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 /* get lov EA only */
289 static int __mdd_lmm_get(const struct lu_env *env,
290 struct mdd_object *mdd_obj, struct md_attr *ma)
295 if (ma->ma_valid & MA_LOV)
298 rc = mdd_get_md(env, mdd_obj, ma->ma_lmm, &ma->ma_lmm_size,
301 ma->ma_valid |= MA_LOV;
307 int mdd_lmm_get_locked(const struct lu_env *env, struct mdd_object *mdd_obj,
313 mdd_read_lock(env, mdd_obj);
314 rc = __mdd_lmm_get(env, mdd_obj, ma);
315 mdd_read_unlock(env, mdd_obj);
320 static int __mdd_lmv_get(const struct lu_env *env,
321 struct mdd_object *mdd_obj, struct md_attr *ma)
326 if (ma->ma_valid & MA_LMV)
329 rc = mdd_get_md(env, mdd_obj, ma->ma_lmv, &ma->ma_lmv_size,
332 ma->ma_valid |= MA_LMV;
338 static int mdd_attr_get_internal(const struct lu_env *env,
339 struct mdd_object *mdd_obj,
345 if (ma->ma_need & MA_INODE)
346 rc = mdd_iattr_get(env, mdd_obj, ma);
348 if (rc == 0 && ma->ma_need & MA_LOV) {
349 if (S_ISREG(mdd_object_type(mdd_obj)) ||
350 S_ISDIR(mdd_object_type(mdd_obj)))
351 rc = __mdd_lmm_get(env, mdd_obj, ma);
353 if (rc == 0 && ma->ma_need & MA_LMV) {
354 if (S_ISDIR(mdd_object_type(mdd_obj)))
355 rc = __mdd_lmv_get(env, mdd_obj, ma);
357 #ifdef CONFIG_FS_POSIX_ACL
358 if (rc == 0 && ma->ma_need & MA_ACL_DEF) {
359 if (S_ISDIR(mdd_object_type(mdd_obj)))
360 rc = mdd_acl_def_get(env, mdd_obj, ma);
363 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64"\n",
368 int mdd_attr_get_internal_locked(const struct lu_env *env,
369 struct mdd_object *mdd_obj, struct md_attr *ma)
372 int needlock = ma->ma_need & (MA_LOV | MA_LMV | MA_ACL_DEF);
375 mdd_read_lock(env, mdd_obj);
376 rc = mdd_attr_get_internal(env, mdd_obj, ma);
378 mdd_read_unlock(env, mdd_obj);
383 * No permission check is needed.
385 static int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
388 struct mdd_object *mdd_obj = md2mdd_obj(obj);
392 rc = mdd_attr_get_internal_locked(env, mdd_obj, ma);
397 * No permission check is needed.
399 static int mdd_xattr_get(const struct lu_env *env,
400 struct md_object *obj, struct lu_buf *buf,
403 struct mdd_object *mdd_obj = md2mdd_obj(obj);
408 LASSERT(mdd_object_exists(mdd_obj));
410 mdd_read_lock(env, mdd_obj);
411 rc = mdo_xattr_get(env, mdd_obj, buf, name,
412 mdd_object_capa(env, mdd_obj));
413 mdd_read_unlock(env, mdd_obj);
419 * Permission check is done when open,
420 * no need check again.
422 static int mdd_readlink(const struct lu_env *env, struct md_object *obj,
425 struct mdd_object *mdd_obj = md2mdd_obj(obj);
426 struct dt_object *next;
431 LASSERT(mdd_object_exists(mdd_obj));
433 next = mdd_object_child(mdd_obj);
434 mdd_read_lock(env, mdd_obj);
435 rc = next->do_body_ops->dbo_read(env, next, buf, &pos,
436 mdd_object_capa(env, mdd_obj));
437 mdd_read_unlock(env, mdd_obj);
442 * No permission check is needed.
444 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
447 struct mdd_object *mdd_obj = md2mdd_obj(obj);
452 mdd_read_lock(env, mdd_obj);
453 rc = mdo_xattr_list(env, mdd_obj, buf, mdd_object_capa(env, mdd_obj));
454 mdd_read_unlock(env, mdd_obj);
459 int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p,
460 struct mdd_object *c, struct md_attr *ma,
461 struct thandle *handle)
463 struct lu_attr *attr = &ma->ma_attr;
464 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
468 if (!mdd_object_exists(c)) {
469 struct dt_object *next = mdd_object_child(c);
472 /* @hint will be initialized by underlying device. */
473 next->do_ops->do_ah_init(env, hint,
474 p ? mdd_object_child(p) : NULL,
475 attr->la_mode & S_IFMT);
476 rc = mdo_create_obj(env, c, attr, hint, handle);
477 LASSERT(ergo(rc == 0, mdd_object_exists(c)));
485 int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj,
486 const struct lu_attr *attr, struct thandle *handle,
492 rc = mdo_attr_set(env, obj, attr, handle, mdd_object_capa(env, obj));
493 #ifdef CONFIG_FS_POSIX_ACL
494 if (!rc && (attr->la_valid & LA_MODE) && needacl)
495 rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
500 int mdd_attr_set_internal_locked(const struct lu_env *env,
501 struct mdd_object *o,
502 const struct lu_attr *attr,
503 struct thandle *handle, int needacl)
508 needacl = needacl && (attr->la_valid & LA_MODE);
511 mdd_write_lock(env, o);
513 rc = mdd_attr_set_internal(env, o, attr, handle, needacl);
516 mdd_write_unlock(env, o);
520 static int __mdd_xattr_set(const struct lu_env *env, struct mdd_object *obj,
521 const struct lu_buf *buf, const char *name,
522 int fl, struct thandle *handle)
524 struct lustre_capa *capa = mdd_object_capa(env, obj);
528 if (buf->lb_buf && buf->lb_len > 0)
529 rc = mdo_xattr_set(env, obj, buf, name, 0, handle, capa);
530 else if (buf->lb_buf == NULL && buf->lb_len == 0)
531 rc = mdo_xattr_del(env, obj, name, handle, capa);
537 * This gives the same functionality as the code between
538 * sys_chmod and inode_setattr
539 * chown_common and inode_setattr
540 * utimes and inode_setattr
541 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
543 static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
544 struct lu_attr *la, const struct md_attr *ma)
546 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
547 struct md_ucred *uc = md_ucred(env);
554 /* Do not permit change file type */
555 if (la->la_valid & LA_TYPE)
558 /* They should not be processed by setattr */
559 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
562 /* This is only for set ctime when rename's source is on remote MDS. */
563 if (unlikely(la->la_valid == LA_CTIME)) {
564 rc = mdd_may_delete(env, NULL, obj, (struct md_attr *)ma, 1, 0);
568 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
572 if (la->la_valid == LA_ATIME) {
573 /* This is atime only set for read atime update on close. */
574 if (la->la_atime <= tmp_la->la_atime + 0/*XXX:mds_atime_diff*/)
575 la->la_valid &= ~LA_ATIME;
579 /* Check if flags change. */
580 if (la->la_valid & LA_FLAGS) {
581 unsigned int oldflags = 0;
582 unsigned int newflags = la->la_flags &
583 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
585 if ((uc->mu_fsuid != tmp_la->la_uid) &&
586 !mdd_capable(uc, CAP_FOWNER))
589 /* XXX: the IMMUTABLE and APPEND_ONLY flags can
590 * only be changed by the relevant capability. */
591 if (mdd_is_immutable(obj))
592 oldflags |= LUSTRE_IMMUTABLE_FL;
593 if (mdd_is_append(obj))
594 oldflags |= LUSTRE_APPEND_FL;
595 if ((oldflags ^ newflags) &&
596 !mdd_capable(uc, CAP_LINUX_IMMUTABLE))
599 if (!S_ISDIR(tmp_la->la_mode))
600 la->la_flags &= ~LUSTRE_DIRSYNC_FL;
603 if ((mdd_is_immutable(obj) || mdd_is_append(obj)) &&
604 (la->la_valid & ~LA_FLAGS) &&
605 !(ma->ma_attr_flags & MDS_PERM_BYPASS))
608 /* Check for setting the obj time. */
609 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
610 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
611 if ((uc->mu_fsuid != tmp_la->la_uid) &&
612 !mdd_capable(uc, CAP_FOWNER)) {
613 rc = mdd_permission_internal_locked(env, obj, tmp_la,
620 /* Make sure a caller can chmod. */
621 if (la->la_valid & LA_MODE) {
623 * Bypass la_vaild == LA_MODE,
624 * this is for changing file with SUID or SGID.
626 if ((la->la_valid & ~LA_MODE) &&
627 (uc->mu_fsuid != tmp_la->la_uid) &&
628 !mdd_capable(uc, CAP_FOWNER))
631 if (la->la_mode == (umode_t) -1)
632 la->la_mode = tmp_la->la_mode;
634 la->la_mode = (la->la_mode & S_IALLUGO) |
635 (tmp_la->la_mode & ~S_IALLUGO);
637 /* Also check the setgid bit! */
638 if (!mdd_in_group_p(uc, (la->la_valid & LA_GID) ? la->la_gid :
639 tmp_la->la_gid) && !mdd_capable(uc, CAP_FSETID))
640 la->la_mode &= ~S_ISGID;
642 la->la_mode = tmp_la->la_mode;
645 /* Make sure a caller can chown. */
646 if (la->la_valid & LA_UID) {
647 if (la->la_uid == (uid_t) -1)
648 la->la_uid = tmp_la->la_uid;
649 if (((uc->mu_fsuid != tmp_la->la_uid) ||
650 (la->la_uid != tmp_la->la_uid)) &&
651 !mdd_capable(uc, CAP_CHOWN))
655 * If the user or group of a non-directory has been
656 * changed by a non-root user, remove the setuid bit.
657 * 19981026 David C Niemi <niemi@tux.org>
659 * Changed this to apply to all users, including root,
660 * to avoid some races. This is the behavior we had in
661 * 2.0. The check for non-root was definitely wrong
662 * for 2.2 anyway, as it should have been using
663 * CAP_FSETID rather than fsuid -- 19990830 SD.
665 if (((tmp_la->la_mode & S_ISUID) == S_ISUID) &&
666 !S_ISDIR(tmp_la->la_mode)) {
667 la->la_mode &= ~S_ISUID;
668 la->la_valid |= LA_MODE;
672 /* Make sure caller can chgrp. */
673 if (la->la_valid & LA_GID) {
674 if (la->la_gid == (gid_t) -1)
675 la->la_gid = tmp_la->la_gid;
676 if (((uc->mu_fsuid != tmp_la->la_uid) ||
677 ((la->la_gid != tmp_la->la_gid) &&
678 !mdd_in_group_p(uc, la->la_gid))) &&
679 !mdd_capable(uc, CAP_CHOWN))
683 * Likewise, if the user or group of a non-directory
684 * has been changed by a non-root user, remove the
685 * setgid bit UNLESS there is no group execute bit
686 * (this would be a file marked for mandatory
687 * locking). 19981026 David C Niemi <niemi@tux.org>
689 * Removed the fsuid check (see the comment above) --
692 if (((tmp_la->la_mode & (S_ISGID | S_IXGRP)) ==
693 (S_ISGID | S_IXGRP)) && !S_ISDIR(tmp_la->la_mode)) {
694 la->la_mode &= ~S_ISGID;
695 la->la_valid |= LA_MODE;
699 /* For truncate (or setsize), we should have MAY_WRITE perm */
700 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
701 if (!((la->la_valid & MDS_OPEN_OWNEROVERRIDE) &&
702 (uc->mu_fsuid == tmp_la->la_uid)) &&
703 !(ma->ma_attr_flags & MDS_PERM_BYPASS)) {
704 rc = mdd_permission_internal_locked(env, obj, tmp_la,
710 /* For the "Size-on-MDS" setattr update, merge coming
711 * attributes with the set in the inode. BUG 10641 */
712 if ((la->la_valid & LA_ATIME) &&
713 (la->la_atime <= tmp_la->la_atime))
714 la->la_valid &= ~LA_ATIME;
716 /* OST attributes do not have a priority over MDS attributes,
717 * so drop times if ctime is equal. */
718 if ((la->la_valid & LA_CTIME) &&
719 (la->la_ctime <= tmp_la->la_ctime))
720 la->la_valid &= ~(LA_MTIME | LA_CTIME);
721 } else if (la->la_valid & LA_CTIME) {
722 /* The pure setattr, it has the priority over what is already
723 * set, do not drop it if ctime is equal. */
724 if (la->la_ctime < tmp_la->la_ctime)
725 la->la_valid &= ~(LA_ATIME | LA_MTIME | LA_CTIME);
731 /* set attr and LOV EA at once, return updated attr */
732 static int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
733 const struct md_attr *ma)
735 struct mdd_object *mdd_obj = md2mdd_obj(obj);
736 struct mdd_device *mdd = mdo2mdd(obj);
737 struct thandle *handle;
738 struct lov_mds_md *lmm = NULL;
739 struct llog_cookie *logcookies = NULL;
740 int rc, lmm_size = 0, cookie_size = 0;
741 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
744 mdd_txn_param_build(env, mdd, MDD_TXN_ATTR_SET_OP);
745 handle = mdd_trans_start(env, mdd);
747 RETURN(PTR_ERR(handle));
748 /*TODO: add lock here*/
749 /* start a log jounal handle if needed */
750 if (S_ISREG(mdd_object_type(mdd_obj)) &&
751 ma->ma_attr.la_valid & (LA_UID | LA_GID)) {
752 lmm_size = mdd_lov_mdsize(env, mdd);
753 lmm = mdd_max_lmm_get(env, mdd);
755 GOTO(cleanup, rc = -ENOMEM);
757 rc = mdd_get_md_locked(env, mdd_obj, lmm, &lmm_size,
764 if (ma->ma_attr.la_valid & (ATTR_MTIME | ATTR_CTIME))
765 CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n",
766 ma->ma_attr.la_mtime, ma->ma_attr.la_ctime);
768 *la_copy = ma->ma_attr;
769 rc = mdd_fix_attr(env, mdd_obj, la_copy, ma);
773 if (la_copy->la_valid & LA_FLAGS) {
774 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
777 mdd_flags_xlate(mdd_obj, la_copy->la_flags);
778 } else if (la_copy->la_valid) { /* setattr */
779 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
781 /* journal chown/chgrp in llog, just like unlink */
782 if (rc == 0 && lmm_size){
783 cookie_size = mdd_lov_cookiesize(env, mdd);
784 logcookies = mdd_max_cookie_get(env, mdd);
785 if (logcookies == NULL)
786 GOTO(cleanup, rc = -ENOMEM);
788 if (mdd_setattr_log(env, mdd, ma, lmm, lmm_size,
789 logcookies, cookie_size) <= 0)
794 if (rc == 0 && ma->ma_valid & MA_LOV) {
797 mode = mdd_object_type(mdd_obj);
798 if (S_ISREG(mode) || S_ISDIR(mode)) {
799 rc = mdd_lsm_sanity_check(env, mdd_obj);
803 rc = mdd_lov_set_md(env, NULL, mdd_obj, ma->ma_lmm,
804 ma->ma_lmm_size, handle, 1);
809 mdd_trans_stop(env, mdd, rc, handle);
810 if (rc == 0 && (lmm != NULL && lmm_size > 0 )) {
811 /*set obd attr, if needed*/
812 rc = mdd_lov_setattr_async(env, mdd_obj, lmm, lmm_size,
818 int mdd_xattr_set_txn(const struct lu_env *env, struct mdd_object *obj,
819 const struct lu_buf *buf, const char *name, int fl,
820 struct thandle *handle)
825 mdd_write_lock(env, obj);
826 rc = __mdd_xattr_set(env, obj, buf, name, fl, handle);
827 mdd_write_unlock(env, obj);
832 static int mdd_xattr_sanity_check(const struct lu_env *env,
833 struct mdd_object *obj)
835 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
836 struct md_ucred *uc = md_ucred(env);
840 if (mdd_is_immutable(obj) || mdd_is_append(obj))
843 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
847 if ((uc->mu_fsuid != tmp_la->la_uid) && !mdd_capable(uc, CAP_FOWNER))
853 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
854 const struct lu_buf *buf, const char *name, int fl)
856 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
857 struct mdd_object *mdd_obj = md2mdd_obj(obj);
858 struct mdd_device *mdd = mdo2mdd(obj);
859 struct thandle *handle;
863 rc = mdd_xattr_sanity_check(env, mdd_obj);
867 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
868 handle = mdd_trans_start(env, mdd);
870 RETURN(PTR_ERR(handle));
872 rc = mdd_xattr_set_txn(env, md2mdd_obj(obj), buf, name,
875 la_copy->la_ctime = CURRENT_SECONDS;
876 la_copy->la_valid = LA_CTIME;
877 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
880 mdd_trans_stop(env, mdd, rc, handle);
885 int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
888 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
889 struct mdd_object *mdd_obj = md2mdd_obj(obj);
890 struct mdd_device *mdd = mdo2mdd(obj);
891 struct thandle *handle;
895 rc = mdd_xattr_sanity_check(env, mdd_obj);
899 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
900 handle = mdd_trans_start(env, mdd);
902 RETURN(PTR_ERR(handle));
904 mdd_write_lock(env, mdd_obj);
905 rc = mdo_xattr_del(env, md2mdd_obj(obj), name, handle,
906 mdd_object_capa(env, mdd_obj));
907 mdd_write_unlock(env, mdd_obj);
909 la_copy->la_ctime = CURRENT_SECONDS;
910 la_copy->la_valid = LA_CTIME;
911 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
915 mdd_trans_stop(env, mdd, rc, handle);
921 static int mdd_ref_del(const struct lu_env *env, struct md_object *obj,
924 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
925 struct mdd_object *mdd_obj = md2mdd_obj(obj);
926 struct mdd_device *mdd = mdo2mdd(obj);
927 struct thandle *handle;
932 * Check -ENOENT early here because we need to get object type
933 * to calculate credits before transaction start
935 if (!mdd_object_exists(mdd_obj))
938 LASSERT(mdd_object_exists(mdd_obj) > 0);
940 rc = mdd_log_txn_param_build(env, obj, ma, MDD_TXN_UNLINK_OP);
944 handle = mdd_trans_start(env, mdd);
948 mdd_write_lock(env, mdd_obj);
950 rc = mdd_unlink_sanity_check(env, NULL, mdd_obj, ma);
954 mdo_ref_del(env, mdd_obj, handle);
956 if (S_ISDIR(lu_object_attr(&obj->mo_lu))) {
958 mdo_ref_del(env, mdd_obj, handle);
961 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
962 la_copy->la_ctime = ma->ma_attr.la_ctime;
964 la_copy->la_valid = LA_CTIME;
965 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 0);
969 rc = mdd_finish_unlink(env, mdd_obj, ma, handle);
973 mdd_write_unlock(env, mdd_obj);
974 mdd_trans_stop(env, mdd, rc, handle);
978 /* partial operation */
979 static int mdd_oc_sanity_check(const struct lu_env *env,
980 struct mdd_object *obj,
986 switch (ma->ma_attr.la_mode & S_IFMT) {
1003 static int mdd_object_create(const struct lu_env *env,
1004 struct md_object *obj,
1005 const struct md_op_spec *spec,
1009 struct mdd_device *mdd = mdo2mdd(obj);
1010 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1011 const struct lu_fid *pfid = spec->u.sp_pfid;
1012 struct thandle *handle;
1016 mdd_txn_param_build(env, mdd, MDD_TXN_OBJECT_CREATE_OP);
1017 handle = mdd_trans_start(env, mdd);
1019 RETURN(PTR_ERR(handle));
1021 mdd_write_lock(env, mdd_obj);
1022 rc = mdd_oc_sanity_check(env, mdd_obj, ma);
1026 rc = mdd_object_create_internal(env, NULL, mdd_obj, ma, handle);
1030 if (spec->sp_cr_flags & MDS_CREATE_SLAVE_OBJ) {
1031 /* If creating the slave object, set slave EA here. */
1032 int lmv_size = spec->u.sp_ea.eadatalen;
1033 struct lmv_stripe_md *lmv;
1035 lmv = (struct lmv_stripe_md *)spec->u.sp_ea.eadata;
1036 LASSERT(lmv != NULL && lmv_size > 0);
1038 rc = __mdd_xattr_set(env, mdd_obj,
1039 mdd_buf_get_const(env, lmv, lmv_size),
1040 MDS_LMV_MD_NAME, 0, handle);
1044 rc = mdd_attr_set_internal(env, mdd_obj, &ma->ma_attr, handle, 0);
1046 #ifdef CONFIG_FS_POSIX_ACL
1047 if (spec->sp_cr_flags & MDS_CREATE_RMT_ACL) {
1048 struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
1050 buf->lb_buf = (void *)spec->u.sp_ea.eadata;
1051 buf->lb_len = spec->u.sp_ea.eadatalen;
1052 if ((buf->lb_len > 0) && (buf->lb_buf != NULL)) {
1053 rc = __mdd_acl_init(env, mdd_obj, buf,
1054 &ma->ma_attr.la_mode,
1059 ma->ma_attr.la_valid |= LA_MODE;
1062 pfid = spec->u.sp_ea.fid;
1065 rc = mdd_object_initialize(env, pfid, mdd_obj, ma, handle);
1069 mdd_write_unlock(env, mdd_obj);
1071 rc = mdd_attr_get_internal_locked(env, mdd_obj, ma);
1073 mdd_trans_stop(env, mdd, rc, handle);
1078 static int mdd_ref_add(const struct lu_env *env, struct md_object *obj,
1079 const struct md_attr *ma)
1081 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
1082 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1083 struct mdd_device *mdd = mdo2mdd(obj);
1084 struct thandle *handle;
1088 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
1089 handle = mdd_trans_start(env, mdd);
1093 mdd_write_lock(env, mdd_obj);
1094 rc = mdd_link_sanity_check(env, NULL, NULL, mdd_obj);
1096 mdo_ref_add(env, mdd_obj, handle);
1097 mdd_write_unlock(env, mdd_obj);
1099 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1100 la_copy->la_ctime = ma->ma_attr.la_ctime;
1102 la_copy->la_valid = LA_CTIME;
1103 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
1106 mdd_trans_stop(env, mdd, 0, handle);
1112 * do NOT or the MAY_*'s, you'll get the weakest
1114 int accmode(const struct lu_env *env, struct lu_attr *la, int flags)
1118 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1119 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1120 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1121 * owner can write to a file even if it is marked readonly to hide
1122 * its brokenness. (bug 5781) */
1123 if (flags & MDS_OPEN_OWNEROVERRIDE) {
1124 struct md_ucred *uc = md_ucred(env);
1126 if ((uc == NULL) || (uc->mu_valid == UCRED_INIT) ||
1127 (la->la_uid == uc->mu_fsuid))
1131 if (flags & FMODE_READ)
1133 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1135 if (flags & MDS_FMODE_EXEC)
1140 static int mdd_open_sanity_check(const struct lu_env *env,
1141 struct mdd_object *obj, int flag)
1143 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
1148 if (mdd_is_dead_obj(obj))
1151 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
1155 if (S_ISLNK(tmp_la->la_mode))
1158 mode = accmode(env, tmp_la, flag);
1160 if (S_ISDIR(tmp_la->la_mode) && (mode & MAY_WRITE))
1163 if (!(flag & MDS_OPEN_CREATED)) {
1164 rc = mdd_permission_internal(env, obj, tmp_la, mode);
1169 if (S_ISFIFO(tmp_la->la_mode) || S_ISSOCK(tmp_la->la_mode) ||
1170 S_ISBLK(tmp_la->la_mode) || S_ISCHR(tmp_la->la_mode))
1171 flag &= ~MDS_OPEN_TRUNC;
1173 /* For writing append-only file must open it with append mode. */
1174 if (mdd_is_append(obj)) {
1175 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1177 if (flag & MDS_OPEN_TRUNC)
1183 * Now, flag -- O_NOATIME does not be packed by client.
1185 if (flag & O_NOATIME) {
1186 struct md_ucred *uc = md_ucred(env);
1188 if (uc && ((uc->mu_valid == UCRED_OLD) ||
1189 (uc->mu_valid == UCRED_NEW)) &&
1190 (uc->mu_fsuid != tmp_la->la_uid) &&
1191 !mdd_capable(uc, CAP_FOWNER))
1199 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1202 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1205 mdd_write_lock(env, mdd_obj);
1207 rc = mdd_open_sanity_check(env, mdd_obj, flags);
1209 mdd_obj->mod_count++;
1211 mdd_write_unlock(env, mdd_obj);
1215 /* return md_attr back,
1216 * if it is last unlink then return lov ea + llog cookie*/
1217 int mdd_object_kill(const struct lu_env *env, struct mdd_object *obj,
1223 if (S_ISREG(mdd_object_type(obj))) {
1224 /* Return LOV & COOKIES unconditionally here. We clean evth up.
1225 * Caller must be ready for that. */
1226 rc = __mdd_lmm_get(env, obj, ma);
1227 if ((ma->ma_valid & MA_LOV))
1228 rc = mdd_unlink_log(env, mdo2mdd(&obj->mod_obj),
1235 * No permission check is needed.
1237 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1241 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1242 struct thandle *handle;
1245 rc = mdd_log_txn_param_build(env, obj, ma, MDD_TXN_UNLINK_OP);
1248 handle = mdd_trans_start(env, mdo2mdd(obj));
1250 RETURN(PTR_ERR(handle));
1252 mdd_write_lock(env, mdd_obj);
1253 /* release open count */
1254 mdd_obj->mod_count --;
1256 rc = mdd_iattr_get(env, mdd_obj, ma);
1257 if (rc == 0 && mdd_obj->mod_count == 0 && ma->ma_attr.la_nlink == 0)
1258 rc = mdd_object_kill(env, mdd_obj, ma);
1260 ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
1262 mdd_write_unlock(env, mdd_obj);
1263 mdd_trans_stop(env, mdo2mdd(obj), rc, handle);
1268 * Permission check is done when open,
1269 * no need check again.
1271 static int mdd_readpage_sanity_check(const struct lu_env *env,
1272 struct mdd_object *obj)
1274 struct dt_object *next = mdd_object_child(obj);
1278 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
1286 static int mdd_dir_page_build(const struct lu_env *env, int first,
1287 void *area, int nob, struct dt_it_ops *iops,
1288 struct dt_it *it, __u32 *start, __u32 *end,
1289 struct lu_dirent **last)
1291 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
1292 struct mdd_thread_info *info = mdd_env_info(env);
1293 struct lu_fid_pack *pack = &info->mti_pack;
1295 struct lu_dirent *ent;
1298 memset(area, 0, sizeof (struct lu_dirpage));
1299 area += sizeof (struct lu_dirpage);
1300 nob -= sizeof (struct lu_dirpage);
1303 LASSERT(nob > sizeof *ent);
1313 name = (char *)iops->key(env, it);
1314 len = iops->key_size(env, it);
1316 pack = (struct lu_fid_pack *)iops->rec(env, it);
1317 result = fid_unpack(pack, fid);
1321 recsize = (sizeof(*ent) + len + 3) & ~3;
1322 hash = iops->store(env, it);
1325 CDEBUG(D_INFO, "%p %p %d "DFID": %#8.8x (%d) \"%*.*s\"\n",
1326 name, ent, nob, PFID(fid), hash, len, len, len, name);
1328 if (nob >= recsize) {
1329 ent->lde_fid = *fid;
1330 fid_cpu_to_le(&ent->lde_fid, &ent->lde_fid);
1331 ent->lde_hash = hash;
1332 ent->lde_namelen = cpu_to_le16(len);
1333 ent->lde_reclen = cpu_to_le16(recsize);
1334 memcpy(ent->lde_name, name, len);
1335 if (first && ent == area)
1338 ent = (void *)ent + recsize;
1340 result = iops->next(env, it);
1343 * record doesn't fit into page, enlarge previous one.
1345 LASSERT(*last != NULL);
1346 (*last)->lde_reclen =
1347 cpu_to_le16(le16_to_cpu((*last)->lde_reclen) +
1351 } while (result == 0);
1356 static int __mdd_readpage(const struct lu_env *env, struct mdd_object *obj,
1357 const struct lu_rdpg *rdpg)
1360 struct dt_object *next = mdd_object_child(obj);
1361 struct dt_it_ops *iops;
1363 struct lu_dirent *last;
1370 LASSERT(rdpg->rp_pages != NULL);
1371 LASSERT(next->do_index_ops != NULL);
1373 if (rdpg->rp_count <= 0)
1377 * iterate through directory and fill pages from @rdpg
1379 iops = &next->do_index_ops->dio_it;
1380 it = iops->init(env, next, 0, mdd_object_capa(env, obj));
1384 rc = iops->load(env, it, rdpg->rp_hash);
1388 * Iterator didn't find record with exactly the key requested.
1390 * It is currently either
1392 * - positioned above record with key less than
1393 * requested---skip it.
1395 * - or not positioned at all (is in IAM_IT_SKEWED
1396 * state)---position it on the next item.
1398 rc = iops->next(env, it);
1403 * At this point and across for-loop:
1405 * rc == 0 -> ok, proceed.
1406 * rc > 0 -> end of directory.
1409 for (i = 0, nob = rdpg->rp_count; rc == 0 && nob > 0;
1410 i++, nob -= CFS_PAGE_SIZE) {
1411 LASSERT(i < rdpg->rp_npages);
1412 pg = rdpg->rp_pages[i];
1413 rc = mdd_dir_page_build(env, !i, kmap(pg),
1414 min_t(int, nob, CFS_PAGE_SIZE), iops,
1415 it, &hash_start, &hash_end, &last);
1416 if (rc != 0 || i == rdpg->rp_npages - 1)
1417 last->lde_reclen = 0;
1424 hash_end = DIR_END_OFF;
1428 struct lu_dirpage *dp;
1430 dp = kmap(rdpg->rp_pages[0]);
1431 dp->ldp_hash_start = rdpg->rp_hash;
1432 dp->ldp_hash_end = hash_end;
1435 * No pages were processed, mark this.
1437 dp->ldp_flags |= LDF_EMPTY;
1438 dp->ldp_flags = cpu_to_le16(dp->ldp_flags);
1439 kunmap(rdpg->rp_pages[0]);
1442 iops->fini(env, it);
1447 static int mdd_readpage(const struct lu_env *env, struct md_object *obj,
1448 const struct lu_rdpg *rdpg)
1450 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1454 LASSERT(mdd_object_exists(mdd_obj));
1456 mdd_read_lock(env, mdd_obj);
1457 rc = mdd_readpage_sanity_check(env, mdd_obj);
1459 GOTO(out_unlock, rc);
1461 if (mdd_is_dead_obj(mdd_obj)) {
1463 struct lu_dirpage *dp;
1466 * According to POSIX, please do not return any entry to client:
1467 * even dot and dotdot should not be returned.
1469 CWARN("readdir from dead object: "DFID"\n",
1470 PFID(mdd_object_fid(mdd_obj)));
1472 if (rdpg->rp_count <= 0)
1473 GOTO(out_unlock, rc = -EFAULT);
1474 LASSERT(rdpg->rp_pages != NULL);
1476 pg = rdpg->rp_pages[0];
1477 dp = (struct lu_dirpage*)kmap(pg);
1478 memset(dp, 0 , sizeof(struct lu_dirpage));
1479 dp->ldp_hash_start = rdpg->rp_hash;
1480 dp->ldp_hash_end = DIR_END_OFF;
1481 dp->ldp_flags |= LDF_EMPTY;
1482 dp->ldp_flags = cpu_to_le16(dp->ldp_flags);
1484 GOTO(out_unlock, rc = 0);
1487 rc = __mdd_readpage(env, mdd_obj, rdpg);
1491 mdd_read_unlock(env, mdd_obj);
1495 struct md_object_operations mdd_obj_ops = {
1496 .moo_permission = mdd_permission,
1497 .moo_attr_get = mdd_attr_get,
1498 .moo_attr_set = mdd_attr_set,
1499 .moo_xattr_get = mdd_xattr_get,
1500 .moo_xattr_set = mdd_xattr_set,
1501 .moo_xattr_list = mdd_xattr_list,
1502 .moo_xattr_del = mdd_xattr_del,
1503 .moo_object_create = mdd_object_create,
1504 .moo_ref_add = mdd_ref_add,
1505 .moo_ref_del = mdd_ref_del,
1506 .moo_open = mdd_open,
1507 .moo_close = mdd_close,
1508 .moo_readpage = mdd_readpage,
1509 .moo_readlink = mdd_readlink,
1510 .moo_capa_get = mdd_capa_get