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_ucred *uc = lu_ucred(env);
642 const struct lu_fid *tfid;
643 struct llog_changelog_rec *rec;
649 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
651 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
654 LASSERT(mdd_obj != NULL);
655 LASSERT(handle != NULL);
657 tfid = mdo2fid(mdd_obj);
659 if ((type >= CL_MTIME) && (type <= CL_ATIME) &&
660 cfs_time_before_64(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) {
661 /* Don't need multiple updates in this log */
662 /* Don't check under lock - no big deal if we get an extra
667 flags = (flags & CLF_FLAGMASK) | CLF_VERSION;
668 if (uc->uc_jobid[0] != '\0')
671 reclen = llog_data_len(changelog_rec_offset(flags & CLF_SUPPORTED));
672 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
673 if (buf->lb_buf == NULL)
677 rec->cr.cr_flags = flags;
678 rec->cr.cr_type = (__u32)type;
679 rec->cr.cr_tfid = *tfid;
680 rec->cr.cr_namelen = 0;
681 mdd_obj->mod_cltime = cfs_time_current_64();
683 if (flags & CLF_JOBID)
684 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
686 rc = mdd_changelog_store(env, mdd, rec, handle);
691 static int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type,
692 int flags, struct md_object *obj)
694 struct thandle *handle;
695 struct mdd_object *mdd_obj = md2mdd_obj(obj);
696 struct mdd_device *mdd = mdo2mdd(obj);
700 handle = mdd_trans_create(env, mdd);
702 RETURN(PTR_ERR(handle));
704 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
708 rc = mdd_trans_start(env, mdd, handle);
712 rc = mdd_changelog_data_store(env, mdd, type, flags, mdd_obj,
716 mdd_trans_stop(env, mdd, rc, handle);
722 * Save LMA extended attributes with data from \a ma.
724 * HSM and Size-On-MDS data will be extracted from \ma if they are valid, if
725 * not, LMA EA will be first read from disk, modified and write back.
728 /* Precedence for choosing record type when multiple
729 * attributes change: setattr > mtime > ctime > atime
730 * (ctime changes when mtime does, plus chmod/chown.
731 * atime and ctime are independent.) */
732 static int mdd_attr_set_changelog(const struct lu_env *env,
733 struct md_object *obj, struct thandle *handle,
736 struct mdd_device *mdd = mdo2mdd(obj);
739 bits = (valid & LA_SIZE) ? 1 << CL_TRUNC : 0;
740 bits |= (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0;
741 bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0;
742 bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0;
743 bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0;
744 bits = bits & mdd->mdd_cl.mc_mask;
745 /* This is an implementation limit rather than a protocol limit */
746 CLASSERT(CL_LAST <= sizeof(int) * 8);
750 /* The record type is the lowest non-masked set bit */
753 /* FYI we only store the first CLF_FLAGMASK bits of la_valid */
754 return mdd_changelog_data_store(env, mdd, type, (int)valid,
755 md2mdd_obj(obj), handle);
758 static int mdd_declare_attr_set(const struct lu_env *env,
759 struct mdd_device *mdd,
760 struct mdd_object *obj,
761 const struct lu_attr *attr,
762 struct thandle *handle)
766 rc = mdo_declare_attr_set(env, obj, attr, handle);
770 #ifdef CONFIG_FS_POSIX_ACL
771 if (attr->la_valid & LA_MODE) {
772 mdd_read_lock(env, obj, MOR_TGT_CHILD);
773 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL,
774 XATTR_NAME_ACL_ACCESS, BYPASS_CAPA);
775 mdd_read_unlock(env, obj);
776 if (rc == -EOPNOTSUPP || rc == -ENODATA)
782 struct lu_buf *buf = mdd_buf_get(env, NULL, rc);
783 rc = mdo_declare_xattr_set(env, obj, buf,
784 XATTR_NAME_ACL_ACCESS, 0,
792 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
799 * permission changes may require sync operation, to mitigate performance
800 * impact, only do this for dir and when permission is reduced.
802 * For regular files, version is updated with permission change (see VBR), async
803 * permission won't cause any issue, while missing permission change on
804 * directory may affect accessibility of other objects after recovery.
806 static inline bool permission_needs_sync(const struct lu_attr *old,
807 const struct lu_attr *new)
809 if (!S_ISDIR(old->la_mode))
812 if (new->la_valid & (LA_UID | LA_GID))
815 if (new->la_valid & LA_MODE &&
816 new->la_mode & (S_ISUID | S_ISGID | S_ISVTX))
819 if ((new->la_valid & LA_MODE) &&
820 ((new->la_mode & old->la_mode) & S_IRWXUGO) !=
821 (old->la_mode & S_IRWXUGO))
827 /* set attr and LOV EA at once, return updated attr */
828 int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
829 const struct md_attr *ma)
831 struct mdd_object *mdd_obj = md2mdd_obj(obj);
832 struct mdd_device *mdd = mdo2mdd(obj);
833 struct thandle *handle;
834 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
835 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
836 const struct lu_attr *la = &ma->ma_attr;
840 /* we do not use ->attr_set() for LOV/SOM/HSM EA any more */
841 LASSERT((ma->ma_valid & MA_LOV) == 0);
842 LASSERT((ma->ma_valid & MA_HSM) == 0);
843 LASSERT((ma->ma_valid & MA_SOM) == 0);
845 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
849 *la_copy = ma->ma_attr;
850 rc = mdd_fix_attr(env, mdd_obj, attr, la_copy, ma->ma_attr_flags);
854 /* setattr on "close" only change atime, or do nothing */
855 if (la->la_valid == LA_ATIME && la_copy->la_valid == 0)
858 handle = mdd_trans_create(env, mdd);
860 RETURN(PTR_ERR(handle));
862 rc = mdd_declare_attr_set(env, mdd, mdd_obj, la, handle);
866 rc = mdd_trans_start(env, mdd, handle);
870 if (mdd->mdd_sync_permission && permission_needs_sync(attr, la))
873 if (la->la_valid & (LA_MTIME | LA_CTIME))
874 CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n",
875 la->la_mtime, la->la_ctime);
877 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
878 if (la_copy->la_valid & LA_FLAGS) {
879 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
881 mdd_flags_xlate(mdd_obj, la_copy->la_flags);
882 } else if (la_copy->la_valid) { /* setattr */
883 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
885 mdd_write_unlock(env, mdd_obj);
888 rc = mdd_attr_set_changelog(env, obj, handle, la->la_valid);
893 mdd_trans_stop(env, mdd, rc, handle);
897 static int mdd_xattr_sanity_check(const struct lu_env *env,
898 struct mdd_object *obj,
899 const struct lu_attr *attr)
901 struct lu_ucred *uc = lu_ucred_assert(env);
904 if (mdd_is_immutable(obj) || mdd_is_append(obj))
907 if ((uc->uc_fsuid != attr->la_uid) && !md_capable(uc, CFS_CAP_FOWNER))
913 static int mdd_declare_xattr_set(const struct lu_env *env,
914 struct mdd_device *mdd,
915 struct mdd_object *obj,
916 const struct lu_buf *buf,
918 int fl, struct thandle *handle)
922 rc = mdo_declare_xattr_set(env, obj, buf, name, fl, handle);
926 /* Only record user and layout xattr changes */
927 if (strncmp(XATTR_USER_PREFIX, name,
928 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
929 strcmp(XATTR_NAME_LOV, name) == 0) {
930 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
935 /* If HSM data is modified, this could add a changelog */
936 if (strcmp(XATTR_NAME_HSM, name) == 0) {
937 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
942 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
947 * Compare current and future data of HSM EA and add a changelog if needed.
949 * Caller should have write-locked \param obj.
951 * \param buf - Future HSM EA content.
952 * \retval 0 if no changelog is needed or changelog was added properly.
953 * \retval -ve errno if there was a problem
955 static int mdd_hsm_update_locked(const struct lu_env *env,
956 struct md_object *obj,
957 const struct lu_buf *buf,
958 struct thandle *handle)
960 struct mdd_thread_info *info = mdd_env_info(env);
961 struct mdd_device *mdd = mdo2mdd(obj);
962 struct mdd_object *mdd_obj = md2mdd_obj(obj);
963 struct lu_buf *current_buf;
964 struct md_hsm *current_mh;
965 struct md_hsm *new_mh;
969 OBD_ALLOC_PTR(current_mh);
970 if (current_mh == NULL)
973 /* Read HSM attrs from disk */
974 CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
975 current_buf = mdd_buf_get(env, info->mti_xattr_buf,
976 sizeof(info->mti_xattr_buf));
977 rc = mdo_xattr_get(env, mdd_obj, current_buf, XATTR_NAME_HSM,
978 mdd_object_capa(env, mdd_obj));
979 rc = lustre_buf2hsm(current_buf->lb_buf, rc, current_mh);
980 if (rc < 0 && rc != -ENODATA)
982 else if (rc == -ENODATA)
983 current_mh->mh_flags = 0;
985 /* Map future HSM xattr */
986 OBD_ALLOC_PTR(new_mh);
988 GOTO(free, rc = -ENOMEM);
989 lustre_buf2hsm(buf->lb_buf, buf->lb_len, new_mh);
991 /* If HSM flags are different, add a changelog */
993 if (current_mh->mh_flags != new_mh->mh_flags) {
995 hsm_set_cl_event(&flags, HE_STATE);
996 if (new_mh->mh_flags & HS_DIRTY)
997 hsm_set_cl_flags(&flags, CLF_HSM_DIRTY);
999 rc = mdd_changelog_data_store(env, mdd, CL_HSM, flags, mdd_obj,
1003 OBD_FREE_PTR(new_mh);
1005 OBD_FREE_PTR(current_mh);
1009 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1013 * The caller should guarantee to update the object ctime
1014 * after xattr_set if needed.
1016 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
1017 const struct lu_buf *buf, const char *name,
1020 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1021 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1022 struct mdd_device *mdd = mdo2mdd(obj);
1023 struct thandle *handle;
1027 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
1031 rc = mdd_xattr_sanity_check(env, mdd_obj, attr);
1035 if (strcmp(name, XATTR_NAME_ACL_ACCESS) == 0 ||
1036 strcmp(name, XATTR_NAME_ACL_DEFAULT) == 0) {
1037 struct posix_acl *acl;
1039 /* user may set empty ACL, which should be treated as removing
1041 acl = posix_acl_from_xattr(&init_user_ns, buf->lb_buf,
1044 rc = mdd_xattr_del(env, obj, name);
1047 posix_acl_release(acl);
1050 if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) {
1051 rc = mdd_acl_set(env, mdd_obj, attr, buf, fl);
1055 handle = mdd_trans_create(env, mdd);
1057 RETURN(PTR_ERR(handle));
1059 rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, fl, handle);
1063 rc = mdd_trans_start(env, mdd, handle);
1067 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1069 if (strcmp(XATTR_NAME_HSM, name) == 0) {
1070 rc = mdd_hsm_update_locked(env, obj, buf, handle);
1072 mdd_write_unlock(env, mdd_obj);
1077 rc = mdo_xattr_set(env, mdd_obj, buf, name, fl, handle,
1078 mdd_object_capa(env, mdd_obj));
1079 mdd_write_unlock(env, mdd_obj);
1083 if (strcmp(XATTR_NAME_LOV, name) == 0)
1084 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, mdd_obj,
1086 else if (strncmp(XATTR_USER_PREFIX, name,
1087 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1088 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1089 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1090 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1091 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1092 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1096 mdd_trans_stop(env, mdd, rc, handle);
1101 static int mdd_declare_xattr_del(const struct lu_env *env,
1102 struct mdd_device *mdd,
1103 struct mdd_object *obj,
1105 struct thandle *handle)
1109 rc = mdo_declare_xattr_del(env, obj, name, handle);
1113 /* Only record user xattr changes */
1114 if ((strncmp(XATTR_USER_PREFIX, name,
1115 sizeof(XATTR_USER_PREFIX) - 1) == 0))
1116 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1122 * The caller should guarantee to update the object ctime
1123 * after xattr_set if needed.
1125 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1128 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1129 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1130 struct mdd_device *mdd = mdo2mdd(obj);
1131 struct thandle *handle;
1135 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
1139 rc = mdd_xattr_sanity_check(env, mdd_obj, attr);
1143 handle = mdd_trans_create(env, mdd);
1145 RETURN(PTR_ERR(handle));
1147 rc = mdd_declare_xattr_del(env, mdd, mdd_obj, name, handle);
1151 rc = mdd_trans_start(env, mdd, handle);
1155 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1156 rc = mdo_xattr_del(env, mdd_obj, name, handle,
1157 mdd_object_capa(env, mdd_obj));
1158 mdd_write_unlock(env, mdd_obj);
1162 /* Only record system & user xattr changes */
1163 if (strncmp(XATTR_USER_PREFIX, name,
1164 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1165 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1166 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1167 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1168 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1169 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1173 mdd_trans_stop(env, mdd, rc, handle);
1179 * read lov EA of an object
1180 * return the lov EA in an allocated lu_buf
1182 int mdd_get_lov_ea(const struct lu_env *env, struct mdd_object *obj,
1183 struct lu_buf *lmm_buf)
1185 struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf;
1190 rc = mdo_xattr_get(env, obj, buf, XATTR_NAME_LOV,
1191 mdd_object_capa(env, obj));
1193 if (rc == -ERANGE) {
1194 /* mti_big_buf is allocated but is too small
1195 * we need to increase it */
1196 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1198 if (buf->lb_buf == NULL)
1199 GOTO(out, rc = -ENOMEM);
1210 if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) {
1211 /* mti_big_buf was not allocated, so we have to
1212 * allocate it based on the ea size */
1213 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1215 if (buf->lb_buf == NULL)
1216 GOTO(out, rc = -ENOMEM);
1220 lu_buf_alloc(lmm_buf, bufsize);
1221 if (lmm_buf->lb_buf == NULL)
1222 GOTO(out, rc = -ENOMEM);
1224 memcpy(lmm_buf->lb_buf, buf->lb_buf, bufsize);
1230 lu_buf_free(lmm_buf);
1234 static int mdd_xattr_hsm_replace(const struct lu_env *env,
1235 struct mdd_object *o, struct lu_buf *buf,
1236 struct thandle *handle)
1238 struct hsm_attrs *attrs;
1244 rc = mdo_xattr_set(env, o, buf, XATTR_NAME_HSM, LU_XATTR_REPLACE,
1245 handle, mdd_object_capa(env, o));
1249 attrs = buf->lb_buf;
1250 hsm_flags = le32_to_cpu(attrs->hsm_flags);
1251 if (!(hsm_flags & HS_RELEASED) || mdd_is_dead_obj(o))
1254 /* Add a changelog record for release. */
1255 hsm_set_cl_event(&flags, HE_RELEASE);
1256 rc = mdd_changelog_data_store(env, mdo2mdd(&o->mod_obj), CL_HSM,
1262 * check if layout swapping between 2 objects is allowed
1264 * - only normal FIDs or non-system IGIFs
1265 * - same type of objects
1266 * - same owner/group (so quotas are still valid)
1268 static int mdd_layout_swap_allowed(const struct lu_env *env,
1269 struct mdd_object *o1,
1270 const struct lu_attr *attr1,
1271 struct mdd_object *o2,
1272 const struct lu_attr *attr2)
1274 const struct lu_fid *fid1, *fid2;
1280 if (!fid_is_norm(fid1) &&
1281 (!fid_is_igif(fid1) || IS_ERR(mdd_links_get(env, o1))))
1284 if (!fid_is_norm(fid2) &&
1285 (!fid_is_igif(fid2) || IS_ERR(mdd_links_get(env, o2))))
1288 if (mdd_object_type(o1) != mdd_object_type(o2)) {
1289 if (S_ISDIR(mdd_object_type(o1)))
1291 if (S_ISREG(mdd_object_type(o1)))
1296 if ((attr1->la_uid != attr2->la_uid) ||
1297 (attr1->la_gid != attr2->la_gid))
1304 * swap layouts between 2 lustre objects
1306 static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1,
1307 struct md_object *obj2, __u64 flags)
1309 struct mdd_thread_info *info = mdd_env_info(env);
1310 struct mdd_object *fst_o = md2mdd_obj(obj1);
1311 struct mdd_object *snd_o = md2mdd_obj(obj2);
1312 struct lu_attr *fst_la = MDD_ENV_VAR(env, cattr);
1313 struct lu_attr *snd_la = MDD_ENV_VAR(env, tattr);
1314 struct mdd_device *mdd = mdo2mdd(obj1);
1315 struct lov_mds_md *fst_lmm, *snd_lmm;
1316 struct lu_buf *fst_buf = &info->mti_buf[0];
1317 struct lu_buf *snd_buf = &info->mti_buf[1];
1318 struct lu_buf *fst_hsm_buf = &info->mti_buf[2];
1319 struct lu_buf *snd_hsm_buf = &info->mti_buf[3];
1320 struct ost_id *saved_oi = NULL;
1321 struct thandle *handle;
1322 __u16 fst_gen, snd_gen;
1328 CLASSERT(ARRAY_SIZE(info->mti_buf) >= 4);
1329 memset(info->mti_buf, 0, sizeof(info->mti_buf));
1331 /* we have to sort the 2 obj, so locking will always
1332 * be in the same order, even in case of 2 concurrent swaps */
1333 rc = lu_fid_cmp(mdo2fid(fst_o), mdo2fid(snd_o));
1334 if (rc == 0) /* same fid ? */
1340 rc = mdd_la_get(env, fst_o, fst_la, BYPASS_CAPA);
1344 rc = mdd_la_get(env, snd_o, snd_la, BYPASS_CAPA);
1348 /* check if layout swapping is allowed */
1349 rc = mdd_layout_swap_allowed(env, fst_o, fst_la, snd_o, snd_la);
1353 handle = mdd_trans_create(env, mdd);
1355 RETURN(PTR_ERR(handle));
1357 /* objects are already sorted */
1358 mdd_write_lock(env, fst_o, MOR_TGT_CHILD);
1359 mdd_write_lock(env, snd_o, MOR_TGT_CHILD);
1361 rc = mdd_get_lov_ea(env, fst_o, fst_buf);
1362 if (rc < 0 && rc != -ENODATA)
1365 rc = mdd_get_lov_ea(env, snd_o, snd_buf);
1366 if (rc < 0 && rc != -ENODATA)
1369 /* swapping 2 non existant layouts is a success */
1370 if (fst_buf->lb_buf == NULL && snd_buf->lb_buf == NULL)
1373 /* to help inode migration between MDT, it is better to
1374 * start by the no layout file (if one), so we order the swap */
1375 if (snd_buf->lb_buf == NULL) {
1377 swap(fst_buf, snd_buf);
1380 /* lmm and generation layout initialization */
1381 if (fst_buf->lb_buf != NULL) {
1382 fst_lmm = fst_buf->lb_buf;
1383 fst_gen = le16_to_cpu(fst_lmm->lmm_layout_gen);
1384 fst_fl = LU_XATTR_REPLACE;
1388 fst_fl = LU_XATTR_CREATE;
1391 snd_lmm = snd_buf->lb_buf;
1392 snd_gen = le16_to_cpu(snd_lmm->lmm_layout_gen);
1394 /* increase the generation layout numbers */
1398 /* set the file specific informations in lmm */
1399 if (fst_lmm != NULL) {
1400 saved_oi = &info->mti_oa.o_oi;
1402 *saved_oi = fst_lmm->lmm_oi;
1403 fst_lmm->lmm_layout_gen = cpu_to_le16(snd_gen);
1404 fst_lmm->lmm_oi = snd_lmm->lmm_oi;
1405 snd_lmm->lmm_oi = *saved_oi;
1407 if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1))
1408 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEF);
1409 else if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))
1410 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V3_DEF);
1412 GOTO(stop, rc = -EPROTO);
1414 snd_lmm->lmm_layout_gen = cpu_to_le16(fst_gen);
1416 /* Prepare HSM attribute if it's required */
1417 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1418 const int buflen = sizeof(struct hsm_attrs);
1420 lu_buf_alloc(fst_hsm_buf, buflen);
1421 lu_buf_alloc(snd_hsm_buf, buflen);
1422 if (fst_hsm_buf->lb_buf == NULL || snd_hsm_buf->lb_buf == NULL)
1423 GOTO(stop, rc = -ENOMEM);
1425 /* Read HSM attribute */
1426 rc = mdo_xattr_get(env, fst_o, fst_hsm_buf, XATTR_NAME_HSM,
1431 rc = mdo_xattr_get(env, snd_o, snd_hsm_buf, XATTR_NAME_HSM,
1436 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_hsm_buf,
1437 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1442 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_hsm_buf,
1443 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1449 /* prepare transaction */
1450 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_buf, XATTR_NAME_LOV,
1455 if (fst_buf->lb_buf != NULL)
1456 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_buf,
1457 XATTR_NAME_LOV, LU_XATTR_REPLACE,
1460 rc = mdd_declare_xattr_del(env, mdd, snd_o, XATTR_NAME_LOV,
1465 rc = mdd_trans_start(env, mdd, handle);
1469 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1470 rc = mdd_xattr_hsm_replace(env, fst_o, snd_hsm_buf, handle);
1474 rc = mdd_xattr_hsm_replace(env, snd_o, fst_hsm_buf, handle);
1476 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf,
1479 CERROR("%s: restore "DFID" HSM error: %d/%d\n",
1480 mdd_obj_dev_name(fst_o),
1481 PFID(mdo2fid(fst_o)), rc, rc2);
1486 rc = mdo_xattr_set(env, fst_o, snd_buf, XATTR_NAME_LOV, fst_fl, handle,
1487 mdd_object_capa(env, fst_o));
1491 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_MDS_HSM_SWAP_LAYOUTS))) {
1494 if (fst_buf->lb_buf != NULL)
1495 rc = mdo_xattr_set(env, snd_o, fst_buf, XATTR_NAME_LOV,
1496 LU_XATTR_REPLACE, handle,
1497 mdd_object_capa(env, snd_o));
1499 rc = mdo_xattr_del(env, snd_o, XATTR_NAME_LOV, handle,
1500 mdd_object_capa(env, snd_o));
1506 /* failure on second file, but first was done, so we have
1507 * to roll back first. */
1508 if (fst_buf->lb_buf != NULL) {
1509 fst_lmm->lmm_oi = *saved_oi;
1510 fst_lmm->lmm_layout_gen = cpu_to_le16(fst_gen - 1);
1511 rc2 = mdo_xattr_set(env, fst_o, fst_buf, XATTR_NAME_LOV,
1512 LU_XATTR_REPLACE, handle,
1513 mdd_object_capa(env, fst_o));
1515 rc2 = mdo_xattr_del(env, fst_o, XATTR_NAME_LOV, handle,
1516 mdd_object_capa(env, fst_o));
1522 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, handle);
1527 rc2 = mdd_xattr_hsm_replace(env, snd_o, snd_hsm_buf, handle);
1532 CERROR("%s: unable to roll back layout swap. FIDs: "
1533 DFID" and "DFID "error: %d/%d, steps: %d\n",
1534 mdd_obj_dev_name(fst_o),
1535 PFID(mdo2fid(snd_o)), PFID(mdo2fid(fst_o)),
1537 /* a solution to avoid journal commit is to panic,
1538 * but it has strong consequences so we use LBUG to
1539 * allow sysdamin to choose to panic or not
1546 /* Issue one changelog record per file */
1547 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle);
1551 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle);
1557 mdd_trans_stop(env, mdd, rc, handle);
1558 mdd_write_unlock(env, snd_o);
1559 mdd_write_unlock(env, fst_o);
1561 lu_buf_free(fst_buf);
1562 lu_buf_free(snd_buf);
1563 lu_buf_free(fst_hsm_buf);
1564 lu_buf_free(snd_hsm_buf);
1568 void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent,
1569 struct mdd_object *child, const struct lu_attr *attr,
1570 const struct md_op_spec *spec,
1571 struct dt_allocation_hint *hint)
1573 struct dt_object *np = parent ? mdd_object_child(parent) : NULL;
1574 struct dt_object *nc = mdd_object_child(child);
1576 memset(hint, 0, sizeof(*hint));
1578 /* For striped directory, give striping EA to lod_ah_init, which will
1579 * decide the stripe_offset and stripe count by it. */
1580 if (S_ISDIR(attr->la_mode) &&
1581 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1582 hint->dah_eadata = spec->u.sp_ea.eadata;
1583 hint->dah_eadata_len = spec->u.sp_ea.eadatalen;
1585 hint->dah_eadata = NULL;
1586 hint->dah_eadata_len = 0;
1589 CDEBUG(D_INFO, DFID" eadata %p len %d\n", PFID(mdd_object_fid(child)),
1590 hint->dah_eadata, hint->dah_eadata_len);
1591 /* @hint will be initialized by underlying device. */
1592 nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT);
1596 * do NOT or the MAY_*'s, you'll get the weakest
1598 int accmode(const struct lu_env *env, const struct lu_attr *la, int flags)
1602 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1603 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1604 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1605 * owner can write to a file even if it is marked readonly to hide
1606 * its brokenness. (bug 5781) */
1607 if (flags & MDS_OPEN_OWNEROVERRIDE) {
1608 struct lu_ucred *uc = lu_ucred_check(env);
1610 if ((uc == NULL) || (la->la_uid == uc->uc_fsuid))
1614 if (flags & FMODE_READ)
1616 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1618 if (flags & MDS_FMODE_EXEC)
1623 static int mdd_open_sanity_check(const struct lu_env *env,
1624 struct mdd_object *obj,
1625 const struct lu_attr *attr, int flag)
1631 if (mdd_is_dead_obj(obj))
1634 if (S_ISLNK(attr->la_mode))
1637 mode = accmode(env, attr, flag);
1639 if (S_ISDIR(attr->la_mode) && (mode & MAY_WRITE))
1642 if (!(flag & MDS_OPEN_CREATED)) {
1643 rc = mdd_permission_internal(env, obj, attr, mode);
1648 if (S_ISFIFO(attr->la_mode) || S_ISSOCK(attr->la_mode) ||
1649 S_ISBLK(attr->la_mode) || S_ISCHR(attr->la_mode))
1650 flag &= ~MDS_OPEN_TRUNC;
1652 /* For writing append-only file must open it with append mode. */
1653 if (mdd_is_append(obj)) {
1654 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1656 if (flag & MDS_OPEN_TRUNC)
1662 * Now, flag -- O_NOATIME does not be packed by client.
1664 if (flag & O_NOATIME) {
1665 struct lu_ucred *uc = lu_ucred(env);
1667 if (uc && ((uc->uc_valid == UCRED_OLD) ||
1668 (uc->uc_valid == UCRED_NEW)) &&
1669 (uc->uc_fsuid != attr->la_uid) &&
1670 !md_capable(uc, CFS_CAP_FOWNER))
1678 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1681 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1682 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1685 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1687 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
1691 rc = mdd_open_sanity_check(env, mdd_obj, attr, flags);
1695 mdd_obj->mod_count++;
1698 mdd_write_unlock(env, mdd_obj);
1702 static int mdd_declare_close(const struct lu_env *env,
1703 struct mdd_object *obj,
1705 struct thandle *handle)
1709 rc = orph_declare_index_delete(env, obj, handle);
1713 return mdo_declare_destroy(env, obj, handle);
1717 * No permission check is needed.
1719 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1720 struct md_attr *ma, int mode)
1722 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1723 struct mdd_device *mdd = mdo2mdd(obj);
1724 struct thandle *handle = NULL;
1725 int rc, is_orphan = 0;
1728 if (ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_KEEP_ORPHAN) {
1729 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1730 mdd_obj->mod_count--;
1731 mdd_write_unlock(env, mdd_obj);
1733 if (mdd_obj->mod_flags & ORPHAN_OBJ && !mdd_obj->mod_count)
1734 CDEBUG(D_HA, "Object "DFID" is retained in orphan "
1735 "list\n", PFID(mdd_object_fid(mdd_obj)));
1739 /* mdd_finish_unlink() will always set orphan object as DEAD_OBJ, but
1740 * it might fail to add the object to orphan list (w/o ORPHAN_OBJ). */
1741 /* check without any lock */
1742 is_orphan = mdd_obj->mod_count == 1 &&
1743 (mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0;
1747 handle = mdd_trans_create(env, mdo2mdd(obj));
1749 RETURN(PTR_ERR(handle));
1751 rc = mdd_declare_close(env, mdd_obj, ma, handle);
1755 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1759 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1764 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1765 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr,
1766 mdd_object_capa(env, mdd_obj));
1768 CERROR("Failed to get lu_attr of "DFID": %d\n",
1769 PFID(mdd_object_fid(mdd_obj)), rc);
1773 /* check again with lock */
1774 is_orphan = (mdd_obj->mod_count == 1) &&
1775 ((mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0 ||
1776 ma->ma_attr.la_nlink == 0);
1778 if (is_orphan && handle == NULL) {
1779 mdd_write_unlock(env, mdd_obj);
1783 mdd_obj->mod_count--; /*release open count */
1789 /* NB: Object maybe not in orphan list originally, it is rare case for
1790 * mdd_finish_unlink() failure, in that case, the object doesn't have
1791 * ORPHAN_OBJ flag */
1792 if ((mdd_obj->mod_flags & ORPHAN_OBJ) != 0) {
1793 /* remove link to object from orphan index */
1794 LASSERT(handle != NULL);
1795 rc = __mdd_orphan_del(env, mdd_obj, handle);
1797 CERROR("%s: unable to delete "DFID" from orphan list: "
1798 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1799 PFID(mdd_object_fid(mdd_obj)), rc);
1800 /* If object was not deleted from orphan list, do not
1801 * destroy OSS objects, which will be done when next
1806 CDEBUG(D_HA, "Object "DFID" is deleted from orphan "
1807 "list, OSS objects to be destroyed.\n",
1808 PFID(mdd_object_fid(mdd_obj)));
1811 rc = mdo_destroy(env, mdd_obj, handle);
1814 CERROR("%s: unable to delete "DFID" from orphan list: "
1815 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1816 PFID(mdd_object_fid(mdd_obj)), rc);
1821 mdd_write_unlock(env, mdd_obj);
1824 (mode & (FMODE_WRITE | MDS_OPEN_APPEND | MDS_OPEN_TRUNC)) &&
1825 !(ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_RECOV_OPEN)) {
1826 if (handle == NULL) {
1827 handle = mdd_trans_create(env, mdo2mdd(obj));
1829 GOTO(stop, rc = PTR_ERR(handle));
1831 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL,
1836 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1841 mdd_changelog_data_store(env, mdd, CL_CLOSE, mode,
1846 if (handle != NULL && !IS_ERR(handle))
1847 mdd_trans_stop(env, mdd, rc, handle);
1853 * Permission check is done when open,
1854 * no need check again.
1856 static int mdd_readpage_sanity_check(const struct lu_env *env,
1857 struct mdd_object *obj)
1859 struct dt_object *next = mdd_object_child(obj);
1863 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
1871 static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp,
1872 int nob, const struct dt_it_ops *iops,
1873 struct dt_it *it, __u32 attr, void *arg)
1875 struct lu_dirpage *dp = &lp->lp_dir;
1879 struct lu_dirent *ent;
1880 struct lu_dirent *last = NULL;
1884 memset(area, 0, sizeof (*dp));
1885 area += sizeof (*dp);
1886 nob -= sizeof (*dp);
1893 len = iops->key_size(env, it);
1895 /* IAM iterator can return record with zero len. */
1899 hash = iops->store(env, it);
1900 if (unlikely(first)) {
1902 dp->ldp_hash_start = cpu_to_le64(hash);
1905 /* calculate max space required for lu_dirent */
1906 recsize = lu_dirent_calc_size(len, attr);
1908 if (nob >= recsize) {
1909 result = iops->rec(env, it, (struct dt_rec *)ent, attr);
1910 if (result == -ESTALE)
1915 /* osd might not able to pack all attributes,
1916 * so recheck rec length */
1917 recsize = le16_to_cpu(ent->lde_reclen);
1919 if (le32_to_cpu(ent->lde_attrs) & LUDA_FID) {
1920 fid_le_to_cpu(&fid, &ent->lde_fid);
1921 if (fid_is_dot_lustre(&fid))
1925 result = (last != NULL) ? 0 :-EINVAL;
1929 ent = (void *)ent + recsize;
1933 result = iops->next(env, it);
1934 if (result == -ESTALE)
1936 } while (result == 0);
1939 dp->ldp_hash_end = cpu_to_le64(hash);
1941 if (last->lde_hash == dp->ldp_hash_end)
1942 dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE);
1943 last->lde_reclen = 0; /* end mark */
1946 /* end of directory */
1947 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1948 else if (result < 0)
1949 CWARN("build page failed: %d!\n", result);
1953 int mdd_readpage(const struct lu_env *env, struct md_object *obj,
1954 const struct lu_rdpg *rdpg)
1956 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1960 if (mdd_object_exists(mdd_obj) == 0) {
1961 CERROR("%s: object "DFID" not found: rc = -2\n",
1962 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
1966 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
1967 rc = mdd_readpage_sanity_check(env, mdd_obj);
1969 GOTO(out_unlock, rc);
1971 if (mdd_is_dead_obj(mdd_obj)) {
1973 struct lu_dirpage *dp;
1976 * According to POSIX, please do not return any entry to client:
1977 * even dot and dotdot should not be returned.
1979 CDEBUG(D_INODE, "readdir from dead object: "DFID"\n",
1980 PFID(mdd_object_fid(mdd_obj)));
1982 if (rdpg->rp_count <= 0)
1983 GOTO(out_unlock, rc = -EFAULT);
1984 LASSERT(rdpg->rp_pages != NULL);
1986 pg = rdpg->rp_pages[0];
1987 dp = (struct lu_dirpage *)kmap(pg);
1988 memset(dp, 0 , sizeof(struct lu_dirpage));
1989 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
1990 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1991 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
1993 GOTO(out_unlock, rc = LU_PAGE_SIZE);
1996 rc = dt_index_walk(env, mdd_object_child(mdd_obj), rdpg,
1997 mdd_dir_page_build, NULL);
1999 struct lu_dirpage *dp;
2001 dp = kmap(rdpg->rp_pages[0]);
2002 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
2005 * No pages were processed, mark this for first page
2008 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
2009 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
2010 rc = min_t(unsigned int, LU_PAGE_SIZE, rdpg->rp_count);
2012 kunmap(rdpg->rp_pages[0]);
2015 GOTO(out_unlock, rc);
2017 mdd_read_unlock(env, mdd_obj);
2021 static int mdd_object_sync(const struct lu_env *env, struct md_object *obj)
2023 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2025 if (mdd_object_exists(mdd_obj) == 0) {
2028 CERROR("%s: object "DFID" not found: rc = %d\n",
2029 mdd_obj_dev_name(mdd_obj),
2030 PFID(mdd_object_fid(mdd_obj)), rc);
2033 return dt_object_sync(env, mdd_object_child(mdd_obj),
2037 static int mdd_object_lock(const struct lu_env *env,
2038 struct md_object *obj,
2039 struct lustre_handle *lh,
2040 struct ldlm_enqueue_info *einfo,
2041 ldlm_policy_data_t *policy)
2043 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2044 return dt_object_lock(env, mdd_object_child(mdd_obj), lh,
2048 static int mdd_object_unlock(const struct lu_env *env,
2049 struct md_object *obj,
2050 struct ldlm_enqueue_info *einfo,
2051 ldlm_policy_data_t *policy)
2053 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2054 return dt_object_unlock(env, mdd_object_child(mdd_obj), einfo, policy);
2057 const struct md_object_operations mdd_obj_ops = {
2058 .moo_permission = mdd_permission,
2059 .moo_attr_get = mdd_attr_get,
2060 .moo_attr_set = mdd_attr_set,
2061 .moo_xattr_get = mdd_xattr_get,
2062 .moo_xattr_set = mdd_xattr_set,
2063 .moo_xattr_list = mdd_xattr_list,
2064 .moo_xattr_del = mdd_xattr_del,
2065 .moo_swap_layouts = mdd_swap_layouts,
2066 .moo_open = mdd_open,
2067 .moo_close = mdd_close,
2068 .moo_readpage = mdd_readpage,
2069 .moo_readlink = mdd_readlink,
2070 .moo_changelog = mdd_changelog,
2071 .moo_capa_get = mdd_capa_get,
2072 .moo_object_sync = mdd_object_sync,
2073 .moo_object_lock = mdd_object_lock,
2074 .moo_object_unlock = mdd_object_unlock,