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, 2013, 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,
66 struct lu_attr *la, struct lustre_capa *capa)
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, capa);
77 void mdd_flags_xlate(struct mdd_object *obj, __u32 flags)
79 obj->mod_flags &= ~(APPEND_OBJ|IMMUTE_OBJ);
81 if (flags & LUSTRE_APPEND_FL)
82 obj->mod_flags |= APPEND_OBJ;
84 if (flags & LUSTRE_IMMUTABLE_FL)
85 obj->mod_flags |= IMMUTE_OBJ;
88 struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
90 struct mdd_thread_info *info;
92 lu_env_refill((struct lu_env *)env);
93 info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
94 LASSERT(info != NULL);
98 struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
102 buf = &mdd_env_info(env)->mti_buf[0];
108 const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
109 const void *area, ssize_t len)
113 buf = &mdd_env_info(env)->mti_buf[0];
114 buf->lb_buf = (void *)area;
119 struct lu_object *mdd_object_alloc(const struct lu_env *env,
120 const struct lu_object_header *hdr,
123 struct mdd_object *mdd_obj;
125 OBD_SLAB_ALLOC_PTR_GFP(mdd_obj, mdd_object_kmem, GFP_NOFS);
126 if (mdd_obj != NULL) {
129 o = mdd2lu_obj(mdd_obj);
130 lu_object_init(o, NULL, d);
131 mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
132 mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
133 mdd_obj->mod_count = 0;
134 o->lo_ops = &mdd_lu_obj_ops;
141 static int mdd_object_init(const struct lu_env *env, struct lu_object *o,
142 const struct lu_object_conf *unused)
144 struct mdd_device *d = lu2mdd_dev(o->lo_dev);
145 struct mdd_object *mdd_obj = lu2mdd_obj(o);
146 struct lu_object *below;
147 struct lu_device *under;
150 mdd_obj->mod_cltime = 0;
151 under = &d->mdd_child->dd_lu_dev;
152 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
154 RETURN(PTR_ERR(below));
156 lu_object_add(o, below);
161 static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
165 if (lu_object_exists(o)) {
166 struct mdd_object *mdd_obj = lu2mdd_obj(o);
167 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
169 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
171 mdd_flags_xlate(mdd_obj, attr->la_flags);
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 mdd_object_capa(env, md2mdd_obj(obj)));
222 if ((ma->ma_need & MA_INODE) != 0 && mdd_is_dead_obj(mdd_obj))
223 ma->ma_attr.la_nlink = 0;
229 * No permission check is needed.
231 static int mdd_xattr_get(const struct lu_env *env,
232 struct md_object *obj, struct lu_buf *buf,
235 struct mdd_object *mdd_obj = md2mdd_obj(obj);
240 if (mdd_object_exists(mdd_obj) == 0) {
241 CERROR("%s: object "DFID" not found: rc = -2\n",
242 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
246 /* If the object has been delete from the namespace, then
247 * get linkEA should return -ENOENT as well */
248 if (unlikely((mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ)) &&
249 strcmp(name, XATTR_NAME_LINK) == 0))
252 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
253 rc = mdo_xattr_get(env, mdd_obj, buf, name,
254 mdd_object_capa(env, mdd_obj));
255 mdd_read_unlock(env, mdd_obj);
261 * Permission check is done when open,
262 * no need check again.
264 int mdd_readlink(const struct lu_env *env, struct md_object *obj,
267 struct mdd_object *mdd_obj = md2mdd_obj(obj);
268 struct dt_object *next;
273 if (mdd_object_exists(mdd_obj) == 0) {
274 CERROR("%s: object "DFID" not found: rc = -2\n",
275 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
279 next = mdd_object_child(mdd_obj);
280 LASSERT(next != NULL);
281 LASSERT(next->do_body_ops != NULL);
282 LASSERT(next->do_body_ops->dbo_read != NULL);
283 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
284 rc = next->do_body_ops->dbo_read(env, next, buf, &pos,
285 mdd_object_capa(env, mdd_obj));
286 mdd_read_unlock(env, mdd_obj);
291 * No permission check is needed.
293 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
296 struct mdd_object *mdd_obj = md2mdd_obj(obj);
301 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
302 rc = mdo_xattr_list(env, mdd_obj, buf, mdd_object_capa(env, mdd_obj));
303 mdd_read_unlock(env, mdd_obj);
308 int mdd_declare_object_create_internal(const struct lu_env *env,
309 struct mdd_object *p,
310 struct mdd_object *c,
311 struct lu_attr *attr,
312 struct thandle *handle,
313 const struct md_op_spec *spec,
314 struct dt_allocation_hint *hint)
316 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
317 const struct dt_index_features *feat = spec->sp_feat;
321 if (feat != &dt_directory_features && feat != NULL) {
322 dof->dof_type = DFT_INDEX;
323 dof->u.dof_idx.di_feat = feat;
326 dof->dof_type = dt_mode_to_dft(attr->la_mode);
327 if (dof->dof_type == DFT_REGULAR) {
328 dof->u.dof_reg.striped =
329 md_should_create(spec->sp_cr_flags);
330 if (spec->sp_cr_flags & MDS_OPEN_HAS_EA)
331 dof->u.dof_reg.striped = 0;
332 /* is this replay? */
334 dof->u.dof_reg.striped = 0;
338 rc = mdo_declare_create_obj(env, c, attr, hint, dof, handle);
343 int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p,
344 struct mdd_object *c, struct lu_attr *attr,
345 struct thandle *handle,
346 const struct md_op_spec *spec,
347 struct dt_allocation_hint *hint)
349 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
353 LASSERT(!mdd_object_exists(c));
355 rc = mdo_create_obj(env, c, attr, hint, dof, handle);
357 LASSERT(ergo(rc == 0, mdd_object_exists(c)));
362 int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj,
363 const struct lu_attr *attr, struct thandle *handle,
369 rc = mdo_attr_set(env, obj, attr, handle, mdd_object_capa(env, obj));
370 #ifdef CONFIG_FS_POSIX_ACL
371 if (!rc && (attr->la_valid & LA_MODE) && needacl)
372 rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
377 int mdd_update_time(const struct lu_env *env, struct mdd_object *obj,
378 const struct lu_attr *oattr, struct lu_attr *attr,
379 struct thandle *handle)
384 LASSERT(attr->la_valid & LA_CTIME);
385 LASSERT(oattr != NULL);
387 /* Make sure the ctime is increased only, however, it's not strictly
388 * reliable at here because there is not guarantee to hold lock on
389 * object, so we just bypass some unnecessary cmtime setting first
390 * and OSD has to check it again. */
391 if (attr->la_ctime < oattr->la_ctime)
392 attr->la_valid &= ~(LA_MTIME | LA_CTIME);
393 else if (attr->la_valid == LA_CTIME &&
394 attr->la_ctime == oattr->la_ctime)
395 attr->la_valid &= ~LA_CTIME;
397 if (attr->la_valid != 0)
398 rc = mdd_attr_set_internal(env, obj, attr, handle, 0);
403 * This gives the same functionality as the code between
404 * sys_chmod and inode_setattr
405 * chown_common and inode_setattr
406 * utimes and inode_setattr
407 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
409 static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
410 const struct lu_attr *oattr, struct lu_attr *la,
411 const unsigned long flags)
420 /* Do not permit change file type */
421 if (la->la_valid & LA_TYPE)
424 /* They should not be processed by setattr */
425 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
428 LASSERT(oattr != NULL);
430 /* export destroy does not have ->le_ses, but we may want
431 * to drop LUSTRE_SOM_FL. */
432 uc = lu_ucred_check(env);
436 if (la->la_valid == LA_CTIME) {
437 if (!(flags & MDS_PERM_BYPASS))
438 /* This is only for set ctime when rename's source is
440 rc = mdd_may_delete(env, NULL, NULL, obj, oattr, NULL,
442 if (rc == 0 && la->la_ctime <= oattr->la_ctime)
443 la->la_valid &= ~LA_CTIME;
447 if (la->la_valid == LA_ATIME) {
448 /* This is atime only set for read atime update on close. */
449 if (la->la_atime >= oattr->la_atime &&
450 la->la_atime < (oattr->la_atime +
451 mdd_obj2mdd_dev(obj)->mdd_atime_diff))
452 la->la_valid &= ~LA_ATIME;
456 /* Check if flags change. */
457 if (la->la_valid & LA_FLAGS) {
458 unsigned int oldflags = 0;
459 unsigned int newflags = la->la_flags &
460 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
462 if ((uc->uc_fsuid != oattr->la_uid) &&
463 !md_capable(uc, CFS_CAP_FOWNER))
466 /* XXX: the IMMUTABLE and APPEND_ONLY flags can
467 * only be changed by the relevant capability. */
468 if (mdd_is_immutable(obj))
469 oldflags |= LUSTRE_IMMUTABLE_FL;
470 if (mdd_is_append(obj))
471 oldflags |= LUSTRE_APPEND_FL;
472 if ((oldflags ^ newflags) &&
473 !md_capable(uc, CFS_CAP_LINUX_IMMUTABLE))
476 if (!S_ISDIR(oattr->la_mode))
477 la->la_flags &= ~LUSTRE_DIRSYNC_FL;
480 if ((mdd_is_immutable(obj) || mdd_is_append(obj)) &&
481 (la->la_valid & ~LA_FLAGS) &&
482 !(flags & MDS_PERM_BYPASS))
485 /* Check for setting the obj time. */
486 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
487 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
488 if ((uc->uc_fsuid != oattr->la_uid) &&
489 !md_capable(uc, CFS_CAP_FOWNER)) {
490 rc = mdd_permission_internal(env, obj, oattr,
497 if (la->la_valid & LA_KILL_SUID) {
498 la->la_valid &= ~LA_KILL_SUID;
499 if ((oattr->la_mode & S_ISUID) &&
500 !(la->la_valid & LA_MODE)) {
501 la->la_mode = oattr->la_mode;
502 la->la_valid |= LA_MODE;
504 la->la_mode &= ~S_ISUID;
507 if (la->la_valid & LA_KILL_SGID) {
508 la->la_valid &= ~LA_KILL_SGID;
509 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
510 (S_ISGID | S_IXGRP)) &&
511 !(la->la_valid & LA_MODE)) {
512 la->la_mode = oattr->la_mode;
513 la->la_valid |= LA_MODE;
515 la->la_mode &= ~S_ISGID;
518 /* Make sure a caller can chmod. */
519 if (la->la_valid & LA_MODE) {
520 if (!(flags & MDS_PERM_BYPASS) &&
521 (uc->uc_fsuid != oattr->la_uid) &&
522 !md_capable(uc, CFS_CAP_FOWNER))
525 if (la->la_mode == (umode_t) -1)
526 la->la_mode = oattr->la_mode;
528 la->la_mode = (la->la_mode & S_IALLUGO) |
529 (oattr->la_mode & ~S_IALLUGO);
531 /* Also check the setgid bit! */
532 if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ?
533 la->la_gid : oattr->la_gid) &&
534 !md_capable(uc, CFS_CAP_FSETID))
535 la->la_mode &= ~S_ISGID;
537 la->la_mode = oattr->la_mode;
540 /* Make sure a caller can chown. */
541 if (la->la_valid & LA_UID) {
542 if (la->la_uid == (uid_t) -1)
543 la->la_uid = oattr->la_uid;
544 if (((uc->uc_fsuid != oattr->la_uid) ||
545 (la->la_uid != oattr->la_uid)) &&
546 !md_capable(uc, CFS_CAP_CHOWN))
549 /* If the user or group of a non-directory has been
550 * changed by a non-root user, remove the setuid bit.
551 * 19981026 David C Niemi <niemi@tux.org>
553 * Changed this to apply to all users, including root,
554 * to avoid some races. This is the behavior we had in
555 * 2.0. The check for non-root was definitely wrong
556 * for 2.2 anyway, as it should have been using
557 * CAP_FSETID rather than fsuid -- 19990830 SD. */
558 if (((oattr->la_mode & S_ISUID) == S_ISUID) &&
559 !S_ISDIR(oattr->la_mode)) {
560 la->la_mode &= ~S_ISUID;
561 la->la_valid |= LA_MODE;
565 /* Make sure caller can chgrp. */
566 if (la->la_valid & LA_GID) {
567 if (la->la_gid == (gid_t) -1)
568 la->la_gid = oattr->la_gid;
569 if (((uc->uc_fsuid != oattr->la_uid) ||
570 ((la->la_gid != oattr->la_gid) &&
571 !lustre_in_group_p(uc, la->la_gid))) &&
572 !md_capable(uc, CFS_CAP_CHOWN))
575 /* Likewise, if the user or group of a non-directory
576 * has been changed by a non-root user, remove the
577 * setgid bit UNLESS there is no group execute bit
578 * (this would be a file marked for mandatory
579 * locking). 19981026 David C Niemi <niemi@tux.org>
581 * Removed the fsuid check (see the comment above) --
583 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
584 (S_ISGID | S_IXGRP)) && !S_ISDIR(oattr->la_mode)) {
585 la->la_mode &= ~S_ISGID;
586 la->la_valid |= LA_MODE;
590 /* For both Size-on-MDS case and truncate case,
591 * "la->la_valid & (LA_SIZE | LA_BLOCKS)" are ture.
592 * We distinguish them by "flags & MDS_SOM".
593 * For SOM case, it is true, the MAY_WRITE perm has been checked
594 * when open, no need check again. For truncate case, it is false,
595 * the MAY_WRITE perm should be checked here. */
596 if (flags & MDS_SOM) {
597 /* For the "Size-on-MDS" setattr update, merge coming
598 * attributes with the set in the inode. BUG 10641 */
599 if ((la->la_valid & LA_ATIME) &&
600 (la->la_atime <= oattr->la_atime))
601 la->la_valid &= ~LA_ATIME;
603 /* OST attributes do not have a priority over MDS attributes,
604 * so drop times if ctime is equal. */
605 if ((la->la_valid & LA_CTIME) &&
606 (la->la_ctime <= oattr->la_ctime))
607 la->la_valid &= ~(LA_MTIME | LA_CTIME);
609 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
610 if (!((flags & MDS_OWNEROVERRIDE) &&
611 (uc->uc_fsuid == oattr->la_uid)) &&
612 !(flags & MDS_PERM_BYPASS)) {
613 rc = mdd_permission_internal(env, obj,
619 if (la->la_valid & LA_CTIME) {
620 /* The pure setattr, it has the priority over what is
621 * already set, do not drop it if ctime is equal. */
622 if (la->la_ctime < oattr->la_ctime)
623 la->la_valid &= ~(LA_ATIME | LA_MTIME |
631 /** Store a data change changelog record
632 * If this fails, we must fail the whole transaction; we don't
633 * want the change to commit without the log entry.
634 * \param mdd_obj - mdd_object of change
635 * \param handle - transacion handle
637 int mdd_changelog_data_store(const struct lu_env *env, struct mdd_device *mdd,
638 enum changelog_rec_type type, int flags,
639 struct mdd_object *mdd_obj, struct thandle *handle)
641 const struct lu_fid *tfid;
642 struct llog_changelog_rec *rec;
648 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
650 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
653 LASSERT(mdd_obj != NULL);
654 LASSERT(handle != NULL);
656 tfid = mdo2fid(mdd_obj);
658 if ((type >= CL_MTIME) && (type <= CL_ATIME) &&
659 cfs_time_before_64(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) {
660 /* Don't need multiple updates in this log */
661 /* Don't check under lock - no big deal if we get an extra
666 reclen = llog_data_len(sizeof(*rec));
667 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
668 if (buf->lb_buf == NULL)
672 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
673 rec->cr.cr_type = (__u32)type;
674 rec->cr.cr_tfid = *tfid;
675 rec->cr.cr_namelen = 0;
676 mdd_obj->mod_cltime = cfs_time_current_64();
678 rc = mdd_changelog_store(env, mdd, rec, handle);
683 static int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type,
684 int flags, struct md_object *obj)
686 struct thandle *handle;
687 struct mdd_object *mdd_obj = md2mdd_obj(obj);
688 struct mdd_device *mdd = mdo2mdd(obj);
692 handle = mdd_trans_create(env, mdd);
694 RETURN(PTR_ERR(handle));
696 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
700 rc = mdd_trans_start(env, mdd, handle);
704 rc = mdd_changelog_data_store(env, mdd, type, flags, mdd_obj,
708 mdd_trans_stop(env, mdd, rc, handle);
714 * Save LMA extended attributes with data from \a ma.
716 * HSM and Size-On-MDS data will be extracted from \ma if they are valid, if
717 * not, LMA EA will be first read from disk, modified and write back.
720 /* Precedence for choosing record type when multiple
721 * attributes change: setattr > mtime > ctime > atime
722 * (ctime changes when mtime does, plus chmod/chown.
723 * atime and ctime are independent.) */
724 static int mdd_attr_set_changelog(const struct lu_env *env,
725 struct md_object *obj, struct thandle *handle,
728 struct mdd_device *mdd = mdo2mdd(obj);
731 bits = (valid & LA_SIZE) ? 1 << CL_TRUNC : 0;
732 bits |= (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0;
733 bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0;
734 bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0;
735 bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0;
736 bits = bits & mdd->mdd_cl.mc_mask;
737 /* This is an implementation limit rather than a protocol limit */
738 CLASSERT(CL_LAST <= sizeof(int) * 8);
742 /* The record type is the lowest non-masked set bit */
745 /* FYI we only store the first CLF_FLAGMASK bits of la_valid */
746 return mdd_changelog_data_store(env, mdd, type, (int)valid,
747 md2mdd_obj(obj), handle);
750 static int mdd_declare_attr_set(const struct lu_env *env,
751 struct mdd_device *mdd,
752 struct mdd_object *obj,
753 const struct lu_attr *attr,
754 struct thandle *handle)
758 rc = mdo_declare_attr_set(env, obj, attr, handle);
762 #ifdef CONFIG_FS_POSIX_ACL
763 if (attr->la_valid & LA_MODE) {
764 mdd_read_lock(env, obj, MOR_TGT_CHILD);
765 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL,
766 XATTR_NAME_ACL_ACCESS, BYPASS_CAPA);
767 mdd_read_unlock(env, obj);
768 if (rc == -EOPNOTSUPP || rc == -ENODATA)
774 struct lu_buf *buf = mdd_buf_get(env, NULL, rc);
775 rc = mdo_declare_xattr_set(env, obj, buf,
776 XATTR_NAME_ACL_ACCESS, 0,
784 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
791 * permission changes may require sync operation, to mitigate performance
792 * impact, only do this for dir and when permission is reduced.
794 * For regular files, version is updated with permission change (see VBR), async
795 * permission won't cause any issue, while missing permission change on
796 * directory may affect accessibility of other objects after recovery.
798 static inline bool permission_needs_sync(const struct lu_attr *old,
799 const struct lu_attr *new)
801 if (!S_ISDIR(old->la_mode))
804 if (new->la_valid & (LA_UID | LA_GID))
807 if (new->la_valid & LA_MODE &&
808 new->la_mode & (S_ISUID | S_ISGID | S_ISVTX))
811 if ((new->la_valid & LA_MODE) &&
812 ((new->la_mode & old->la_mode) & S_IRWXUGO) !=
813 (old->la_mode & S_IRWXUGO))
819 /* set attr and LOV EA at once, return updated attr */
820 int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
821 const struct md_attr *ma)
823 struct mdd_object *mdd_obj = md2mdd_obj(obj);
824 struct mdd_device *mdd = mdo2mdd(obj);
825 struct thandle *handle;
826 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
827 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
828 const struct lu_attr *la = &ma->ma_attr;
832 /* we do not use ->attr_set() for LOV/SOM/HSM EA any more */
833 LASSERT((ma->ma_valid & MA_LOV) == 0);
834 LASSERT((ma->ma_valid & MA_HSM) == 0);
835 LASSERT((ma->ma_valid & MA_SOM) == 0);
837 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
841 *la_copy = ma->ma_attr;
842 rc = mdd_fix_attr(env, mdd_obj, attr, la_copy, ma->ma_attr_flags);
846 /* setattr on "close" only change atime, or do nothing */
847 if (la->la_valid == LA_ATIME && la_copy->la_valid == 0)
850 handle = mdd_trans_create(env, mdd);
852 RETURN(PTR_ERR(handle));
854 rc = mdd_declare_attr_set(env, mdd, mdd_obj, la, handle);
858 rc = mdd_trans_start(env, mdd, handle);
862 if (mdd->mdd_sync_permission && permission_needs_sync(attr, la))
865 if (la->la_valid & (LA_MTIME | LA_CTIME))
866 CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n",
867 la->la_mtime, la->la_ctime);
869 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
870 if (la_copy->la_valid & LA_FLAGS) {
871 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
873 mdd_flags_xlate(mdd_obj, la_copy->la_flags);
874 } else if (la_copy->la_valid) { /* setattr */
875 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
877 mdd_write_unlock(env, mdd_obj);
880 rc = mdd_attr_set_changelog(env, obj, handle, la->la_valid);
885 mdd_trans_stop(env, mdd, rc, handle);
889 static int mdd_xattr_sanity_check(const struct lu_env *env,
890 struct mdd_object *obj,
891 const struct lu_attr *attr)
893 struct lu_ucred *uc = lu_ucred_assert(env);
896 if (mdd_is_immutable(obj) || mdd_is_append(obj))
899 if ((uc->uc_fsuid != attr->la_uid) && !md_capable(uc, CFS_CAP_FOWNER))
905 static int mdd_declare_xattr_set(const struct lu_env *env,
906 struct mdd_device *mdd,
907 struct mdd_object *obj,
908 const struct lu_buf *buf,
910 int fl, struct thandle *handle)
914 rc = mdo_declare_xattr_set(env, obj, buf, name, fl, handle);
918 /* Only record user and layout xattr changes */
919 if (strncmp(XATTR_USER_PREFIX, name,
920 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
921 strcmp(XATTR_NAME_LOV, name) == 0) {
922 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
927 /* If HSM data is modified, this could add a changelog */
928 if (strcmp(XATTR_NAME_HSM, name) == 0) {
929 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
934 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
939 * Compare current and future data of HSM EA and add a changelog if needed.
941 * Caller should have write-locked \param obj.
943 * \param buf - Future HSM EA content.
944 * \retval 0 if no changelog is needed or changelog was added properly.
945 * \retval -ve errno if there was a problem
947 static int mdd_hsm_update_locked(const struct lu_env *env,
948 struct md_object *obj,
949 const struct lu_buf *buf,
950 struct thandle *handle)
952 struct mdd_thread_info *info = mdd_env_info(env);
953 struct mdd_device *mdd = mdo2mdd(obj);
954 struct mdd_object *mdd_obj = md2mdd_obj(obj);
955 struct lu_buf *current_buf;
956 struct md_hsm *current_mh;
957 struct md_hsm *new_mh;
961 OBD_ALLOC_PTR(current_mh);
962 if (current_mh == NULL)
965 /* Read HSM attrs from disk */
966 CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
967 current_buf = mdd_buf_get(env, info->mti_xattr_buf,
968 sizeof(info->mti_xattr_buf));
969 rc = mdo_xattr_get(env, mdd_obj, current_buf, XATTR_NAME_HSM,
970 mdd_object_capa(env, mdd_obj));
971 rc = lustre_buf2hsm(current_buf->lb_buf, rc, current_mh);
972 if (rc < 0 && rc != -ENODATA)
974 else if (rc == -ENODATA)
975 current_mh->mh_flags = 0;
977 /* Map future HSM xattr */
978 OBD_ALLOC_PTR(new_mh);
980 GOTO(free, rc = -ENOMEM);
981 lustre_buf2hsm(buf->lb_buf, buf->lb_len, new_mh);
983 /* If HSM flags are different, add a changelog */
985 if (current_mh->mh_flags != new_mh->mh_flags) {
987 hsm_set_cl_event(&flags, HE_STATE);
988 if (new_mh->mh_flags & HS_DIRTY)
989 hsm_set_cl_flags(&flags, CLF_HSM_DIRTY);
991 rc = mdd_changelog_data_store(env, mdd, CL_HSM, flags, mdd_obj,
995 OBD_FREE_PTR(new_mh);
997 OBD_FREE_PTR(current_mh);
1002 * The caller should guarantee to update the object ctime
1003 * after xattr_set if needed.
1005 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
1006 const struct lu_buf *buf, const char *name,
1009 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1010 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1011 struct mdd_device *mdd = mdo2mdd(obj);
1012 struct thandle *handle;
1016 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
1020 rc = mdd_xattr_sanity_check(env, mdd_obj, attr);
1024 if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) {
1025 rc = mdd_acl_set(env, mdd_obj, attr, buf, fl);
1029 handle = mdd_trans_create(env, mdd);
1031 RETURN(PTR_ERR(handle));
1033 rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, fl, handle);
1037 rc = mdd_trans_start(env, mdd, handle);
1041 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1043 if (strcmp(XATTR_NAME_HSM, name) == 0) {
1044 rc = mdd_hsm_update_locked(env, obj, buf, handle);
1046 mdd_write_unlock(env, mdd_obj);
1051 rc = mdo_xattr_set(env, mdd_obj, buf, name, fl, handle,
1052 mdd_object_capa(env, mdd_obj));
1053 mdd_write_unlock(env, mdd_obj);
1057 if (strcmp(XATTR_NAME_LOV, name) == 0)
1058 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, mdd_obj,
1060 else if (strncmp(XATTR_USER_PREFIX, name,
1061 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1062 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1063 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1064 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1065 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1066 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1070 mdd_trans_stop(env, mdd, rc, handle);
1075 static int mdd_declare_xattr_del(const struct lu_env *env,
1076 struct mdd_device *mdd,
1077 struct mdd_object *obj,
1079 struct thandle *handle)
1083 rc = mdo_declare_xattr_del(env, obj, name, handle);
1087 /* Only record user xattr changes */
1088 if ((strncmp(XATTR_USER_PREFIX, name,
1089 sizeof(XATTR_USER_PREFIX) - 1) == 0))
1090 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1096 * The caller should guarantee to update the object ctime
1097 * after xattr_set if needed.
1099 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1102 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1103 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1104 struct mdd_device *mdd = mdo2mdd(obj);
1105 struct thandle *handle;
1109 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
1113 rc = mdd_xattr_sanity_check(env, mdd_obj, attr);
1117 handle = mdd_trans_create(env, mdd);
1119 RETURN(PTR_ERR(handle));
1121 rc = mdd_declare_xattr_del(env, mdd, mdd_obj, name, handle);
1125 rc = mdd_trans_start(env, mdd, handle);
1129 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1130 rc = mdo_xattr_del(env, mdd_obj, name, handle,
1131 mdd_object_capa(env, mdd_obj));
1132 mdd_write_unlock(env, mdd_obj);
1136 /* Only record system & user xattr changes */
1137 if (strncmp(XATTR_USER_PREFIX, name,
1138 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1139 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1140 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1141 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1142 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1143 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1147 mdd_trans_stop(env, mdd, rc, handle);
1153 * read lov EA of an object
1154 * return the lov EA in an allocated lu_buf
1156 int mdd_get_lov_ea(const struct lu_env *env, struct mdd_object *obj,
1157 struct lu_buf *lmm_buf)
1159 struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf;
1164 rc = mdo_xattr_get(env, obj, buf, XATTR_NAME_LOV,
1165 mdd_object_capa(env, obj));
1167 if (rc == -ERANGE) {
1168 /* mti_big_buf is allocated but is too small
1169 * we need to increase it */
1170 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1172 if (buf->lb_buf == NULL)
1173 GOTO(out, rc = -ENOMEM);
1184 if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) {
1185 /* mti_big_buf was not allocated, so we have to
1186 * allocate it based on the ea size */
1187 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1189 if (buf->lb_buf == NULL)
1190 GOTO(out, rc = -ENOMEM);
1194 lu_buf_alloc(lmm_buf, bufsize);
1195 if (lmm_buf->lb_buf == NULL)
1196 GOTO(out, rc = -ENOMEM);
1198 memcpy(lmm_buf->lb_buf, buf->lb_buf, bufsize);
1204 lu_buf_free(lmm_buf);
1208 static int mdd_xattr_hsm_replace(const struct lu_env *env,
1209 struct mdd_object *o, struct lu_buf *buf,
1210 struct thandle *handle)
1212 struct hsm_attrs *attrs;
1218 rc = mdo_xattr_set(env, o, buf, XATTR_NAME_HSM, LU_XATTR_REPLACE,
1219 handle, mdd_object_capa(env, o));
1223 attrs = buf->lb_buf;
1224 hsm_flags = le32_to_cpu(attrs->hsm_flags);
1225 if (!(hsm_flags & HS_RELEASED) || mdd_is_dead_obj(o))
1228 /* Add a changelog record for release. */
1229 hsm_set_cl_event(&flags, HE_RELEASE);
1230 rc = mdd_changelog_data_store(env, mdo2mdd(&o->mod_obj), CL_HSM,
1236 * check if layout swapping between 2 objects is allowed
1238 * - only normal FIDs or non-system IGIFs
1239 * - same type of objects
1240 * - same owner/group (so quotas are still valid)
1242 static int mdd_layout_swap_allowed(const struct lu_env *env,
1243 struct mdd_object *o1,
1244 const struct lu_attr *attr1,
1245 struct mdd_object *o2,
1246 const struct lu_attr *attr2)
1248 const struct lu_fid *fid1, *fid2;
1254 if (!fid_is_norm(fid1) &&
1255 (!fid_is_igif(fid1) || IS_ERR(mdd_links_get(env, o1))))
1258 if (!fid_is_norm(fid2) &&
1259 (!fid_is_igif(fid2) || IS_ERR(mdd_links_get(env, o2))))
1262 if (mdd_object_type(o1) != mdd_object_type(o2)) {
1263 if (S_ISDIR(mdd_object_type(o1)))
1265 if (S_ISREG(mdd_object_type(o1)))
1270 if ((attr1->la_uid != attr2->la_uid) ||
1271 (attr1->la_gid != attr2->la_gid))
1278 * swap layouts between 2 lustre objects
1280 static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1,
1281 struct md_object *obj2, __u64 flags)
1283 struct mdd_thread_info *info = mdd_env_info(env);
1284 struct mdd_object *fst_o = md2mdd_obj(obj1);
1285 struct mdd_object *snd_o = md2mdd_obj(obj2);
1286 struct lu_attr *fst_la = MDD_ENV_VAR(env, cattr);
1287 struct lu_attr *snd_la = MDD_ENV_VAR(env, tattr);
1288 struct mdd_device *mdd = mdo2mdd(obj1);
1289 struct lov_mds_md *fst_lmm, *snd_lmm;
1290 struct lu_buf *fst_buf = &info->mti_buf[0];
1291 struct lu_buf *snd_buf = &info->mti_buf[1];
1292 struct lu_buf *fst_hsm_buf = &info->mti_buf[2];
1293 struct lu_buf *snd_hsm_buf = &info->mti_buf[3];
1294 struct ost_id *saved_oi = NULL;
1295 struct thandle *handle;
1296 __u16 fst_gen, snd_gen;
1302 CLASSERT(ARRAY_SIZE(info->mti_buf) >= 4);
1303 memset(info->mti_buf, 0, sizeof(info->mti_buf));
1305 /* we have to sort the 2 obj, so locking will always
1306 * be in the same order, even in case of 2 concurrent swaps */
1307 rc = lu_fid_cmp(mdo2fid(fst_o), mdo2fid(snd_o));
1308 if (rc == 0) /* same fid ? */
1314 rc = mdd_la_get(env, fst_o, fst_la, BYPASS_CAPA);
1318 rc = mdd_la_get(env, snd_o, snd_la, BYPASS_CAPA);
1322 /* check if layout swapping is allowed */
1323 rc = mdd_layout_swap_allowed(env, fst_o, fst_la, snd_o, snd_la);
1327 handle = mdd_trans_create(env, mdd);
1329 RETURN(PTR_ERR(handle));
1331 /* objects are already sorted */
1332 mdd_write_lock(env, fst_o, MOR_TGT_CHILD);
1333 mdd_write_lock(env, snd_o, MOR_TGT_CHILD);
1335 rc = mdd_get_lov_ea(env, fst_o, fst_buf);
1336 if (rc < 0 && rc != -ENODATA)
1339 rc = mdd_get_lov_ea(env, snd_o, snd_buf);
1340 if (rc < 0 && rc != -ENODATA)
1343 /* swapping 2 non existant layouts is a success */
1344 if (fst_buf->lb_buf == NULL && snd_buf->lb_buf == NULL)
1347 /* to help inode migration between MDT, it is better to
1348 * start by the no layout file (if one), so we order the swap */
1349 if (snd_buf->lb_buf == NULL) {
1351 swap(fst_buf, snd_buf);
1354 /* lmm and generation layout initialization */
1355 if (fst_buf->lb_buf != NULL) {
1356 fst_lmm = fst_buf->lb_buf;
1357 fst_gen = le16_to_cpu(fst_lmm->lmm_layout_gen);
1358 fst_fl = LU_XATTR_REPLACE;
1362 fst_fl = LU_XATTR_CREATE;
1365 snd_lmm = snd_buf->lb_buf;
1366 snd_gen = le16_to_cpu(snd_lmm->lmm_layout_gen);
1368 /* increase the generation layout numbers */
1372 /* set the file specific informations in lmm */
1373 if (fst_lmm != NULL) {
1374 saved_oi = &info->mti_oa.o_oi;
1376 *saved_oi = fst_lmm->lmm_oi;
1377 fst_lmm->lmm_layout_gen = cpu_to_le16(snd_gen);
1378 fst_lmm->lmm_oi = snd_lmm->lmm_oi;
1379 snd_lmm->lmm_oi = *saved_oi;
1381 if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1))
1382 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEF);
1383 else if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))
1384 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V3_DEF);
1386 GOTO(stop, rc = -EPROTO);
1388 snd_lmm->lmm_layout_gen = cpu_to_le16(fst_gen);
1390 /* Prepare HSM attribute if it's required */
1391 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1392 const int buflen = sizeof(struct hsm_attrs);
1394 lu_buf_alloc(fst_hsm_buf, buflen);
1395 lu_buf_alloc(snd_hsm_buf, buflen);
1396 if (fst_hsm_buf->lb_buf == NULL || snd_hsm_buf->lb_buf == NULL)
1397 GOTO(stop, rc = -ENOMEM);
1399 /* Read HSM attribute */
1400 rc = mdo_xattr_get(env, fst_o, fst_hsm_buf, XATTR_NAME_HSM,
1405 rc = mdo_xattr_get(env, snd_o, snd_hsm_buf, XATTR_NAME_HSM,
1410 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_hsm_buf,
1411 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1416 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_hsm_buf,
1417 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1423 /* prepare transaction */
1424 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_buf, XATTR_NAME_LOV,
1429 if (fst_buf->lb_buf != NULL)
1430 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_buf,
1431 XATTR_NAME_LOV, LU_XATTR_REPLACE,
1434 rc = mdd_declare_xattr_del(env, mdd, snd_o, XATTR_NAME_LOV,
1439 rc = mdd_trans_start(env, mdd, handle);
1443 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1444 rc = mdd_xattr_hsm_replace(env, fst_o, snd_hsm_buf, handle);
1448 rc = mdd_xattr_hsm_replace(env, snd_o, fst_hsm_buf, handle);
1450 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf,
1453 CERROR("%s: restore "DFID" HSM error: %d/%d\n",
1454 mdd_obj_dev_name(fst_o),
1455 PFID(mdo2fid(fst_o)), rc, rc2);
1460 rc = mdo_xattr_set(env, fst_o, snd_buf, XATTR_NAME_LOV, fst_fl, handle,
1461 mdd_object_capa(env, fst_o));
1465 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_MDS_HSM_SWAP_LAYOUTS))) {
1468 if (fst_buf->lb_buf != NULL)
1469 rc = mdo_xattr_set(env, snd_o, fst_buf, XATTR_NAME_LOV,
1470 LU_XATTR_REPLACE, handle,
1471 mdd_object_capa(env, snd_o));
1473 rc = mdo_xattr_del(env, snd_o, XATTR_NAME_LOV, handle,
1474 mdd_object_capa(env, snd_o));
1480 /* failure on second file, but first was done, so we have
1481 * to roll back first. */
1482 if (fst_buf->lb_buf != NULL) {
1483 fst_lmm->lmm_oi = *saved_oi;
1484 fst_lmm->lmm_layout_gen = cpu_to_le16(fst_gen - 1);
1485 rc2 = mdo_xattr_set(env, fst_o, fst_buf, XATTR_NAME_LOV,
1486 LU_XATTR_REPLACE, handle,
1487 mdd_object_capa(env, fst_o));
1489 rc2 = mdo_xattr_del(env, fst_o, XATTR_NAME_LOV, handle,
1490 mdd_object_capa(env, fst_o));
1496 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, handle);
1501 rc2 = mdd_xattr_hsm_replace(env, snd_o, snd_hsm_buf, handle);
1506 CERROR("%s: unable to roll back layout swap. FIDs: "
1507 DFID" and "DFID "error: %d/%d, steps: %d\n",
1508 mdd_obj_dev_name(fst_o),
1509 PFID(mdo2fid(snd_o)), PFID(mdo2fid(fst_o)),
1511 /* a solution to avoid journal commit is to panic,
1512 * but it has strong consequences so we use LBUG to
1513 * allow sysdamin to choose to panic or not
1520 /* Issue one changelog record per file */
1521 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle);
1525 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle);
1531 mdd_trans_stop(env, mdd, rc, handle);
1532 mdd_write_unlock(env, snd_o);
1533 mdd_write_unlock(env, fst_o);
1535 lu_buf_free(fst_buf);
1536 lu_buf_free(snd_buf);
1537 lu_buf_free(fst_hsm_buf);
1538 lu_buf_free(snd_hsm_buf);
1542 void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent,
1543 struct mdd_object *child, const struct lu_attr *attr,
1544 const struct md_op_spec *spec,
1545 struct dt_allocation_hint *hint)
1547 struct dt_object *np = parent ? mdd_object_child(parent) : NULL;
1548 struct dt_object *nc = mdd_object_child(child);
1550 memset(hint, 0, sizeof(*hint));
1552 /* For striped directory, give striping EA to lod_ah_init, which will
1553 * decide the stripe_offset and stripe count by it. */
1554 if (S_ISDIR(attr->la_mode) &&
1555 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1556 hint->dah_eadata = spec->u.sp_ea.eadata;
1557 hint->dah_eadata_len = spec->u.sp_ea.eadatalen;
1559 hint->dah_eadata = NULL;
1560 hint->dah_eadata_len = 0;
1563 CDEBUG(D_INFO, DFID" eadata %p len %d\n", PFID(mdd_object_fid(child)),
1564 hint->dah_eadata, hint->dah_eadata_len);
1565 /* @hint will be initialized by underlying device. */
1566 nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT);
1570 * do NOT or the MAY_*'s, you'll get the weakest
1572 int accmode(const struct lu_env *env, const struct lu_attr *la, int flags)
1576 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1577 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1578 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1579 * owner can write to a file even if it is marked readonly to hide
1580 * its brokenness. (bug 5781) */
1581 if (flags & MDS_OPEN_OWNEROVERRIDE) {
1582 struct lu_ucred *uc = lu_ucred_check(env);
1584 if ((uc == NULL) || (la->la_uid == uc->uc_fsuid))
1588 if (flags & FMODE_READ)
1590 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1592 if (flags & MDS_FMODE_EXEC)
1597 static int mdd_open_sanity_check(const struct lu_env *env,
1598 struct mdd_object *obj,
1599 const struct lu_attr *attr, int flag)
1605 if (mdd_is_dead_obj(obj))
1608 if (S_ISLNK(attr->la_mode))
1611 mode = accmode(env, attr, flag);
1613 if (S_ISDIR(attr->la_mode) && (mode & MAY_WRITE))
1616 if (!(flag & MDS_OPEN_CREATED)) {
1617 rc = mdd_permission_internal(env, obj, attr, mode);
1622 if (S_ISFIFO(attr->la_mode) || S_ISSOCK(attr->la_mode) ||
1623 S_ISBLK(attr->la_mode) || S_ISCHR(attr->la_mode))
1624 flag &= ~MDS_OPEN_TRUNC;
1626 /* For writing append-only file must open it with append mode. */
1627 if (mdd_is_append(obj)) {
1628 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1630 if (flag & MDS_OPEN_TRUNC)
1636 * Now, flag -- O_NOATIME does not be packed by client.
1638 if (flag & O_NOATIME) {
1639 struct lu_ucred *uc = lu_ucred(env);
1641 if (uc && ((uc->uc_valid == UCRED_OLD) ||
1642 (uc->uc_valid == UCRED_NEW)) &&
1643 (uc->uc_fsuid != attr->la_uid) &&
1644 !md_capable(uc, CFS_CAP_FOWNER))
1652 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1655 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1656 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1659 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1661 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
1665 rc = mdd_open_sanity_check(env, mdd_obj, attr, flags);
1667 mdd_obj->mod_count++;
1669 mdd_write_unlock(env, mdd_obj);
1673 static int mdd_declare_close(const struct lu_env *env,
1674 struct mdd_object *obj,
1676 struct thandle *handle)
1680 rc = orph_declare_index_delete(env, obj, handle);
1684 return mdo_declare_destroy(env, obj, handle);
1688 * No permission check is needed.
1690 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1691 struct md_attr *ma, int mode)
1693 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1694 struct mdd_device *mdd = mdo2mdd(obj);
1695 struct thandle *handle = NULL;
1696 int rc, is_orphan = 0;
1699 if (ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_KEEP_ORPHAN) {
1700 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1701 mdd_obj->mod_count--;
1702 mdd_write_unlock(env, mdd_obj);
1704 if (mdd_obj->mod_flags & ORPHAN_OBJ && !mdd_obj->mod_count)
1705 CDEBUG(D_HA, "Object "DFID" is retained in orphan "
1706 "list\n", PFID(mdd_object_fid(mdd_obj)));
1710 /* mdd_finish_unlink() will always set orphan object as DEAD_OBJ, but
1711 * it might fail to add the object to orphan list (w/o ORPHAN_OBJ). */
1712 /* check without any lock */
1713 is_orphan = mdd_obj->mod_count == 1 &&
1714 (mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0;
1718 handle = mdd_trans_create(env, mdo2mdd(obj));
1720 RETURN(PTR_ERR(handle));
1722 rc = mdd_declare_close(env, mdd_obj, ma, handle);
1726 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1730 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1735 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1736 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr,
1737 mdd_object_capa(env, mdd_obj));
1739 CERROR("Failed to get lu_attr of "DFID": %d\n",
1740 PFID(mdd_object_fid(mdd_obj)), rc);
1744 /* check again with lock */
1745 is_orphan = (mdd_obj->mod_count == 1) &&
1746 ((mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0 ||
1747 ma->ma_attr.la_nlink == 0);
1749 if (is_orphan && handle == NULL) {
1750 mdd_write_unlock(env, mdd_obj);
1754 mdd_obj->mod_count--; /*release open count */
1760 /* NB: Object maybe not in orphan list originally, it is rare case for
1761 * mdd_finish_unlink() failure, in that case, the object doesn't have
1762 * ORPHAN_OBJ flag */
1763 if ((mdd_obj->mod_flags & ORPHAN_OBJ) != 0) {
1764 /* remove link to object from orphan index */
1765 LASSERT(handle != NULL);
1766 rc = __mdd_orphan_del(env, mdd_obj, handle);
1768 CERROR("%s: unable to delete "DFID" from orphan list: "
1769 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1770 PFID(mdd_object_fid(mdd_obj)), rc);
1771 /* If object was not deleted from orphan list, do not
1772 * destroy OSS objects, which will be done when next
1777 CDEBUG(D_HA, "Object "DFID" is deleted from orphan "
1778 "list, OSS objects to be destroyed.\n",
1779 PFID(mdd_object_fid(mdd_obj)));
1782 rc = mdo_destroy(env, mdd_obj, handle);
1785 CERROR("%s: unable to delete "DFID" from orphan list: "
1786 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1787 PFID(mdd_object_fid(mdd_obj)), rc);
1792 mdd_write_unlock(env, mdd_obj);
1795 (mode & (FMODE_WRITE | MDS_OPEN_APPEND | MDS_OPEN_TRUNC)) &&
1796 !(ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_RECOV_OPEN)) {
1797 if (handle == NULL) {
1798 handle = mdd_trans_create(env, mdo2mdd(obj));
1800 GOTO(stop, rc = PTR_ERR(handle));
1802 rc = mdd_declare_changelog_store(env, mdd, NULL,
1807 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1812 mdd_changelog_data_store(env, mdd, CL_CLOSE, mode,
1817 if (handle != NULL && !IS_ERR(handle))
1818 mdd_trans_stop(env, mdd, rc, handle);
1824 * Permission check is done when open,
1825 * no need check again.
1827 static int mdd_readpage_sanity_check(const struct lu_env *env,
1828 struct mdd_object *obj)
1830 struct dt_object *next = mdd_object_child(obj);
1834 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
1842 static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp,
1843 int nob, const struct dt_it_ops *iops,
1844 struct dt_it *it, __u32 attr, void *arg)
1846 struct lu_dirpage *dp = &lp->lp_dir;
1850 struct lu_dirent *ent;
1851 struct lu_dirent *last = NULL;
1854 memset(area, 0, sizeof (*dp));
1855 area += sizeof (*dp);
1856 nob -= sizeof (*dp);
1863 len = iops->key_size(env, it);
1865 /* IAM iterator can return record with zero len. */
1869 hash = iops->store(env, it);
1870 if (unlikely(first)) {
1872 dp->ldp_hash_start = cpu_to_le64(hash);
1875 /* calculate max space required for lu_dirent */
1876 recsize = lu_dirent_calc_size(len, attr);
1878 if (nob >= recsize) {
1879 result = iops->rec(env, it, (struct dt_rec *)ent, attr);
1880 if (result == -ESTALE)
1885 /* osd might not able to pack all attributes,
1886 * so recheck rec length */
1887 recsize = le16_to_cpu(ent->lde_reclen);
1889 result = (last != NULL) ? 0 :-EINVAL;
1893 ent = (void *)ent + recsize;
1897 result = iops->next(env, it);
1898 if (result == -ESTALE)
1900 } while (result == 0);
1903 dp->ldp_hash_end = cpu_to_le64(hash);
1905 if (last->lde_hash == dp->ldp_hash_end)
1906 dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE);
1907 last->lde_reclen = 0; /* end mark */
1910 /* end of directory */
1911 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1912 else if (result < 0)
1913 CWARN("build page failed: %d!\n", result);
1917 int mdd_readpage(const struct lu_env *env, struct md_object *obj,
1918 const struct lu_rdpg *rdpg)
1920 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1924 if (mdd_object_exists(mdd_obj) == 0) {
1925 CERROR("%s: object "DFID" not found: rc = -2\n",
1926 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
1930 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
1931 rc = mdd_readpage_sanity_check(env, mdd_obj);
1933 GOTO(out_unlock, rc);
1935 if (mdd_is_dead_obj(mdd_obj)) {
1937 struct lu_dirpage *dp;
1940 * According to POSIX, please do not return any entry to client:
1941 * even dot and dotdot should not be returned.
1943 CDEBUG(D_INODE, "readdir from dead object: "DFID"\n",
1944 PFID(mdd_object_fid(mdd_obj)));
1946 if (rdpg->rp_count <= 0)
1947 GOTO(out_unlock, rc = -EFAULT);
1948 LASSERT(rdpg->rp_pages != NULL);
1950 pg = rdpg->rp_pages[0];
1951 dp = (struct lu_dirpage *)kmap(pg);
1952 memset(dp, 0 , sizeof(struct lu_dirpage));
1953 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
1954 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1955 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
1957 GOTO(out_unlock, rc = LU_PAGE_SIZE);
1960 rc = dt_index_walk(env, mdd_object_child(mdd_obj), rdpg,
1961 mdd_dir_page_build, NULL);
1963 struct lu_dirpage *dp;
1965 dp = kmap(rdpg->rp_pages[0]);
1966 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
1969 * No pages were processed, mark this for first page
1972 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1973 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
1974 rc = min_t(unsigned int, LU_PAGE_SIZE, rdpg->rp_count);
1976 kunmap(rdpg->rp_pages[0]);
1979 GOTO(out_unlock, rc);
1981 mdd_read_unlock(env, mdd_obj);
1985 static int mdd_object_sync(const struct lu_env *env, struct md_object *obj)
1987 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1989 if (mdd_object_exists(mdd_obj) == 0) {
1992 CERROR("%s: object "DFID" not found: rc = %d\n",
1993 mdd_obj_dev_name(mdd_obj),
1994 PFID(mdd_object_fid(mdd_obj)), rc);
1997 return dt_object_sync(env, mdd_object_child(mdd_obj),
2001 static int mdd_object_lock(const struct lu_env *env,
2002 struct md_object *obj,
2003 struct lustre_handle *lh,
2004 struct ldlm_enqueue_info *einfo,
2005 ldlm_policy_data_t *policy)
2007 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2008 return dt_object_lock(env, mdd_object_child(mdd_obj), lh,
2012 static int mdd_object_unlock(const struct lu_env *env,
2013 struct md_object *obj,
2014 struct ldlm_enqueue_info *einfo,
2015 ldlm_policy_data_t *policy)
2017 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2018 return dt_object_unlock(env, mdd_object_child(mdd_obj), einfo, policy);
2021 const struct md_object_operations mdd_obj_ops = {
2022 .moo_permission = mdd_permission,
2023 .moo_attr_get = mdd_attr_get,
2024 .moo_attr_set = mdd_attr_set,
2025 .moo_xattr_get = mdd_xattr_get,
2026 .moo_xattr_set = mdd_xattr_set,
2027 .moo_xattr_list = mdd_xattr_list,
2028 .moo_xattr_del = mdd_xattr_del,
2029 .moo_swap_layouts = mdd_swap_layouts,
2030 .moo_open = mdd_open,
2031 .moo_close = mdd_close,
2032 .moo_readpage = mdd_readpage,
2033 .moo_readlink = mdd_readlink,
2034 .moo_changelog = mdd_changelog,
2035 .moo_capa_get = mdd_capa_get,
2036 .moo_object_sync = mdd_object_sync,
2037 .moo_object_lock = mdd_object_lock,
2038 .moo_object_unlock = mdd_object_unlock,