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>
53 #include <lustre_idmap.h>
54 #include <lustre_param.h>
55 #include <lustre_mds.h>
56 #include <lustre/lustre_idl.h>
58 #include "mdd_internal.h"
60 static const struct lu_object_operations mdd_lu_obj_ops;
61 extern struct kmem_cache *mdd_object_kmem;
63 static int mdd_xattr_get(const struct lu_env *env,
64 struct md_object *obj, struct lu_buf *buf,
67 int mdd_data_get(const struct lu_env *env, struct mdd_object *obj,
70 if (mdd_object_exists(obj) == 0) {
71 CERROR("%s: object "DFID" not found: rc = -2\n",
72 mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj)));
75 mdo_data_get(env, obj, data);
79 int mdd_la_get(const struct lu_env *env, struct mdd_object *obj,
80 struct lu_attr *la, struct lustre_capa *capa)
82 if (mdd_object_exists(obj) == 0) {
83 CERROR("%s: object "DFID" not found: rc = -2\n",
84 mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj)));
88 return mdo_attr_get(env, obj, la, capa);
91 static void mdd_flags_xlate(struct mdd_object *obj, __u32 flags)
93 obj->mod_flags &= ~(APPEND_OBJ|IMMUTE_OBJ);
95 if (flags & LUSTRE_APPEND_FL)
96 obj->mod_flags |= APPEND_OBJ;
98 if (flags & LUSTRE_IMMUTABLE_FL)
99 obj->mod_flags |= IMMUTE_OBJ;
102 struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
104 struct mdd_thread_info *info;
106 lu_env_refill((struct lu_env *)env);
107 info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
108 LASSERT(info != NULL);
112 const struct lu_name *mdd_name_get_const(const struct lu_env *env,
113 const void *area, ssize_t len)
115 struct lu_name *lname;
117 lname = &mdd_env_info(env)->mti_name;
118 lname->ln_name = area;
119 lname->ln_namelen = len;
123 struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
127 buf = &mdd_env_info(env)->mti_buf[0];
133 const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
134 const void *area, ssize_t len)
138 buf = &mdd_env_info(env)->mti_buf[0];
139 buf->lb_buf = (void *)area;
144 struct lu_object *mdd_object_alloc(const struct lu_env *env,
145 const struct lu_object_header *hdr,
148 struct mdd_object *mdd_obj;
150 OBD_SLAB_ALLOC_PTR_GFP(mdd_obj, mdd_object_kmem, GFP_NOFS);
151 if (mdd_obj != NULL) {
154 o = mdd2lu_obj(mdd_obj);
155 lu_object_init(o, NULL, d);
156 mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
157 mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
158 mdd_obj->mod_count = 0;
159 o->lo_ops = &mdd_lu_obj_ops;
166 static int mdd_object_init(const struct lu_env *env, struct lu_object *o,
167 const struct lu_object_conf *unused)
169 struct mdd_device *d = lu2mdd_dev(o->lo_dev);
170 struct mdd_object *mdd_obj = lu2mdd_obj(o);
171 struct lu_object *below;
172 struct lu_device *under;
175 mdd_obj->mod_cltime = 0;
176 under = &d->mdd_child->dd_lu_dev;
177 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
179 RETURN(PTR_ERR(below));
181 lu_object_add(o, below);
186 static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
190 if (lu_object_exists(o)) {
191 struct mdd_object *mdd_obj = lu2mdd_obj(o);
192 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
194 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
196 mdd_flags_xlate(mdd_obj, attr->la_flags);
202 static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
204 struct mdd_object *mdd = lu2mdd_obj(o);
207 OBD_SLAB_FREE_PTR(mdd, mdd_object_kmem);
210 static int mdd_object_print(const struct lu_env *env, void *cookie,
211 lu_printer_t p, const struct lu_object *o)
213 struct mdd_object *mdd = lu2mdd_obj((struct lu_object *)o);
214 return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p(open_count=%d, "
215 "valid=%x, cltime="LPU64", flags=%lx)",
216 mdd, mdd->mod_count, mdd->mod_valid,
217 mdd->mod_cltime, mdd->mod_flags);
220 static const struct lu_object_operations mdd_lu_obj_ops = {
221 .loo_object_init = mdd_object_init,
222 .loo_object_start = mdd_object_start,
223 .loo_object_free = mdd_object_free,
224 .loo_object_print = mdd_object_print,
227 struct mdd_object *mdd_object_find(const struct lu_env *env,
228 struct mdd_device *d,
229 const struct lu_fid *f)
231 return md2mdd_obj(md_object_find_slice(env, &d->mdd_md_dev, f));
235 * No permission check is needed.
237 int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
240 struct mdd_object *mdd_obj = md2mdd_obj(obj);
245 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr,
246 mdd_object_capa(env, md2mdd_obj(obj)));
247 if ((ma->ma_need & MA_INODE) != 0 && mdd_is_dead_obj(mdd_obj))
248 ma->ma_attr.la_nlink = 0;
254 * No permission check is needed.
256 static int mdd_xattr_get(const struct lu_env *env,
257 struct md_object *obj, struct lu_buf *buf,
260 struct mdd_object *mdd_obj = md2mdd_obj(obj);
265 if (mdd_object_exists(mdd_obj) == 0) {
266 CERROR("%s: object "DFID" not found: rc = -2\n",
267 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
271 /* If the object has been delete from the namespace, then
272 * get linkEA should return -ENOENT as well */
273 if (unlikely((mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ)) &&
274 strcmp(name, XATTR_NAME_LINK) == 0))
277 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
278 rc = mdo_xattr_get(env, mdd_obj, buf, name,
279 mdd_object_capa(env, mdd_obj));
280 mdd_read_unlock(env, mdd_obj);
286 * Permission check is done when open,
287 * no need check again.
289 static int mdd_readlink(const struct lu_env *env, struct md_object *obj,
292 struct mdd_object *mdd_obj = md2mdd_obj(obj);
293 struct dt_object *next;
298 if (mdd_object_exists(mdd_obj) == 0) {
299 CERROR("%s: object "DFID" not found: rc = -2\n",
300 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
304 next = mdd_object_child(mdd_obj);
305 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
306 rc = next->do_body_ops->dbo_read(env, next, buf, &pos,
307 mdd_object_capa(env, mdd_obj));
308 mdd_read_unlock(env, mdd_obj);
313 * No permission check is needed.
315 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
318 struct mdd_object *mdd_obj = md2mdd_obj(obj);
323 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
324 rc = mdo_xattr_list(env, mdd_obj, buf, mdd_object_capa(env, mdd_obj));
325 mdd_read_unlock(env, mdd_obj);
330 int mdd_declare_object_create_internal(const struct lu_env *env,
331 struct mdd_object *p,
332 struct mdd_object *c,
333 struct lu_attr *attr,
334 struct thandle *handle,
335 const struct md_op_spec *spec)
337 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
338 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
339 const struct dt_index_features *feat = spec->sp_feat;
343 if (feat != &dt_directory_features && feat != NULL) {
344 dof->dof_type = DFT_INDEX;
345 dof->u.dof_idx.di_feat = feat;
348 dof->dof_type = dt_mode_to_dft(attr->la_mode);
349 if (dof->dof_type == DFT_REGULAR) {
350 dof->u.dof_reg.striped =
351 md_should_create(spec->sp_cr_flags);
352 if (spec->sp_cr_flags & MDS_OPEN_HAS_EA)
353 dof->u.dof_reg.striped = 0;
354 /* is this replay? */
356 dof->u.dof_reg.striped = 0;
360 rc = mdo_declare_create_obj(env, c, attr, hint, dof, handle);
365 int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p,
366 struct mdd_object *c, struct lu_attr *attr,
367 struct thandle *handle,
368 const struct md_op_spec *spec)
370 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
371 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
375 LASSERT(!mdd_object_exists(c));
377 rc = mdo_create_obj(env, c, attr, hint, dof, handle);
379 LASSERT(ergo(rc == 0, mdd_object_exists(c)));
384 int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj,
385 const struct lu_attr *attr, struct thandle *handle,
391 rc = mdo_attr_set(env, obj, attr, handle, mdd_object_capa(env, obj));
392 #ifdef CONFIG_FS_POSIX_ACL
393 if (!rc && (attr->la_valid & LA_MODE) && needacl)
394 rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
399 int mdd_update_time(const struct lu_env *env, struct mdd_object *obj,
400 const struct lu_attr *oattr, struct lu_attr *attr,
401 struct thandle *handle)
406 LASSERT(attr->la_valid & LA_CTIME);
407 LASSERT(oattr != NULL);
409 /* Make sure the ctime is increased only, however, it's not strictly
410 * reliable at here because there is not guarantee to hold lock on
411 * object, so we just bypass some unnecessary cmtime setting first
412 * and OSD has to check it again. */
413 if (attr->la_ctime < oattr->la_ctime)
414 attr->la_valid &= ~(LA_MTIME | LA_CTIME);
415 else if (attr->la_valid == LA_CTIME &&
416 attr->la_ctime == oattr->la_ctime)
417 attr->la_valid &= ~LA_CTIME;
419 if (attr->la_valid != 0)
420 rc = mdd_attr_set_internal(env, obj, attr, handle, 0);
425 * This gives the same functionality as the code between
426 * sys_chmod and inode_setattr
427 * chown_common and inode_setattr
428 * utimes and inode_setattr
429 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
431 static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
432 const struct lu_attr *oattr, struct lu_attr *la,
433 const unsigned long flags)
442 /* Do not permit change file type */
443 if (la->la_valid & LA_TYPE)
446 /* They should not be processed by setattr */
447 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
450 LASSERT(oattr != NULL);
452 /* export destroy does not have ->le_ses, but we may want
453 * to drop LUSTRE_SOM_FL. */
454 uc = lu_ucred_check(env);
458 if (la->la_valid == LA_CTIME) {
459 if (!(flags & MDS_PERM_BYPASS))
460 /* This is only for set ctime when rename's source is
462 rc = mdd_may_delete(env, NULL, NULL, obj, oattr, NULL,
464 if (rc == 0 && la->la_ctime <= oattr->la_ctime)
465 la->la_valid &= ~LA_CTIME;
469 if (la->la_valid == LA_ATIME) {
470 /* This is atime only set for read atime update on close. */
471 if (la->la_atime >= oattr->la_atime &&
472 la->la_atime < (oattr->la_atime +
473 mdd_obj2mdd_dev(obj)->mdd_atime_diff))
474 la->la_valid &= ~LA_ATIME;
478 /* Check if flags change. */
479 if (la->la_valid & LA_FLAGS) {
480 unsigned int oldflags = 0;
481 unsigned int newflags = la->la_flags &
482 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
484 if ((uc->uc_fsuid != oattr->la_uid) &&
485 !md_capable(uc, CFS_CAP_FOWNER))
488 /* XXX: the IMMUTABLE and APPEND_ONLY flags can
489 * only be changed by the relevant capability. */
490 if (mdd_is_immutable(obj))
491 oldflags |= LUSTRE_IMMUTABLE_FL;
492 if (mdd_is_append(obj))
493 oldflags |= LUSTRE_APPEND_FL;
494 if ((oldflags ^ newflags) &&
495 !md_capable(uc, CFS_CAP_LINUX_IMMUTABLE))
498 if (!S_ISDIR(oattr->la_mode))
499 la->la_flags &= ~LUSTRE_DIRSYNC_FL;
502 if ((mdd_is_immutable(obj) || mdd_is_append(obj)) &&
503 (la->la_valid & ~LA_FLAGS) &&
504 !(flags & MDS_PERM_BYPASS))
507 /* Check for setting the obj time. */
508 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
509 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
510 if ((uc->uc_fsuid != oattr->la_uid) &&
511 !md_capable(uc, CFS_CAP_FOWNER)) {
512 rc = mdd_permission_internal(env, obj, oattr,
519 if (la->la_valid & LA_KILL_SUID) {
520 la->la_valid &= ~LA_KILL_SUID;
521 if ((oattr->la_mode & S_ISUID) &&
522 !(la->la_valid & LA_MODE)) {
523 la->la_mode = oattr->la_mode;
524 la->la_valid |= LA_MODE;
526 la->la_mode &= ~S_ISUID;
529 if (la->la_valid & LA_KILL_SGID) {
530 la->la_valid &= ~LA_KILL_SGID;
531 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
532 (S_ISGID | S_IXGRP)) &&
533 !(la->la_valid & LA_MODE)) {
534 la->la_mode = oattr->la_mode;
535 la->la_valid |= LA_MODE;
537 la->la_mode &= ~S_ISGID;
540 /* Make sure a caller can chmod. */
541 if (la->la_valid & LA_MODE) {
542 if (!(flags & MDS_PERM_BYPASS) &&
543 (uc->uc_fsuid != oattr->la_uid) &&
544 !md_capable(uc, CFS_CAP_FOWNER))
547 if (la->la_mode == (umode_t) -1)
548 la->la_mode = oattr->la_mode;
550 la->la_mode = (la->la_mode & S_IALLUGO) |
551 (oattr->la_mode & ~S_IALLUGO);
553 /* Also check the setgid bit! */
554 if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ?
555 la->la_gid : oattr->la_gid) &&
556 !md_capable(uc, CFS_CAP_FSETID))
557 la->la_mode &= ~S_ISGID;
559 la->la_mode = oattr->la_mode;
562 /* Make sure a caller can chown. */
563 if (la->la_valid & LA_UID) {
564 if (la->la_uid == (uid_t) -1)
565 la->la_uid = oattr->la_uid;
566 if (((uc->uc_fsuid != oattr->la_uid) ||
567 (la->la_uid != oattr->la_uid)) &&
568 !md_capable(uc, CFS_CAP_CHOWN))
571 /* If the user or group of a non-directory has been
572 * changed by a non-root user, remove the setuid bit.
573 * 19981026 David C Niemi <niemi@tux.org>
575 * Changed this to apply to all users, including root,
576 * to avoid some races. This is the behavior we had in
577 * 2.0. The check for non-root was definitely wrong
578 * for 2.2 anyway, as it should have been using
579 * CAP_FSETID rather than fsuid -- 19990830 SD. */
580 if (((oattr->la_mode & S_ISUID) == S_ISUID) &&
581 !S_ISDIR(oattr->la_mode)) {
582 la->la_mode &= ~S_ISUID;
583 la->la_valid |= LA_MODE;
587 /* Make sure caller can chgrp. */
588 if (la->la_valid & LA_GID) {
589 if (la->la_gid == (gid_t) -1)
590 la->la_gid = oattr->la_gid;
591 if (((uc->uc_fsuid != oattr->la_uid) ||
592 ((la->la_gid != oattr->la_gid) &&
593 !lustre_in_group_p(uc, la->la_gid))) &&
594 !md_capable(uc, CFS_CAP_CHOWN))
597 /* Likewise, if the user or group of a non-directory
598 * has been changed by a non-root user, remove the
599 * setgid bit UNLESS there is no group execute bit
600 * (this would be a file marked for mandatory
601 * locking). 19981026 David C Niemi <niemi@tux.org>
603 * Removed the fsuid check (see the comment above) --
605 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
606 (S_ISGID | S_IXGRP)) && !S_ISDIR(oattr->la_mode)) {
607 la->la_mode &= ~S_ISGID;
608 la->la_valid |= LA_MODE;
612 /* For both Size-on-MDS case and truncate case,
613 * "la->la_valid & (LA_SIZE | LA_BLOCKS)" are ture.
614 * We distinguish them by "flags & MDS_SOM".
615 * For SOM case, it is true, the MAY_WRITE perm has been checked
616 * when open, no need check again. For truncate case, it is false,
617 * the MAY_WRITE perm should be checked here. */
618 if (flags & MDS_SOM) {
619 /* For the "Size-on-MDS" setattr update, merge coming
620 * attributes with the set in the inode. BUG 10641 */
621 if ((la->la_valid & LA_ATIME) &&
622 (la->la_atime <= oattr->la_atime))
623 la->la_valid &= ~LA_ATIME;
625 /* OST attributes do not have a priority over MDS attributes,
626 * so drop times if ctime is equal. */
627 if ((la->la_valid & LA_CTIME) &&
628 (la->la_ctime <= oattr->la_ctime))
629 la->la_valid &= ~(LA_MTIME | LA_CTIME);
631 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
632 if (!((flags & MDS_OWNEROVERRIDE) &&
633 (uc->uc_fsuid == oattr->la_uid)) &&
634 !(flags & MDS_PERM_BYPASS)) {
635 rc = mdd_permission_internal(env, obj,
641 if (la->la_valid & LA_CTIME) {
642 /* The pure setattr, it has the priority over what is
643 * already set, do not drop it if ctime is equal. */
644 if (la->la_ctime < oattr->la_ctime)
645 la->la_valid &= ~(LA_ATIME | LA_MTIME |
653 /** Store a data change changelog record
654 * If this fails, we must fail the whole transaction; we don't
655 * want the change to commit without the log entry.
656 * \param mdd_obj - mdd_object of change
657 * \param handle - transacion handle
659 int mdd_changelog_data_store(const struct lu_env *env, struct mdd_device *mdd,
660 enum changelog_rec_type type, int flags,
661 struct mdd_object *mdd_obj, struct thandle *handle)
663 const struct lu_fid *tfid;
664 struct llog_changelog_rec *rec;
670 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
672 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
675 LASSERT(mdd_obj != NULL);
676 LASSERT(handle != NULL);
678 tfid = mdo2fid(mdd_obj);
680 if ((type >= CL_MTIME) && (type <= CL_ATIME) &&
681 cfs_time_before_64(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) {
682 /* Don't need multiple updates in this log */
683 /* Don't check under lock - no big deal if we get an extra
688 reclen = llog_data_len(sizeof(*rec));
689 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
690 if (buf->lb_buf == NULL)
694 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
695 rec->cr.cr_type = (__u32)type;
696 rec->cr.cr_tfid = *tfid;
697 rec->cr.cr_namelen = 0;
698 mdd_obj->mod_cltime = cfs_time_current_64();
700 rc = mdd_changelog_store(env, mdd, rec, handle);
705 int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type,
706 int flags, struct md_object *obj)
708 struct thandle *handle;
709 struct mdd_object *mdd_obj = md2mdd_obj(obj);
710 struct mdd_device *mdd = mdo2mdd(obj);
714 handle = mdd_trans_create(env, mdd);
716 RETURN(PTR_ERR(handle));
718 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
722 rc = mdd_trans_start(env, mdd, handle);
726 rc = mdd_changelog_data_store(env, mdd, type, flags, mdd_obj,
730 mdd_trans_stop(env, mdd, rc, handle);
736 * Save LMA extended attributes with data from \a ma.
738 * HSM and Size-On-MDS data will be extracted from \ma if they are valid, if
739 * not, LMA EA will be first read from disk, modified and write back.
742 /* Precedence for choosing record type when multiple
743 * attributes change: setattr > mtime > ctime > atime
744 * (ctime changes when mtime does, plus chmod/chown.
745 * atime and ctime are independent.) */
746 static int mdd_attr_set_changelog(const struct lu_env *env,
747 struct md_object *obj, struct thandle *handle,
750 struct mdd_device *mdd = mdo2mdd(obj);
753 bits = (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0;
754 bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0;
755 bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0;
756 bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0;
757 bits = bits & mdd->mdd_cl.mc_mask;
761 /* The record type is the lowest non-masked set bit */
762 while (bits && ((bits & 1) == 0)) {
767 /* FYI we only store the first CLF_FLAGMASK bits of la_valid */
768 return mdd_changelog_data_store(env, mdd, type, (int)valid,
769 md2mdd_obj(obj), handle);
772 static int mdd_declare_attr_set(const struct lu_env *env,
773 struct mdd_device *mdd,
774 struct mdd_object *obj,
775 const struct lu_attr *attr,
776 struct thandle *handle)
780 rc = mdo_declare_attr_set(env, obj, attr, handle);
784 #ifdef CONFIG_FS_POSIX_ACL
785 if (attr->la_valid & LA_MODE) {
786 mdd_read_lock(env, obj, MOR_TGT_CHILD);
787 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL,
788 XATTR_NAME_ACL_ACCESS, BYPASS_CAPA);
789 mdd_read_unlock(env, obj);
790 if (rc == -EOPNOTSUPP || rc == -ENODATA)
796 struct lu_buf *buf = mdd_buf_get(env, NULL, rc);
797 rc = mdo_declare_xattr_set(env, obj, buf,
798 XATTR_NAME_ACL_ACCESS, 0,
806 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
813 * permission changes may require sync operation, to mitigate performance
814 * impact, only do this for dir and when permission is reduced.
816 * For regular files, version is updated with permission change (see VBR), async
817 * permission won't cause any issue, while missing permission change on
818 * directory may affect accessibility of other objects after recovery.
820 static inline bool permission_needs_sync(const struct lu_attr *old,
821 const struct lu_attr *new)
823 if (!S_ISDIR(old->la_mode))
826 if (new->la_valid & (LA_UID | LA_GID))
829 if (new->la_valid & LA_MODE &&
830 new->la_mode & (S_ISUID | S_ISGID | S_ISVTX))
833 if ((new->la_valid & LA_MODE) &&
834 ((new->la_mode & old->la_mode) & S_IRWXUGO) !=
835 (old->la_mode & S_IRWXUGO))
841 /* set attr and LOV EA at once, return updated attr */
842 int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
843 const struct md_attr *ma)
845 struct mdd_object *mdd_obj = md2mdd_obj(obj);
846 struct mdd_device *mdd = mdo2mdd(obj);
847 struct thandle *handle;
848 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
849 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
850 const struct lu_attr *la = &ma->ma_attr;
854 /* we do not use ->attr_set() for LOV/SOM/HSM EA any more */
855 LASSERT((ma->ma_valid & MA_LOV) == 0);
856 LASSERT((ma->ma_valid & MA_HSM) == 0);
857 LASSERT((ma->ma_valid & MA_SOM) == 0);
859 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
863 *la_copy = ma->ma_attr;
864 rc = mdd_fix_attr(env, mdd_obj, attr, la_copy, ma->ma_attr_flags);
868 /* setattr on "close" only change atime, or do nothing */
869 if (la->la_valid == LA_ATIME && la_copy->la_valid == 0)
872 handle = mdd_trans_create(env, mdd);
874 RETURN(PTR_ERR(handle));
876 rc = mdd_declare_attr_set(env, mdd, mdd_obj, la, handle);
880 rc = mdd_trans_start(env, mdd, handle);
884 if (mdd->mdd_sync_permission && permission_needs_sync(attr, la))
887 if (la->la_valid & (LA_MTIME | LA_CTIME))
888 CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n",
889 la->la_mtime, la->la_ctime);
891 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
892 if (la_copy->la_valid & LA_FLAGS) {
893 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
895 mdd_flags_xlate(mdd_obj, la_copy->la_flags);
896 } else if (la_copy->la_valid) { /* setattr */
897 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
899 mdd_write_unlock(env, mdd_obj);
902 rc = mdd_attr_set_changelog(env, obj, handle, la->la_valid);
907 mdd_trans_stop(env, mdd, rc, handle);
911 static int mdd_xattr_sanity_check(const struct lu_env *env,
912 struct mdd_object *obj,
913 const struct lu_attr *attr)
915 struct lu_ucred *uc = lu_ucred_assert(env);
918 if (mdd_is_immutable(obj) || mdd_is_append(obj))
921 if ((uc->uc_fsuid != attr->la_uid) && !md_capable(uc, CFS_CAP_FOWNER))
927 static int mdd_declare_xattr_set(const struct lu_env *env,
928 struct mdd_device *mdd,
929 struct mdd_object *obj,
930 const struct lu_buf *buf,
932 int fl, struct thandle *handle)
936 rc = mdo_declare_xattr_set(env, obj, buf, name, fl, handle);
940 /* Only record user and layout xattr changes */
941 if (strncmp(XATTR_USER_PREFIX, name,
942 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
943 strcmp(XATTR_NAME_LOV, name) == 0) {
944 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
949 /* If HSM data is modified, this could add a changelog */
950 if (strcmp(XATTR_NAME_HSM, name) == 0) {
951 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
956 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
961 * Compare current and future data of HSM EA and add a changelog if needed.
963 * Caller should have write-locked \param obj.
965 * \param buf - Future HSM EA content.
966 * \retval 0 if no changelog is needed or changelog was added properly.
967 * \retval -ve errno if there was a problem
969 static int mdd_hsm_update_locked(const struct lu_env *env,
970 struct md_object *obj,
971 const struct lu_buf *buf,
972 struct thandle *handle)
974 struct mdd_thread_info *info = mdd_env_info(env);
975 struct mdd_device *mdd = mdo2mdd(obj);
976 struct mdd_object *mdd_obj = md2mdd_obj(obj);
977 struct lu_buf *current_buf;
978 struct md_hsm *current_mh;
979 struct md_hsm *new_mh;
983 OBD_ALLOC_PTR(current_mh);
984 if (current_mh == NULL)
987 /* Read HSM attrs from disk */
988 CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
989 current_buf = mdd_buf_get(env, info->mti_xattr_buf,
990 sizeof(info->mti_xattr_buf));
991 rc = mdo_xattr_get(env, mdd_obj, current_buf, XATTR_NAME_HSM,
992 mdd_object_capa(env, mdd_obj));
993 rc = lustre_buf2hsm(current_buf->lb_buf, rc, current_mh);
994 if (rc < 0 && rc != -ENODATA)
996 else if (rc == -ENODATA)
997 current_mh->mh_flags = 0;
999 /* Map future HSM xattr */
1000 OBD_ALLOC_PTR(new_mh);
1002 GOTO(free, rc = -ENOMEM);
1003 lustre_buf2hsm(buf->lb_buf, buf->lb_len, new_mh);
1005 /* If HSM flags are different, add a changelog */
1007 if (current_mh->mh_flags != new_mh->mh_flags) {
1009 hsm_set_cl_event(&flags, HE_STATE);
1010 if (new_mh->mh_flags & HS_DIRTY)
1011 hsm_set_cl_flags(&flags, CLF_HSM_DIRTY);
1013 rc = mdd_changelog_data_store(env, mdd, CL_HSM, flags, mdd_obj,
1017 OBD_FREE_PTR(new_mh);
1019 OBD_FREE_PTR(current_mh);
1024 * The caller should guarantee to update the object ctime
1025 * after xattr_set if needed.
1027 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
1028 const struct lu_buf *buf, const char *name,
1031 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1032 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1033 struct mdd_device *mdd = mdo2mdd(obj);
1034 struct thandle *handle;
1038 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
1042 if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) {
1043 rc = mdd_acl_set(env, mdd_obj, attr, buf, fl);
1047 rc = mdd_xattr_sanity_check(env, mdd_obj, attr);
1051 handle = mdd_trans_create(env, mdd);
1053 RETURN(PTR_ERR(handle));
1055 rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, fl, handle);
1059 rc = mdd_trans_start(env, mdd, handle);
1063 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1065 if (strcmp(XATTR_NAME_HSM, name) == 0) {
1066 rc = mdd_hsm_update_locked(env, obj, buf, handle);
1068 mdd_write_unlock(env, mdd_obj);
1073 rc = mdo_xattr_set(env, mdd_obj, buf, name, fl, handle,
1074 mdd_object_capa(env, mdd_obj));
1075 mdd_write_unlock(env, mdd_obj);
1079 if (strcmp(XATTR_NAME_LOV, name) == 0)
1080 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, mdd_obj,
1082 else if (strncmp(XATTR_USER_PREFIX, name,
1083 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1084 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1085 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1086 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1087 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1088 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1092 mdd_trans_stop(env, mdd, rc, handle);
1097 static int mdd_declare_xattr_del(const struct lu_env *env,
1098 struct mdd_device *mdd,
1099 struct mdd_object *obj,
1101 struct thandle *handle)
1105 rc = mdo_declare_xattr_del(env, obj, name, handle);
1109 /* Only record user xattr changes */
1110 if ((strncmp(XATTR_USER_PREFIX, name,
1111 sizeof(XATTR_USER_PREFIX) - 1) == 0))
1112 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1118 * The caller should guarantee to update the object ctime
1119 * after xattr_set if needed.
1121 int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1124 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1125 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1126 struct mdd_device *mdd = mdo2mdd(obj);
1127 struct thandle *handle;
1131 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
1135 rc = mdd_xattr_sanity_check(env, mdd_obj, attr);
1139 handle = mdd_trans_create(env, mdd);
1141 RETURN(PTR_ERR(handle));
1143 rc = mdd_declare_xattr_del(env, mdd, mdd_obj, name, handle);
1147 rc = mdd_trans_start(env, mdd, handle);
1151 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1152 rc = mdo_xattr_del(env, mdd_obj, name, handle,
1153 mdd_object_capa(env, mdd_obj));
1154 mdd_write_unlock(env, mdd_obj);
1158 /* Only record system & user xattr changes */
1159 if (strncmp(XATTR_USER_PREFIX, name,
1160 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1161 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1162 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1163 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1164 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1165 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1169 mdd_trans_stop(env, mdd, rc, handle);
1175 * read lov EA of an object
1176 * return the lov EA in an allocated lu_buf
1178 static int mdd_get_lov_ea(const struct lu_env *env,
1179 struct mdd_object *obj,
1180 struct lu_buf *lmm_buf)
1182 struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf;
1187 rc = mdo_xattr_get(env, obj, buf, XATTR_NAME_LOV,
1188 mdd_object_capa(env, obj));
1190 if (rc == -ERANGE) {
1191 /* mti_big_buf is allocated but is too small
1192 * we need to increase it */
1193 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1195 if (buf->lb_buf == NULL)
1196 GOTO(out, rc = -ENOMEM);
1204 GOTO(out, rc = -ENODATA);
1207 if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) {
1208 /* mti_big_buf was not allocated, so we have to
1209 * allocate it based on the ea size */
1210 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1212 if (buf->lb_buf == NULL)
1213 GOTO(out, rc = -ENOMEM);
1217 lu_buf_alloc(lmm_buf, sz);
1218 if (lmm_buf->lb_buf == NULL)
1219 GOTO(out, rc = -ENOMEM);
1221 memcpy(lmm_buf->lb_buf, buf->lb_buf, sz);
1227 lu_buf_free(lmm_buf);
1231 static int mdd_xattr_hsm_replace(const struct lu_env *env,
1232 struct mdd_object *o, struct lu_buf *buf,
1233 struct thandle *handle)
1235 struct hsm_attrs *attrs;
1241 rc = mdo_xattr_set(env, o, buf, XATTR_NAME_HSM, LU_XATTR_REPLACE,
1242 handle, mdd_object_capa(env, o));
1246 attrs = buf->lb_buf;
1247 hsm_flags = le32_to_cpu(attrs->hsm_flags);
1248 if (!(hsm_flags & HS_RELEASED) || mdd_is_dead_obj(o))
1251 /* Add a changelog record for release. */
1252 hsm_set_cl_event(&flags, HE_RELEASE);
1253 rc = mdd_changelog_data_store(env, mdo2mdd(&o->mod_obj), CL_HSM,
1259 * check if layout swapping between 2 objects is allowed
1261 * - only normal FIDs or non-system IGIFs
1262 * - same type of objects
1263 * - same owner/group (so quotas are still valid)
1265 static int mdd_layout_swap_allowed(const struct lu_env *env,
1266 struct mdd_object *o1,
1267 const struct lu_attr *attr1,
1268 struct mdd_object *o2,
1269 const struct lu_attr *attr2)
1271 const struct lu_fid *fid1, *fid2;
1277 if (!fid_is_norm(fid1) &&
1278 (!fid_is_igif(fid1) || IS_ERR(mdd_links_get(env, o1))))
1281 if (!fid_is_norm(fid2) &&
1282 (!fid_is_igif(fid2) || IS_ERR(mdd_links_get(env, o2))))
1285 if (mdd_object_type(o1) != mdd_object_type(o2)) {
1286 if (S_ISDIR(mdd_object_type(o1)))
1288 if (S_ISREG(mdd_object_type(o1)))
1293 if ((attr1->la_uid != attr2->la_uid) ||
1294 (attr1->la_gid != attr2->la_gid))
1301 * swap layouts between 2 lustre objects
1303 static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1,
1304 struct md_object *obj2, __u64 flags)
1306 struct mdd_thread_info *info = mdd_env_info(env);
1307 struct mdd_object *fst_o = md2mdd_obj(obj1);
1308 struct mdd_object *snd_o = md2mdd_obj(obj2);
1309 struct lu_attr *fst_la = MDD_ENV_VAR(env, cattr);
1310 struct lu_attr *snd_la = MDD_ENV_VAR(env, tattr);
1311 struct mdd_device *mdd = mdo2mdd(obj1);
1312 struct lov_mds_md *fst_lmm, *snd_lmm;
1313 struct lu_buf *fst_buf = &info->mti_buf[0];
1314 struct lu_buf *snd_buf = &info->mti_buf[1];
1315 struct lu_buf *fst_hsm_buf = &info->mti_buf[2];
1316 struct lu_buf *snd_hsm_buf = &info->mti_buf[3];
1317 struct ost_id *saved_oi = NULL;
1318 struct thandle *handle;
1319 __u16 fst_gen, snd_gen;
1325 CLASSERT(ARRAY_SIZE(info->mti_buf) >= 4);
1326 memset(info->mti_buf, 0, sizeof(info->mti_buf));
1328 /* we have to sort the 2 obj, so locking will always
1329 * be in the same order, even in case of 2 concurrent swaps */
1330 rc = lu_fid_cmp(mdo2fid(fst_o), mdo2fid(snd_o));
1331 if (rc == 0) /* same fid ? */
1337 rc = mdd_la_get(env, fst_o, fst_la, BYPASS_CAPA);
1341 rc = mdd_la_get(env, snd_o, snd_la, BYPASS_CAPA);
1345 /* check if layout swapping is allowed */
1346 rc = mdd_layout_swap_allowed(env, fst_o, fst_la, snd_o, snd_la);
1350 handle = mdd_trans_create(env, mdd);
1352 RETURN(PTR_ERR(handle));
1354 /* objects are already sorted */
1355 mdd_write_lock(env, fst_o, MOR_TGT_CHILD);
1356 mdd_write_lock(env, snd_o, MOR_TGT_CHILD);
1358 rc = mdd_get_lov_ea(env, fst_o, fst_buf);
1359 if (rc < 0 && rc != -ENODATA)
1362 rc = mdd_get_lov_ea(env, snd_o, snd_buf);
1363 if (rc < 0 && rc != -ENODATA)
1366 /* swapping 2 non existant layouts is a success */
1367 if (fst_buf->lb_buf == NULL && snd_buf->lb_buf == NULL)
1370 /* to help inode migration between MDT, it is better to
1371 * start by the no layout file (if one), so we order the swap */
1372 if (snd_buf->lb_buf == NULL) {
1374 swap(fst_buf, snd_buf);
1377 /* lmm and generation layout initialization */
1378 if (fst_buf->lb_buf != NULL) {
1379 fst_lmm = fst_buf->lb_buf;
1380 fst_gen = le16_to_cpu(fst_lmm->lmm_layout_gen);
1381 fst_fl = LU_XATTR_REPLACE;
1385 fst_fl = LU_XATTR_CREATE;
1388 snd_lmm = snd_buf->lb_buf;
1389 snd_gen = le16_to_cpu(snd_lmm->lmm_layout_gen);
1391 /* increase the generation layout numbers */
1395 /* set the file specific informations in lmm */
1396 if (fst_lmm != NULL) {
1397 saved_oi = &info->mti_oa.o_oi;
1399 *saved_oi = fst_lmm->lmm_oi;
1400 fst_lmm->lmm_layout_gen = cpu_to_le16(snd_gen);
1401 fst_lmm->lmm_oi = snd_lmm->lmm_oi;
1402 snd_lmm->lmm_oi = *saved_oi;
1404 if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1))
1405 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEF);
1406 else if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))
1407 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V3_DEF);
1409 GOTO(stop, rc = -EPROTO);
1411 snd_lmm->lmm_layout_gen = cpu_to_le16(fst_gen);
1413 /* Prepare HSM attribute if it's required */
1414 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1415 const int buflen = sizeof(struct hsm_attrs);
1417 lu_buf_alloc(fst_hsm_buf, buflen);
1418 lu_buf_alloc(snd_hsm_buf, buflen);
1419 if (fst_hsm_buf->lb_buf == NULL || snd_hsm_buf->lb_buf == NULL)
1420 GOTO(stop, rc = -ENOMEM);
1422 /* Read HSM attribute */
1423 rc = mdo_xattr_get(env, fst_o, fst_hsm_buf, XATTR_NAME_HSM,
1428 rc = mdo_xattr_get(env, snd_o, snd_hsm_buf, XATTR_NAME_HSM,
1433 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_hsm_buf,
1434 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1439 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_hsm_buf,
1440 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1446 /* prepare transaction */
1447 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_buf, XATTR_NAME_LOV,
1452 if (fst_buf->lb_buf != NULL)
1453 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_buf,
1454 XATTR_NAME_LOV, LU_XATTR_REPLACE,
1457 rc = mdd_declare_xattr_del(env, mdd, snd_o, XATTR_NAME_LOV,
1462 rc = mdd_trans_start(env, mdd, handle);
1466 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1467 rc = mdd_xattr_hsm_replace(env, fst_o, snd_hsm_buf, handle);
1471 rc = mdd_xattr_hsm_replace(env, snd_o, fst_hsm_buf, handle);
1473 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf,
1476 CERROR("%s: restore "DFID" HSM error: %d/%d\n",
1477 mdd_obj_dev_name(fst_o),
1478 PFID(mdo2fid(fst_o)), rc, rc2);
1483 rc = mdo_xattr_set(env, fst_o, snd_buf, XATTR_NAME_LOV, fst_fl, handle,
1484 mdd_object_capa(env, fst_o));
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,
1494 mdd_object_capa(env, snd_o));
1496 rc = mdo_xattr_del(env, snd_o, XATTR_NAME_LOV, handle,
1497 mdd_object_capa(env, snd_o));
1503 /* failure on second file, but first was done, so we have
1504 * to roll back first. */
1505 if (fst_buf->lb_buf != NULL) {
1506 fst_lmm->lmm_oi = *saved_oi;
1507 fst_lmm->lmm_layout_gen = cpu_to_le16(fst_gen - 1);
1508 rc2 = mdo_xattr_set(env, fst_o, fst_buf, XATTR_NAME_LOV,
1509 LU_XATTR_REPLACE, handle,
1510 mdd_object_capa(env, fst_o));
1512 rc2 = mdo_xattr_del(env, fst_o, XATTR_NAME_LOV, handle,
1513 mdd_object_capa(env, fst_o));
1519 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, handle);
1524 rc2 = mdd_xattr_hsm_replace(env, snd_o, snd_hsm_buf, handle);
1529 CERROR("%s: unable to roll back layout swap. FIDs: "
1530 DFID" and "DFID "error: %d/%d, steps: %d\n",
1531 mdd_obj_dev_name(fst_o),
1532 PFID(mdo2fid(snd_o)), PFID(mdo2fid(fst_o)),
1534 /* a solution to avoid journal commit is to panic,
1535 * but it has strong consequences so we use LBUG to
1536 * allow sysdamin to choose to panic or not
1543 /* Issue one changelog record per file */
1544 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle);
1548 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle);
1554 mdd_trans_stop(env, mdd, rc, handle);
1555 mdd_write_unlock(env, snd_o);
1556 mdd_write_unlock(env, fst_o);
1558 lu_buf_free(fst_buf);
1559 lu_buf_free(snd_buf);
1560 lu_buf_free(fst_hsm_buf);
1561 lu_buf_free(snd_hsm_buf);
1565 void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent,
1566 struct mdd_object *child, const struct lu_attr *attr,
1567 const struct md_op_spec *spec)
1569 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1570 struct dt_object *np = parent ? mdd_object_child(parent) : NULL;
1571 struct dt_object *nc = mdd_object_child(child);
1573 memset(hint, 0, sizeof(*hint));
1575 /* For striped directory, give striping EA to lod_ah_init, which will
1576 * decide the stripe_offset and stripe count by it. */
1577 if (S_ISDIR(attr->la_mode) &&
1578 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1579 hint->dah_eadata = spec->u.sp_ea.eadata;
1580 hint->dah_eadata_len = spec->u.sp_ea.eadatalen;
1582 hint->dah_eadata = NULL;
1583 hint->dah_eadata_len = 0;
1586 CDEBUG(D_INFO, DFID" eadata %p, len %d\n", PFID(mdd_object_fid(child)),
1587 hint->dah_eadata, hint->dah_eadata_len);
1588 /* @hint will be initialized by underlying device. */
1589 nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT);
1593 * do NOT or the MAY_*'s, you'll get the weakest
1595 int accmode(const struct lu_env *env, const struct lu_attr *la, int flags)
1599 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1600 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1601 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1602 * owner can write to a file even if it is marked readonly to hide
1603 * its brokenness. (bug 5781) */
1604 if (flags & MDS_OPEN_OWNEROVERRIDE) {
1605 struct lu_ucred *uc = lu_ucred_check(env);
1607 if ((uc == NULL) || (la->la_uid == uc->uc_fsuid))
1611 if (flags & FMODE_READ)
1613 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1615 if (flags & MDS_FMODE_EXEC)
1620 static int mdd_open_sanity_check(const struct lu_env *env,
1621 struct mdd_object *obj,
1622 const struct lu_attr *attr, int flag)
1628 if (mdd_is_dead_obj(obj))
1631 if (S_ISLNK(attr->la_mode))
1634 mode = accmode(env, attr, flag);
1636 if (S_ISDIR(attr->la_mode) && (mode & MAY_WRITE))
1639 if (!(flag & MDS_OPEN_CREATED)) {
1640 rc = mdd_permission_internal(env, obj, attr, mode);
1645 if (S_ISFIFO(attr->la_mode) || S_ISSOCK(attr->la_mode) ||
1646 S_ISBLK(attr->la_mode) || S_ISCHR(attr->la_mode))
1647 flag &= ~MDS_OPEN_TRUNC;
1649 /* For writing append-only file must open it with append mode. */
1650 if (mdd_is_append(obj)) {
1651 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1653 if (flag & MDS_OPEN_TRUNC)
1659 * Now, flag -- O_NOATIME does not be packed by client.
1661 if (flag & O_NOATIME) {
1662 struct lu_ucred *uc = lu_ucred(env);
1664 if (uc && ((uc->uc_valid == UCRED_OLD) ||
1665 (uc->uc_valid == UCRED_NEW)) &&
1666 (uc->uc_fsuid != attr->la_uid) &&
1667 !md_capable(uc, CFS_CAP_FOWNER))
1675 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1678 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1679 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1682 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1684 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
1688 rc = mdd_open_sanity_check(env, mdd_obj, attr, flags);
1690 mdd_obj->mod_count++;
1692 mdd_write_unlock(env, mdd_obj);
1696 int mdd_declare_object_kill(const struct lu_env *env, struct mdd_object *obj,
1697 struct md_attr *ma, struct thandle *handle)
1699 return mdo_declare_destroy(env, obj, handle);
1702 /* return md_attr back,
1703 * if it is last unlink then return lov ea + llog cookie*/
1704 int mdd_object_kill(const struct lu_env *env, struct mdd_object *obj,
1705 struct md_attr *ma, struct thandle *handle)
1710 rc = mdo_destroy(env, obj, handle);
1715 static int mdd_declare_close(const struct lu_env *env,
1716 struct mdd_object *obj,
1718 struct thandle *handle)
1722 rc = orph_declare_index_delete(env, obj, handle);
1726 return mdo_declare_destroy(env, obj, handle);
1730 * No permission check is needed.
1732 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1733 struct md_attr *ma, int mode)
1735 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1736 struct mdd_device *mdd = mdo2mdd(obj);
1737 struct thandle *handle = NULL;
1738 int rc, is_orphan = 0;
1741 if (ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_KEEP_ORPHAN) {
1742 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1743 mdd_obj->mod_count--;
1744 mdd_write_unlock(env, mdd_obj);
1746 if (mdd_obj->mod_flags & ORPHAN_OBJ && !mdd_obj->mod_count)
1747 CDEBUG(D_HA, "Object "DFID" is retained in orphan "
1748 "list\n", PFID(mdd_object_fid(mdd_obj)));
1752 /* mdd_finish_unlink() will always set orphan object as DEAD_OBJ, but
1753 * it might fail to add the object to orphan list (w/o ORPHAN_OBJ). */
1754 /* check without any lock */
1755 is_orphan = mdd_obj->mod_count == 1 &&
1756 (mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0;
1760 handle = mdd_trans_create(env, mdo2mdd(obj));
1762 RETURN(PTR_ERR(handle));
1764 rc = mdd_declare_close(env, mdd_obj, ma, handle);
1768 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1772 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1777 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1778 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr,
1779 mdd_object_capa(env, mdd_obj));
1781 CERROR("Failed to get lu_attr of "DFID": %d\n",
1782 PFID(mdd_object_fid(mdd_obj)), rc);
1786 /* check again with lock */
1787 is_orphan = (mdd_obj->mod_count == 1) &&
1788 ((mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0 ||
1789 ma->ma_attr.la_nlink == 0);
1791 if (is_orphan && handle == NULL) {
1792 mdd_write_unlock(env, mdd_obj);
1796 mdd_obj->mod_count--; /*release open count */
1802 /* NB: Object maybe not in orphan list originally, it is rare case for
1803 * mdd_finish_unlink() failure, in that case, the object doesn't have
1804 * ORPHAN_OBJ flag */
1805 if ((mdd_obj->mod_flags & ORPHAN_OBJ) != 0) {
1806 /* remove link to object from orphan index */
1807 LASSERT(handle != NULL);
1808 rc = __mdd_orphan_del(env, mdd_obj, handle);
1810 CERROR("%s: unable to delete "DFID" from orphan list: "
1811 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1812 PFID(mdd_object_fid(mdd_obj)), rc);
1813 /* If object was not deleted from orphan list, do not
1814 * destroy OSS objects, which will be done when next
1819 CDEBUG(D_HA, "Object "DFID" is deleted from orphan "
1820 "list, OSS objects to be destroyed.\n",
1821 PFID(mdd_object_fid(mdd_obj)));
1824 rc = mdo_destroy(env, mdd_obj, handle);
1827 CERROR("%s: unable to delete "DFID" from orphan list: "
1828 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1829 PFID(mdd_object_fid(mdd_obj)), rc);
1834 mdd_write_unlock(env, mdd_obj);
1837 (mode & (FMODE_WRITE | MDS_OPEN_APPEND | MDS_OPEN_TRUNC)) &&
1838 !(ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_RECOV_OPEN)) {
1839 if (handle == NULL) {
1840 handle = mdd_trans_create(env, mdo2mdd(obj));
1842 GOTO(stop, rc = PTR_ERR(handle));
1844 rc = mdd_declare_changelog_store(env, mdd, NULL,
1849 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1854 mdd_changelog_data_store(env, mdd, CL_CLOSE, mode,
1859 if (handle != NULL && !IS_ERR(handle))
1860 mdd_trans_stop(env, mdd, rc, handle);
1866 * Permission check is done when open,
1867 * no need check again.
1869 static int mdd_readpage_sanity_check(const struct lu_env *env,
1870 struct mdd_object *obj)
1872 struct dt_object *next = mdd_object_child(obj);
1876 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
1884 static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp,
1885 int nob, const struct dt_it_ops *iops,
1886 struct dt_it *it, __u32 attr, void *arg)
1888 struct lu_dirpage *dp = &lp->lp_dir;
1892 struct lu_dirent *ent;
1893 struct lu_dirent *last = NULL;
1896 memset(area, 0, sizeof (*dp));
1897 area += sizeof (*dp);
1898 nob -= sizeof (*dp);
1905 len = iops->key_size(env, it);
1907 /* IAM iterator can return record with zero len. */
1911 hash = iops->store(env, it);
1912 if (unlikely(first)) {
1914 dp->ldp_hash_start = cpu_to_le64(hash);
1917 /* calculate max space required for lu_dirent */
1918 recsize = lu_dirent_calc_size(len, attr);
1920 if (nob >= recsize) {
1921 result = iops->rec(env, it, (struct dt_rec *)ent, attr);
1922 if (result == -ESTALE)
1927 /* osd might not able to pack all attributes,
1928 * so recheck rec length */
1929 recsize = le16_to_cpu(ent->lde_reclen);
1931 result = (last != NULL) ? 0 :-EINVAL;
1935 ent = (void *)ent + recsize;
1939 result = iops->next(env, it);
1940 if (result == -ESTALE)
1942 } while (result == 0);
1945 dp->ldp_hash_end = cpu_to_le64(hash);
1947 if (last->lde_hash == dp->ldp_hash_end)
1948 dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE);
1949 last->lde_reclen = 0; /* end mark */
1952 /* end of directory */
1953 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1954 else if (result < 0)
1955 CWARN("build page failed: %d!\n", result);
1959 int mdd_readpage(const struct lu_env *env, struct md_object *obj,
1960 const struct lu_rdpg *rdpg)
1962 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1966 if (mdd_object_exists(mdd_obj) == 0) {
1967 CERROR("%s: object "DFID" not found: rc = -2\n",
1968 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
1972 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
1973 rc = mdd_readpage_sanity_check(env, mdd_obj);
1975 GOTO(out_unlock, rc);
1977 if (mdd_is_dead_obj(mdd_obj)) {
1979 struct lu_dirpage *dp;
1982 * According to POSIX, please do not return any entry to client:
1983 * even dot and dotdot should not be returned.
1985 CDEBUG(D_INODE, "readdir from dead object: "DFID"\n",
1986 PFID(mdd_object_fid(mdd_obj)));
1988 if (rdpg->rp_count <= 0)
1989 GOTO(out_unlock, rc = -EFAULT);
1990 LASSERT(rdpg->rp_pages != NULL);
1992 pg = rdpg->rp_pages[0];
1993 dp = (struct lu_dirpage *)kmap(pg);
1994 memset(dp, 0 , sizeof(struct lu_dirpage));
1995 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
1996 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1997 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
1999 GOTO(out_unlock, rc = LU_PAGE_SIZE);
2002 rc = dt_index_walk(env, mdd_object_child(mdd_obj), rdpg,
2003 mdd_dir_page_build, NULL);
2005 struct lu_dirpage *dp;
2007 dp = kmap(rdpg->rp_pages[0]);
2008 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
2011 * No pages were processed, mark this for first page
2014 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
2015 rc = min_t(unsigned int, LU_PAGE_SIZE, rdpg->rp_count);
2017 kunmap(rdpg->rp_pages[0]);
2020 GOTO(out_unlock, rc);
2022 mdd_read_unlock(env, mdd_obj);
2026 static int mdd_object_sync(const struct lu_env *env, struct md_object *obj)
2028 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2030 if (mdd_object_exists(mdd_obj) == 0) {
2031 CERROR("%s: object "DFID" not found: rc = -2\n",
2032 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
2035 return dt_object_sync(env, mdd_object_child(mdd_obj));
2038 static int mdd_object_lock(const struct lu_env *env,
2039 struct md_object *obj,
2040 struct lustre_handle *lh,
2041 struct ldlm_enqueue_info *einfo,
2042 ldlm_policy_data_t *policy)
2044 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2045 return dt_object_lock(env, mdd_object_child(mdd_obj), lh,
2049 static int mdd_object_unlock(const struct lu_env *env,
2050 struct md_object *obj,
2051 struct ldlm_enqueue_info *einfo,
2052 ldlm_policy_data_t *policy)
2054 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2055 return dt_object_unlock(env, mdd_object_child(mdd_obj), einfo, policy);
2058 const struct md_object_operations mdd_obj_ops = {
2059 .moo_permission = mdd_permission,
2060 .moo_attr_get = mdd_attr_get,
2061 .moo_attr_set = mdd_attr_set,
2062 .moo_xattr_get = mdd_xattr_get,
2063 .moo_xattr_set = mdd_xattr_set,
2064 .moo_xattr_list = mdd_xattr_list,
2065 .moo_xattr_del = mdd_xattr_del,
2066 .moo_swap_layouts = mdd_swap_layouts,
2067 .moo_open = mdd_open,
2068 .moo_close = mdd_close,
2069 .moo_readpage = mdd_readpage,
2070 .moo_readlink = mdd_readlink,
2071 .moo_changelog = mdd_changelog,
2072 .moo_capa_get = mdd_capa_get,
2073 .moo_object_sync = mdd_object_sync,
2074 .moo_object_lock = mdd_object_lock,
2075 .moo_object_unlock = mdd_object_unlock,