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 <lustre_ver.h>
38 #include <obd_support.h>
39 #include <lprocfs_status.h>
40 /* fid_be_cpu(), fid_cpu_to_be(). */
41 #include <lustre_fid.h>
43 #include <linux/ldiskfs_fs.h>
44 #include <lustre_mds.h>
45 #include <lustre/lustre_idl.h>
47 #include "mdd_internal.h"
49 static struct lu_object_operations mdd_lu_obj_ops;
51 static inline void mdd_set_dead_obj(struct mdd_object *obj)
54 obj->mod_flags |= DEAD_OBJ;
57 int mdd_la_get(const struct lu_env *env, struct mdd_object *obj,
58 struct lu_attr *la, struct lustre_capa *capa)
60 struct dt_object *next = mdd_object_child(obj);
61 LASSERT(lu_object_exists(mdd2lu_obj(obj)));
62 return next->do_ops->do_attr_get(env, next, la, capa);
65 static void mdd_flags_xlate(struct mdd_object *obj, __u32 flags)
67 obj->mod_flags &= ~(APPEND_OBJ|IMMUTE_OBJ);
69 if (flags & LUSTRE_APPEND_FL)
70 obj->mod_flags |= APPEND_OBJ;
72 if (flags & LUSTRE_IMMUTABLE_FL)
73 obj->mod_flags |= IMMUTE_OBJ;
76 struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
80 buf = &mdd_env_info(env)->mti_buf;
86 const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
87 const void *area, ssize_t len)
91 buf = &mdd_env_info(env)->mti_buf;
92 buf->lb_buf = (void *)area;
97 struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
99 struct mdd_thread_info *info;
101 info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
102 LASSERT(info != NULL);
106 struct lu_object *mdd_object_alloc(const struct lu_env *env,
107 const struct lu_object_header *hdr,
110 struct mdd_object *mdd_obj;
112 OBD_ALLOC_PTR(mdd_obj);
113 if (mdd_obj != NULL) {
116 o = mdd2lu_obj(mdd_obj);
117 lu_object_init(o, NULL, d);
118 mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
119 mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
120 mdd_obj->mod_count = 0;
121 o->lo_ops = &mdd_lu_obj_ops;
128 static int mdd_object_init(const struct lu_env *env, struct lu_object *o)
130 struct mdd_device *d = lu2mdd_dev(o->lo_dev);
131 struct lu_object *below;
132 struct lu_device *under;
135 under = &d->mdd_child->dd_lu_dev;
136 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
137 mdd_pdlock_init(lu2mdd_obj(o));
141 lu_object_add(o, below);
145 static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
147 if (lu_object_exists(o))
148 return mdd_get_flags(env, lu2mdd_obj(o));
153 static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
155 struct mdd_object *mdd = lu2mdd_obj(o);
161 static int mdd_object_print(const struct lu_env *env, void *cookie,
162 lu_printer_t p, const struct lu_object *o)
164 return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p", o);
167 /* orphan handling is here */
168 static void mdd_object_delete(const struct lu_env *env,
171 struct mdd_object *mdd_obj = lu2mdd_obj(o);
172 struct thandle *handle = NULL;
175 if (lu2mdd_dev(o->lo_dev)->mdd_orphans == NULL)
178 if (mdd_obj->mod_flags & ORPHAN_OBJ) {
179 mdd_txn_param_build(env, lu2mdd_dev(o->lo_dev),
180 MDD_TXN_INDEX_DELETE_OP);
181 handle = mdd_trans_start(env, lu2mdd_dev(o->lo_dev));
183 CERROR("Cannot get thandle\n");
185 mdd_write_lock(env, mdd_obj);
186 /* let's remove obj from the orphan list */
187 __mdd_orphan_del(env, mdd_obj, handle);
188 mdd_write_unlock(env, mdd_obj);
189 mdd_trans_stop(env, lu2mdd_dev(o->lo_dev),
195 static 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,
200 .loo_object_delete = mdd_object_delete
203 struct mdd_object *mdd_object_find(const struct lu_env *env,
204 struct mdd_device *d,
205 const struct lu_fid *f)
207 struct lu_object *o, *lo;
208 struct mdd_object *m;
211 o = lu_object_find(env, mdd2lu_dev(d)->ld_site, f);
213 m = (struct mdd_object *)o;
215 lo = lu_object_locate(o->lo_header, mdd2lu_dev(d)->ld_type);
216 /* remote object can't be located and should be put then */
218 lu_object_put(env, o);
224 int mdd_get_flags(const struct lu_env *env, struct mdd_object *obj)
226 struct lu_attr *la = &mdd_env_info(env)->mti_la;
230 mdd_read_lock(env, obj);
231 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
232 mdd_read_unlock(env, obj);
234 mdd_flags_xlate(obj, la->la_flags);
238 void mdd_ref_add_internal(const struct lu_env *env, struct mdd_object *obj,
239 struct thandle *handle)
241 struct dt_object *next;
243 LASSERT(lu_object_exists(mdd2lu_obj(obj)));
244 next = mdd_object_child(obj);
245 next->do_ops->do_ref_add(env, next, handle);
248 void mdd_ref_del_internal(const struct lu_env *env, struct mdd_object *obj,
249 struct thandle *handle)
251 struct dt_object *next = mdd_object_child(obj);
254 LASSERT(lu_object_exists(mdd2lu_obj(obj)));
256 next->do_ops->do_ref_del(env, next, handle);
260 /* get only inode attributes */
261 int mdd_iattr_get(const struct lu_env *env, struct mdd_object *mdd_obj,
267 if (ma->ma_valid & MA_INODE)
270 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr,
271 mdd_object_capa(env, mdd_obj));
273 ma->ma_valid |= MA_INODE;
277 /* get lov EA only */
278 static int __mdd_lmm_get(const struct lu_env *env,
279 struct mdd_object *mdd_obj, struct md_attr *ma)
284 if (ma->ma_valid & MA_LOV)
287 lmm_size = ma->ma_lmm_size;
288 rc = mdd_get_md(env, mdd_obj, ma->ma_lmm, &lmm_size,
291 ma->ma_valid |= MA_LOV;
292 ma->ma_lmm_size = lmm_size;
298 int mdd_lmm_get_locked(const struct lu_env *env, struct mdd_object *mdd_obj,
304 mdd_read_lock(env, mdd_obj);
305 rc = __mdd_lmm_get(env, mdd_obj, ma);
306 mdd_read_unlock(env, mdd_obj);
311 static int __mdd_lmv_get(const struct lu_env *env,
312 struct mdd_object *mdd_obj, struct md_attr *ma)
316 if (ma->ma_valid & MA_LMV)
319 rc = mdd_get_md(env, mdd_obj, ma->ma_lmv, &ma->ma_lmv_size,
322 ma->ma_valid |= MA_LMV;
328 static int mdd_attr_get_internal(const struct lu_env *env,
329 struct mdd_object *mdd_obj,
335 if (ma->ma_need & MA_INODE)
336 rc = mdd_iattr_get(env, mdd_obj, ma);
338 if (rc == 0 && ma->ma_need & MA_LOV) {
339 if (S_ISREG(mdd_object_type(mdd_obj)) ||
340 S_ISDIR(mdd_object_type(mdd_obj)))
341 rc = __mdd_lmm_get(env, mdd_obj, ma);
343 if (rc == 0 && ma->ma_need & MA_LMV) {
344 if (S_ISDIR(mdd_object_type(mdd_obj)))
345 rc = __mdd_lmv_get(env, mdd_obj, ma);
347 #ifdef CONFIG_FS_POSIX_ACL
348 if (rc == 0 && ma->ma_need & MA_ACL_DEF) {
349 if (S_ISDIR(mdd_object_type(mdd_obj)))
350 rc = mdd_acl_def_get(env, mdd_obj, ma);
353 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64"\n",
358 int mdd_attr_get_internal_locked(const struct lu_env *env,
359 struct mdd_object *mdd_obj, struct md_attr *ma)
362 mdd_read_lock(env, mdd_obj);
363 rc = mdd_attr_get_internal(env, mdd_obj, ma);
364 mdd_read_unlock(env, mdd_obj);
369 * No permission check is needed.
371 static int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
374 struct mdd_object *mdd_obj = md2mdd_obj(obj);
378 rc = mdd_attr_get_internal_locked(env, mdd_obj, ma);
383 * No permission check is needed.
385 static int mdd_xattr_get(const struct lu_env *env,
386 struct md_object *obj, struct lu_buf *buf,
389 struct mdd_object *mdd_obj = md2mdd_obj(obj);
390 struct dt_object *next;
395 LASSERT(lu_object_exists(&obj->mo_lu));
397 next = mdd_object_child(mdd_obj);
398 mdd_read_lock(env, mdd_obj);
399 rc = next->do_ops->do_xattr_get(env, next, buf, name,
400 mdd_object_capa(env, mdd_obj));
401 mdd_read_unlock(env, mdd_obj);
407 * Permission check is done when open,
408 * no need check again.
410 static int mdd_readlink(const struct lu_env *env, struct md_object *obj,
413 struct mdd_object *mdd_obj = md2mdd_obj(obj);
414 struct dt_object *next;
419 LASSERT(lu_object_exists(&obj->mo_lu));
421 next = mdd_object_child(mdd_obj);
422 mdd_read_lock(env, mdd_obj);
423 rc = next->do_body_ops->dbo_read(env, next, buf, &pos,
424 mdd_object_capa(env, mdd_obj));
425 mdd_read_unlock(env, mdd_obj);
429 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
432 struct mdd_object *mdd_obj = md2mdd_obj(obj);
433 struct dt_object *next;
438 LASSERT(lu_object_exists(&obj->mo_lu));
440 next = mdd_object_child(mdd_obj);
441 mdd_read_lock(env, mdd_obj);
442 rc = next->do_ops->do_xattr_list(env, next, buf,
443 mdd_object_capa(env, mdd_obj));
444 mdd_read_unlock(env, mdd_obj);
449 int mdd_object_create_internal(const struct lu_env *env,
450 struct mdd_object *obj, struct md_attr *ma,
451 struct thandle *handle)
453 struct dt_object *next;
454 struct lu_attr *attr = &ma->ma_attr;
458 if (!lu_object_exists(mdd2lu_obj(obj))) {
459 next = mdd_object_child(obj);
460 rc = next->do_ops->do_create(env, next, attr, handle);
464 LASSERT(ergo(rc == 0, lu_object_exists(mdd2lu_obj(obj))));
470 int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *o,
471 const struct lu_attr *attr, struct thandle *handle,
474 struct dt_object *next;
477 LASSERT(lu_object_exists(mdd2lu_obj(o)));
478 next = mdd_object_child(o);
479 rc = next->do_ops->do_attr_set(env, next, attr, handle,
480 mdd_object_capa(env, o));
481 #ifdef CONFIG_FS_POSIX_ACL
482 if (!rc && (attr->la_valid & LA_MODE) && needacl)
483 rc = mdd_acl_chmod(env, o, attr->la_mode, handle);
488 int mdd_attr_set_internal_locked(const struct lu_env *env,
489 struct mdd_object *o,
490 const struct lu_attr *attr,
491 struct thandle *handle, int needacl)
494 mdd_write_lock(env, o);
495 rc = mdd_attr_set_internal(env, o, attr, handle, needacl);
496 mdd_write_unlock(env, o);
500 static int __mdd_xattr_set(const struct lu_env *env, struct mdd_object *o,
501 const struct lu_buf *buf, const char *name,
502 int fl, struct thandle *handle)
504 struct dt_object *next;
505 struct lustre_capa *capa = mdd_object_capa(env, o);
509 LASSERT(lu_object_exists(mdd2lu_obj(o)));
510 next = mdd_object_child(o);
511 if (buf->lb_buf && buf->lb_len > 0) {
512 rc = next->do_ops->do_xattr_set(env, next, buf, name, 0, handle,
514 } else if (buf->lb_buf == NULL && buf->lb_len == 0) {
515 rc = next->do_ops->do_xattr_del(env, next, name, handle, capa);
520 /* this gives the same functionality as the code between
521 * sys_chmod and inode_setattr
522 * chown_common and inode_setattr
523 * utimes and inode_setattr
524 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
527 int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
530 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
531 struct md_ucred *uc = md_ucred(env);
532 time_t now = CURRENT_SECONDS;
539 /* Do not permit change file type */
540 if (la->la_valid & LA_TYPE)
543 /* They should not be processed by setattr */
544 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
547 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
551 if (mdd_is_immutable(obj) || mdd_is_append(obj)) {
554 * If only change flags of the object, we should
555 * let it pass, but also need capability check
556 * here if (!capable(CAP_LINUX_IMMUTABLE)),
557 * fix it, when implement capable in mds
559 if (la->la_valid & ~LA_FLAGS)
562 if (!mdd_capable(uc, CAP_LINUX_IMMUTABLE))
565 if ((uc->mu_fsuid != tmp_la->la_uid) &&
566 !mdd_capable(uc, CAP_FOWNER))
570 * According to Ext3 implementation on this, the
571 * Ctime will be changed, but not clear why?
574 la->la_valid |= LA_CTIME;
578 /* Check for setting the obj time. */
579 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
580 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
581 rc = __mdd_permission_internal(env, obj, MAY_WRITE, 0);
586 /* Make sure a caller can chmod. */
587 if (la->la_valid & LA_MODE) {
589 * Bypass la_vaild == LA_MODE,
590 * this is for changing file with SUID or SGID.
592 if ((la->la_valid & ~LA_MODE) &&
593 (uc->mu_fsuid != tmp_la->la_uid) &&
594 !mdd_capable(uc, CAP_FOWNER))
597 if (la->la_mode == (umode_t) -1)
598 la->la_mode = tmp_la->la_mode;
600 la->la_mode = (la->la_mode & S_IALLUGO) |
601 (tmp_la->la_mode & ~S_IALLUGO);
603 /* Also check the setgid bit! */
604 if (!mdd_in_group_p(uc, (la->la_valid & LA_GID) ? la->la_gid :
605 tmp_la->la_gid) && !mdd_capable(uc, CAP_FSETID))
606 la->la_mode &= ~S_ISGID;
608 la->la_mode = tmp_la->la_mode;
611 /* Make sure a caller can chown. */
612 if (la->la_valid & LA_UID) {
613 if (la->la_uid == (uid_t) -1)
614 la->la_uid = tmp_la->la_uid;
615 if (((uc->mu_fsuid != tmp_la->la_uid) ||
616 (la->la_uid != tmp_la->la_uid)) &&
617 !mdd_capable(uc, CAP_CHOWN))
621 * If the user or group of a non-directory has been
622 * changed by a non-root user, remove the setuid bit.
623 * 19981026 David C Niemi <niemi@tux.org>
625 * Changed this to apply to all users, including root,
626 * to avoid some races. This is the behavior we had in
627 * 2.0. The check for non-root was definitely wrong
628 * for 2.2 anyway, as it should have been using
629 * CAP_FSETID rather than fsuid -- 19990830 SD.
631 if (((tmp_la->la_mode & S_ISUID) == S_ISUID) &&
632 !S_ISDIR(tmp_la->la_mode)) {
633 la->la_mode &= ~S_ISUID;
634 la->la_valid |= LA_MODE;
638 /* Make sure caller can chgrp. */
639 if (la->la_valid & LA_GID) {
640 if (la->la_gid == (gid_t) -1)
641 la->la_gid = tmp_la->la_gid;
642 if (((uc->mu_fsuid != tmp_la->la_uid) ||
643 ((la->la_gid != tmp_la->la_gid) &&
644 !mdd_in_group_p(uc, la->la_gid))) &&
645 !mdd_capable(uc, CAP_CHOWN))
649 * Likewise, if the user or group of a non-directory
650 * has been changed by a non-root user, remove the
651 * setgid bit UNLESS there is no group execute bit
652 * (this would be a file marked for mandatory
653 * locking). 19981026 David C Niemi <niemi@tux.org>
655 * Removed the fsuid check (see the comment above) --
658 if (((tmp_la->la_mode & (S_ISGID | S_IXGRP)) ==
659 (S_ISGID | S_IXGRP)) && !S_ISDIR(tmp_la->la_mode)) {
660 la->la_mode &= ~S_ISGID;
661 la->la_valid |= LA_MODE;
665 /* For tuncate (or setsize), we should have MAY_WRITE perm */
666 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
667 rc = __mdd_permission_internal(env, obj, MAY_WRITE, 0);
672 * For the "Size-on-MDS" setattr update, merge coming
673 * attributes with the set in the inode. BUG 10641
675 if ((la->la_valid & LA_ATIME) &&
676 (la->la_atime < tmp_la->la_atime))
677 la->la_valid &= ~LA_ATIME;
679 if ((la->la_valid & LA_CTIME) &&
680 (la->la_ctime < tmp_la->la_ctime))
681 la->la_valid &= ~(LA_MTIME | LA_CTIME);
683 if (!(la->la_valid & LA_MTIME) && (now > tmp_la->la_mtime)) {
685 la->la_valid |= LA_MTIME;
689 /* For last, ctime must be fixed */
690 if (!(la->la_valid & LA_CTIME) && (now > tmp_la->la_ctime)) {
692 la->la_valid |= LA_CTIME;
698 /* set attr and LOV EA at once, return updated attr */
699 static int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
700 const struct md_attr *ma)
702 struct mdd_object *mdd_obj = md2mdd_obj(obj);
703 struct mdd_device *mdd = mdo2mdd(obj);
704 struct thandle *handle;
705 struct lov_mds_md *lmm = NULL;
706 int rc, lmm_size = 0, max_size = 0;
707 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
710 mdd_txn_param_build(env, mdd, MDD_TXN_ATTR_SET_OP);
711 handle = mdd_trans_start(env, mdd);
713 RETURN(PTR_ERR(handle));
714 /*TODO: add lock here*/
715 /* start a log jounal handle if needed */
716 if (S_ISREG(mdd_object_type(mdd_obj)) &&
717 ma->ma_attr.la_valid & (LA_UID | LA_GID)) {
718 max_size = mdd_lov_mdsize(env, mdd);
719 OBD_ALLOC(lmm, max_size);
722 GOTO(cleanup, rc = -ENOMEM);
724 rc = mdd_get_md_locked(env, mdd_obj, lmm, &lmm_size,
731 if (ma->ma_attr.la_valid & (ATTR_MTIME | ATTR_CTIME))
732 CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n",
733 ma->ma_attr.la_mtime, ma->ma_attr.la_ctime);
735 *la_copy = ma->ma_attr;
736 mdd_write_lock(env, mdd_obj);
737 rc = mdd_fix_attr(env, mdd_obj, la_copy);
738 mdd_write_unlock(env, mdd_obj);
742 if (la_copy->la_valid & LA_FLAGS) {
743 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
746 mdd_flags_xlate(mdd_obj, la_copy->la_flags);
747 } else if (la_copy->la_valid) { /* setattr */
748 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
750 /* journal chown/chgrp in llog, just like unlink */
751 if (rc == 0 && lmm_size){
752 /*TODO set_attr llog */
756 if (rc == 0 && ma->ma_valid & MA_LOV) {
759 mode = mdd_object_type(mdd_obj);
760 if (S_ISREG(mode) || S_ISDIR(mode)) {
761 /*TODO check permission*/
762 rc = mdd_lov_set_md(env, NULL, mdd_obj, ma->ma_lmm,
763 ma->ma_lmm_size, handle, 1);
768 mdd_trans_stop(env, mdd, rc, handle);
769 if (rc == 0 && (lmm != NULL && lmm_size > 0 )) {
770 /*set obd attr, if needed*/
771 rc = mdd_lov_setattr_async(env, mdd_obj, lmm, lmm_size);
774 OBD_FREE(lmm, max_size);
780 int mdd_xattr_set_txn(const struct lu_env *env, struct mdd_object *obj,
781 const struct lu_buf *buf, const char *name, int fl,
782 struct thandle *handle)
787 mdd_write_lock(env, obj);
788 rc = __mdd_xattr_set(env, obj, buf, name, fl, handle);
789 mdd_write_unlock(env, obj);
794 static int mdd_xattr_sanity_check(const struct lu_env *env,
795 struct mdd_object *obj)
797 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
798 struct md_ucred *uc = md_ucred(env);
802 if (mdd_is_immutable(obj) || mdd_is_append(obj))
805 mdd_read_lock(env, obj);
806 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
807 mdd_read_unlock(env, obj);
811 if ((uc->mu_fsuid != tmp_la->la_uid) && !mdd_capable(uc, CAP_FOWNER))
817 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
818 const struct lu_buf *buf, const char *name, int fl)
820 struct mdd_object *mdd_obj = md2mdd_obj(obj);
821 struct mdd_device *mdd = mdo2mdd(obj);
822 struct thandle *handle;
826 rc = mdd_xattr_sanity_check(env, mdd_obj);
830 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
831 handle = mdd_trans_start(env, mdd);
833 RETURN(PTR_ERR(handle));
835 rc = mdd_xattr_set_txn(env, md2mdd_obj(obj), buf, name,
838 mdd_trans_stop(env, mdd, rc, handle);
843 static int __mdd_xattr_del(const struct lu_env *env,struct mdd_device *mdd,
844 struct mdd_object *obj,
845 const char *name, struct thandle *handle)
847 struct dt_object *next;
849 LASSERT(lu_object_exists(mdd2lu_obj(obj)));
850 next = mdd_object_child(obj);
851 return next->do_ops->do_xattr_del(env, next, name, handle,
852 mdd_object_capa(env, obj));
855 int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
858 struct mdd_object *mdd_obj = md2mdd_obj(obj);
859 struct mdd_device *mdd = mdo2mdd(obj);
860 struct thandle *handle;
864 rc = mdd_xattr_sanity_check(env, mdd_obj);
868 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
869 handle = mdd_trans_start(env, mdd);
871 RETURN(PTR_ERR(handle));
873 mdd_write_lock(env, mdd_obj);
874 rc = __mdd_xattr_del(env, mdd, md2mdd_obj(obj), name, handle);
875 mdd_write_unlock(env, mdd_obj);
877 mdd_trans_stop(env, mdd, rc, handle);
883 static int mdd_ref_del(const struct lu_env *env, struct md_object *obj,
886 struct mdd_object *mdd_obj = md2mdd_obj(obj);
887 struct mdd_device *mdd = mdo2mdd(obj);
888 struct thandle *handle;
892 rc = mdd_log_txn_param_build(env, obj, ma, MDD_TXN_UNLINK_OP);
896 handle = mdd_trans_start(env, mdd);
900 mdd_write_lock(env, mdd_obj);
902 rc = mdd_unlink_sanity_check(env, NULL, mdd_obj, ma);
906 mdd_ref_del_internal(env, mdd_obj, handle);
908 if (S_ISDIR(lu_object_attr(&obj->mo_lu))) {
910 mdd_ref_del_internal(env, mdd_obj, handle);
913 rc = mdd_finish_unlink(env, mdd_obj, ma, handle);
917 mdd_write_unlock(env, mdd_obj);
918 mdd_trans_stop(env, mdd, rc, handle);
922 /* partial operation */
923 static int mdd_oc_sanity_check(const struct lu_env *env,
924 struct mdd_object *obj,
930 switch (ma->ma_attr.la_mode & S_IFMT) {
947 static int mdd_object_create(const struct lu_env *env,
948 struct md_object *obj,
949 const struct md_create_spec *spec,
953 struct mdd_device *mdd = mdo2mdd(obj);
954 struct mdd_object *mdd_obj = md2mdd_obj(obj);
955 const struct lu_fid *pfid = spec->u.sp_pfid;
956 struct thandle *handle;
960 rc = mdd_oc_sanity_check(env, mdd_obj, ma);
964 mdd_txn_param_build(env, mdd, MDD_TXN_OBJECT_CREATE_OP);
965 handle = mdd_trans_start(env, mdd);
967 RETURN(PTR_ERR(handle));
969 mdd_write_lock(env, mdd_obj);
970 rc = mdd_object_create_internal(env, mdd_obj, ma, handle);
974 if (spec->sp_cr_flags & MDS_CREATE_SLAVE_OBJ) {
975 /* If creating the slave object, set slave EA here. */
976 int lmv_size = spec->u.sp_ea.eadatalen;
977 struct lmv_stripe_md *lmv;
979 lmv = (struct lmv_stripe_md *)spec->u.sp_ea.eadata;
980 LASSERT(lmv != NULL && lmv_size > 0);
982 rc = __mdd_xattr_set(env, mdd_obj,
983 mdd_buf_get_const(env, lmv, lmv_size),
984 MDS_LMV_MD_NAME, 0, handle);
987 pfid = spec->u.sp_ea.fid;
989 CDEBUG(D_INFO, "Set slave ea "DFID", eadatalen %d, rc %d\n",
990 PFID(mdo2fid(mdd_obj)), spec->u.sp_ea.eadatalen, rc);
991 rc = mdd_attr_set_internal(env, mdd_obj, &ma->ma_attr, handle, 0);
993 #ifdef CONFIG_FS_POSIX_ACL
994 if (spec->sp_cr_flags & MDS_CREATE_RMT_ACL) {
995 struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
997 buf->lb_buf = (void *)spec->u.sp_ea.eadata;
998 buf->lb_len = spec->u.sp_ea.eadatalen;
999 if ((buf->lb_len > 0) && (buf->lb_buf != NULL)) {
1000 rc = __mdd_acl_init(env, mdd_obj, buf,
1001 &ma->ma_attr.la_mode,
1006 ma->ma_attr.la_valid |= LA_MODE;
1010 rc = mdd_object_initialize(env, pfid, mdd_obj, ma, handle);
1014 mdd_write_unlock(env, mdd_obj);
1016 rc = mdd_attr_get_internal_locked(env, mdd_obj, ma);
1018 mdd_trans_stop(env, mdd, rc, handle);
1023 * XXX: if permission check is needed here?
1025 static int mdd_ref_add(const struct lu_env *env,
1026 struct md_object *obj)
1028 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1029 struct mdd_device *mdd = mdo2mdd(obj);
1030 struct thandle *handle;
1034 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
1035 handle = mdd_trans_start(env, mdd);
1039 mdd_write_lock(env, mdd_obj);
1040 rc = mdd_link_sanity_check(env, NULL, mdd_obj);
1042 mdd_ref_add_internal(env, mdd_obj, handle);
1043 mdd_write_unlock(env, mdd_obj);
1045 mdd_trans_stop(env, mdd, 0, handle);
1050 /* do NOT or the MAY_*'s, you'll get the weakest */
1051 static int accmode(struct mdd_object *mdd_obj, int flags)
1056 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1057 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1058 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1059 * owner can write to a file even if it is marked readonly to hide
1060 * its brokenness. (bug 5781) */
1061 if (flags & MDS_OPEN_OWNEROVERRIDE && inode->i_uid == current->fsuid)
1064 if (flags & FMODE_READ)
1066 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1068 if (flags & MDS_FMODE_EXEC)
1073 static int mdd_open_sanity_check(const struct lu_env *env,
1074 struct mdd_object *obj, int flag)
1076 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
1077 int mode = accmode(obj, flag);
1082 if (mdd_is_dead_obj(obj))
1085 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
1089 if (S_ISLNK(tmp_la->la_mode))
1092 if (S_ISDIR(tmp_la->la_mode) && (mode & MAY_WRITE))
1095 if (!(flag & MDS_OPEN_CREATED)) {
1096 rc = __mdd_permission_internal(env, obj, mode, 0);
1102 * FIFO's, sockets and device files are special: they don't
1103 * actually live on the filesystem itself, and as such you
1104 * can write to them even if the filesystem is read-only.
1106 if (S_ISFIFO(tmp_la->la_mode) || S_ISSOCK(tmp_la->la_mode) ||
1107 S_ISBLK(tmp_la->la_mode) || S_ISCHR(tmp_la->la_mode))
1111 * An append-only file must be opened in append mode for writing.
1113 if (mdd_is_append(obj)) {
1114 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1120 /* O_NOATIME can only be set by the owner or superuser */
1121 if (flag & O_NOATIME) {
1122 struct md_ucred *uc = md_ucred(env);
1124 if (uc->mu_fsuid != tmp_la->la_uid &&
1125 !mdd_capable(uc, CAP_FOWNER))
1132 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1135 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1138 mdd_write_lock(env, mdd_obj);
1140 rc = mdd_open_sanity_check(env, mdd_obj, flags);
1142 mdd_obj->mod_count++;
1144 mdd_write_unlock(env, mdd_obj);
1148 /* return md_attr back,
1149 * if it is last unlink then return lov ea + llog cookie*/
1150 int mdd_object_kill(const struct lu_env *env, struct mdd_object *obj,
1156 mdd_set_dead_obj(obj);
1157 if (S_ISREG(mdd_object_type(obj))) {
1158 /* Return LOV & COOKIES unconditionally here. We clean evth up.
1159 * Caller must be ready for that. */
1160 rc = __mdd_lmm_get(env, obj, ma);
1161 if ((ma->ma_valid & MA_LOV))
1162 rc = mdd_unlink_log(env, mdo2mdd(&obj->mod_obj),
1169 * No permission check is needed.
1171 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1175 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1178 mdd_write_lock(env, mdd_obj);
1179 /* release open count */
1180 mdd_obj->mod_count --;
1182 rc = mdd_iattr_get(env, mdd_obj, ma);
1183 if (rc == 0 && mdd_obj->mod_count == 0) {
1184 if (ma->ma_attr.la_nlink == 0)
1185 rc = mdd_object_kill(env, mdd_obj, ma);
1187 mdd_write_unlock(env, mdd_obj);
1192 * Permission check is done when open,
1193 * no need check again.
1195 static int mdd_readpage_sanity_check(const struct lu_env *env,
1196 struct mdd_object *obj)
1198 struct dt_object *next = mdd_object_child(obj);
1202 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
1204 rc = mdd_permission_internal(env, obj, MAY_READ);
1214 static int mdd_dir_page_build(const struct lu_env *env, int first,
1215 void *area, int nob,
1216 struct dt_it_ops *iops, struct dt_it *it,
1217 __u32 *start, __u32 *end, struct lu_dirent **last)
1220 struct mdd_thread_info *info = mdd_env_info(env);
1221 struct lu_fid *fid = &info->mti_fid;
1222 struct lu_dirent *ent;
1225 memset(area, 0, sizeof (struct lu_dirpage));
1226 area += sizeof (struct lu_dirpage);
1227 nob -= sizeof (struct lu_dirpage);
1230 LASSERT(nob > sizeof *ent);
1240 name = (char *)iops->key(env, it);
1241 len = iops->key_size(env, it);
1243 fid = (struct lu_fid *)iops->rec(env, it);
1245 recsize = (sizeof *ent + len + 3) & ~3;
1246 hash = iops->store(env, it);
1248 CDEBUG(D_INFO, "%p %p %d "DFID": %#8.8x (%d) \"%*.*s\"\n",
1249 name, ent, nob, PFID(fid), hash, len, len, len, name);
1251 if (nob >= recsize) {
1252 fid_be_to_cpu(&ent->lde_fid, fid);
1253 fid_cpu_to_le(&ent->lde_fid, &ent->lde_fid);
1254 ent->lde_hash = hash;
1255 ent->lde_namelen = cpu_to_le16(len);
1256 ent->lde_reclen = cpu_to_le16(recsize);
1257 memcpy(ent->lde_name, name, len);
1258 if (first && ent == area)
1261 ent = (void *)ent + recsize;
1263 result = iops->next(env, it);
1266 * record doesn't fit into page, enlarge previous one.
1268 LASSERT(*last != NULL);
1269 (*last)->lde_reclen =
1270 cpu_to_le16(le16_to_cpu((*last)->lde_reclen) +
1274 } while (result == 0);
1279 static int __mdd_readpage(const struct lu_env *env, struct mdd_object *obj,
1280 const struct lu_rdpg *rdpg)
1283 struct dt_object *next = mdd_object_child(obj);
1284 struct dt_it_ops *iops;
1286 struct lu_dirent *last;
1293 LASSERT(rdpg->rp_pages != NULL);
1294 LASSERT(next->do_index_ops != NULL);
1296 if (rdpg->rp_count <= 0)
1300 * iterate through directory and fill pages from @rdpg
1302 iops = &next->do_index_ops->dio_it;
1303 it = iops->init(env, next, 0, mdd_object_capa(env, obj));
1307 rc = iops->load(env, it, rdpg->rp_hash);
1311 * Iterator didn't find record with exactly the key requested.
1313 * It is currently either
1315 * - positioned above record with key less than
1316 * requested---skip it.
1318 * - or not positioned at all (is in IAM_IT_SKEWED
1319 * state)---position it on the next item.
1321 rc = iops->next(env, it);
1326 * At this point and across for-loop:
1328 * rc == 0 -> ok, proceed.
1329 * rc > 0 -> end of directory.
1332 for (i = 0, nob = rdpg->rp_count; rc == 0 && nob > 0;
1333 i++, nob -= CFS_PAGE_SIZE) {
1334 LASSERT(i < rdpg->rp_npages);
1335 pg = rdpg->rp_pages[i];
1336 rc = mdd_dir_page_build(env, !i, kmap(pg),
1337 min_t(int, nob, CFS_PAGE_SIZE), iops,
1338 it, &hash_start, &hash_end, &last);
1339 if (rc != 0 || i == rdpg->rp_npages - 1)
1340 last->lde_reclen = 0;
1351 struct lu_dirpage *dp;
1353 dp = kmap(rdpg->rp_pages[0]);
1354 dp->ldp_hash_start = rdpg->rp_hash;
1355 dp->ldp_hash_end = hash_end;
1358 * No pages were processed, mark this.
1360 dp->ldp_flags |= LDF_EMPTY;
1361 dp->ldp_flags = cpu_to_le16(dp->ldp_flags);
1362 kunmap(rdpg->rp_pages[0]);
1365 iops->fini(env, it);
1370 static int mdd_readpage(const struct lu_env *env, struct md_object *obj,
1371 const struct lu_rdpg *rdpg)
1373 struct dt_object *next;
1374 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1378 LASSERT(lu_object_exists(mdd2lu_obj(mdd_obj)));
1379 next = mdd_object_child(mdd_obj);
1381 mdd_read_lock(env, mdd_obj);
1382 rc = mdd_readpage_sanity_check(env, mdd_obj);
1384 GOTO(out_unlock, rc);
1386 rc = __mdd_readpage(env, mdd_obj, rdpg);
1389 mdd_read_unlock(env, mdd_obj);
1393 struct md_object_operations mdd_obj_ops = {
1394 .moo_permission = mdd_permission,
1395 .moo_attr_get = mdd_attr_get,
1396 .moo_attr_set = mdd_attr_set,
1397 .moo_xattr_get = mdd_xattr_get,
1398 .moo_xattr_set = mdd_xattr_set,
1399 .moo_xattr_list = mdd_xattr_list,
1400 .moo_xattr_del = mdd_xattr_del,
1401 .moo_object_create = mdd_object_create,
1402 .moo_ref_add = mdd_ref_add,
1403 .moo_ref_del = mdd_ref_del,
1404 .moo_open = mdd_open,
1405 .moo_close = mdd_close,
1406 .moo_readpage = mdd_readpage,
1407 .moo_readlink = mdd_readlink,
1408 .moo_capa_get = mdd_capa_get