4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdd/mdd_object.c
38 * Lustre Metadata Server (mdd) routines
40 * Author: Wang Di <wangdi@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_MDS
45 #include <linux/module.h>
47 #include <obd_class.h>
48 #include <obd_support.h>
49 #include <lprocfs_status.h>
50 /* fid_be_cpu(), fid_cpu_to_be(). */
51 #include <lustre_fid.h>
52 #include <lustre_idmap.h>
53 #include <lustre_param.h>
54 #include <lustre_mds.h>
55 #include <lustre/lustre_idl.h>
57 #include "mdd_internal.h"
59 static const struct lu_object_operations mdd_lu_obj_ops;
61 static int mdd_xattr_get(const struct lu_env *env,
62 struct md_object *obj, struct lu_buf *buf,
65 int mdd_la_get(const struct lu_env *env, struct mdd_object *obj,
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 struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
79 struct mdd_thread_info *info;
81 lu_env_refill((struct lu_env *)env);
82 info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
83 LASSERT(info != NULL);
87 struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
91 buf = &mdd_env_info(env)->mti_buf[0];
97 const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
98 const void *area, ssize_t len)
102 buf = &mdd_env_info(env)->mti_buf[0];
103 buf->lb_buf = (void *)area;
108 struct lu_object *mdd_object_alloc(const struct lu_env *env,
109 const struct lu_object_header *hdr,
112 struct mdd_object *mdd_obj;
114 OBD_SLAB_ALLOC_PTR_GFP(mdd_obj, mdd_object_kmem, GFP_NOFS);
115 if (mdd_obj != NULL) {
118 o = mdd2lu_obj(mdd_obj);
119 lu_object_init(o, NULL, d);
120 mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
121 mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
122 mdd_obj->mod_count = 0;
123 o->lo_ops = &mdd_lu_obj_ops;
130 static int mdd_object_init(const struct lu_env *env, struct lu_object *o,
131 const struct lu_object_conf *unused)
133 struct mdd_device *d = lu2mdd_dev(o->lo_dev);
134 struct mdd_object *mdd_obj = lu2mdd_obj(o);
135 struct lu_object *below;
136 struct lu_device *under;
139 mdd_obj->mod_cltime = 0;
140 under = &d->mdd_child->dd_lu_dev;
141 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
143 RETURN(PTR_ERR(below));
145 lu_object_add(o, below);
150 static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
154 if (lu_object_exists(o)) {
155 struct mdd_object *mdd_obj = lu2mdd_obj(o);
156 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
158 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
164 static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
166 struct mdd_object *mdd = lu2mdd_obj(o);
169 OBD_SLAB_FREE_PTR(mdd, mdd_object_kmem);
172 static int mdd_object_print(const struct lu_env *env, void *cookie,
173 lu_printer_t p, const struct lu_object *o)
175 struct mdd_object *mdd = lu2mdd_obj((struct lu_object *)o);
176 return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p(open_count=%d, "
177 "valid=%x, cltime="LPU64", flags=%lx)",
178 mdd, mdd->mod_count, mdd->mod_valid,
179 mdd->mod_cltime, mdd->mod_flags);
182 static const struct lu_object_operations mdd_lu_obj_ops = {
183 .loo_object_init = mdd_object_init,
184 .loo_object_start = mdd_object_start,
185 .loo_object_free = mdd_object_free,
186 .loo_object_print = mdd_object_print,
189 struct mdd_object *mdd_object_find(const struct lu_env *env,
190 struct mdd_device *d,
191 const struct lu_fid *f)
193 return md2mdd_obj(md_object_find_slice(env, &d->mdd_md_dev, f));
197 * No permission check is needed.
199 int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
202 struct mdd_object *mdd_obj = md2mdd_obj(obj);
207 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr,
208 mdd_object_capa(env, md2mdd_obj(obj)));
209 if ((ma->ma_need & MA_INODE) != 0 && mdd_is_dead_obj(mdd_obj))
210 ma->ma_attr.la_nlink = 0;
216 * No permission check is needed.
218 static int mdd_xattr_get(const struct lu_env *env,
219 struct md_object *obj, struct lu_buf *buf,
222 struct mdd_object *mdd_obj = md2mdd_obj(obj);
227 if (mdd_object_exists(mdd_obj) == 0) {
228 CERROR("%s: object "DFID" not found: rc = -2\n",
229 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
233 /* If the object has been delete from the namespace, then
234 * get linkEA should return -ENOENT as well */
235 if (unlikely((mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ)) &&
236 strcmp(name, XATTR_NAME_LINK) == 0))
239 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
240 rc = mdo_xattr_get(env, mdd_obj, buf, name,
241 mdd_object_capa(env, mdd_obj));
242 mdd_read_unlock(env, mdd_obj);
248 * Permission check is done when open,
249 * no need check again.
251 int mdd_readlink(const struct lu_env *env, struct md_object *obj,
254 struct mdd_object *mdd_obj = md2mdd_obj(obj);
255 struct dt_object *next;
260 if (mdd_object_exists(mdd_obj) == 0) {
261 CERROR("%s: object "DFID" not found: rc = -2\n",
262 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
266 next = mdd_object_child(mdd_obj);
267 LASSERT(next != NULL);
268 LASSERT(next->do_body_ops != NULL);
269 LASSERT(next->do_body_ops->dbo_read != NULL);
270 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
271 rc = next->do_body_ops->dbo_read(env, next, buf, &pos,
272 mdd_object_capa(env, mdd_obj));
273 mdd_read_unlock(env, mdd_obj);
278 * No permission check is needed.
280 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
283 struct mdd_object *mdd_obj = md2mdd_obj(obj);
288 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
289 rc = mdo_xattr_list(env, mdd_obj, buf, mdd_object_capa(env, mdd_obj));
290 mdd_read_unlock(env, mdd_obj);
296 * Filter out XATTR_NAME_LINK if this is an orphan object. See
299 if (unlikely(mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ))) {
300 char *end = (char *)buf->lb_buf + rc;
301 char *p = buf->lb_buf;
304 char *next = p + strlen(p) + 1;
306 if (strcmp(p, XATTR_NAME_LINK) == 0) {
308 memmove(p, next, end - next);
310 CDEBUG(D_INFO, "Filtered out "XATTR_NAME_LINK
311 " of orphan "DFID"\n",
312 PFID(mdd_object_fid(mdd_obj)));
323 int mdd_declare_object_create_internal(const struct lu_env *env,
324 struct mdd_object *p,
325 struct mdd_object *c,
326 struct lu_attr *attr,
327 struct thandle *handle,
328 const struct md_op_spec *spec,
329 struct dt_allocation_hint *hint)
331 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
332 const struct dt_index_features *feat = spec->sp_feat;
336 if (feat != &dt_directory_features && feat != NULL) {
337 dof->dof_type = DFT_INDEX;
338 dof->u.dof_idx.di_feat = feat;
341 dof->dof_type = dt_mode_to_dft(attr->la_mode);
342 if (dof->dof_type == DFT_REGULAR) {
343 dof->u.dof_reg.striped =
344 md_should_create(spec->sp_cr_flags);
345 if (spec->sp_cr_flags & MDS_OPEN_HAS_EA)
346 dof->u.dof_reg.striped = 0;
347 /* is this replay? */
349 dof->u.dof_reg.striped = 0;
353 rc = mdo_declare_create_obj(env, c, attr, hint, dof, handle);
358 int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p,
359 struct mdd_object *c, struct lu_attr *attr,
360 struct thandle *handle,
361 const struct md_op_spec *spec,
362 struct dt_allocation_hint *hint)
364 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
368 LASSERT(!mdd_object_exists(c));
370 rc = mdo_create_obj(env, c, attr, hint, dof, handle);
372 LASSERT(ergo(rc == 0, mdd_object_exists(c)));
377 int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj,
378 const struct lu_attr *attr, struct thandle *handle,
384 rc = mdo_attr_set(env, obj, attr, handle, mdd_object_capa(env, obj));
385 #ifdef CONFIG_FS_POSIX_ACL
386 if (!rc && (attr->la_valid & LA_MODE) && needacl)
387 rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
392 int mdd_update_time(const struct lu_env *env, struct mdd_object *obj,
393 const struct lu_attr *oattr, struct lu_attr *attr,
394 struct thandle *handle)
399 LASSERT(attr->la_valid & LA_CTIME);
400 LASSERT(oattr != NULL);
402 /* Make sure the ctime is increased only, however, it's not strictly
403 * reliable at here because there is not guarantee to hold lock on
404 * object, so we just bypass some unnecessary cmtime setting first
405 * and OSD has to check it again. */
406 if (attr->la_ctime < oattr->la_ctime)
407 attr->la_valid &= ~(LA_MTIME | LA_CTIME);
408 else if (attr->la_valid == LA_CTIME &&
409 attr->la_ctime == oattr->la_ctime)
410 attr->la_valid &= ~LA_CTIME;
412 if (attr->la_valid != 0)
413 rc = mdd_attr_set_internal(env, obj, attr, handle, 0);
418 * This gives the same functionality as the code between
419 * sys_chmod and inode_setattr
420 * chown_common and inode_setattr
421 * utimes and inode_setattr
422 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
424 static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
425 const struct lu_attr *oattr, struct lu_attr *la,
426 const unsigned long flags)
435 /* Do not permit change file type */
436 if (la->la_valid & LA_TYPE)
439 /* They should not be processed by setattr */
440 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
443 LASSERT(oattr != NULL);
445 /* export destroy does not have ->le_ses, but we may want
446 * to drop LUSTRE_SOM_FL. */
447 uc = lu_ucred_check(env);
451 if (la->la_valid == LA_CTIME) {
452 if (!(flags & MDS_PERM_BYPASS))
453 /* This is only for set ctime when rename's source is
455 rc = mdd_may_delete(env, NULL, NULL, obj, oattr, NULL,
457 if (rc == 0 && la->la_ctime <= oattr->la_ctime)
458 la->la_valid &= ~LA_CTIME;
462 if (la->la_valid == LA_ATIME) {
463 /* This is atime only set for read atime update on close. */
464 if (la->la_atime >= oattr->la_atime &&
465 la->la_atime < (oattr->la_atime +
466 mdd_obj2mdd_dev(obj)->mdd_atime_diff))
467 la->la_valid &= ~LA_ATIME;
471 /* Check if flags change. */
472 if (la->la_valid & LA_FLAGS) {
473 unsigned int oldflags = oattr->la_flags &
474 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
475 unsigned int newflags = la->la_flags &
476 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
478 if ((uc->uc_fsuid != oattr->la_uid) &&
479 !md_capable(uc, CFS_CAP_FOWNER))
482 /* The IMMUTABLE and APPEND_ONLY flags can
483 * only be changed by the relevant capability. */
484 if ((oldflags ^ newflags) &&
485 !md_capable(uc, CFS_CAP_LINUX_IMMUTABLE))
488 if (!S_ISDIR(oattr->la_mode))
489 la->la_flags &= ~LUSTRE_DIRSYNC_FL;
492 if (oattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL) &&
493 (la->la_valid & ~LA_FLAGS) &&
494 !(flags & MDS_PERM_BYPASS))
497 /* Check for setting the obj time. */
498 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
499 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
500 if ((uc->uc_fsuid != oattr->la_uid) &&
501 !md_capable(uc, CFS_CAP_FOWNER)) {
502 rc = mdd_permission_internal(env, obj, oattr,
509 if (la->la_valid & LA_KILL_SUID) {
510 la->la_valid &= ~LA_KILL_SUID;
511 if ((oattr->la_mode & S_ISUID) &&
512 !(la->la_valid & LA_MODE)) {
513 la->la_mode = oattr->la_mode;
514 la->la_valid |= LA_MODE;
516 la->la_mode &= ~S_ISUID;
519 if (la->la_valid & LA_KILL_SGID) {
520 la->la_valid &= ~LA_KILL_SGID;
521 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
522 (S_ISGID | S_IXGRP)) &&
523 !(la->la_valid & LA_MODE)) {
524 la->la_mode = oattr->la_mode;
525 la->la_valid |= LA_MODE;
527 la->la_mode &= ~S_ISGID;
530 /* Make sure a caller can chmod. */
531 if (la->la_valid & LA_MODE) {
532 if (!(flags & MDS_PERM_BYPASS) &&
533 (uc->uc_fsuid != oattr->la_uid) &&
534 !md_capable(uc, CFS_CAP_FOWNER))
537 if (la->la_mode == (umode_t) -1)
538 la->la_mode = oattr->la_mode;
540 la->la_mode = (la->la_mode & S_IALLUGO) |
541 (oattr->la_mode & ~S_IALLUGO);
543 /* Also check the setgid bit! */
544 if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ?
545 la->la_gid : oattr->la_gid) &&
546 !md_capable(uc, CFS_CAP_FSETID))
547 la->la_mode &= ~S_ISGID;
549 la->la_mode = oattr->la_mode;
552 /* Make sure a caller can chown. */
553 if (la->la_valid & LA_UID) {
554 if (la->la_uid == (uid_t) -1)
555 la->la_uid = oattr->la_uid;
556 if (((uc->uc_fsuid != oattr->la_uid) ||
557 (la->la_uid != oattr->la_uid)) &&
558 !md_capable(uc, CFS_CAP_CHOWN))
561 /* If the user or group of a non-directory has been
562 * changed by a non-root user, remove the setuid bit.
563 * 19981026 David C Niemi <niemi@tux.org>
565 * Changed this to apply to all users, including root,
566 * to avoid some races. This is the behavior we had in
567 * 2.0. The check for non-root was definitely wrong
568 * for 2.2 anyway, as it should have been using
569 * CAP_FSETID rather than fsuid -- 19990830 SD. */
570 if (((oattr->la_mode & S_ISUID) == S_ISUID) &&
571 !S_ISDIR(oattr->la_mode)) {
572 la->la_mode &= ~S_ISUID;
573 la->la_valid |= LA_MODE;
577 /* Make sure caller can chgrp. */
578 if (la->la_valid & LA_GID) {
579 if (la->la_gid == (gid_t) -1)
580 la->la_gid = oattr->la_gid;
581 if (((uc->uc_fsuid != oattr->la_uid) ||
582 ((la->la_gid != oattr->la_gid) &&
583 !lustre_in_group_p(uc, la->la_gid))) &&
584 !md_capable(uc, CFS_CAP_CHOWN))
587 /* Likewise, if the user or group of a non-directory
588 * has been changed by a non-root user, remove the
589 * setgid bit UNLESS there is no group execute bit
590 * (this would be a file marked for mandatory
591 * locking). 19981026 David C Niemi <niemi@tux.org>
593 * Removed the fsuid check (see the comment above) --
595 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
596 (S_ISGID | S_IXGRP)) && !S_ISDIR(oattr->la_mode)) {
597 la->la_mode &= ~S_ISGID;
598 la->la_valid |= LA_MODE;
602 /* For both Size-on-MDS case and truncate case,
603 * "la->la_valid & (LA_SIZE | LA_BLOCKS)" are ture.
604 * We distinguish them by "flags & MDS_SOM".
605 * For SOM case, it is true, the MAY_WRITE perm has been checked
606 * when open, no need check again. For truncate case, it is false,
607 * the MAY_WRITE perm should be checked here. */
608 if (flags & MDS_SOM) {
609 /* For the "Size-on-MDS" setattr update, merge coming
610 * attributes with the set in the inode. BUG 10641 */
611 if ((la->la_valid & LA_ATIME) &&
612 (la->la_atime <= oattr->la_atime))
613 la->la_valid &= ~LA_ATIME;
615 /* OST attributes do not have a priority over MDS attributes,
616 * so drop times if ctime is equal. */
617 if ((la->la_valid & LA_CTIME) &&
618 (la->la_ctime <= oattr->la_ctime))
619 la->la_valid &= ~(LA_MTIME | LA_CTIME);
621 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
622 if (!((flags & MDS_OWNEROVERRIDE) &&
623 (uc->uc_fsuid == oattr->la_uid)) &&
624 !(flags & MDS_PERM_BYPASS)) {
625 rc = mdd_permission_internal(env, obj,
631 if (la->la_valid & LA_CTIME) {
632 /* The pure setattr, it has the priority over what is
633 * already set, do not drop it if ctime is equal. */
634 if (la->la_ctime < oattr->la_ctime)
635 la->la_valid &= ~(LA_ATIME | LA_MTIME |
643 /** Store a data change changelog record
644 * If this fails, we must fail the whole transaction; we don't
645 * want the change to commit without the log entry.
646 * \param mdd_obj - mdd_object of change
647 * \param handle - transacion handle
649 int mdd_changelog_data_store(const struct lu_env *env, struct mdd_device *mdd,
650 enum changelog_rec_type type, int flags,
651 struct mdd_object *mdd_obj, struct thandle *handle)
653 const struct lu_ucred *uc = lu_ucred(env);
654 const struct lu_fid *tfid;
655 struct llog_changelog_rec *rec;
661 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
663 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
666 LASSERT(mdd_obj != NULL);
667 LASSERT(handle != NULL);
669 tfid = mdo2fid(mdd_obj);
671 if ((type >= CL_MTIME) && (type <= CL_ATIME) &&
672 cfs_time_before_64(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) {
673 /* Don't need multiple updates in this log */
674 /* Don't check under lock - no big deal if we get an extra
679 flags = (flags & CLF_FLAGMASK) | CLF_VERSION;
680 if (uc->uc_jobid[0] != '\0')
683 reclen = llog_data_len(changelog_rec_offset(flags & CLF_SUPPORTED));
684 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
685 if (buf->lb_buf == NULL)
689 rec->cr.cr_flags = flags;
690 rec->cr.cr_type = (__u32)type;
691 rec->cr.cr_tfid = *tfid;
692 rec->cr.cr_namelen = 0;
693 mdd_obj->mod_cltime = cfs_time_current_64();
695 if (flags & CLF_JOBID)
696 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
698 rc = mdd_changelog_store(env, mdd, rec, handle);
703 static int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type,
704 int flags, struct md_object *obj)
706 struct thandle *handle;
707 struct mdd_object *mdd_obj = md2mdd_obj(obj);
708 struct mdd_device *mdd = mdo2mdd(obj);
712 handle = mdd_trans_create(env, mdd);
714 RETURN(PTR_ERR(handle));
716 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
720 rc = mdd_trans_start(env, mdd, handle);
724 rc = mdd_changelog_data_store(env, mdd, type, flags, mdd_obj,
728 mdd_trans_stop(env, mdd, rc, handle);
734 * Save LMA extended attributes with data from \a ma.
736 * HSM and Size-On-MDS data will be extracted from \ma if they are valid, if
737 * not, LMA EA will be first read from disk, modified and write back.
740 /* Precedence for choosing record type when multiple
741 * attributes change: setattr > mtime > ctime > atime
742 * (ctime changes when mtime does, plus chmod/chown.
743 * atime and ctime are independent.) */
744 static int mdd_attr_set_changelog(const struct lu_env *env,
745 struct md_object *obj, struct thandle *handle,
748 struct mdd_device *mdd = mdo2mdd(obj);
751 bits = (valid & LA_SIZE) ? 1 << CL_TRUNC : 0;
752 bits |= (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0;
753 bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0;
754 bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0;
755 bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0;
756 bits = bits & mdd->mdd_cl.mc_mask;
757 /* This is an implementation limit rather than a protocol limit */
758 CLASSERT(CL_LAST <= sizeof(int) * 8);
762 /* The record type is the lowest non-masked set bit */
765 /* FYI we only store the first CLF_FLAGMASK bits of la_valid */
766 return mdd_changelog_data_store(env, mdd, type, (int)valid,
767 md2mdd_obj(obj), handle);
770 static int mdd_declare_attr_set(const struct lu_env *env,
771 struct mdd_device *mdd,
772 struct mdd_object *obj,
773 const struct lu_attr *attr,
774 struct thandle *handle)
778 rc = mdo_declare_attr_set(env, obj, attr, handle);
782 #ifdef CONFIG_FS_POSIX_ACL
783 if (attr->la_valid & LA_MODE) {
784 mdd_read_lock(env, obj, MOR_TGT_CHILD);
785 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL,
786 XATTR_NAME_ACL_ACCESS, BYPASS_CAPA);
787 mdd_read_unlock(env, obj);
788 if (rc == -EOPNOTSUPP || rc == -ENODATA)
794 struct lu_buf *buf = mdd_buf_get(env, NULL, rc);
795 rc = mdo_declare_xattr_set(env, obj, buf,
796 XATTR_NAME_ACL_ACCESS, 0,
804 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
811 * permission changes may require sync operation, to mitigate performance
812 * impact, only do this for dir and when permission is reduced.
814 * For regular files, version is updated with permission change (see VBR), async
815 * permission won't cause any issue, while missing permission change on
816 * directory may affect accessibility of other objects after recovery.
818 static inline bool permission_needs_sync(const struct lu_attr *old,
819 const struct lu_attr *new)
821 if (!S_ISDIR(old->la_mode))
824 if (new->la_valid & (LA_UID | LA_GID))
827 if (new->la_valid & LA_MODE &&
828 new->la_mode & (S_ISUID | S_ISGID | S_ISVTX))
831 if ((new->la_valid & LA_MODE) &&
832 ((new->la_mode & old->la_mode) & S_IRWXUGO) !=
833 (old->la_mode & S_IRWXUGO))
839 /* set attr and LOV EA at once, return updated attr */
840 int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
841 const struct md_attr *ma)
843 struct mdd_object *mdd_obj = md2mdd_obj(obj);
844 struct mdd_device *mdd = mdo2mdd(obj);
845 struct thandle *handle;
846 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
847 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
848 const struct lu_attr *la = &ma->ma_attr;
852 /* we do not use ->attr_set() for LOV/SOM/HSM EA any more */
853 LASSERT((ma->ma_valid & MA_LOV) == 0);
854 LASSERT((ma->ma_valid & MA_HSM) == 0);
855 LASSERT((ma->ma_valid & MA_SOM) == 0);
857 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
861 *la_copy = ma->ma_attr;
862 rc = mdd_fix_attr(env, mdd_obj, attr, la_copy, ma->ma_attr_flags);
866 /* setattr on "close" only change atime, or do nothing */
867 if (la->la_valid == LA_ATIME && la_copy->la_valid == 0)
870 handle = mdd_trans_create(env, mdd);
872 RETURN(PTR_ERR(handle));
874 rc = mdd_declare_attr_set(env, mdd, mdd_obj, la, handle);
878 rc = mdd_trans_start(env, mdd, handle);
882 if (mdd->mdd_sync_permission && permission_needs_sync(attr, la))
885 if (la->la_valid & (LA_MTIME | LA_CTIME))
886 CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n",
887 la->la_mtime, la->la_ctime);
889 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
890 if (la_copy->la_valid & LA_FLAGS)
891 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
892 else if (la_copy->la_valid) /* setattr */
893 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
894 mdd_write_unlock(env, mdd_obj);
897 rc = mdd_attr_set_changelog(env, obj, handle, la->la_valid);
902 mdd_trans_stop(env, mdd, rc, handle);
906 static int mdd_xattr_sanity_check(const struct lu_env *env,
907 struct mdd_object *obj,
908 const struct lu_attr *attr)
910 struct lu_ucred *uc = lu_ucred_assert(env);
913 if (attr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
916 if ((uc->uc_fsuid != attr->la_uid) && !md_capable(uc, CFS_CAP_FOWNER))
922 static int mdd_declare_xattr_set(const struct lu_env *env,
923 struct mdd_device *mdd,
924 struct mdd_object *obj,
925 const struct lu_buf *buf,
927 int fl, struct thandle *handle)
931 rc = mdo_declare_xattr_set(env, obj, buf, name, fl, handle);
935 /* Only record user and layout xattr changes */
936 if (strncmp(XATTR_USER_PREFIX, name,
937 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
938 strcmp(XATTR_NAME_LOV, name) == 0) {
939 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
944 /* If HSM data is modified, this could add a changelog */
945 if (strcmp(XATTR_NAME_HSM, name) == 0) {
946 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
951 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
956 * Compare current and future data of HSM EA and add a changelog if needed.
958 * Caller should have write-locked \param obj.
960 * \param buf - Future HSM EA content.
961 * \retval 0 if no changelog is needed or changelog was added properly.
962 * \retval -ve errno if there was a problem
964 static int mdd_hsm_update_locked(const struct lu_env *env,
965 struct md_object *obj,
966 const struct lu_buf *buf,
967 struct thandle *handle)
969 struct mdd_thread_info *info = mdd_env_info(env);
970 struct mdd_device *mdd = mdo2mdd(obj);
971 struct mdd_object *mdd_obj = md2mdd_obj(obj);
972 struct lu_buf *current_buf;
973 struct md_hsm *current_mh;
974 struct md_hsm *new_mh;
978 OBD_ALLOC_PTR(current_mh);
979 if (current_mh == NULL)
982 /* Read HSM attrs from disk */
983 CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
984 current_buf = mdd_buf_get(env, info->mti_xattr_buf,
985 sizeof(info->mti_xattr_buf));
986 rc = mdo_xattr_get(env, mdd_obj, current_buf, XATTR_NAME_HSM,
987 mdd_object_capa(env, mdd_obj));
988 rc = lustre_buf2hsm(current_buf->lb_buf, rc, current_mh);
989 if (rc < 0 && rc != -ENODATA)
991 else if (rc == -ENODATA)
992 current_mh->mh_flags = 0;
994 /* Map future HSM xattr */
995 OBD_ALLOC_PTR(new_mh);
997 GOTO(free, rc = -ENOMEM);
998 lustre_buf2hsm(buf->lb_buf, buf->lb_len, new_mh);
1000 /* If HSM flags are different, add a changelog */
1002 if (current_mh->mh_flags != new_mh->mh_flags) {
1004 hsm_set_cl_event(&flags, HE_STATE);
1005 if (new_mh->mh_flags & HS_DIRTY)
1006 hsm_set_cl_flags(&flags, CLF_HSM_DIRTY);
1008 rc = mdd_changelog_data_store(env, mdd, CL_HSM, flags, mdd_obj,
1012 OBD_FREE_PTR(new_mh);
1014 OBD_FREE_PTR(current_mh);
1018 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1022 * The caller should guarantee to update the object ctime
1023 * after xattr_set if needed.
1025 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
1026 const struct lu_buf *buf, const char *name,
1029 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1030 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1031 struct mdd_device *mdd = mdo2mdd(obj);
1032 struct thandle *handle;
1036 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
1040 rc = mdd_xattr_sanity_check(env, mdd_obj, attr);
1044 if (strcmp(name, XATTR_NAME_ACL_ACCESS) == 0 ||
1045 strcmp(name, XATTR_NAME_ACL_DEFAULT) == 0) {
1046 struct posix_acl *acl;
1048 /* user may set empty ACL, which should be treated as removing
1050 acl = posix_acl_from_xattr(&init_user_ns, buf->lb_buf,
1053 rc = mdd_xattr_del(env, obj, name);
1056 posix_acl_release(acl);
1059 if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) {
1060 rc = mdd_acl_set(env, mdd_obj, attr, buf, fl);
1064 handle = mdd_trans_create(env, mdd);
1066 RETURN(PTR_ERR(handle));
1068 rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, fl, handle);
1072 rc = mdd_trans_start(env, mdd, handle);
1076 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1078 if (strcmp(XATTR_NAME_HSM, name) == 0) {
1079 rc = mdd_hsm_update_locked(env, obj, buf, handle);
1081 mdd_write_unlock(env, mdd_obj);
1086 rc = mdo_xattr_set(env, mdd_obj, buf, name, fl, handle,
1087 mdd_object_capa(env, mdd_obj));
1088 mdd_write_unlock(env, mdd_obj);
1092 if (strcmp(XATTR_NAME_LOV, name) == 0)
1093 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, mdd_obj,
1095 else if (strncmp(XATTR_USER_PREFIX, name,
1096 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1097 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1098 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1099 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1100 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1101 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1105 mdd_trans_stop(env, mdd, rc, handle);
1110 static int mdd_declare_xattr_del(const struct lu_env *env,
1111 struct mdd_device *mdd,
1112 struct mdd_object *obj,
1114 struct thandle *handle)
1118 rc = mdo_declare_xattr_del(env, obj, name, handle);
1122 /* Only record user xattr changes */
1123 if ((strncmp(XATTR_USER_PREFIX, name,
1124 sizeof(XATTR_USER_PREFIX) - 1) == 0))
1125 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1131 * The caller should guarantee to update the object ctime
1132 * after xattr_set if needed.
1134 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1137 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1138 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1139 struct mdd_device *mdd = mdo2mdd(obj);
1140 struct thandle *handle;
1144 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
1148 rc = mdd_xattr_sanity_check(env, mdd_obj, attr);
1152 handle = mdd_trans_create(env, mdd);
1154 RETURN(PTR_ERR(handle));
1156 rc = mdd_declare_xattr_del(env, mdd, mdd_obj, name, handle);
1160 rc = mdd_trans_start(env, mdd, handle);
1164 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1165 rc = mdo_xattr_del(env, mdd_obj, name, handle,
1166 mdd_object_capa(env, mdd_obj));
1167 mdd_write_unlock(env, mdd_obj);
1171 /* Only record system & user xattr changes */
1172 if (strncmp(XATTR_USER_PREFIX, name,
1173 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1174 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1175 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1176 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1177 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1178 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1182 mdd_trans_stop(env, mdd, rc, handle);
1188 * read lov EA of an object
1189 * return the lov EA in an allocated lu_buf
1191 int mdd_get_lov_ea(const struct lu_env *env, struct mdd_object *obj,
1192 struct lu_buf *lmm_buf)
1194 struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf;
1199 rc = mdo_xattr_get(env, obj, buf, XATTR_NAME_LOV,
1200 mdd_object_capa(env, obj));
1202 if (rc == -ERANGE) {
1203 /* mti_big_buf is allocated but is too small
1204 * we need to increase it */
1205 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1207 if (buf->lb_buf == NULL)
1208 GOTO(out, rc = -ENOMEM);
1219 if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) {
1220 /* mti_big_buf was not allocated, so we have to
1221 * allocate it based on the ea size */
1222 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1224 if (buf->lb_buf == NULL)
1225 GOTO(out, rc = -ENOMEM);
1229 lu_buf_alloc(lmm_buf, bufsize);
1230 if (lmm_buf->lb_buf == NULL)
1231 GOTO(out, rc = -ENOMEM);
1233 memcpy(lmm_buf->lb_buf, buf->lb_buf, bufsize);
1239 lu_buf_free(lmm_buf);
1243 static int mdd_xattr_hsm_replace(const struct lu_env *env,
1244 struct mdd_object *o, struct lu_buf *buf,
1245 struct thandle *handle)
1247 struct hsm_attrs *attrs;
1253 rc = mdo_xattr_set(env, o, buf, XATTR_NAME_HSM, LU_XATTR_REPLACE,
1254 handle, mdd_object_capa(env, o));
1258 attrs = buf->lb_buf;
1259 hsm_flags = le32_to_cpu(attrs->hsm_flags);
1260 if (!(hsm_flags & HS_RELEASED) || mdd_is_dead_obj(o))
1263 /* Add a changelog record for release. */
1264 hsm_set_cl_event(&flags, HE_RELEASE);
1265 rc = mdd_changelog_data_store(env, mdo2mdd(&o->mod_obj), CL_HSM,
1271 * check if layout swapping between 2 objects is allowed
1273 * - only normal FIDs or non-system IGIFs
1274 * - same type of objects
1275 * - same owner/group (so quotas are still valid)
1277 static int mdd_layout_swap_allowed(const struct lu_env *env,
1278 struct mdd_object *o1,
1279 const struct lu_attr *attr1,
1280 struct mdd_object *o2,
1281 const struct lu_attr *attr2)
1283 const struct lu_fid *fid1, *fid2;
1289 if (!fid_is_norm(fid1) &&
1290 (!fid_is_igif(fid1) || IS_ERR(mdd_links_get(env, o1))))
1293 if (!fid_is_norm(fid2) &&
1294 (!fid_is_igif(fid2) || IS_ERR(mdd_links_get(env, o2))))
1297 if (mdd_object_type(o1) != mdd_object_type(o2)) {
1298 if (S_ISDIR(mdd_object_type(o1)))
1300 if (S_ISREG(mdd_object_type(o1)))
1305 if ((attr1->la_uid != attr2->la_uid) ||
1306 (attr1->la_gid != attr2->la_gid))
1313 * swap layouts between 2 lustre objects
1315 static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1,
1316 struct md_object *obj2, __u64 flags)
1318 struct mdd_thread_info *info = mdd_env_info(env);
1319 struct mdd_object *fst_o = md2mdd_obj(obj1);
1320 struct mdd_object *snd_o = md2mdd_obj(obj2);
1321 struct lu_attr *fst_la = MDD_ENV_VAR(env, cattr);
1322 struct lu_attr *snd_la = MDD_ENV_VAR(env, tattr);
1323 struct mdd_device *mdd = mdo2mdd(obj1);
1324 struct lov_mds_md *fst_lmm, *snd_lmm;
1325 struct lu_buf *fst_buf = &info->mti_buf[0];
1326 struct lu_buf *snd_buf = &info->mti_buf[1];
1327 struct lu_buf *fst_hsm_buf = &info->mti_buf[2];
1328 struct lu_buf *snd_hsm_buf = &info->mti_buf[3];
1329 struct ost_id *saved_oi = NULL;
1330 struct thandle *handle;
1331 __u16 fst_gen, snd_gen;
1337 CLASSERT(ARRAY_SIZE(info->mti_buf) >= 4);
1338 memset(info->mti_buf, 0, sizeof(info->mti_buf));
1340 /* we have to sort the 2 obj, so locking will always
1341 * be in the same order, even in case of 2 concurrent swaps */
1342 rc = lu_fid_cmp(mdo2fid(fst_o), mdo2fid(snd_o));
1343 if (rc == 0) /* same fid ? */
1349 rc = mdd_la_get(env, fst_o, fst_la, BYPASS_CAPA);
1353 rc = mdd_la_get(env, snd_o, snd_la, BYPASS_CAPA);
1357 /* check if layout swapping is allowed */
1358 rc = mdd_layout_swap_allowed(env, fst_o, fst_la, snd_o, snd_la);
1362 handle = mdd_trans_create(env, mdd);
1364 RETURN(PTR_ERR(handle));
1366 /* objects are already sorted */
1367 mdd_write_lock(env, fst_o, MOR_TGT_CHILD);
1368 mdd_write_lock(env, snd_o, MOR_TGT_CHILD);
1370 rc = mdd_get_lov_ea(env, fst_o, fst_buf);
1371 if (rc < 0 && rc != -ENODATA)
1374 rc = mdd_get_lov_ea(env, snd_o, snd_buf);
1375 if (rc < 0 && rc != -ENODATA)
1378 /* swapping 2 non existant layouts is a success */
1379 if (fst_buf->lb_buf == NULL && snd_buf->lb_buf == NULL)
1382 /* to help inode migration between MDT, it is better to
1383 * start by the no layout file (if one), so we order the swap */
1384 if (snd_buf->lb_buf == NULL) {
1386 swap(fst_buf, snd_buf);
1389 /* lmm and generation layout initialization */
1390 if (fst_buf->lb_buf != NULL) {
1391 fst_lmm = fst_buf->lb_buf;
1392 fst_gen = le16_to_cpu(fst_lmm->lmm_layout_gen);
1393 fst_fl = LU_XATTR_REPLACE;
1397 fst_fl = LU_XATTR_CREATE;
1400 snd_lmm = snd_buf->lb_buf;
1401 snd_gen = le16_to_cpu(snd_lmm->lmm_layout_gen);
1403 /* increase the generation layout numbers */
1407 /* set the file specific informations in lmm */
1408 if (fst_lmm != NULL) {
1409 saved_oi = &info->mti_oa.o_oi;
1411 *saved_oi = fst_lmm->lmm_oi;
1412 fst_lmm->lmm_layout_gen = cpu_to_le16(snd_gen);
1413 fst_lmm->lmm_oi = snd_lmm->lmm_oi;
1414 snd_lmm->lmm_oi = *saved_oi;
1416 if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1))
1417 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEF);
1418 else if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))
1419 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V3_DEF);
1421 GOTO(stop, rc = -EPROTO);
1423 snd_lmm->lmm_layout_gen = cpu_to_le16(fst_gen);
1425 /* Prepare HSM attribute if it's required */
1426 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1427 const int buflen = sizeof(struct hsm_attrs);
1429 lu_buf_alloc(fst_hsm_buf, buflen);
1430 lu_buf_alloc(snd_hsm_buf, buflen);
1431 if (fst_hsm_buf->lb_buf == NULL || snd_hsm_buf->lb_buf == NULL)
1432 GOTO(stop, rc = -ENOMEM);
1434 /* Read HSM attribute */
1435 rc = mdo_xattr_get(env, fst_o, fst_hsm_buf, XATTR_NAME_HSM,
1440 rc = mdo_xattr_get(env, snd_o, snd_hsm_buf, XATTR_NAME_HSM,
1445 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_hsm_buf,
1446 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1451 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_hsm_buf,
1452 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1458 /* prepare transaction */
1459 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_buf, XATTR_NAME_LOV,
1464 if (fst_buf->lb_buf != NULL)
1465 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_buf,
1466 XATTR_NAME_LOV, LU_XATTR_REPLACE,
1469 rc = mdd_declare_xattr_del(env, mdd, snd_o, XATTR_NAME_LOV,
1474 rc = mdd_trans_start(env, mdd, handle);
1478 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1479 rc = mdd_xattr_hsm_replace(env, fst_o, snd_hsm_buf, handle);
1483 rc = mdd_xattr_hsm_replace(env, snd_o, fst_hsm_buf, handle);
1485 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf,
1488 CERROR("%s: restore "DFID" HSM error: %d/%d\n",
1489 mdd_obj_dev_name(fst_o),
1490 PFID(mdo2fid(fst_o)), rc, rc2);
1495 rc = mdo_xattr_set(env, fst_o, snd_buf, XATTR_NAME_LOV, fst_fl, handle,
1496 mdd_object_capa(env, fst_o));
1500 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_MDS_HSM_SWAP_LAYOUTS))) {
1503 if (fst_buf->lb_buf != NULL)
1504 rc = mdo_xattr_set(env, snd_o, fst_buf, XATTR_NAME_LOV,
1505 LU_XATTR_REPLACE, handle,
1506 mdd_object_capa(env, snd_o));
1508 rc = mdo_xattr_del(env, snd_o, XATTR_NAME_LOV, handle,
1509 mdd_object_capa(env, snd_o));
1515 /* failure on second file, but first was done, so we have
1516 * to roll back first. */
1517 if (fst_buf->lb_buf != NULL) {
1518 fst_lmm->lmm_oi = *saved_oi;
1519 fst_lmm->lmm_layout_gen = cpu_to_le16(fst_gen - 1);
1520 rc2 = mdo_xattr_set(env, fst_o, fst_buf, XATTR_NAME_LOV,
1521 LU_XATTR_REPLACE, handle,
1522 mdd_object_capa(env, fst_o));
1524 rc2 = mdo_xattr_del(env, fst_o, XATTR_NAME_LOV, handle,
1525 mdd_object_capa(env, fst_o));
1531 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, handle);
1536 rc2 = mdd_xattr_hsm_replace(env, snd_o, snd_hsm_buf, handle);
1541 CERROR("%s: unable to roll back layout swap. FIDs: "
1542 DFID" and "DFID "error: %d/%d, steps: %d\n",
1543 mdd_obj_dev_name(fst_o),
1544 PFID(mdo2fid(snd_o)), PFID(mdo2fid(fst_o)),
1546 /* a solution to avoid journal commit is to panic,
1547 * but it has strong consequences so we use LBUG to
1548 * allow sysdamin to choose to panic or not
1555 /* Issue one changelog record per file */
1556 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle);
1560 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle);
1566 mdd_trans_stop(env, mdd, rc, handle);
1567 mdd_write_unlock(env, snd_o);
1568 mdd_write_unlock(env, fst_o);
1570 lu_buf_free(fst_buf);
1571 lu_buf_free(snd_buf);
1572 lu_buf_free(fst_hsm_buf);
1573 lu_buf_free(snd_hsm_buf);
1577 void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent,
1578 struct mdd_object *child, const struct lu_attr *attr,
1579 const struct md_op_spec *spec,
1580 struct dt_allocation_hint *hint)
1582 struct dt_object *np = parent ? mdd_object_child(parent) : NULL;
1583 struct dt_object *nc = mdd_object_child(child);
1585 memset(hint, 0, sizeof(*hint));
1587 /* For striped directory, give striping EA to lod_ah_init, which will
1588 * decide the stripe_offset and stripe count by it. */
1589 if (S_ISDIR(attr->la_mode) &&
1590 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1591 hint->dah_eadata = spec->u.sp_ea.eadata;
1592 hint->dah_eadata_len = spec->u.sp_ea.eadatalen;
1594 hint->dah_eadata = NULL;
1595 hint->dah_eadata_len = 0;
1598 CDEBUG(D_INFO, DFID" eadata %p len %d\n", PFID(mdd_object_fid(child)),
1599 hint->dah_eadata, hint->dah_eadata_len);
1600 /* @hint will be initialized by underlying device. */
1601 nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT);
1605 * do NOT or the MAY_*'s, you'll get the weakest
1607 int accmode(const struct lu_env *env, const struct lu_attr *la, int flags)
1611 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1612 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1613 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1614 * owner can write to a file even if it is marked readonly to hide
1615 * its brokenness. (bug 5781) */
1616 if (flags & MDS_OPEN_OWNEROVERRIDE) {
1617 struct lu_ucred *uc = lu_ucred_check(env);
1619 if ((uc == NULL) || (la->la_uid == uc->uc_fsuid))
1623 if (flags & FMODE_READ)
1625 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1627 if (flags & MDS_FMODE_EXEC)
1632 static int mdd_open_sanity_check(const struct lu_env *env,
1633 struct mdd_object *obj,
1634 const struct lu_attr *attr, int flag)
1640 if (mdd_is_dead_obj(obj))
1643 if (S_ISLNK(attr->la_mode))
1646 mode = accmode(env, attr, flag);
1648 if (S_ISDIR(attr->la_mode) && (mode & MAY_WRITE))
1651 if (!(flag & MDS_OPEN_CREATED)) {
1652 rc = mdd_permission_internal(env, obj, attr, mode);
1657 if (S_ISFIFO(attr->la_mode) || S_ISSOCK(attr->la_mode) ||
1658 S_ISBLK(attr->la_mode) || S_ISCHR(attr->la_mode))
1659 flag &= ~MDS_OPEN_TRUNC;
1661 /* For writing append-only file must open it with append mode. */
1662 if (attr->la_flags & LUSTRE_APPEND_FL) {
1663 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1665 if (flag & MDS_OPEN_TRUNC)
1671 * Now, flag -- O_NOATIME does not be packed by client.
1673 if (flag & O_NOATIME) {
1674 struct lu_ucred *uc = lu_ucred(env);
1676 if (uc && ((uc->uc_valid == UCRED_OLD) ||
1677 (uc->uc_valid == UCRED_NEW)) &&
1678 (uc->uc_fsuid != attr->la_uid) &&
1679 !md_capable(uc, CFS_CAP_FOWNER))
1687 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1690 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1691 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1694 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1696 rc = mdd_la_get(env, mdd_obj, attr, BYPASS_CAPA);
1700 rc = mdd_open_sanity_check(env, mdd_obj, attr, flags);
1704 mdd_obj->mod_count++;
1707 mdd_write_unlock(env, mdd_obj);
1711 static int mdd_declare_close(const struct lu_env *env,
1712 struct mdd_object *obj,
1714 struct thandle *handle)
1718 rc = orph_declare_index_delete(env, obj, handle);
1722 return mdo_declare_destroy(env, obj, handle);
1726 * No permission check is needed.
1728 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1729 struct md_attr *ma, int mode)
1731 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1732 struct mdd_device *mdd = mdo2mdd(obj);
1733 struct thandle *handle = NULL;
1734 int rc, is_orphan = 0;
1737 if (ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_KEEP_ORPHAN) {
1738 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1739 mdd_obj->mod_count--;
1740 mdd_write_unlock(env, mdd_obj);
1742 if (mdd_obj->mod_flags & ORPHAN_OBJ && !mdd_obj->mod_count)
1743 CDEBUG(D_HA, "Object "DFID" is retained in orphan "
1744 "list\n", PFID(mdd_object_fid(mdd_obj)));
1748 /* mdd_finish_unlink() will always set orphan object as DEAD_OBJ, but
1749 * it might fail to add the object to orphan list (w/o ORPHAN_OBJ). */
1750 /* check without any lock */
1751 is_orphan = mdd_obj->mod_count == 1 &&
1752 (mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0;
1756 handle = mdd_trans_create(env, mdo2mdd(obj));
1758 RETURN(PTR_ERR(handle));
1760 rc = mdd_declare_close(env, mdd_obj, ma, handle);
1764 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1768 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1773 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1774 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr,
1775 mdd_object_capa(env, mdd_obj));
1777 CERROR("Failed to get lu_attr of "DFID": %d\n",
1778 PFID(mdd_object_fid(mdd_obj)), rc);
1782 /* check again with lock */
1783 is_orphan = (mdd_obj->mod_count == 1) &&
1784 ((mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0 ||
1785 ma->ma_attr.la_nlink == 0);
1787 if (is_orphan && handle == NULL) {
1788 mdd_write_unlock(env, mdd_obj);
1792 mdd_obj->mod_count--; /*release open count */
1798 /* NB: Object maybe not in orphan list originally, it is rare case for
1799 * mdd_finish_unlink() failure, in that case, the object doesn't have
1800 * ORPHAN_OBJ flag */
1801 if ((mdd_obj->mod_flags & ORPHAN_OBJ) != 0) {
1802 /* remove link to object from orphan index */
1803 LASSERT(handle != NULL);
1804 rc = __mdd_orphan_del(env, mdd_obj, handle);
1806 CERROR("%s: unable to delete "DFID" from orphan list: "
1807 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1808 PFID(mdd_object_fid(mdd_obj)), rc);
1809 /* If object was not deleted from orphan list, do not
1810 * destroy OSS objects, which will be done when next
1815 CDEBUG(D_HA, "Object "DFID" is deleted from orphan "
1816 "list, OSS objects to be destroyed.\n",
1817 PFID(mdd_object_fid(mdd_obj)));
1820 rc = mdo_destroy(env, mdd_obj, handle);
1823 CERROR("%s: unable to delete "DFID" from orphan list: "
1824 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1825 PFID(mdd_object_fid(mdd_obj)), rc);
1830 mdd_write_unlock(env, mdd_obj);
1833 (mode & (FMODE_WRITE | MDS_OPEN_APPEND | MDS_OPEN_TRUNC)) &&
1834 !(ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_RECOV_OPEN)) {
1835 if (handle == NULL) {
1836 handle = mdd_trans_create(env, mdo2mdd(obj));
1838 GOTO(stop, rc = PTR_ERR(handle));
1840 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL,
1845 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1850 mdd_changelog_data_store(env, mdd, CL_CLOSE, mode,
1855 if (handle != NULL && !IS_ERR(handle))
1856 mdd_trans_stop(env, mdd, rc, handle);
1862 * Permission check is done when open,
1863 * no need check again.
1865 static int mdd_readpage_sanity_check(const struct lu_env *env,
1866 struct mdd_object *obj)
1868 struct dt_object *next = mdd_object_child(obj);
1872 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
1880 static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp,
1881 size_t nob, const struct dt_it_ops *iops,
1882 struct dt_it *it, __u32 attr, void *arg)
1884 struct lu_dirpage *dp = &lp->lp_dir;
1888 struct lu_dirent *ent;
1889 struct lu_dirent *last = NULL;
1893 if (nob < sizeof(*dp))
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 if (le32_to_cpu(ent->lde_attrs) & LUDA_FID) {
1932 fid_le_to_cpu(&fid, &ent->lde_fid);
1933 if (fid_is_dot_lustre(&fid))
1937 result = (last != NULL) ? 0 :-EINVAL;
1941 ent = (void *)ent + recsize;
1945 result = iops->next(env, it);
1946 if (result == -ESTALE)
1948 } while (result == 0);
1951 dp->ldp_hash_end = cpu_to_le64(hash);
1953 if (last->lde_hash == dp->ldp_hash_end)
1954 dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE);
1955 last->lde_reclen = 0; /* end mark */
1958 /* end of directory */
1959 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1960 else if (result < 0)
1961 CWARN("build page failed: %d!\n", result);
1965 int mdd_readpage(const struct lu_env *env, struct md_object *obj,
1966 const struct lu_rdpg *rdpg)
1968 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1972 if (mdd_object_exists(mdd_obj) == 0) {
1973 CERROR("%s: object "DFID" not found: rc = -2\n",
1974 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
1978 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
1979 rc = mdd_readpage_sanity_check(env, mdd_obj);
1981 GOTO(out_unlock, rc);
1983 if (mdd_is_dead_obj(mdd_obj)) {
1985 struct lu_dirpage *dp;
1988 * According to POSIX, please do not return any entry to client:
1989 * even dot and dotdot should not be returned.
1991 CDEBUG(D_INODE, "readdir from dead object: "DFID"\n",
1992 PFID(mdd_object_fid(mdd_obj)));
1994 if (rdpg->rp_count <= 0)
1995 GOTO(out_unlock, rc = -EFAULT);
1996 LASSERT(rdpg->rp_pages != NULL);
1998 pg = rdpg->rp_pages[0];
1999 dp = (struct lu_dirpage *)kmap(pg);
2000 memset(dp, 0 , sizeof(struct lu_dirpage));
2001 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
2002 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
2003 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
2005 GOTO(out_unlock, rc = LU_PAGE_SIZE);
2008 rc = dt_index_walk(env, mdd_object_child(mdd_obj), rdpg,
2009 mdd_dir_page_build, NULL);
2011 struct lu_dirpage *dp;
2013 dp = kmap(rdpg->rp_pages[0]);
2014 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
2017 * No pages were processed, mark this for first page
2020 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
2021 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
2022 rc = min_t(unsigned int, LU_PAGE_SIZE, rdpg->rp_count);
2024 kunmap(rdpg->rp_pages[0]);
2027 GOTO(out_unlock, rc);
2029 mdd_read_unlock(env, mdd_obj);
2033 static int mdd_object_sync(const struct lu_env *env, struct md_object *obj)
2035 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2037 if (mdd_object_exists(mdd_obj) == 0) {
2040 CERROR("%s: object "DFID" not found: rc = %d\n",
2041 mdd_obj_dev_name(mdd_obj),
2042 PFID(mdd_object_fid(mdd_obj)), rc);
2045 return dt_object_sync(env, mdd_object_child(mdd_obj),
2049 static int mdd_object_lock(const struct lu_env *env,
2050 struct md_object *obj,
2051 struct lustre_handle *lh,
2052 struct ldlm_enqueue_info *einfo,
2053 ldlm_policy_data_t *policy)
2055 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2056 return dt_object_lock(env, mdd_object_child(mdd_obj), lh,
2060 static int mdd_object_unlock(const struct lu_env *env,
2061 struct md_object *obj,
2062 struct ldlm_enqueue_info *einfo,
2063 ldlm_policy_data_t *policy)
2065 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2066 return dt_object_unlock(env, mdd_object_child(mdd_obj), einfo, policy);
2069 const struct md_object_operations mdd_obj_ops = {
2070 .moo_permission = mdd_permission,
2071 .moo_attr_get = mdd_attr_get,
2072 .moo_attr_set = mdd_attr_set,
2073 .moo_xattr_get = mdd_xattr_get,
2074 .moo_xattr_set = mdd_xattr_set,
2075 .moo_xattr_list = mdd_xattr_list,
2076 .moo_xattr_del = mdd_xattr_del,
2077 .moo_swap_layouts = mdd_swap_layouts,
2078 .moo_open = mdd_open,
2079 .moo_close = mdd_close,
2080 .moo_readpage = mdd_readpage,
2081 .moo_readlink = mdd_readlink,
2082 .moo_changelog = mdd_changelog,
2083 .moo_capa_get = mdd_capa_get,
2084 .moo_object_sync = mdd_object_sync,
2085 .moo_object_lock = mdd_object_lock,
2086 .moo_object_unlock = mdd_object_unlock,