1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdd/mdd_object.c
38 * Lustre Metadata Server (mdd) routines
40 * Author: Wang Di <wangdi@clusterfs.com>
44 # define EXPORT_SYMTAB
46 #define DEBUG_SUBSYSTEM S_MDS
48 #include <linux/module.h>
49 #ifdef HAVE_EXT4_LDISKFS
50 #include <ldiskfs/ldiskfs_jbd2.h>
52 #include <linux/jbd.h>
55 #include <obd_class.h>
56 #include <obd_support.h>
57 #include <lprocfs_status.h>
58 /* fid_be_cpu(), fid_cpu_to_be(). */
59 #include <lustre_fid.h>
61 #include <lustre_param.h>
62 #ifdef HAVE_EXT4_LDISKFS
63 #include <ldiskfs/ldiskfs.h>
65 #include <linux/ldiskfs_fs.h>
67 #include <lustre_mds.h>
68 #include <lustre/lustre_idl.h>
70 #include "mdd_internal.h"
72 static const struct lu_object_operations mdd_lu_obj_ops;
74 static int mdd_xattr_get(const struct lu_env *env,
75 struct md_object *obj, struct lu_buf *buf,
78 int mdd_data_get(const struct lu_env *env, struct mdd_object *obj,
81 LASSERTF(mdd_object_exists(obj), "FID is "DFID"\n",
82 PFID(mdd_object_fid(obj)));
83 mdo_data_get(env, obj, data);
87 int mdd_la_get(const struct lu_env *env, struct mdd_object *obj,
88 struct lu_attr *la, struct lustre_capa *capa)
90 LASSERTF(mdd_object_exists(obj), "FID is "DFID"\n",
91 PFID(mdd_object_fid(obj)));
92 return mdo_attr_get(env, obj, la, capa);
95 static void mdd_flags_xlate(struct mdd_object *obj, __u32 flags)
97 obj->mod_flags &= ~(APPEND_OBJ|IMMUTE_OBJ);
99 if (flags & LUSTRE_APPEND_FL)
100 obj->mod_flags |= APPEND_OBJ;
102 if (flags & LUSTRE_IMMUTABLE_FL)
103 obj->mod_flags |= IMMUTE_OBJ;
106 struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
108 struct mdd_thread_info *info;
110 info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
111 LASSERT(info != NULL);
115 struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
119 buf = &mdd_env_info(env)->mti_buf;
125 void mdd_buf_put(struct lu_buf *buf)
127 if (buf == NULL || buf->lb_buf == NULL)
130 OBD_VFREE(buf->lb_buf, buf->lb_len);
132 OBD_FREE(buf->lb_buf, buf->lb_len);
137 const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
138 const void *area, ssize_t len)
142 buf = &mdd_env_info(env)->mti_buf;
143 buf->lb_buf = (void *)area;
148 #define BUF_VMALLOC_SIZE (CFS_PAGE_SIZE<<2) /* 16k */
149 struct lu_buf *mdd_buf_alloc(const struct lu_env *env, ssize_t len)
151 struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf;
153 if ((len > buf->lb_len) && (buf->lb_buf != NULL)) {
155 OBD_VFREE(buf->lb_buf, buf->lb_len);
157 OBD_FREE(buf->lb_buf, buf->lb_len);
160 if (buf->lb_buf == NULL) {
162 if (buf->lb_len <= BUF_VMALLOC_SIZE) {
163 OBD_ALLOC(buf->lb_buf, buf->lb_len);
166 if (buf->lb_buf == NULL) {
167 OBD_VMALLOC(buf->lb_buf, buf->lb_len);
170 if (buf->lb_buf == NULL)
176 /** Increase the size of the \a mti_big_buf.
177 * preserves old data in buffer
178 * old buffer remains unchanged on error
179 * \retval 0 or -ENOMEM
181 int mdd_buf_grow(const struct lu_env *env, ssize_t len)
183 struct lu_buf *oldbuf = &mdd_env_info(env)->mti_big_buf;
186 LASSERT(len >= oldbuf->lb_len);
187 if (len > BUF_VMALLOC_SIZE) {
188 OBD_VMALLOC(buf.lb_buf, len);
191 OBD_ALLOC(buf.lb_buf, len);
194 if (buf.lb_buf == NULL)
198 memcpy(buf.lb_buf, oldbuf->lb_buf, oldbuf->lb_len);
200 if (oldbuf->lb_vmalloc)
201 OBD_VFREE(oldbuf->lb_buf, oldbuf->lb_len);
203 OBD_FREE(oldbuf->lb_buf, oldbuf->lb_len);
205 memcpy(oldbuf, &buf, sizeof(buf));
210 struct llog_cookie *mdd_max_cookie_get(const struct lu_env *env,
211 struct mdd_device *mdd)
213 struct mdd_thread_info *mti = mdd_env_info(env);
216 max_cookie_size = mdd_lov_cookiesize(env, mdd);
217 if (unlikely(mti->mti_max_cookie_size < max_cookie_size)) {
218 if (mti->mti_max_cookie)
219 OBD_FREE(mti->mti_max_cookie, mti->mti_max_cookie_size);
220 mti->mti_max_cookie = NULL;
221 mti->mti_max_cookie_size = 0;
223 if (unlikely(mti->mti_max_cookie == NULL)) {
224 OBD_ALLOC(mti->mti_max_cookie, max_cookie_size);
225 if (likely(mti->mti_max_cookie != NULL))
226 mti->mti_max_cookie_size = max_cookie_size;
228 if (likely(mti->mti_max_cookie != NULL))
229 memset(mti->mti_max_cookie, 0, mti->mti_max_cookie_size);
230 return mti->mti_max_cookie;
233 struct lov_mds_md *mdd_max_lmm_get(const struct lu_env *env,
234 struct mdd_device *mdd)
236 struct mdd_thread_info *mti = mdd_env_info(env);
239 max_lmm_size = mdd_lov_mdsize(env, mdd);
240 if (unlikely(mti->mti_max_lmm_size < max_lmm_size)) {
241 if (mti->mti_max_lmm)
242 OBD_FREE(mti->mti_max_lmm, mti->mti_max_lmm_size);
243 mti->mti_max_lmm = NULL;
244 mti->mti_max_lmm_size = 0;
246 if (unlikely(mti->mti_max_lmm == NULL)) {
247 OBD_ALLOC(mti->mti_max_lmm, max_lmm_size);
248 if (unlikely(mti->mti_max_lmm != NULL))
249 mti->mti_max_lmm_size = max_lmm_size;
251 return mti->mti_max_lmm;
254 struct lu_object *mdd_object_alloc(const struct lu_env *env,
255 const struct lu_object_header *hdr,
258 struct mdd_object *mdd_obj;
260 OBD_ALLOC_PTR(mdd_obj);
261 if (mdd_obj != NULL) {
264 o = mdd2lu_obj(mdd_obj);
265 lu_object_init(o, NULL, d);
266 mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
267 mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
268 mdd_obj->mod_count = 0;
269 o->lo_ops = &mdd_lu_obj_ops;
276 static int mdd_object_init(const struct lu_env *env, struct lu_object *o,
277 const struct lu_object_conf *unused)
279 struct mdd_device *d = lu2mdd_dev(o->lo_dev);
280 struct mdd_object *mdd_obj = lu2mdd_obj(o);
281 struct lu_object *below;
282 struct lu_device *under;
285 mdd_obj->mod_cltime = 0;
286 under = &d->mdd_child->dd_lu_dev;
287 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
288 mdd_pdlock_init(mdd_obj);
292 lu_object_add(o, below);
297 static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
299 if (lu_object_exists(o))
300 return mdd_get_flags(env, lu2mdd_obj(o));
305 static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
307 struct mdd_object *mdd = lu2mdd_obj(o);
313 static int mdd_object_print(const struct lu_env *env, void *cookie,
314 lu_printer_t p, const struct lu_object *o)
316 struct mdd_object *mdd = lu2mdd_obj((struct lu_object *)o);
317 return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p(open_count=%d, "
318 "valid=%x, cltime="LPU64", flags=%lx)",
319 mdd, mdd->mod_count, mdd->mod_valid,
320 mdd->mod_cltime, mdd->mod_flags);
323 static const struct lu_object_operations mdd_lu_obj_ops = {
324 .loo_object_init = mdd_object_init,
325 .loo_object_start = mdd_object_start,
326 .loo_object_free = mdd_object_free,
327 .loo_object_print = mdd_object_print,
330 struct mdd_object *mdd_object_find(const struct lu_env *env,
331 struct mdd_device *d,
332 const struct lu_fid *f)
334 return md2mdd_obj(md_object_find_slice(env, &d->mdd_md_dev, f));
337 static int mdd_path2fid(const struct lu_env *env, struct mdd_device *mdd,
338 const char *path, struct lu_fid *fid)
341 struct lu_fid *f = &mdd_env_info(env)->mti_fid;
342 struct mdd_object *obj;
343 struct lu_name *lname = &mdd_env_info(env)->mti_name;
348 /* temp buffer for path element */
349 buf = mdd_buf_alloc(env, PATH_MAX);
350 if (buf->lb_buf == NULL)
353 lname->ln_name = name = buf->lb_buf;
354 lname->ln_namelen = 0;
355 *f = mdd->mdd_root_fid;
362 while (*path != '/' && *path != '\0') {
370 /* find obj corresponding to fid */
371 obj = mdd_object_find(env, mdd, f);
373 GOTO(out, rc = -EREMOTE);
375 GOTO(out, rc = -PTR_ERR(obj));
376 /* get child fid from parent and name */
377 rc = mdd_lookup(env, &obj->mod_obj, lname, f, NULL);
378 mdd_object_put(env, obj);
383 lname->ln_namelen = 0;
392 /** The maximum depth that fid2path() will search.
393 * This is limited only because we want to store the fids for
394 * historical path lookup purposes.
396 #define MAX_PATH_DEPTH 100
398 /** mdd_path() lookup structure. */
399 struct path_lookup_info {
400 __u64 pli_recno; /**< history point */
401 __u64 pli_currec; /**< current record */
402 struct lu_fid pli_fid;
403 struct lu_fid pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
404 struct mdd_object *pli_mdd_obj;
405 char *pli_path; /**< full path */
407 int pli_linkno; /**< which hardlink to follow */
408 int pli_fidcount; /**< number of \a pli_fids */
411 static int mdd_path_current(const struct lu_env *env,
412 struct path_lookup_info *pli)
414 struct mdd_device *mdd = mdo2mdd(&pli->pli_mdd_obj->mod_obj);
415 struct mdd_object *mdd_obj;
416 struct lu_buf *buf = NULL;
417 struct link_ea_header *leh;
418 struct link_ea_entry *lee;
419 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
420 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
426 ptr = pli->pli_path + pli->pli_pathlen - 1;
429 pli->pli_fidcount = 0;
430 pli->pli_fids[0] = *(struct lu_fid *)mdd_object_fid(pli->pli_mdd_obj);
432 while (!mdd_is_root(mdd, &pli->pli_fids[pli->pli_fidcount])) {
433 mdd_obj = mdd_object_find(env, mdd,
434 &pli->pli_fids[pli->pli_fidcount]);
436 GOTO(out, rc = -EREMOTE);
438 GOTO(out, rc = -PTR_ERR(mdd_obj));
439 rc = lu_object_exists(&mdd_obj->mod_obj.mo_lu);
441 mdd_object_put(env, mdd_obj);
445 /* Do I need to error out here? */
450 /* Get parent fid and object name */
451 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
452 buf = mdd_links_get(env, mdd_obj);
453 mdd_read_unlock(env, mdd_obj);
454 mdd_object_put(env, mdd_obj);
456 GOTO(out, rc = PTR_ERR(buf));
459 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
460 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
462 /* If set, use link #linkno for path lookup, otherwise use
463 link #0. Only do this for the final path element. */
464 if ((pli->pli_fidcount == 0) &&
465 (pli->pli_linkno < leh->leh_reccount)) {
467 for (count = 0; count < pli->pli_linkno; count++) {
468 lee = (struct link_ea_entry *)
469 ((char *)lee + reclen);
470 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
472 if (pli->pli_linkno < leh->leh_reccount - 1)
473 /* indicate to user there are more links */
477 /* Pack the name in the end of the buffer */
478 ptr -= tmpname->ln_namelen;
479 if (ptr - 1 <= pli->pli_path)
480 GOTO(out, rc = -EOVERFLOW);
481 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
484 /* Store the parent fid for historic lookup */
485 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
486 GOTO(out, rc = -EOVERFLOW);
487 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
490 /* Verify that our path hasn't changed since we started the lookup.
491 Record the current index, and verify the path resolves to the
492 same fid. If it does, then the path is correct as of this index. */
493 cfs_spin_lock(&mdd->mdd_cl.mc_lock);
494 pli->pli_currec = mdd->mdd_cl.mc_index;
495 cfs_spin_unlock(&mdd->mdd_cl.mc_lock);
496 rc = mdd_path2fid(env, mdd, ptr, &pli->pli_fid);
498 CDEBUG(D_INFO, "mdd_path2fid(%s) failed %d\n", ptr, rc);
499 GOTO (out, rc = -EAGAIN);
501 if (!lu_fid_eq(&pli->pli_fids[0], &pli->pli_fid)) {
502 CDEBUG(D_INFO, "mdd_path2fid(%s) found another FID o="DFID
503 " n="DFID"\n", ptr, PFID(&pli->pli_fids[0]),
504 PFID(&pli->pli_fid));
505 GOTO(out, rc = -EAGAIN);
507 ptr++; /* skip leading / */
508 memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
512 if (buf && !IS_ERR(buf) && buf->lb_vmalloc)
513 /* if we vmalloced a large buffer drop it */
519 static int mdd_path_historic(const struct lu_env *env,
520 struct path_lookup_info *pli)
525 /* Returns the full path to this fid, as of changelog record recno. */
526 static int mdd_path(const struct lu_env *env, struct md_object *obj,
527 char *path, int pathlen, __u64 *recno, int *linkno)
529 struct path_lookup_info *pli;
537 if (mdd_is_root(mdo2mdd(obj), mdd_object_fid(md2mdd_obj(obj)))) {
546 pli->pli_mdd_obj = md2mdd_obj(obj);
547 pli->pli_recno = *recno;
548 pli->pli_path = path;
549 pli->pli_pathlen = pathlen;
550 pli->pli_linkno = *linkno;
552 /* Retry multiple times in case file is being moved */
553 while (tries-- && rc == -EAGAIN)
554 rc = mdd_path_current(env, pli);
556 /* For historical path lookup, the current links may not have existed
557 * at "recno" time. We must switch over to earlier links/parents
558 * by using the changelog records. If the earlier parent doesn't
559 * exist, we must search back through the changelog to reconstruct
560 * its parents, then check if it exists, etc.
561 * We may ignore this problem for the initial implementation and
562 * state that an "original" hardlink must still exist for us to find
563 * historic path name. */
564 if (pli->pli_recno != -1) {
565 rc = mdd_path_historic(env, pli);
567 *recno = pli->pli_currec;
568 /* Return next link index to caller */
569 *linkno = pli->pli_linkno;
577 int mdd_get_flags(const struct lu_env *env, struct mdd_object *obj)
579 struct lu_attr *la = &mdd_env_info(env)->mti_la;
583 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
585 mdd_flags_xlate(obj, la->la_flags);
586 if (S_ISDIR(la->la_mode) && la->la_nlink == 1)
587 obj->mod_flags |= MNLINK_OBJ;
592 /* get only inode attributes */
593 int mdd_iattr_get(const struct lu_env *env, struct mdd_object *mdd_obj,
599 if (ma->ma_valid & MA_INODE)
602 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr,
603 mdd_object_capa(env, mdd_obj));
605 ma->ma_valid |= MA_INODE;
609 int mdd_get_default_md(struct mdd_object *mdd_obj, struct lov_mds_md *lmm)
611 struct lov_desc *ldesc;
612 struct mdd_device *mdd = mdo2mdd(&mdd_obj->mod_obj);
613 struct lov_user_md *lum = (struct lov_user_md*)lmm;
619 ldesc = &mdd->mdd_obd_dev->u.mds.mds_lov_desc;
620 LASSERT(ldesc != NULL);
622 lum->lmm_magic = LOV_MAGIC_V1;
623 lum->lmm_object_seq = LOV_OBJECT_GROUP_DEFAULT;
624 lum->lmm_pattern = ldesc->ld_pattern;
625 lum->lmm_stripe_size = ldesc->ld_default_stripe_size;
626 lum->lmm_stripe_count = ldesc->ld_default_stripe_count;
627 lum->lmm_stripe_offset = ldesc->ld_default_stripe_offset;
629 RETURN(sizeof(*lum));
632 /* get lov EA only */
633 static int __mdd_lmm_get(const struct lu_env *env,
634 struct mdd_object *mdd_obj, struct md_attr *ma)
639 if (ma->ma_valid & MA_LOV)
642 rc = mdd_get_md(env, mdd_obj, ma->ma_lmm, &ma->ma_lmm_size,
644 if (rc == 0 && (ma->ma_need & MA_LOV_DEF))
645 rc = mdd_get_default_md(mdd_obj, ma->ma_lmm);
647 ma->ma_lmm_size = rc;
648 ma->ma_valid |= MA_LOV;
654 int mdd_lmm_get_locked(const struct lu_env *env, struct mdd_object *mdd_obj,
660 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
661 rc = __mdd_lmm_get(env, mdd_obj, ma);
662 mdd_read_unlock(env, mdd_obj);
667 static int __mdd_lmv_get(const struct lu_env *env,
668 struct mdd_object *mdd_obj, struct md_attr *ma)
673 if (ma->ma_valid & MA_LMV)
676 rc = mdd_get_md(env, mdd_obj, ma->ma_lmv, &ma->ma_lmv_size,
679 ma->ma_valid |= MA_LMV;
685 static int __mdd_lma_get(const struct lu_env *env, struct mdd_object *mdd_obj,
688 struct mdd_thread_info *info = mdd_env_info(env);
689 struct lustre_mdt_attrs *lma =
690 (struct lustre_mdt_attrs *)info->mti_xattr_buf;
695 /* If all needed data are already valid, nothing to do */
696 if ((ma->ma_valid & (MA_HSM | MA_SOM)) ==
697 (ma->ma_need & (MA_HSM | MA_SOM)))
700 /* Read LMA from disk EA */
701 lma_size = sizeof(info->mti_xattr_buf);
702 rc = mdd_get_md(env, mdd_obj, lma, &lma_size, XATTR_NAME_LMA);
706 /* Useless to check LMA incompatibility because this is already done in
707 * osd_ea_fid_get(), and this will fail long before this code is
709 * So, if we are here, LMA is compatible.
712 lustre_lma_swab(lma);
714 /* Swab and copy LMA */
715 if (ma->ma_need & MA_HSM) {
716 if (lma->lma_compat & LMAC_HSM)
717 ma->ma_hsm_flags = lma->lma_flags & HSM_FLAGS_MASK;
719 ma->ma_hsm_flags = 0;
720 ma->ma_valid |= MA_HSM;
724 if (ma->ma_need & MA_SOM && lma->lma_compat & LMAC_SOM) {
725 LASSERT(ma->ma_som != NULL);
726 ma->ma_som->msd_ioepoch = lma->lma_ioepoch;
727 ma->ma_som->msd_size = lma->lma_som_size;
728 ma->ma_som->msd_blocks = lma->lma_som_blocks;
729 ma->ma_som->msd_mountid = lma->lma_som_mountid;
730 ma->ma_valid |= MA_SOM;
736 static int mdd_attr_get_internal(const struct lu_env *env,
737 struct mdd_object *mdd_obj,
743 if (ma->ma_need & MA_INODE)
744 rc = mdd_iattr_get(env, mdd_obj, ma);
746 if (rc == 0 && ma->ma_need & MA_LOV) {
747 if (S_ISREG(mdd_object_type(mdd_obj)) ||
748 S_ISDIR(mdd_object_type(mdd_obj)))
749 rc = __mdd_lmm_get(env, mdd_obj, ma);
751 if (rc == 0 && ma->ma_need & MA_LMV) {
752 if (S_ISDIR(mdd_object_type(mdd_obj)))
753 rc = __mdd_lmv_get(env, mdd_obj, ma);
755 if (rc == 0 && ma->ma_need & (MA_HSM | MA_SOM)) {
756 if (S_ISREG(mdd_object_type(mdd_obj)))
757 rc = __mdd_lma_get(env, mdd_obj, ma);
759 #ifdef CONFIG_FS_POSIX_ACL
760 if (rc == 0 && ma->ma_need & MA_ACL_DEF) {
761 if (S_ISDIR(mdd_object_type(mdd_obj)))
762 rc = mdd_def_acl_get(env, mdd_obj, ma);
765 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64"\n",
770 int mdd_attr_get_internal_locked(const struct lu_env *env,
771 struct mdd_object *mdd_obj, struct md_attr *ma)
774 int needlock = ma->ma_need &
775 (MA_LOV | MA_LMV | MA_ACL_DEF | MA_HSM | MA_SOM);
778 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
779 rc = mdd_attr_get_internal(env, mdd_obj, ma);
781 mdd_read_unlock(env, mdd_obj);
786 * No permission check is needed.
788 static int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
791 struct mdd_object *mdd_obj = md2mdd_obj(obj);
795 rc = mdd_attr_get_internal_locked(env, mdd_obj, ma);
800 * No permission check is needed.
802 static int mdd_xattr_get(const struct lu_env *env,
803 struct md_object *obj, struct lu_buf *buf,
806 struct mdd_object *mdd_obj = md2mdd_obj(obj);
811 LASSERT(mdd_object_exists(mdd_obj));
813 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
814 rc = mdo_xattr_get(env, mdd_obj, buf, name,
815 mdd_object_capa(env, mdd_obj));
816 mdd_read_unlock(env, mdd_obj);
822 * Permission check is done when open,
823 * no need check again.
825 static int mdd_readlink(const struct lu_env *env, struct md_object *obj,
828 struct mdd_object *mdd_obj = md2mdd_obj(obj);
829 struct dt_object *next;
834 LASSERT(mdd_object_exists(mdd_obj));
836 next = mdd_object_child(mdd_obj);
837 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
838 rc = next->do_body_ops->dbo_read(env, next, buf, &pos,
839 mdd_object_capa(env, mdd_obj));
840 mdd_read_unlock(env, mdd_obj);
845 * No permission check is needed.
847 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
850 struct mdd_object *mdd_obj = md2mdd_obj(obj);
855 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
856 rc = mdo_xattr_list(env, mdd_obj, buf, mdd_object_capa(env, mdd_obj));
857 mdd_read_unlock(env, mdd_obj);
862 int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p,
863 struct mdd_object *c, struct md_attr *ma,
864 struct thandle *handle,
865 const struct md_op_spec *spec)
867 struct lu_attr *attr = &ma->ma_attr;
868 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
869 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
870 const struct dt_index_features *feat = spec->sp_feat;
874 if (!mdd_object_exists(c)) {
875 struct dt_object *next = mdd_object_child(c);
878 if (feat != &dt_directory_features && feat != NULL)
879 dof->dof_type = DFT_INDEX;
881 dof->dof_type = dt_mode_to_dft(attr->la_mode);
883 dof->u.dof_idx.di_feat = feat;
885 /* @hint will be initialized by underlying device. */
886 next->do_ops->do_ah_init(env, hint,
887 p ? mdd_object_child(p) : NULL,
888 attr->la_mode & S_IFMT);
890 rc = mdo_create_obj(env, c, attr, hint, dof, handle);
891 LASSERT(ergo(rc == 0, mdd_object_exists(c)));
899 * Make sure the ctime is increased only.
901 static inline int mdd_attr_check(const struct lu_env *env,
902 struct mdd_object *obj,
903 struct lu_attr *attr)
905 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
909 if (attr->la_valid & LA_CTIME) {
910 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
914 if (attr->la_ctime < tmp_la->la_ctime)
915 attr->la_valid &= ~(LA_MTIME | LA_CTIME);
916 else if (attr->la_valid == LA_CTIME &&
917 attr->la_ctime == tmp_la->la_ctime)
918 attr->la_valid &= ~LA_CTIME;
923 int mdd_attr_set_internal(const struct lu_env *env,
924 struct mdd_object *obj,
925 struct lu_attr *attr,
926 struct thandle *handle,
932 rc = mdo_attr_set(env, obj, attr, handle, mdd_object_capa(env, obj));
933 #ifdef CONFIG_FS_POSIX_ACL
934 if (!rc && (attr->la_valid & LA_MODE) && needacl)
935 rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
940 int mdd_attr_check_set_internal(const struct lu_env *env,
941 struct mdd_object *obj,
942 struct lu_attr *attr,
943 struct thandle *handle,
949 rc = mdd_attr_check(env, obj, attr);
954 rc = mdd_attr_set_internal(env, obj, attr, handle, needacl);
958 static int mdd_attr_set_internal_locked(const struct lu_env *env,
959 struct mdd_object *obj,
960 struct lu_attr *attr,
961 struct thandle *handle,
967 needacl = needacl && (attr->la_valid & LA_MODE);
969 mdd_write_lock(env, obj, MOR_TGT_CHILD);
970 rc = mdd_attr_set_internal(env, obj, attr, handle, needacl);
972 mdd_write_unlock(env, obj);
976 int mdd_attr_check_set_internal_locked(const struct lu_env *env,
977 struct mdd_object *obj,
978 struct lu_attr *attr,
979 struct thandle *handle,
985 needacl = needacl && (attr->la_valid & LA_MODE);
987 mdd_write_lock(env, obj, MOR_TGT_CHILD);
988 rc = mdd_attr_check_set_internal(env, obj, attr, handle, needacl);
990 mdd_write_unlock(env, obj);
994 int __mdd_xattr_set(const struct lu_env *env, struct mdd_object *obj,
995 const struct lu_buf *buf, const char *name,
996 int fl, struct thandle *handle)
998 struct lustre_capa *capa = mdd_object_capa(env, obj);
1002 if (buf->lb_buf && buf->lb_len > 0)
1003 rc = mdo_xattr_set(env, obj, buf, name, 0, handle, capa);
1004 else if (buf->lb_buf == NULL && buf->lb_len == 0)
1005 rc = mdo_xattr_del(env, obj, name, handle, capa);
1011 * This gives the same functionality as the code between
1012 * sys_chmod and inode_setattr
1013 * chown_common and inode_setattr
1014 * utimes and inode_setattr
1015 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
1017 static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
1018 struct lu_attr *la, const struct md_attr *ma)
1020 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
1021 struct md_ucred *uc;
1028 /* Do not permit change file type */
1029 if (la->la_valid & LA_TYPE)
1032 /* They should not be processed by setattr */
1033 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
1036 /* export destroy does not have ->le_ses, but we may want
1037 * to drop LUSTRE_SOM_FL. */
1043 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
1047 if (la->la_valid == LA_CTIME) {
1048 if (!(ma->ma_attr_flags & MDS_PERM_BYPASS))
1049 /* This is only for set ctime when rename's source is
1051 rc = mdd_may_delete(env, NULL, obj,
1052 (struct md_attr *)ma, 1, 0);
1053 if (rc == 0 && la->la_ctime <= tmp_la->la_ctime)
1054 la->la_valid &= ~LA_CTIME;
1058 if (la->la_valid == LA_ATIME) {
1059 /* This is atime only set for read atime update on close. */
1060 if (la->la_atime <= tmp_la->la_atime +
1061 mdd_obj2mdd_dev(obj)->mdd_atime_diff)
1062 la->la_valid &= ~LA_ATIME;
1066 /* Check if flags change. */
1067 if (la->la_valid & LA_FLAGS) {
1068 unsigned int oldflags = 0;
1069 unsigned int newflags = la->la_flags &
1070 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
1072 if ((uc->mu_fsuid != tmp_la->la_uid) &&
1073 !mdd_capable(uc, CFS_CAP_FOWNER))
1076 /* XXX: the IMMUTABLE and APPEND_ONLY flags can
1077 * only be changed by the relevant capability. */
1078 if (mdd_is_immutable(obj))
1079 oldflags |= LUSTRE_IMMUTABLE_FL;
1080 if (mdd_is_append(obj))
1081 oldflags |= LUSTRE_APPEND_FL;
1082 if ((oldflags ^ newflags) &&
1083 !mdd_capable(uc, CFS_CAP_LINUX_IMMUTABLE))
1086 if (!S_ISDIR(tmp_la->la_mode))
1087 la->la_flags &= ~LUSTRE_DIRSYNC_FL;
1090 if ((mdd_is_immutable(obj) || mdd_is_append(obj)) &&
1091 (la->la_valid & ~LA_FLAGS) &&
1092 !(ma->ma_attr_flags & MDS_PERM_BYPASS))
1095 /* Check for setting the obj time. */
1096 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
1097 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
1098 if ((uc->mu_fsuid != tmp_la->la_uid) &&
1099 !mdd_capable(uc, CFS_CAP_FOWNER)) {
1100 rc = mdd_permission_internal_locked(env, obj, tmp_la,
1108 /* Make sure a caller can chmod. */
1109 if (la->la_valid & LA_MODE) {
1110 /* Bypass la_vaild == LA_MODE,
1111 * this is for changing file with SUID or SGID. */
1112 if ((la->la_valid & ~LA_MODE) &&
1113 !(ma->ma_attr_flags & MDS_PERM_BYPASS) &&
1114 (uc->mu_fsuid != tmp_la->la_uid) &&
1115 !mdd_capable(uc, CFS_CAP_FOWNER))
1118 if (la->la_mode == (cfs_umode_t) -1)
1119 la->la_mode = tmp_la->la_mode;
1121 la->la_mode = (la->la_mode & S_IALLUGO) |
1122 (tmp_la->la_mode & ~S_IALLUGO);
1124 /* Also check the setgid bit! */
1125 if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ?
1126 la->la_gid : tmp_la->la_gid) &&
1127 !mdd_capable(uc, CFS_CAP_FSETID))
1128 la->la_mode &= ~S_ISGID;
1130 la->la_mode = tmp_la->la_mode;
1133 /* Make sure a caller can chown. */
1134 if (la->la_valid & LA_UID) {
1135 if (la->la_uid == (uid_t) -1)
1136 la->la_uid = tmp_la->la_uid;
1137 if (((uc->mu_fsuid != tmp_la->la_uid) ||
1138 (la->la_uid != tmp_la->la_uid)) &&
1139 !mdd_capable(uc, CFS_CAP_CHOWN))
1142 /* If the user or group of a non-directory has been
1143 * changed by a non-root user, remove the setuid bit.
1144 * 19981026 David C Niemi <niemi@tux.org>
1146 * Changed this to apply to all users, including root,
1147 * to avoid some races. This is the behavior we had in
1148 * 2.0. The check for non-root was definitely wrong
1149 * for 2.2 anyway, as it should have been using
1150 * CAP_FSETID rather than fsuid -- 19990830 SD. */
1151 if (((tmp_la->la_mode & S_ISUID) == S_ISUID) &&
1152 !S_ISDIR(tmp_la->la_mode)) {
1153 la->la_mode &= ~S_ISUID;
1154 la->la_valid |= LA_MODE;
1158 /* Make sure caller can chgrp. */
1159 if (la->la_valid & LA_GID) {
1160 if (la->la_gid == (gid_t) -1)
1161 la->la_gid = tmp_la->la_gid;
1162 if (((uc->mu_fsuid != tmp_la->la_uid) ||
1163 ((la->la_gid != tmp_la->la_gid) &&
1164 !lustre_in_group_p(uc, la->la_gid))) &&
1165 !mdd_capable(uc, CFS_CAP_CHOWN))
1168 /* Likewise, if the user or group of a non-directory
1169 * has been changed by a non-root user, remove the
1170 * setgid bit UNLESS there is no group execute bit
1171 * (this would be a file marked for mandatory
1172 * locking). 19981026 David C Niemi <niemi@tux.org>
1174 * Removed the fsuid check (see the comment above) --
1176 if (((tmp_la->la_mode & (S_ISGID | S_IXGRP)) ==
1177 (S_ISGID | S_IXGRP)) && !S_ISDIR(tmp_la->la_mode)) {
1178 la->la_mode &= ~S_ISGID;
1179 la->la_valid |= LA_MODE;
1183 /* For both Size-on-MDS case and truncate case,
1184 * "la->la_valid & (LA_SIZE | LA_BLOCKS)" are ture.
1185 * We distinguish them by "ma->ma_attr_flags & MDS_SOM".
1186 * For SOM case, it is true, the MAY_WRITE perm has been checked
1187 * when open, no need check again. For truncate case, it is false,
1188 * the MAY_WRITE perm should be checked here. */
1189 if (ma->ma_attr_flags & MDS_SOM) {
1190 /* For the "Size-on-MDS" setattr update, merge coming
1191 * attributes with the set in the inode. BUG 10641 */
1192 if ((la->la_valid & LA_ATIME) &&
1193 (la->la_atime <= tmp_la->la_atime))
1194 la->la_valid &= ~LA_ATIME;
1196 /* OST attributes do not have a priority over MDS attributes,
1197 * so drop times if ctime is equal. */
1198 if ((la->la_valid & LA_CTIME) &&
1199 (la->la_ctime <= tmp_la->la_ctime))
1200 la->la_valid &= ~(LA_MTIME | LA_CTIME);
1202 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
1203 if (!((ma->ma_attr_flags & MDS_OPEN_OWNEROVERRIDE) &&
1204 (uc->mu_fsuid == tmp_la->la_uid)) &&
1205 !(ma->ma_attr_flags & MDS_PERM_BYPASS)) {
1206 rc = mdd_permission_internal_locked(env, obj,
1213 if (la->la_valid & LA_CTIME) {
1214 /* The pure setattr, it has the priority over what is
1215 * already set, do not drop it if ctime is equal. */
1216 if (la->la_ctime < tmp_la->la_ctime)
1217 la->la_valid &= ~(LA_ATIME | LA_MTIME |
1225 /** Store a data change changelog record
1226 * If this fails, we must fail the whole transaction; we don't
1227 * want the change to commit without the log entry.
1228 * \param mdd_obj - mdd_object of change
1229 * \param handle - transacion handle
1231 static int mdd_changelog_data_store(const struct lu_env *env,
1232 struct mdd_device *mdd,
1233 enum changelog_rec_type type,
1235 struct mdd_object *mdd_obj,
1236 struct thandle *handle)
1238 const struct lu_fid *tfid = mdo2fid(mdd_obj);
1239 struct llog_changelog_rec *rec;
1245 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
1247 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
1250 LASSERT(handle != NULL);
1251 LASSERT(mdd_obj != NULL);
1253 if ((type >= CL_MTIME) && (type <= CL_ATIME) &&
1254 cfs_time_before_64(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) {
1255 /* Don't need multiple updates in this log */
1256 /* Don't check under lock - no big deal if we get an extra
1261 reclen = llog_data_len(sizeof(*rec));
1262 buf = mdd_buf_alloc(env, reclen);
1263 if (buf->lb_buf == NULL)
1265 rec = (struct llog_changelog_rec *)buf->lb_buf;
1267 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
1268 rec->cr.cr_type = (__u32)type;
1269 rec->cr.cr_tfid = *tfid;
1270 rec->cr.cr_namelen = 0;
1271 mdd_obj->mod_cltime = cfs_time_current_64();
1273 rc = mdd_changelog_llog_write(mdd, rec, handle);
1275 CERROR("changelog failed: rc=%d op%d t"DFID"\n",
1276 rc, type, PFID(tfid));
1284 * Should be called with write lock held.
1286 * \see mdd_lma_set_locked().
1288 static int __mdd_lma_set(const struct lu_env *env, struct mdd_object *mdd_obj,
1289 const struct md_attr *ma, struct thandle *handle)
1291 struct mdd_thread_info *info = mdd_env_info(env);
1293 struct lustre_mdt_attrs *lma =
1294 (struct lustre_mdt_attrs *) info->mti_xattr_buf;
1295 int lmasize = sizeof(struct lustre_mdt_attrs);
1300 /* Either HSM or SOM part is not valid, we need to read it before */
1301 if ((!ma->ma_valid) & (MA_HSM | MA_SOM)) {
1302 rc = mdd_get_md(env, mdd_obj, lma, &lmasize, XATTR_NAME_LMA);
1306 lustre_lma_swab(lma);
1308 memset(lma, 0, lmasize);
1312 if (ma->ma_valid & MA_HSM) {
1313 lma->lma_flags |= ma->ma_hsm_flags & HSM_FLAGS_MASK;
1314 lma->lma_compat |= LMAC_HSM;
1318 if (ma->ma_valid & MA_SOM) {
1319 LASSERT(ma->ma_som != NULL);
1320 if (ma->ma_som->msd_ioepoch == IOEPOCH_INVAL) {
1321 lma->lma_compat &= ~LMAC_SOM;
1323 lma->lma_compat |= LMAC_SOM;
1324 lma->lma_ioepoch = ma->ma_som->msd_ioepoch;
1325 lma->lma_som_size = ma->ma_som->msd_size;
1326 lma->lma_som_blocks = ma->ma_som->msd_blocks;
1327 lma->lma_som_mountid = ma->ma_som->msd_mountid;
1332 memcpy(&lma->lma_self_fid, mdo2fid(mdd_obj), sizeof(lma->lma_self_fid));
1334 lustre_lma_swab(lma);
1335 buf = mdd_buf_get(env, lma, lmasize);
1336 rc = __mdd_xattr_set(env, mdd_obj, buf, XATTR_NAME_LMA, 0, handle);
1342 * Save LMA extended attributes with data from \a ma.
1344 * HSM and Size-On-MDS data will be extracted from \ma if they are valid, if
1345 * not, LMA EA will be first read from disk, modified and write back.
1348 static int mdd_lma_set_locked(const struct lu_env *env,
1349 struct mdd_object *mdd_obj,
1350 const struct md_attr *ma, struct thandle *handle)
1354 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1355 rc = __mdd_lma_set(env, mdd_obj, ma, handle);
1356 mdd_write_unlock(env, mdd_obj);
1360 /* Precedence for choosing record type when multiple
1361 * attributes change: setattr > mtime > ctime > atime
1362 * (ctime changes when mtime does, plus chmod/chown.
1363 * atime and ctime are independent.) */
1364 static int mdd_attr_set_changelog(const struct lu_env *env,
1365 struct md_object *obj, struct thandle *handle,
1368 struct mdd_device *mdd = mdo2mdd(obj);
1371 bits = (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0;
1372 bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0;
1373 bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0;
1374 bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0;
1375 bits = bits & mdd->mdd_cl.mc_mask;
1379 /* The record type is the lowest non-masked set bit */
1380 while (bits && ((bits & 1) == 0)) {
1385 /* FYI we only store the first CLF_FLAGMASK bits of la_valid */
1386 return mdd_changelog_data_store(env, mdd, type, (int)valid,
1387 md2mdd_obj(obj), handle);
1390 /* set attr and LOV EA at once, return updated attr */
1391 static int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
1392 const struct md_attr *ma)
1394 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1395 struct mdd_device *mdd = mdo2mdd(obj);
1396 struct thandle *handle;
1397 struct lov_mds_md *lmm = NULL;
1398 struct llog_cookie *logcookies = NULL;
1399 int rc, lmm_size = 0, cookie_size = 0;
1400 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
1401 #ifdef HAVE_QUOTA_SUPPORT
1402 struct obd_device *obd = mdd->mdd_obd_dev;
1403 struct mds_obd *mds = &obd->u.mds;
1404 unsigned int qnids[MAXQUOTAS] = { 0, 0 };
1405 unsigned int qoids[MAXQUOTAS] = { 0, 0 };
1406 int quota_opc = 0, block_count = 0;
1407 int inode_pending[MAXQUOTAS] = { 0, 0 };
1408 int block_pending[MAXQUOTAS] = { 0, 0 };
1412 mdd_setattr_txn_param_build(env, obj, (struct md_attr *)ma,
1413 MDD_TXN_ATTR_SET_OP);
1414 handle = mdd_trans_start(env, mdd);
1416 RETURN(PTR_ERR(handle));
1417 /*TODO: add lock here*/
1418 /* start a log jounal handle if needed */
1419 if (S_ISREG(mdd_object_type(mdd_obj)) &&
1420 ma->ma_attr.la_valid & (LA_UID | LA_GID)) {
1421 lmm_size = mdd_lov_mdsize(env, mdd);
1422 lmm = mdd_max_lmm_get(env, mdd);
1424 GOTO(cleanup, rc = -ENOMEM);
1426 rc = mdd_get_md_locked(env, mdd_obj, lmm, &lmm_size,
1433 if (ma->ma_attr.la_valid & (LA_MTIME | LA_CTIME))
1434 CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n",
1435 ma->ma_attr.la_mtime, ma->ma_attr.la_ctime);
1437 *la_copy = ma->ma_attr;
1438 rc = mdd_fix_attr(env, mdd_obj, la_copy, ma);
1442 #ifdef HAVE_QUOTA_SUPPORT
1443 if (mds->mds_quota && la_copy->la_valid & (LA_UID | LA_GID)) {
1444 struct obd_export *exp = md_quota(env)->mq_exp;
1445 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1447 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
1449 quota_opc = FSFILT_OP_SETATTR;
1450 mdd_quota_wrapper(la_copy, qnids);
1451 mdd_quota_wrapper(la_tmp, qoids);
1452 /* get file quota for new owner */
1453 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1454 qnids, inode_pending, 1, NULL, 0,
1456 block_count = (la_tmp->la_blocks + 7) >> 3;
1459 mdd_data_get(env, mdd_obj, &data);
1460 /* get block quota for new owner */
1461 lquota_chkquota(mds_quota_interface_ref, obd,
1462 exp, qnids, block_pending,
1464 LQUOTA_FLAGS_BLK, data, 1);
1470 if (la_copy->la_valid & LA_FLAGS) {
1471 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
1474 mdd_flags_xlate(mdd_obj, la_copy->la_flags);
1475 } else if (la_copy->la_valid) { /* setattr */
1476 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
1478 /* journal chown/chgrp in llog, just like unlink */
1479 if (rc == 0 && lmm_size){
1480 cookie_size = mdd_lov_cookiesize(env, mdd);
1481 logcookies = mdd_max_cookie_get(env, mdd);
1482 if (logcookies == NULL)
1483 GOTO(cleanup, rc = -ENOMEM);
1485 if (mdd_setattr_log(env, mdd, ma, lmm, lmm_size,
1486 logcookies, cookie_size) <= 0)
1491 if (rc == 0 && ma->ma_valid & MA_LOV) {
1494 mode = mdd_object_type(mdd_obj);
1495 if (S_ISREG(mode) || S_ISDIR(mode)) {
1496 rc = mdd_lsm_sanity_check(env, mdd_obj);
1500 rc = mdd_lov_set_md(env, NULL, mdd_obj, ma->ma_lmm,
1501 ma->ma_lmm_size, handle, 1);
1505 if (rc == 0 && ma->ma_valid & (MA_HSM | MA_SOM)) {
1508 mode = mdd_object_type(mdd_obj);
1510 rc = mdd_lma_set_locked(env, mdd_obj, ma, handle);
1515 rc = mdd_attr_set_changelog(env, obj, handle,
1516 ma->ma_attr.la_valid);
1517 mdd_trans_stop(env, mdd, rc, handle);
1518 if (rc == 0 && (lmm != NULL && lmm_size > 0 )) {
1519 /*set obd attr, if needed*/
1520 rc = mdd_lov_setattr_async(env, mdd_obj, lmm, lmm_size,
1523 #ifdef HAVE_QUOTA_SUPPORT
1525 lquota_pending_commit(mds_quota_interface_ref, obd, qnids,
1527 lquota_pending_commit(mds_quota_interface_ref, obd, qnids,
1529 /* Trigger dqrel/dqacq for original owner and new owner.
1530 * If failed, the next call for lquota_chkquota will
1532 lquota_adjust(mds_quota_interface_ref, obd, qnids, qoids, rc,
1539 int mdd_xattr_set_txn(const struct lu_env *env, struct mdd_object *obj,
1540 const struct lu_buf *buf, const char *name, int fl,
1541 struct thandle *handle)
1546 mdd_write_lock(env, obj, MOR_TGT_CHILD);
1547 rc = __mdd_xattr_set(env, obj, buf, name, fl, handle);
1548 mdd_write_unlock(env, obj);
1553 static int mdd_xattr_sanity_check(const struct lu_env *env,
1554 struct mdd_object *obj)
1556 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
1557 struct md_ucred *uc = md_ucred(env);
1561 if (mdd_is_immutable(obj) || mdd_is_append(obj))
1564 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
1568 if ((uc->mu_fsuid != tmp_la->la_uid) &&
1569 !mdd_capable(uc, CFS_CAP_FOWNER))
1576 * The caller should guarantee to update the object ctime
1577 * after xattr_set if needed.
1579 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
1580 const struct lu_buf *buf, const char *name,
1583 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1584 struct mdd_device *mdd = mdo2mdd(obj);
1585 struct thandle *handle;
1589 rc = mdd_xattr_sanity_check(env, mdd_obj);
1593 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
1594 /* security-replated changes may require sync */
1595 if (!strcmp(name, XATTR_NAME_ACL_ACCESS) &&
1596 mdd->mdd_sync_permission == 1)
1597 txn_param_sync(&mdd_env_info(env)->mti_param);
1599 handle = mdd_trans_start(env, mdd);
1601 RETURN(PTR_ERR(handle));
1603 rc = mdd_xattr_set_txn(env, mdd_obj, buf, name, fl, handle);
1605 /* Only record user xattr changes */
1606 if ((rc == 0) && (strncmp("user.", name, 5) == 0))
1607 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1609 mdd_trans_stop(env, mdd, rc, handle);
1615 * The caller should guarantee to update the object ctime
1616 * after xattr_set if needed.
1618 int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1621 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1622 struct mdd_device *mdd = mdo2mdd(obj);
1623 struct thandle *handle;
1627 rc = mdd_xattr_sanity_check(env, mdd_obj);
1631 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
1632 handle = mdd_trans_start(env, mdd);
1634 RETURN(PTR_ERR(handle));
1636 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1637 rc = mdo_xattr_del(env, mdd_obj, name, handle,
1638 mdd_object_capa(env, mdd_obj));
1639 mdd_write_unlock(env, mdd_obj);
1641 /* Only record user xattr changes */
1642 if ((rc == 0) && (strncmp("user.", name, 5) != 0))
1643 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1646 mdd_trans_stop(env, mdd, rc, handle);
1651 /* partial unlink */
1652 static int mdd_ref_del(const struct lu_env *env, struct md_object *obj,
1655 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
1656 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1657 struct mdd_device *mdd = mdo2mdd(obj);
1658 struct thandle *handle;
1659 #ifdef HAVE_QUOTA_SUPPORT
1660 struct obd_device *obd = mdd->mdd_obd_dev;
1661 struct mds_obd *mds = &obd->u.mds;
1662 unsigned int qids[MAXQUOTAS] = { 0, 0 };
1669 * Check -ENOENT early here because we need to get object type
1670 * to calculate credits before transaction start
1672 if (!mdd_object_exists(mdd_obj))
1675 LASSERT(mdd_object_exists(mdd_obj) > 0);
1677 rc = mdd_log_txn_param_build(env, obj, ma, MDD_TXN_UNLINK_OP);
1681 handle = mdd_trans_start(env, mdd);
1685 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1687 rc = mdd_unlink_sanity_check(env, NULL, mdd_obj, ma);
1691 __mdd_ref_del(env, mdd_obj, handle, 0);
1693 if (S_ISDIR(lu_object_attr(&obj->mo_lu))) {
1695 __mdd_ref_del(env, mdd_obj, handle, 1);
1698 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1699 la_copy->la_ctime = ma->ma_attr.la_ctime;
1701 la_copy->la_valid = LA_CTIME;
1702 rc = mdd_attr_check_set_internal(env, mdd_obj, la_copy, handle, 0);
1706 rc = mdd_finish_unlink(env, mdd_obj, ma, handle);
1707 #ifdef HAVE_QUOTA_SUPPORT
1708 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
1709 ma->ma_attr.la_nlink == 0 && mdd_obj->mod_count == 0) {
1710 quota_opc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
1711 mdd_quota_wrapper(&ma->ma_attr, qids);
1718 mdd_write_unlock(env, mdd_obj);
1719 mdd_trans_stop(env, mdd, rc, handle);
1720 #ifdef HAVE_QUOTA_SUPPORT
1722 /* Trigger dqrel on the owner of child. If failed,
1723 * the next call for lquota_chkquota will process it */
1724 lquota_adjust(mds_quota_interface_ref, obd, qids, 0, rc,
1730 /* partial operation */
1731 static int mdd_oc_sanity_check(const struct lu_env *env,
1732 struct mdd_object *obj,
1738 switch (ma->ma_attr.la_mode & S_IFMT) {
1755 static int mdd_object_create(const struct lu_env *env,
1756 struct md_object *obj,
1757 const struct md_op_spec *spec,
1761 struct mdd_device *mdd = mdo2mdd(obj);
1762 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1763 const struct lu_fid *pfid = spec->u.sp_pfid;
1764 struct thandle *handle;
1765 #ifdef HAVE_QUOTA_SUPPORT
1766 struct obd_device *obd = mdd->mdd_obd_dev;
1767 struct obd_export *exp = md_quota(env)->mq_exp;
1768 struct mds_obd *mds = &obd->u.mds;
1769 unsigned int qids[MAXQUOTAS] = { 0, 0 };
1770 int quota_opc = 0, block_count = 0;
1771 int inode_pending[MAXQUOTAS] = { 0, 0 };
1772 int block_pending[MAXQUOTAS] = { 0, 0 };
1777 #ifdef HAVE_QUOTA_SUPPORT
1778 if (mds->mds_quota) {
1779 quota_opc = FSFILT_OP_CREATE_PARTIAL_CHILD;
1780 mdd_quota_wrapper(&ma->ma_attr, qids);
1781 /* get file quota for child */
1782 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1783 qids, inode_pending, 1, NULL, 0,
1785 switch (ma->ma_attr.la_mode & S_IFMT) {
1794 /* get block quota for child */
1796 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1797 qids, block_pending, block_count,
1798 NULL, LQUOTA_FLAGS_BLK, NULL, 0);
1802 mdd_txn_param_build(env, mdd, MDD_TXN_OBJECT_CREATE_OP);
1803 handle = mdd_trans_start(env, mdd);
1805 GOTO(out_pending, rc = PTR_ERR(handle));
1807 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1808 rc = mdd_oc_sanity_check(env, mdd_obj, ma);
1812 rc = mdd_object_create_internal(env, NULL, mdd_obj, ma, handle, spec);
1816 if (spec->sp_cr_flags & MDS_CREATE_SLAVE_OBJ) {
1817 /* If creating the slave object, set slave EA here. */
1818 int lmv_size = spec->u.sp_ea.eadatalen;
1819 struct lmv_stripe_md *lmv;
1821 lmv = (struct lmv_stripe_md *)spec->u.sp_ea.eadata;
1822 LASSERT(lmv != NULL && lmv_size > 0);
1824 rc = __mdd_xattr_set(env, mdd_obj,
1825 mdd_buf_get_const(env, lmv, lmv_size),
1826 XATTR_NAME_LMV, 0, handle);
1830 rc = mdd_attr_set_internal(env, mdd_obj, &ma->ma_attr,
1833 #ifdef CONFIG_FS_POSIX_ACL
1834 if (spec->sp_cr_flags & MDS_CREATE_RMT_ACL) {
1835 struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
1837 buf->lb_buf = (void *)spec->u.sp_ea.eadata;
1838 buf->lb_len = spec->u.sp_ea.eadatalen;
1839 if ((buf->lb_len > 0) && (buf->lb_buf != NULL)) {
1840 rc = __mdd_acl_init(env, mdd_obj, buf,
1841 &ma->ma_attr.la_mode,
1846 ma->ma_attr.la_valid |= LA_MODE;
1849 pfid = spec->u.sp_ea.fid;
1852 rc = mdd_object_initialize(env, pfid, NULL, mdd_obj, ma, handle,
1858 rc = mdd_attr_get_internal(env, mdd_obj, ma);
1859 mdd_write_unlock(env, mdd_obj);
1861 mdd_trans_stop(env, mdd, rc, handle);
1863 #ifdef HAVE_QUOTA_SUPPORT
1865 lquota_pending_commit(mds_quota_interface_ref, obd, qids,
1867 lquota_pending_commit(mds_quota_interface_ref, obd, qids,
1869 /* Trigger dqacq on the owner of child. If failed,
1870 * the next call for lquota_chkquota will process it. */
1871 lquota_adjust(mds_quota_interface_ref, obd, qids, 0, rc,
1879 static int mdd_ref_add(const struct lu_env *env, struct md_object *obj,
1880 const struct md_attr *ma)
1882 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
1883 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1884 struct mdd_device *mdd = mdo2mdd(obj);
1885 struct thandle *handle;
1889 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
1890 handle = mdd_trans_start(env, mdd);
1894 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1895 rc = mdd_link_sanity_check(env, NULL, NULL, mdd_obj);
1897 __mdd_ref_add(env, mdd_obj, handle);
1898 mdd_write_unlock(env, mdd_obj);
1900 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1901 la_copy->la_ctime = ma->ma_attr.la_ctime;
1903 la_copy->la_valid = LA_CTIME;
1904 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la_copy,
1907 mdd_trans_stop(env, mdd, 0, handle);
1913 * do NOT or the MAY_*'s, you'll get the weakest
1915 int accmode(const struct lu_env *env, struct lu_attr *la, int flags)
1919 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1920 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1921 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1922 * owner can write to a file even if it is marked readonly to hide
1923 * its brokenness. (bug 5781) */
1924 if (flags & MDS_OPEN_OWNEROVERRIDE) {
1925 struct md_ucred *uc = md_ucred(env);
1927 if ((uc == NULL) || (uc->mu_valid == UCRED_INIT) ||
1928 (la->la_uid == uc->mu_fsuid))
1932 if (flags & FMODE_READ)
1934 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1936 if (flags & MDS_FMODE_EXEC)
1941 static int mdd_open_sanity_check(const struct lu_env *env,
1942 struct mdd_object *obj, int flag)
1944 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
1949 if (mdd_is_dead_obj(obj))
1952 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
1956 if (S_ISLNK(tmp_la->la_mode))
1959 mode = accmode(env, tmp_la, flag);
1961 if (S_ISDIR(tmp_la->la_mode) && (mode & MAY_WRITE))
1964 if (!(flag & MDS_OPEN_CREATED)) {
1965 rc = mdd_permission_internal(env, obj, tmp_la, mode);
1970 if (S_ISFIFO(tmp_la->la_mode) || S_ISSOCK(tmp_la->la_mode) ||
1971 S_ISBLK(tmp_la->la_mode) || S_ISCHR(tmp_la->la_mode))
1972 flag &= ~MDS_OPEN_TRUNC;
1974 /* For writing append-only file must open it with append mode. */
1975 if (mdd_is_append(obj)) {
1976 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1978 if (flag & MDS_OPEN_TRUNC)
1984 * Now, flag -- O_NOATIME does not be packed by client.
1986 if (flag & O_NOATIME) {
1987 struct md_ucred *uc = md_ucred(env);
1989 if (uc && ((uc->mu_valid == UCRED_OLD) ||
1990 (uc->mu_valid == UCRED_NEW)) &&
1991 (uc->mu_fsuid != tmp_la->la_uid) &&
1992 !mdd_capable(uc, CFS_CAP_FOWNER))
2000 static int mdd_open(const struct lu_env *env, struct md_object *obj,
2003 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2006 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
2008 rc = mdd_open_sanity_check(env, mdd_obj, flags);
2010 mdd_obj->mod_count++;
2012 mdd_write_unlock(env, mdd_obj);
2016 /* return md_attr back,
2017 * if it is last unlink then return lov ea + llog cookie*/
2018 int mdd_object_kill(const struct lu_env *env, struct mdd_object *obj,
2024 if (S_ISREG(mdd_object_type(obj))) {
2025 /* Return LOV & COOKIES unconditionally here. We clean evth up.
2026 * Caller must be ready for that. */
2028 rc = __mdd_lmm_get(env, obj, ma);
2029 if ((ma->ma_valid & MA_LOV))
2030 rc = mdd_unlink_log(env, mdo2mdd(&obj->mod_obj),
2037 * No permission check is needed.
2039 static int mdd_close(const struct lu_env *env, struct md_object *obj,
2042 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2043 struct mdd_device *mdd = mdo2mdd(obj);
2044 struct thandle *handle;
2048 #ifdef HAVE_QUOTA_SUPPORT
2049 struct obd_device *obd = mdo2mdd(obj)->mdd_obd_dev;
2050 struct mds_obd *mds = &obd->u.mds;
2051 unsigned int qids[MAXQUOTAS] = { 0, 0 };
2056 rc = mdd_log_txn_param_build(env, obj, ma, MDD_TXN_UNLINK_OP);
2059 handle = mdd_trans_start(env, mdo2mdd(obj));
2061 RETURN(PTR_ERR(handle));
2063 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
2064 /* release open count */
2065 mdd_obj->mod_count --;
2067 if (mdd_obj->mod_count == 0 && mdd_obj->mod_flags & ORPHAN_OBJ) {
2068 /* remove link to object from orphan index */
2069 rc = __mdd_orphan_del(env, mdd_obj, handle);
2071 CDEBUG(D_HA, "Object "DFID" is deleted from orphan "
2072 "list, OSS objects to be destroyed.\n",
2073 PFID(mdd_object_fid(mdd_obj)));
2075 CERROR("Object "DFID" can not be deleted from orphan "
2076 "list, maybe cause OST objects can not be "
2077 "destroyed (err: %d).\n",
2078 PFID(mdd_object_fid(mdd_obj)), rc);
2079 /* If object was not deleted from orphan list, do not
2080 * destroy OSS objects, which will be done when next
2086 rc = mdd_iattr_get(env, mdd_obj, ma);
2087 /* Object maybe not in orphan list originally, it is rare case for
2088 * mdd_finish_unlink() failure. */
2089 if (rc == 0 && ma->ma_attr.la_nlink == 0) {
2090 #ifdef HAVE_QUOTA_SUPPORT
2091 if (mds->mds_quota) {
2092 quota_opc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
2093 mdd_quota_wrapper(&ma->ma_attr, qids);
2096 /* MDS_CLOSE_CLEANUP means destroy OSS objects by MDS. */
2097 if (ma->ma_valid & MA_FLAGS &&
2098 ma->ma_attr_flags & MDS_CLOSE_CLEANUP) {
2099 rc = mdd_lov_destroy(env, mdd, mdd_obj, &ma->ma_attr);
2101 rc = mdd_object_kill(env, mdd_obj, ma);
2107 CERROR("Error when prepare to delete Object "DFID" , "
2108 "which will cause OST objects can not be "
2109 "destroyed.\n", PFID(mdd_object_fid(mdd_obj)));
2115 ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
2117 mdd_write_unlock(env, mdd_obj);
2118 mdd_trans_stop(env, mdo2mdd(obj), rc, handle);
2119 #ifdef HAVE_QUOTA_SUPPORT
2121 /* Trigger dqrel on the owner of child. If failed,
2122 * the next call for lquota_chkquota will process it */
2123 lquota_adjust(mds_quota_interface_ref, obd, qids, 0, rc,
2130 * Permission check is done when open,
2131 * no need check again.
2133 static int mdd_readpage_sanity_check(const struct lu_env *env,
2134 struct mdd_object *obj)
2136 struct dt_object *next = mdd_object_child(obj);
2140 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
2148 static int mdd_dir_page_build(const struct lu_env *env, struct mdd_device *mdd,
2149 int first, void *area, int nob,
2150 const struct dt_it_ops *iops, struct dt_it *it,
2151 __u64 *start, __u64 *end,
2152 struct lu_dirent **last, __u32 attr)
2156 struct lu_dirent *ent;
2159 memset(area, 0, sizeof (struct lu_dirpage));
2160 area += sizeof (struct lu_dirpage);
2161 nob -= sizeof (struct lu_dirpage);
2169 len = iops->key_size(env, it);
2171 /* IAM iterator can return record with zero len. */
2175 hash = iops->store(env, it);
2176 if (unlikely(first)) {
2181 /* calculate max space required for lu_dirent */
2182 recsize = lu_dirent_calc_size(len, attr);
2184 if (nob >= recsize) {
2185 result = iops->rec(env, it, ent, attr);
2186 if (result == -ESTALE)
2191 /* osd might not able to pack all attributes,
2192 * so recheck rec length */
2193 recsize = le16_to_cpu(ent->lde_reclen);
2196 * record doesn't fit into page, enlarge previous one.
2199 (*last)->lde_reclen =
2200 cpu_to_le16(le16_to_cpu((*last)->lde_reclen) +
2209 ent = (void *)ent + recsize;
2213 result = iops->next(env, it);
2214 if (result == -ESTALE)
2216 } while (result == 0);
2223 static int __mdd_readpage(const struct lu_env *env, struct mdd_object *obj,
2224 const struct lu_rdpg *rdpg)
2227 struct dt_object *next = mdd_object_child(obj);
2228 const struct dt_it_ops *iops;
2230 struct lu_dirent *last = NULL;
2231 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
2238 LASSERT(rdpg->rp_pages != NULL);
2239 LASSERT(next->do_index_ops != NULL);
2241 if (rdpg->rp_count <= 0)
2245 * iterate through directory and fill pages from @rdpg
2247 iops = &next->do_index_ops->dio_it;
2248 it = iops->init(env, next, mdd_object_capa(env, obj));
2252 rc = iops->load(env, it, rdpg->rp_hash);
2256 * Iterator didn't find record with exactly the key requested.
2258 * It is currently either
2260 * - positioned above record with key less than
2261 * requested---skip it.
2263 * - or not positioned at all (is in IAM_IT_SKEWED
2264 * state)---position it on the next item.
2266 rc = iops->next(env, it);
2271 * At this point and across for-loop:
2273 * rc == 0 -> ok, proceed.
2274 * rc > 0 -> end of directory.
2277 for (i = 0, nob = rdpg->rp_count; rc == 0 && nob > 0;
2278 i++, nob -= CFS_PAGE_SIZE) {
2279 LASSERT(i < rdpg->rp_npages);
2280 pg = rdpg->rp_pages[i];
2281 rc = mdd_dir_page_build(env, mdd, !i, cfs_kmap(pg),
2282 min_t(int, nob, CFS_PAGE_SIZE), iops,
2283 it, &hash_start, &hash_end, &last,
2285 if (rc != 0 || i == rdpg->rp_npages - 1) {
2287 last->lde_reclen = 0;
2295 hash_end = DIR_END_OFF;
2299 struct lu_dirpage *dp;
2301 dp = cfs_kmap(rdpg->rp_pages[0]);
2302 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
2303 dp->ldp_hash_end = cpu_to_le64(hash_end);
2306 * No pages were processed, mark this.
2308 dp->ldp_flags |= LDF_EMPTY;
2310 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2311 cfs_kunmap(rdpg->rp_pages[0]);
2314 iops->fini(env, it);
2319 int mdd_readpage(const struct lu_env *env, struct md_object *obj,
2320 const struct lu_rdpg *rdpg)
2322 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2326 LASSERT(mdd_object_exists(mdd_obj));
2328 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
2329 rc = mdd_readpage_sanity_check(env, mdd_obj);
2331 GOTO(out_unlock, rc);
2333 if (mdd_is_dead_obj(mdd_obj)) {
2335 struct lu_dirpage *dp;
2338 * According to POSIX, please do not return any entry to client:
2339 * even dot and dotdot should not be returned.
2341 CWARN("readdir from dead object: "DFID"\n",
2342 PFID(mdd_object_fid(mdd_obj)));
2344 if (rdpg->rp_count <= 0)
2345 GOTO(out_unlock, rc = -EFAULT);
2346 LASSERT(rdpg->rp_pages != NULL);
2348 pg = rdpg->rp_pages[0];
2349 dp = (struct lu_dirpage*)cfs_kmap(pg);
2350 memset(dp, 0 , sizeof(struct lu_dirpage));
2351 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
2352 dp->ldp_hash_end = cpu_to_le64(DIR_END_OFF);
2353 dp->ldp_flags |= LDF_EMPTY;
2354 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2356 GOTO(out_unlock, rc = 0);
2359 rc = __mdd_readpage(env, mdd_obj, rdpg);
2363 mdd_read_unlock(env, mdd_obj);
2367 static int mdd_object_sync(const struct lu_env *env, struct md_object *obj)
2369 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2370 struct dt_object *next;
2372 LASSERT(mdd_object_exists(mdd_obj));
2373 next = mdd_object_child(mdd_obj);
2374 return next->do_ops->do_object_sync(env, next);
2377 static dt_obj_version_t mdd_version_get(const struct lu_env *env,
2378 struct md_object *obj)
2380 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2382 LASSERT(mdd_object_exists(mdd_obj));
2383 return do_version_get(env, mdd_object_child(mdd_obj));
2386 static void mdd_version_set(const struct lu_env *env, struct md_object *obj,
2387 dt_obj_version_t version)
2389 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2391 LASSERT(mdd_object_exists(mdd_obj));
2392 do_version_set(env, mdd_object_child(mdd_obj), version);
2395 const struct md_object_operations mdd_obj_ops = {
2396 .moo_permission = mdd_permission,
2397 .moo_attr_get = mdd_attr_get,
2398 .moo_attr_set = mdd_attr_set,
2399 .moo_xattr_get = mdd_xattr_get,
2400 .moo_xattr_set = mdd_xattr_set,
2401 .moo_xattr_list = mdd_xattr_list,
2402 .moo_xattr_del = mdd_xattr_del,
2403 .moo_object_create = mdd_object_create,
2404 .moo_ref_add = mdd_ref_add,
2405 .moo_ref_del = mdd_ref_del,
2406 .moo_open = mdd_open,
2407 .moo_close = mdd_close,
2408 .moo_readpage = mdd_readpage,
2409 .moo_readlink = mdd_readlink,
2410 .moo_capa_get = mdd_capa_get,
2411 .moo_object_sync = mdd_object_sync,
2412 .moo_version_get = mdd_version_get,
2413 .moo_version_set = mdd_version_set,
2414 .moo_path = mdd_path,