1 /* -*- MODE: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
5 * Lustre Metadata Server (mdd) routines
7 * Copyright (C) 2006 Cluster File Systems, Inc.
8 * Author: Wang Di <wangdi@clusterfs.com>
10 * This file is part of the Lustre file system, http://www.lustre.org
11 * Lustre is a trademark of Cluster File Systems, Inc.
13 * You may have signed or agreed to another license before downloading
14 * this software. If so, you are bound by the terms and conditions
15 * of that agreement, and the following does not apply to you. See the
16 * LICENSE file included with this distribution for more information.
18 * If you did not agree to a different license, then this copy of Lustre
19 * is open source software; you can redistribute it and/or modify it
20 * under the terms of version 2 of the GNU General Public License as
21 * published by the Free Software Foundation.
23 * In either case, Lustre is distributed in the hope that it will be
24 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
25 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
26 * license text for more details.
29 # define EXPORT_SYMTAB
31 #define DEBUG_SUBSYSTEM S_MDS
33 #include <linux/module.h>
34 #include <linux/jbd.h>
36 #include <obd_class.h>
37 #include <obd_support.h>
38 #include <lprocfs_status.h>
39 /* fid_be_cpu(), fid_cpu_to_be(). */
40 #include <lustre_fid.h>
42 #include <linux/ldiskfs_fs.h>
43 #include <lustre_mds.h>
44 #include <lustre/lustre_idl.h>
46 #include "mdd_internal.h"
48 static struct lu_object_operations mdd_lu_obj_ops;
50 int mdd_la_get(const struct lu_env *env, struct mdd_object *obj,
51 struct lu_attr *la, struct lustre_capa *capa)
53 LASSERTF(mdd_object_exists(obj), "FID is "DFID"\n",
54 PFID(mdd_object_fid(obj)));
55 return mdo_attr_get(env, obj, la, capa);
58 static void mdd_flags_xlate(struct mdd_object *obj, __u32 flags)
60 obj->mod_flags &= ~(APPEND_OBJ|IMMUTE_OBJ);
62 if (flags & LUSTRE_APPEND_FL)
63 obj->mod_flags |= APPEND_OBJ;
65 if (flags & LUSTRE_IMMUTABLE_FL)
66 obj->mod_flags |= IMMUTE_OBJ;
69 struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
73 buf = &mdd_env_info(env)->mti_buf;
79 struct llog_cookie *mdd_max_cookie_get(const struct lu_env *env,
80 struct mdd_device *mdd)
82 struct mdd_thread_info *mti = mdd_env_info(env);
85 max_cookie_size = mdd_lov_cookiesize(env, mdd);
86 if (unlikely(mti->mti_max_cookie_size < max_cookie_size)) {
87 if (mti->mti_max_cookie)
88 OBD_FREE(mti->mti_max_cookie, mti->mti_max_cookie_size);
89 mti->mti_max_cookie = NULL;
90 mti->mti_max_cookie_size = 0;
92 if (unlikely(mti->mti_max_cookie == NULL)) {
93 OBD_ALLOC(mti->mti_max_cookie, max_cookie_size);
94 if (unlikely(mti->mti_max_cookie != NULL))
95 mti->mti_max_cookie_size = max_cookie_size;
97 return mti->mti_max_cookie;
100 struct lov_mds_md *mdd_max_lmm_get(const struct lu_env *env,
101 struct mdd_device *mdd)
103 struct mdd_thread_info *mti = mdd_env_info(env);
106 max_lmm_size = mdd_lov_mdsize(env, mdd);
107 if (unlikely(mti->mti_max_lmm_size < max_lmm_size)) {
108 if (mti->mti_max_lmm)
109 OBD_FREE(mti->mti_max_lmm, mti->mti_max_lmm_size);
110 mti->mti_max_lmm = NULL;
111 mti->mti_max_lmm_size = 0;
113 if (unlikely(mti->mti_max_lmm == NULL)) {
114 OBD_ALLOC(mti->mti_max_lmm, max_lmm_size);
115 if (unlikely(mti->mti_max_lmm != NULL))
116 mti->mti_max_lmm_size = max_lmm_size;
118 return mti->mti_max_lmm;
121 const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
122 const void *area, ssize_t len)
126 buf = &mdd_env_info(env)->mti_buf;
127 buf->lb_buf = (void *)area;
132 struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
134 struct mdd_thread_info *info;
136 info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
137 LASSERT(info != NULL);
141 struct lu_object *mdd_object_alloc(const struct lu_env *env,
142 const struct lu_object_header *hdr,
145 struct mdd_object *mdd_obj;
147 OBD_ALLOC_PTR(mdd_obj);
148 if (mdd_obj != NULL) {
151 o = mdd2lu_obj(mdd_obj);
152 lu_object_init(o, NULL, d);
153 mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
154 mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
155 mdd_obj->mod_count = 0;
156 o->lo_ops = &mdd_lu_obj_ops;
163 static int mdd_object_init(const struct lu_env *env, struct lu_object *o)
165 struct mdd_device *d = lu2mdd_dev(o->lo_dev);
166 struct lu_object *below;
167 struct lu_device *under;
170 under = &d->mdd_child->dd_lu_dev;
171 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
172 mdd_pdlock_init(lu2mdd_obj(o));
176 lu_object_add(o, below);
180 static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
182 if (lu_object_exists(o))
183 return mdd_get_flags(env, lu2mdd_obj(o));
188 static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
190 struct mdd_object *mdd = lu2mdd_obj(o);
196 static int mdd_object_print(const struct lu_env *env, void *cookie,
197 lu_printer_t p, const struct lu_object *o)
199 return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p", o);
202 /* orphan handling is here */
203 static void mdd_object_delete(const struct lu_env *env,
206 struct mdd_object *mdd_obj = lu2mdd_obj(o);
207 struct thandle *handle = NULL;
210 if (lu2mdd_dev(o->lo_dev)->mdd_orphans == NULL)
213 if (mdd_obj->mod_flags & ORPHAN_OBJ) {
214 mdd_txn_param_build(env, lu2mdd_dev(o->lo_dev),
215 MDD_TXN_INDEX_DELETE_OP);
216 handle = mdd_trans_start(env, lu2mdd_dev(o->lo_dev));
218 CERROR("Cannot get thandle\n");
220 mdd_write_lock(env, mdd_obj);
221 /* let's remove obj from the orphan list */
222 __mdd_orphan_del(env, mdd_obj, handle);
223 mdd_write_unlock(env, mdd_obj);
224 mdd_trans_stop(env, lu2mdd_dev(o->lo_dev),
230 static struct lu_object_operations mdd_lu_obj_ops = {
231 .loo_object_init = mdd_object_init,
232 .loo_object_start = mdd_object_start,
233 .loo_object_free = mdd_object_free,
234 .loo_object_print = mdd_object_print,
235 .loo_object_delete = mdd_object_delete
238 struct mdd_object *mdd_object_find(const struct lu_env *env,
239 struct mdd_device *d,
240 const struct lu_fid *f)
242 struct lu_object *o, *lo;
243 struct mdd_object *m;
246 o = lu_object_find(env, mdd2lu_dev(d)->ld_site, f);
248 m = (struct mdd_object *)o;
250 lo = lu_object_locate(o->lo_header, mdd2lu_dev(d)->ld_type);
251 /* remote object can't be located and should be put then */
253 lu_object_put(env, o);
259 int mdd_get_flags(const struct lu_env *env, struct mdd_object *obj)
261 struct lu_attr *la = &mdd_env_info(env)->mti_la;
265 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
267 mdd_flags_xlate(obj, la->la_flags);
271 /* get only inode attributes */
272 int mdd_iattr_get(const struct lu_env *env, struct mdd_object *mdd_obj,
278 if (ma->ma_valid & MA_INODE)
281 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr,
282 mdd_object_capa(env, mdd_obj));
284 ma->ma_valid |= MA_INODE;
288 static int mdd_get_default_md(struct mdd_object *mdd_obj,
289 struct lov_mds_md *lmm, int *size)
291 struct lov_desc *ldesc;
292 struct mdd_device *mdd = mdo2mdd(&mdd_obj->mod_obj);
295 ldesc = &mdd->mdd_obd_dev->u.mds.mds_lov_desc;
296 LASSERT(ldesc != NULL);
301 lmm->lmm_magic = LOV_MAGIC_V1;
302 lmm->lmm_object_gr = LOV_OBJECT_GROUP_DEFAULT;
303 lmm->lmm_pattern = ldesc->ld_pattern;
304 lmm->lmm_stripe_size = ldesc->ld_default_stripe_size;
305 lmm->lmm_stripe_count = ldesc->ld_default_stripe_count;
306 *size = sizeof(struct lov_mds_md);
308 RETURN(sizeof(struct lov_mds_md));
311 /* get lov EA only */
312 static int __mdd_lmm_get(const struct lu_env *env,
313 struct mdd_object *mdd_obj, struct md_attr *ma)
318 if (ma->ma_valid & MA_LOV)
321 rc = mdd_get_md(env, mdd_obj, ma->ma_lmm, &ma->ma_lmm_size,
324 if (rc == 0 && (ma->ma_need & MA_LOV_DEF)) {
325 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm,
330 ma->ma_valid |= MA_LOV;
336 int mdd_lmm_get_locked(const struct lu_env *env, struct mdd_object *mdd_obj,
342 mdd_read_lock(env, mdd_obj);
343 rc = __mdd_lmm_get(env, mdd_obj, ma);
344 mdd_read_unlock(env, mdd_obj);
349 static int __mdd_lmv_get(const struct lu_env *env,
350 struct mdd_object *mdd_obj, struct md_attr *ma)
355 if (ma->ma_valid & MA_LMV)
358 rc = mdd_get_md(env, mdd_obj, ma->ma_lmv, &ma->ma_lmv_size,
361 ma->ma_valid |= MA_LMV;
367 static int mdd_attr_get_internal(const struct lu_env *env,
368 struct mdd_object *mdd_obj,
374 if (ma->ma_need & MA_INODE)
375 rc = mdd_iattr_get(env, mdd_obj, ma);
377 if (rc == 0 && ma->ma_need & MA_LOV) {
378 if (S_ISREG(mdd_object_type(mdd_obj)) ||
379 S_ISDIR(mdd_object_type(mdd_obj)))
380 rc = __mdd_lmm_get(env, mdd_obj, ma);
382 if (rc == 0 && ma->ma_need & MA_LMV) {
383 if (S_ISDIR(mdd_object_type(mdd_obj)))
384 rc = __mdd_lmv_get(env, mdd_obj, ma);
386 #ifdef CONFIG_FS_POSIX_ACL
387 if (rc == 0 && ma->ma_need & MA_ACL_DEF) {
388 if (S_ISDIR(mdd_object_type(mdd_obj)))
389 rc = mdd_acl_def_get(env, mdd_obj, ma);
392 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64"\n",
397 int mdd_attr_get_internal_locked(const struct lu_env *env,
398 struct mdd_object *mdd_obj, struct md_attr *ma)
401 int needlock = ma->ma_need & (MA_LOV | MA_LMV | MA_ACL_DEF);
404 mdd_read_lock(env, mdd_obj);
405 rc = mdd_attr_get_internal(env, mdd_obj, ma);
407 mdd_read_unlock(env, mdd_obj);
412 * No permission check is needed.
414 static int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
417 struct mdd_object *mdd_obj = md2mdd_obj(obj);
421 rc = mdd_attr_get_internal_locked(env, mdd_obj, ma);
426 * No permission check is needed.
428 static int mdd_xattr_get(const struct lu_env *env,
429 struct md_object *obj, struct lu_buf *buf,
432 struct mdd_object *mdd_obj = md2mdd_obj(obj);
437 LASSERT(mdd_object_exists(mdd_obj));
439 mdd_read_lock(env, mdd_obj);
440 rc = mdo_xattr_get(env, mdd_obj, buf, name,
441 mdd_object_capa(env, mdd_obj));
442 mdd_read_unlock(env, mdd_obj);
448 * Permission check is done when open,
449 * no need check again.
451 static int mdd_readlink(const struct lu_env *env, struct md_object *obj,
454 struct mdd_object *mdd_obj = md2mdd_obj(obj);
455 struct dt_object *next;
460 LASSERT(mdd_object_exists(mdd_obj));
462 next = mdd_object_child(mdd_obj);
463 mdd_read_lock(env, mdd_obj);
464 rc = next->do_body_ops->dbo_read(env, next, buf, &pos,
465 mdd_object_capa(env, mdd_obj));
466 mdd_read_unlock(env, mdd_obj);
471 * No permission check is needed.
473 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
476 struct mdd_object *mdd_obj = md2mdd_obj(obj);
481 mdd_read_lock(env, mdd_obj);
482 rc = mdo_xattr_list(env, mdd_obj, buf, mdd_object_capa(env, mdd_obj));
483 mdd_read_unlock(env, mdd_obj);
488 int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p,
489 struct mdd_object *c, struct md_attr *ma,
490 struct thandle *handle)
492 struct lu_attr *attr = &ma->ma_attr;
493 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
497 if (!mdd_object_exists(c)) {
498 struct dt_object *next = mdd_object_child(c);
501 /* @hint will be initialized by underlying device. */
502 next->do_ops->do_ah_init(env, hint,
503 p ? mdd_object_child(p) : NULL,
504 attr->la_mode & S_IFMT);
505 rc = mdo_create_obj(env, c, attr, hint, handle);
506 LASSERT(ergo(rc == 0, mdd_object_exists(c)));
514 int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj,
515 const struct lu_attr *attr, struct thandle *handle,
521 rc = mdo_attr_set(env, obj, attr, handle, mdd_object_capa(env, obj));
522 #ifdef CONFIG_FS_POSIX_ACL
523 if (!rc && (attr->la_valid & LA_MODE) && needacl)
524 rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
529 int mdd_attr_set_internal_locked(const struct lu_env *env,
530 struct mdd_object *o,
531 const struct lu_attr *attr,
532 struct thandle *handle, int needacl)
537 needacl = needacl && (attr->la_valid & LA_MODE);
540 mdd_write_lock(env, o);
542 rc = mdd_attr_set_internal(env, o, attr, handle, needacl);
545 mdd_write_unlock(env, o);
549 static int __mdd_xattr_set(const struct lu_env *env, struct mdd_object *obj,
550 const struct lu_buf *buf, const char *name,
551 int fl, struct thandle *handle)
553 struct lustre_capa *capa = mdd_object_capa(env, obj);
557 if (buf->lb_buf && buf->lb_len > 0)
558 rc = mdo_xattr_set(env, obj, buf, name, 0, handle, capa);
559 else if (buf->lb_buf == NULL && buf->lb_len == 0)
560 rc = mdo_xattr_del(env, obj, name, handle, capa);
566 * This gives the same functionality as the code between
567 * sys_chmod and inode_setattr
568 * chown_common and inode_setattr
569 * utimes and inode_setattr
570 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
572 static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
573 struct lu_attr *la, const struct md_attr *ma)
575 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
576 struct md_ucred *uc = md_ucred(env);
583 /* Do not permit change file type */
584 if (la->la_valid & LA_TYPE)
587 /* They should not be processed by setattr */
588 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
591 /* This is only for set ctime when rename's source is on remote MDS. */
592 if (unlikely(la->la_valid == LA_CTIME)) {
593 rc = mdd_may_delete(env, NULL, obj, (struct md_attr *)ma, 1, 0);
597 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
601 if (la->la_valid == LA_ATIME) {
602 /* This is atime only set for read atime update on close. */
603 if (la->la_atime <= tmp_la->la_atime +
604 mdd_obj2mdd_dev(obj)->mdd_atime_diff)
605 la->la_valid &= ~LA_ATIME;
609 /* Check if flags change. */
610 if (la->la_valid & LA_FLAGS) {
611 unsigned int oldflags = 0;
612 unsigned int newflags = la->la_flags &
613 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
615 if ((uc->mu_fsuid != tmp_la->la_uid) &&
616 !mdd_capable(uc, CAP_FOWNER))
619 /* XXX: the IMMUTABLE and APPEND_ONLY flags can
620 * only be changed by the relevant capability. */
621 if (mdd_is_immutable(obj))
622 oldflags |= LUSTRE_IMMUTABLE_FL;
623 if (mdd_is_append(obj))
624 oldflags |= LUSTRE_APPEND_FL;
625 if ((oldflags ^ newflags) &&
626 !mdd_capable(uc, CAP_LINUX_IMMUTABLE))
629 if (!S_ISDIR(tmp_la->la_mode))
630 la->la_flags &= ~LUSTRE_DIRSYNC_FL;
633 if ((mdd_is_immutable(obj) || mdd_is_append(obj)) &&
634 (la->la_valid & ~LA_FLAGS) &&
635 !(ma->ma_attr_flags & MDS_PERM_BYPASS))
638 /* Check for setting the obj time. */
639 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
640 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
641 if ((uc->mu_fsuid != tmp_la->la_uid) &&
642 !mdd_capable(uc, CAP_FOWNER)) {
643 rc = mdd_permission_internal_locked(env, obj, tmp_la,
650 /* Make sure a caller can chmod. */
651 if (la->la_valid & LA_MODE) {
653 * Bypass la_vaild == LA_MODE,
654 * this is for changing file with SUID or SGID.
656 if ((la->la_valid & ~LA_MODE) &&
657 (uc->mu_fsuid != tmp_la->la_uid) &&
658 !mdd_capable(uc, CAP_FOWNER))
661 if (la->la_mode == (umode_t) -1)
662 la->la_mode = tmp_la->la_mode;
664 la->la_mode = (la->la_mode & S_IALLUGO) |
665 (tmp_la->la_mode & ~S_IALLUGO);
667 /* Also check the setgid bit! */
668 if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ? la->la_gid :
669 tmp_la->la_gid) && !mdd_capable(uc, CAP_FSETID))
670 la->la_mode &= ~S_ISGID;
672 la->la_mode = tmp_la->la_mode;
675 /* Make sure a caller can chown. */
676 if (la->la_valid & LA_UID) {
677 if (la->la_uid == (uid_t) -1)
678 la->la_uid = tmp_la->la_uid;
679 if (((uc->mu_fsuid != tmp_la->la_uid) ||
680 (la->la_uid != tmp_la->la_uid)) &&
681 !mdd_capable(uc, CAP_CHOWN))
685 * If the user or group of a non-directory has been
686 * changed by a non-root user, remove the setuid bit.
687 * 19981026 David C Niemi <niemi@tux.org>
689 * Changed this to apply to all users, including root,
690 * to avoid some races. This is the behavior we had in
691 * 2.0. The check for non-root was definitely wrong
692 * for 2.2 anyway, as it should have been using
693 * CAP_FSETID rather than fsuid -- 19990830 SD.
695 if (((tmp_la->la_mode & S_ISUID) == S_ISUID) &&
696 !S_ISDIR(tmp_la->la_mode)) {
697 la->la_mode &= ~S_ISUID;
698 la->la_valid |= LA_MODE;
702 /* Make sure caller can chgrp. */
703 if (la->la_valid & LA_GID) {
704 if (la->la_gid == (gid_t) -1)
705 la->la_gid = tmp_la->la_gid;
706 if (((uc->mu_fsuid != tmp_la->la_uid) ||
707 ((la->la_gid != tmp_la->la_gid) &&
708 !lustre_in_group_p(uc, la->la_gid))) &&
709 !mdd_capable(uc, CAP_CHOWN))
713 * Likewise, if the user or group of a non-directory
714 * has been changed by a non-root user, remove the
715 * setgid bit UNLESS there is no group execute bit
716 * (this would be a file marked for mandatory
717 * locking). 19981026 David C Niemi <niemi@tux.org>
719 * Removed the fsuid check (see the comment above) --
722 if (((tmp_la->la_mode & (S_ISGID | S_IXGRP)) ==
723 (S_ISGID | S_IXGRP)) && !S_ISDIR(tmp_la->la_mode)) {
724 la->la_mode &= ~S_ISGID;
725 la->la_valid |= LA_MODE;
729 /* For truncate (or setsize), we should have MAY_WRITE perm */
730 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
731 if (!((la->la_valid & MDS_OPEN_OWNEROVERRIDE) &&
732 (uc->mu_fsuid == tmp_la->la_uid)) &&
733 !(ma->ma_attr_flags & MDS_PERM_BYPASS)) {
734 rc = mdd_permission_internal_locked(env, obj, tmp_la,
740 /* For the "Size-on-MDS" setattr update, merge coming
741 * attributes with the set in the inode. BUG 10641 */
742 if ((la->la_valid & LA_ATIME) &&
743 (la->la_atime <= tmp_la->la_atime))
744 la->la_valid &= ~LA_ATIME;
746 /* OST attributes do not have a priority over MDS attributes,
747 * so drop times if ctime is equal. */
748 if ((la->la_valid & LA_CTIME) &&
749 (la->la_ctime <= tmp_la->la_ctime))
750 la->la_valid &= ~(LA_MTIME | LA_CTIME);
751 } else if (la->la_valid & LA_CTIME) {
752 /* The pure setattr, it has the priority over what is already
753 * set, do not drop it if ctime is equal. */
754 if (la->la_ctime < tmp_la->la_ctime)
755 la->la_valid &= ~(LA_ATIME | LA_MTIME | LA_CTIME);
761 /* set attr and LOV EA at once, return updated attr */
762 static int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
763 const struct md_attr *ma)
765 struct mdd_object *mdd_obj = md2mdd_obj(obj);
766 struct mdd_device *mdd = mdo2mdd(obj);
767 struct thandle *handle;
768 struct lov_mds_md *lmm = NULL;
769 struct llog_cookie *logcookies = NULL;
770 int rc, lmm_size = 0, cookie_size = 0;
771 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
774 mdd_txn_param_build(env, mdd, MDD_TXN_ATTR_SET_OP);
775 handle = mdd_trans_start(env, mdd);
777 RETURN(PTR_ERR(handle));
778 /*TODO: add lock here*/
779 /* start a log jounal handle if needed */
780 if (S_ISREG(mdd_object_type(mdd_obj)) &&
781 ma->ma_attr.la_valid & (LA_UID | LA_GID)) {
782 lmm_size = mdd_lov_mdsize(env, mdd);
783 lmm = mdd_max_lmm_get(env, mdd);
785 GOTO(cleanup, rc = -ENOMEM);
787 rc = mdd_get_md_locked(env, mdd_obj, lmm, &lmm_size,
794 if (ma->ma_attr.la_valid & (ATTR_MTIME | ATTR_CTIME))
795 CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n",
796 ma->ma_attr.la_mtime, ma->ma_attr.la_ctime);
798 *la_copy = ma->ma_attr;
799 rc = mdd_fix_attr(env, mdd_obj, la_copy, ma);
803 if (la_copy->la_valid & LA_FLAGS) {
804 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
807 mdd_flags_xlate(mdd_obj, la_copy->la_flags);
808 } else if (la_copy->la_valid) { /* setattr */
809 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
811 /* journal chown/chgrp in llog, just like unlink */
812 if (rc == 0 && lmm_size){
813 cookie_size = mdd_lov_cookiesize(env, mdd);
814 logcookies = mdd_max_cookie_get(env, mdd);
815 if (logcookies == NULL)
816 GOTO(cleanup, rc = -ENOMEM);
818 if (mdd_setattr_log(env, mdd, ma, lmm, lmm_size,
819 logcookies, cookie_size) <= 0)
824 if (rc == 0 && ma->ma_valid & MA_LOV) {
827 mode = mdd_object_type(mdd_obj);
828 if (S_ISREG(mode) || S_ISDIR(mode)) {
829 rc = mdd_lsm_sanity_check(env, mdd_obj);
833 rc = mdd_lov_set_md(env, NULL, mdd_obj, ma->ma_lmm,
834 ma->ma_lmm_size, handle, 1);
839 mdd_trans_stop(env, mdd, rc, handle);
840 if (rc == 0 && (lmm != NULL && lmm_size > 0 )) {
841 /*set obd attr, if needed*/
842 rc = mdd_lov_setattr_async(env, mdd_obj, lmm, lmm_size,
848 int mdd_xattr_set_txn(const struct lu_env *env, struct mdd_object *obj,
849 const struct lu_buf *buf, const char *name, int fl,
850 struct thandle *handle)
855 mdd_write_lock(env, obj);
856 rc = __mdd_xattr_set(env, obj, buf, name, fl, handle);
857 mdd_write_unlock(env, obj);
862 static int mdd_xattr_sanity_check(const struct lu_env *env,
863 struct mdd_object *obj)
865 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
866 struct md_ucred *uc = md_ucred(env);
870 if (mdd_is_immutable(obj) || mdd_is_append(obj))
873 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
877 if ((uc->mu_fsuid != tmp_la->la_uid) && !mdd_capable(uc, CAP_FOWNER))
883 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
884 const struct lu_buf *buf, const char *name, int fl)
886 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
887 struct mdd_object *mdd_obj = md2mdd_obj(obj);
888 struct mdd_device *mdd = mdo2mdd(obj);
889 struct thandle *handle;
893 rc = mdd_xattr_sanity_check(env, mdd_obj);
897 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
898 handle = mdd_trans_start(env, mdd);
900 RETURN(PTR_ERR(handle));
902 rc = mdd_xattr_set_txn(env, md2mdd_obj(obj), buf, name,
905 la_copy->la_ctime = cfs_time_current_sec();
906 la_copy->la_valid = LA_CTIME;
907 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
910 mdd_trans_stop(env, mdd, rc, handle);
915 int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
918 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
919 struct mdd_object *mdd_obj = md2mdd_obj(obj);
920 struct mdd_device *mdd = mdo2mdd(obj);
921 struct thandle *handle;
925 rc = mdd_xattr_sanity_check(env, mdd_obj);
929 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
930 handle = mdd_trans_start(env, mdd);
932 RETURN(PTR_ERR(handle));
934 mdd_write_lock(env, mdd_obj);
935 rc = mdo_xattr_del(env, md2mdd_obj(obj), name, handle,
936 mdd_object_capa(env, mdd_obj));
937 mdd_write_unlock(env, mdd_obj);
939 la_copy->la_ctime = cfs_time_current_sec();
940 la_copy->la_valid = LA_CTIME;
941 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
945 mdd_trans_stop(env, mdd, rc, handle);
951 static int mdd_ref_del(const struct lu_env *env, struct md_object *obj,
954 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
955 struct mdd_object *mdd_obj = md2mdd_obj(obj);
956 struct mdd_device *mdd = mdo2mdd(obj);
957 struct thandle *handle;
962 * Check -ENOENT early here because we need to get object type
963 * to calculate credits before transaction start
965 if (!mdd_object_exists(mdd_obj))
968 LASSERT(mdd_object_exists(mdd_obj) > 0);
970 rc = mdd_log_txn_param_build(env, obj, ma, MDD_TXN_UNLINK_OP);
974 handle = mdd_trans_start(env, mdd);
978 mdd_write_lock(env, mdd_obj);
980 rc = mdd_unlink_sanity_check(env, NULL, mdd_obj, ma);
984 mdo_ref_del(env, mdd_obj, handle);
986 if (S_ISDIR(lu_object_attr(&obj->mo_lu))) {
988 mdo_ref_del(env, mdd_obj, handle);
991 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
992 la_copy->la_ctime = ma->ma_attr.la_ctime;
994 la_copy->la_valid = LA_CTIME;
995 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 0);
999 rc = mdd_finish_unlink(env, mdd_obj, ma, handle);
1003 mdd_write_unlock(env, mdd_obj);
1004 mdd_trans_stop(env, mdd, rc, handle);
1008 /* partial operation */
1009 static int mdd_oc_sanity_check(const struct lu_env *env,
1010 struct mdd_object *obj,
1016 switch (ma->ma_attr.la_mode & S_IFMT) {
1033 static int mdd_object_create(const struct lu_env *env,
1034 struct md_object *obj,
1035 const struct md_op_spec *spec,
1039 struct mdd_device *mdd = mdo2mdd(obj);
1040 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1041 const struct lu_fid *pfid = spec->u.sp_pfid;
1042 struct thandle *handle;
1046 mdd_txn_param_build(env, mdd, MDD_TXN_OBJECT_CREATE_OP);
1047 handle = mdd_trans_start(env, mdd);
1049 RETURN(PTR_ERR(handle));
1051 mdd_write_lock(env, mdd_obj);
1052 rc = mdd_oc_sanity_check(env, mdd_obj, ma);
1056 rc = mdd_object_create_internal(env, NULL, mdd_obj, ma, handle);
1060 if (spec->sp_cr_flags & MDS_CREATE_SLAVE_OBJ) {
1061 /* If creating the slave object, set slave EA here. */
1062 int lmv_size = spec->u.sp_ea.eadatalen;
1063 struct lmv_stripe_md *lmv;
1065 lmv = (struct lmv_stripe_md *)spec->u.sp_ea.eadata;
1066 LASSERT(lmv != NULL && lmv_size > 0);
1068 rc = __mdd_xattr_set(env, mdd_obj,
1069 mdd_buf_get_const(env, lmv, lmv_size),
1070 MDS_LMV_MD_NAME, 0, handle);
1074 rc = mdd_attr_set_internal(env, mdd_obj, &ma->ma_attr, handle, 0);
1076 #ifdef CONFIG_FS_POSIX_ACL
1077 if (spec->sp_cr_flags & MDS_CREATE_RMT_ACL) {
1078 struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
1080 buf->lb_buf = (void *)spec->u.sp_ea.eadata;
1081 buf->lb_len = spec->u.sp_ea.eadatalen;
1082 if ((buf->lb_len > 0) && (buf->lb_buf != NULL)) {
1083 rc = __mdd_acl_init(env, mdd_obj, buf,
1084 &ma->ma_attr.la_mode,
1089 ma->ma_attr.la_valid |= LA_MODE;
1092 pfid = spec->u.sp_ea.fid;
1095 rc = mdd_object_initialize(env, pfid, mdd_obj, ma, handle);
1099 mdd_write_unlock(env, mdd_obj);
1101 rc = mdd_attr_get_internal_locked(env, mdd_obj, ma);
1103 mdd_trans_stop(env, mdd, rc, handle);
1108 static int mdd_ref_add(const struct lu_env *env, struct md_object *obj,
1109 const struct md_attr *ma)
1111 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
1112 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1113 struct mdd_device *mdd = mdo2mdd(obj);
1114 struct thandle *handle;
1118 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
1119 handle = mdd_trans_start(env, mdd);
1123 mdd_write_lock(env, mdd_obj);
1124 rc = mdd_link_sanity_check(env, NULL, NULL, mdd_obj);
1126 mdo_ref_add(env, mdd_obj, handle);
1127 mdd_write_unlock(env, mdd_obj);
1129 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1130 la_copy->la_ctime = ma->ma_attr.la_ctime;
1132 la_copy->la_valid = LA_CTIME;
1133 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
1136 mdd_trans_stop(env, mdd, 0, handle);
1142 * do NOT or the MAY_*'s, you'll get the weakest
1144 int accmode(const struct lu_env *env, struct lu_attr *la, int flags)
1148 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1149 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1150 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1151 * owner can write to a file even if it is marked readonly to hide
1152 * its brokenness. (bug 5781) */
1153 if (flags & MDS_OPEN_OWNEROVERRIDE) {
1154 struct md_ucred *uc = md_ucred(env);
1156 if ((uc == NULL) || (uc->mu_valid == UCRED_INIT) ||
1157 (la->la_uid == uc->mu_fsuid))
1161 if (flags & FMODE_READ)
1163 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1165 if (flags & MDS_FMODE_EXEC)
1170 static int mdd_open_sanity_check(const struct lu_env *env,
1171 struct mdd_object *obj, int flag)
1173 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
1178 if (mdd_is_dead_obj(obj))
1181 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
1185 if (S_ISLNK(tmp_la->la_mode))
1188 mode = accmode(env, tmp_la, flag);
1190 if (S_ISDIR(tmp_la->la_mode) && (mode & MAY_WRITE))
1193 if (!(flag & MDS_OPEN_CREATED)) {
1194 rc = mdd_permission_internal(env, obj, tmp_la, mode);
1199 if (S_ISFIFO(tmp_la->la_mode) || S_ISSOCK(tmp_la->la_mode) ||
1200 S_ISBLK(tmp_la->la_mode) || S_ISCHR(tmp_la->la_mode))
1201 flag &= ~MDS_OPEN_TRUNC;
1203 /* For writing append-only file must open it with append mode. */
1204 if (mdd_is_append(obj)) {
1205 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1207 if (flag & MDS_OPEN_TRUNC)
1213 * Now, flag -- O_NOATIME does not be packed by client.
1215 if (flag & O_NOATIME) {
1216 struct md_ucred *uc = md_ucred(env);
1218 if (uc && ((uc->mu_valid == UCRED_OLD) ||
1219 (uc->mu_valid == UCRED_NEW)) &&
1220 (uc->mu_fsuid != tmp_la->la_uid) &&
1221 !mdd_capable(uc, CAP_FOWNER))
1229 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1232 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1235 mdd_write_lock(env, mdd_obj);
1237 rc = mdd_open_sanity_check(env, mdd_obj, flags);
1239 mdd_obj->mod_count++;
1241 mdd_write_unlock(env, mdd_obj);
1245 /* return md_attr back,
1246 * if it is last unlink then return lov ea + llog cookie*/
1247 int mdd_object_kill(const struct lu_env *env, struct mdd_object *obj,
1253 if (S_ISREG(mdd_object_type(obj))) {
1254 /* Return LOV & COOKIES unconditionally here. We clean evth up.
1255 * Caller must be ready for that. */
1256 rc = __mdd_lmm_get(env, obj, ma);
1257 if ((ma->ma_valid & MA_LOV))
1258 rc = mdd_unlink_log(env, mdo2mdd(&obj->mod_obj),
1265 * No permission check is needed.
1267 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1271 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1272 struct thandle *handle;
1275 rc = mdd_log_txn_param_build(env, obj, ma, MDD_TXN_UNLINK_OP);
1278 handle = mdd_trans_start(env, mdo2mdd(obj));
1280 RETURN(PTR_ERR(handle));
1282 mdd_write_lock(env, mdd_obj);
1283 /* release open count */
1284 mdd_obj->mod_count --;
1286 rc = mdd_iattr_get(env, mdd_obj, ma);
1287 if (rc == 0 && mdd_obj->mod_count == 0 && ma->ma_attr.la_nlink == 0)
1288 rc = mdd_object_kill(env, mdd_obj, ma);
1290 ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
1292 mdd_write_unlock(env, mdd_obj);
1293 mdd_trans_stop(env, mdo2mdd(obj), rc, handle);
1298 * Permission check is done when open,
1299 * no need check again.
1301 static int mdd_readpage_sanity_check(const struct lu_env *env,
1302 struct mdd_object *obj)
1304 struct dt_object *next = mdd_object_child(obj);
1308 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
1316 static int mdd_dir_page_build(const struct lu_env *env, int first,
1317 void *area, int nob, struct dt_it_ops *iops,
1318 struct dt_it *it, __u64 *start, __u64 *end,
1319 struct lu_dirent **last)
1321 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
1322 struct mdd_thread_info *info = mdd_env_info(env);
1323 struct lu_fid_pack *pack = &info->mti_pack;
1325 struct lu_dirent *ent;
1328 memset(area, 0, sizeof (struct lu_dirpage));
1329 area += sizeof (struct lu_dirpage);
1330 nob -= sizeof (struct lu_dirpage);
1333 LASSERT(nob > sizeof *ent);
1343 name = (char *)iops->key(env, it);
1344 len = iops->key_size(env, it);
1346 pack = (struct lu_fid_pack *)iops->rec(env, it);
1347 result = fid_unpack(pack, fid);
1351 recsize = (sizeof(*ent) + len + 3) & ~3;
1352 hash = iops->store(env, it);
1355 CDEBUG(D_INFO, "%p %p %d "DFID": "LPU64" (%d) \"%*.*s\"\n",
1356 name, ent, nob, PFID(fid), hash, len, len, len, name);
1358 if (nob >= recsize) {
1359 ent->lde_fid = *fid;
1360 fid_cpu_to_le(&ent->lde_fid, &ent->lde_fid);
1361 ent->lde_hash = hash;
1362 ent->lde_namelen = cpu_to_le16(len);
1363 ent->lde_reclen = cpu_to_le16(recsize);
1364 memcpy(ent->lde_name, name, len);
1365 if (first && ent == area)
1368 ent = (void *)ent + recsize;
1370 result = iops->next(env, it);
1373 * record doesn't fit into page, enlarge previous one.
1375 LASSERT(*last != NULL);
1376 (*last)->lde_reclen =
1377 cpu_to_le16(le16_to_cpu((*last)->lde_reclen) +
1381 } while (result == 0);
1386 static int __mdd_readpage(const struct lu_env *env, struct mdd_object *obj,
1387 const struct lu_rdpg *rdpg)
1390 struct dt_object *next = mdd_object_child(obj);
1391 struct dt_it_ops *iops;
1393 struct lu_dirent *last;
1400 LASSERT(rdpg->rp_pages != NULL);
1401 LASSERT(next->do_index_ops != NULL);
1403 if (rdpg->rp_count <= 0)
1407 * iterate through directory and fill pages from @rdpg
1409 iops = &next->do_index_ops->dio_it;
1410 it = iops->init(env, next, 0, mdd_object_capa(env, obj));
1414 rc = iops->load(env, it, rdpg->rp_hash);
1418 * Iterator didn't find record with exactly the key requested.
1420 * It is currently either
1422 * - positioned above record with key less than
1423 * requested---skip it.
1425 * - or not positioned at all (is in IAM_IT_SKEWED
1426 * state)---position it on the next item.
1428 rc = iops->next(env, it);
1433 * At this point and across for-loop:
1435 * rc == 0 -> ok, proceed.
1436 * rc > 0 -> end of directory.
1439 for (i = 0, nob = rdpg->rp_count; rc == 0 && nob > 0;
1440 i++, nob -= CFS_PAGE_SIZE) {
1441 LASSERT(i < rdpg->rp_npages);
1442 pg = rdpg->rp_pages[i];
1443 rc = mdd_dir_page_build(env, !i, cfs_kmap(pg),
1444 min_t(int, nob, CFS_PAGE_SIZE), iops,
1445 it, &hash_start, &hash_end, &last);
1446 if (rc != 0 || i == rdpg->rp_npages - 1)
1447 last->lde_reclen = 0;
1454 hash_end = DIR_END_OFF;
1458 struct lu_dirpage *dp;
1460 dp = cfs_kmap(rdpg->rp_pages[0]);
1461 dp->ldp_hash_start = rdpg->rp_hash;
1462 dp->ldp_hash_end = hash_end;
1465 * No pages were processed, mark this.
1467 dp->ldp_flags |= LDF_EMPTY;
1468 dp->ldp_flags = cpu_to_le16(dp->ldp_flags);
1469 cfs_kunmap(rdpg->rp_pages[0]);
1472 iops->fini(env, it);
1477 static int mdd_readpage(const struct lu_env *env, struct md_object *obj,
1478 const struct lu_rdpg *rdpg)
1480 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1484 LASSERT(mdd_object_exists(mdd_obj));
1486 mdd_read_lock(env, mdd_obj);
1487 rc = mdd_readpage_sanity_check(env, mdd_obj);
1489 GOTO(out_unlock, rc);
1491 if (mdd_is_dead_obj(mdd_obj)) {
1493 struct lu_dirpage *dp;
1496 * According to POSIX, please do not return any entry to client:
1497 * even dot and dotdot should not be returned.
1499 CWARN("readdir from dead object: "DFID"\n",
1500 PFID(mdd_object_fid(mdd_obj)));
1502 if (rdpg->rp_count <= 0)
1503 GOTO(out_unlock, rc = -EFAULT);
1504 LASSERT(rdpg->rp_pages != NULL);
1506 pg = rdpg->rp_pages[0];
1507 dp = (struct lu_dirpage*)cfs_kmap(pg);
1508 memset(dp, 0 , sizeof(struct lu_dirpage));
1509 dp->ldp_hash_start = rdpg->rp_hash;
1510 dp->ldp_hash_end = DIR_END_OFF;
1511 dp->ldp_flags |= LDF_EMPTY;
1512 dp->ldp_flags = cpu_to_le16(dp->ldp_flags);
1514 GOTO(out_unlock, rc = 0);
1517 rc = __mdd_readpage(env, mdd_obj, rdpg);
1521 mdd_read_unlock(env, mdd_obj);
1525 struct md_object_operations mdd_obj_ops = {
1526 .moo_permission = mdd_permission,
1527 .moo_attr_get = mdd_attr_get,
1528 .moo_attr_set = mdd_attr_set,
1529 .moo_xattr_get = mdd_xattr_get,
1530 .moo_xattr_set = mdd_xattr_set,
1531 .moo_xattr_list = mdd_xattr_list,
1532 .moo_xattr_del = mdd_xattr_del,
1533 .moo_object_create = mdd_object_create,
1534 .moo_ref_add = mdd_ref_add,
1535 .moo_ref_del = mdd_ref_del,
1536 .moo_open = mdd_open,
1537 .moo_close = mdd_close,
1538 .moo_readpage = mdd_readpage,
1539 .moo_readlink = mdd_readlink,
1540 .moo_capa_get = mdd_capa_get