doesn't have
1783 * ORPHAN_OBJ flag */
1784 if ((mdd_obj->mod_flags & ORPHAN_OBJ) != 0) {
1785 /* remove link to object from orphan index */
1786 LASSERT(handle != NULL);
1787 rc = __mdd_orphan_del(env, mdd_obj, handle);
1789 CERROR("%s: unable to delete "DFID" from orphan list: "
1790 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1791 PFID(mdd_object_fid(mdd_obj)), rc);
1792 /* If object was not deleted from orphan list, do not
1793 * destroy OSS objects, which will be done when next
1798 CDEBUG(D_HA, "Object "DFID" is deleted from orphan "
1799 "list, OSS objects to be destroyed.\n",
1800 PFID(mdd_object_fid(mdd_obj)));
1803 rc = mdo_destroy(env, mdd_obj, handle);
1806 CERROR("%s: unable to delete "DFID" from orphan list: "
1807 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1808 PFID(mdd_object_fid(mdd_obj)), rc);
1813 mdd_write_unlock(env, mdd_obj);
1816 (mode & (FMODE_WRITE | MDS_OPEN_APPEND | MDS_OPEN_TRUNC)) &&
1817 !(ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_RECOV_OPEN)) {
1818 if (handle == NULL) {
1819 handle = mdd_trans_create(env, mdo2mdd(obj));
1821 GOTO(stop, rc = PTR_ERR(handle));
1823 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL,
1828 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1833 mdd_changelog_data_store(env, mdd, CL_CLOSE, mode,
1838 if (handle != NULL && !IS_ERR(handle))
1839 mdd_trans_stop(env, mdd, rc, handle);
1845 * Permission check is done when open,
1846 * no need check again.
1848 static int mdd_readpage_sanity_check(const struct lu_env *env,
1849 struct mdd_object *obj)
1851 struct dt_object *next = mdd_object_child(obj);
1855 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
1863 static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp,
1864 size_t nob, const struct dt_it_ops *iops,
1865 struct dt_it *it, __u32 attr, void *arg)
1867 struct lu_dirpage *dp = &lp->lp_dir;
1871 struct lu_dirent *ent;
1872 struct lu_dirent *last = NULL;
1876 if (nob < sizeof(*dp))
1879 memset(area, 0, sizeof (*dp));
1880 area += sizeof (*dp);
1881 nob -= sizeof (*dp);
1888 len = iops->key_size(env, it);
1890 /* IAM iterator can return record with zero len. */
1894 hash = iops->store(env, it);
1895 if (unlikely(first)) {
1897 dp->ldp_hash_start = cpu_to_le64(hash);
1900 /* calculate max space required for lu_dirent */
1901 recsize = lu_dirent_calc_size(len, attr);
1903 if (nob >= recsize) {
1904 result = iops->rec(env, it, (struct dt_rec *)ent, attr);
1905 if (result == -ESTALE)
1910 /* osd might not able to pack all attributes,
1911 * so recheck rec length */
1912 recsize = le16_to_cpu(ent->lde_reclen);
1914 if (le32_to_cpu(ent->lde_attrs) & LUDA_FID) {
1915 fid_le_to_cpu(&fid, &ent->lde_fid);
1916 if (fid_is_dot_lustre(&fid))
1920 result = (last != NULL) ? 0 :-EINVAL;
1924 ent = (void *)ent + recsize;
1928 result = iops->next(env, it);
1929 if (result == -ESTALE)
1931 } while (result == 0);
1934 dp->ldp_hash_end = cpu_to_le64(hash);
1936 if (last->lde_hash == dp->ldp_hash_end)
1937 dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE);
1938 last->lde_reclen = 0; /* end mark */
1941 /* end of directory */
1942 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1943 else if (result < 0)
1944 CWARN("build page failed: %d!\n", result);
1948 int mdd_readpage(const struct lu_env *env, struct md_object *obj,
1949 const struct lu_rdpg *rdpg)
1951 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1955 if (mdd_object_exists(mdd_obj) == 0) {
1956 CERROR("%s: object "DFID" not found: rc = -2\n",
1957 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
1961 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
1962 rc = mdd_readpage_sanity_check(env, mdd_obj);
1964 GOTO(out_unlock, rc);
1966 if (mdd_is_dead_obj(mdd_obj)) {
1968 struct lu_dirpage *dp;
1971 * According to POSIX, please do not return any entry to client:
1972 * even dot and dotdot should not be returned.
1974 CDEBUG(D_INODE, "readdir from dead object: "DFID"\n",
1975 PFID(mdd_object_fid(mdd_obj)));
1977 if (rdpg->rp_count <= 0)
1978 GOTO(out_unlock, rc = -EFAULT);
1979 LASSERT(rdpg->rp_pages != NULL);
1981 pg = rdpg->rp_pages[0];
1982 dp = (struct lu_dirpage *)kmap(pg);
1983 memset(dp, 0 , sizeof(struct lu_dirpage));
1984 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
1985 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1986 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
1988 GOTO(out_unlock, rc = LU_PAGE_SIZE);
1991 rc = dt_index_walk(env, mdd_object_child(mdd_obj), rdpg,
1992 mdd_dir_page_build, NULL);
1994 struct lu_dirpage *dp;
1996 dp = kmap(rdpg->rp_pages[0]);
1997 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
2000 * No pages were processed, mark this for first page
2003 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
2004 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
2005 rc = min_t(unsigned int, LU_PAGE_SIZE, rdpg->rp_count);
2007 kunmap(rdpg->rp_pages[0]);
2010 GOTO(out_unlock, rc);
2012 mdd_read_unlock(env, mdd_obj);
2016 static int mdd_object_sync(const struct lu_env *env, struct md_object *obj)
2018 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2020 if (mdd_object_exists(mdd_obj) == 0) {
2023 CERROR("%s: object "DFID" not found: rc = %d\n",
2024 mdd_obj_dev_name(mdd_obj),
2025 PFID(mdd_object_fid(mdd_obj)), rc);
2028 return dt_object_sync(env, mdd_object_child(mdd_obj),
2032 static int mdd_object_lock(const struct lu_env *env,
2033 struct md_object *obj,
2034 struct lustre_handle *lh,
2035 struct ldlm_enqueue_info *einfo,
2036 ldlm_policy_data_t *policy)
2038 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2039 return dt_object_lock(env, mdd_object_child(mdd_obj), lh,
2043 static int mdd_object_unlock(const struct lu_env *env,
2044 struct md_object *obj,
2045 struct ldlm_enqueue_info *einfo,
2046 ldlm_policy_data_t *policy)
2048 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2049 return dt_object_unlock(env, mdd_object_child(mdd_obj), einfo, policy);
2052 const struct md_object_operations mdd_obj_ops = {
2053 .moo_permission = mdd_permission,
2054 .moo_attr_get = mdd_attr_get,
2055 .moo_attr_set = mdd_attr_set,
2056 .moo_xattr_get = mdd_xattr_get,
2057 .moo_xattr_set = mdd_xattr_set,
2058 .moo_xattr_list = mdd_xattr_list,
2059 .moo_xattr_del = mdd_xattr_del,
2060 .moo_swap_layouts = mdd_swap_layouts,
2061 .moo_open = mdd_open,
2062 .moo_close = mdd_close,
2063 .moo_readpage = mdd_readpage,
2064 .moo_readlink = mdd_readlink,
2065 .moo_changelog = mdd_changelog,
2066 .moo_object_sync = mdd_object_sync,
2067 .moo_object_lock = mdd_object_lock,
2068 .moo_object_unlock = mdd_object_unlock,