4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdd/mdd_object.c
38 * Lustre Metadata Server (mdd) routines
40 * Author: Wang Di <wangdi@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_MDS
45 #include <linux/module.h>
47 #include <obd_class.h>
48 #include <obd_support.h>
49 #include <lprocfs_status.h>
50 /* fid_be_cpu(), fid_cpu_to_be(). */
51 #include <lustre_fid.h>
54 #include <lustre_param.h>
55 #include <lustre_mds.h>
56 #include <lustre/lustre_idl.h>
58 #include "mdd_internal.h"
60 static const struct lu_object_operations mdd_lu_obj_ops;
61 extern cfs_mem_cache_t *mdd_object_kmem;
63 static int mdd_xattr_get(const struct lu_env *env,
64 struct md_object *obj, struct lu_buf *buf,
67 int mdd_data_get(const struct lu_env *env, struct mdd_object *obj,
70 if (mdd_object_exists(obj) == 0) {
71 CERROR("%s: object "DFID" not found: rc = -2\n",
72 mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj)));
75 mdo_data_get(env, obj, data);
79 int mdd_la_get(const struct lu_env *env, struct mdd_object *obj,
80 struct lu_attr *la, struct lustre_capa *capa)
82 if (mdd_object_exists(obj) == 0) {
83 CERROR("%s: object "DFID" not found: rc = -2\n",
84 mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj)));
87 return mdo_attr_get(env, obj, la, capa);
90 static void mdd_flags_xlate(struct mdd_object *obj, __u32 flags)
92 obj->mod_flags &= ~(APPEND_OBJ|IMMUTE_OBJ);
94 if (flags & LUSTRE_APPEND_FL)
95 obj->mod_flags |= APPEND_OBJ;
97 if (flags & LUSTRE_IMMUTABLE_FL)
98 obj->mod_flags |= IMMUTE_OBJ;
101 struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
103 struct mdd_thread_info *info;
105 info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
106 LASSERT(info != NULL);
110 struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
114 buf = &mdd_env_info(env)->mti_buf;
120 void mdd_buf_put(struct lu_buf *buf)
122 if (buf == NULL || buf->lb_buf == NULL)
124 OBD_FREE_LARGE(buf->lb_buf, buf->lb_len);
128 const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
129 const void *area, ssize_t len)
133 buf = &mdd_env_info(env)->mti_buf;
134 buf->lb_buf = (void *)area;
139 struct lu_buf *mdd_buf_alloc(const struct lu_env *env, ssize_t len)
141 struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf;
143 if ((len > buf->lb_len) && (buf->lb_buf != NULL)) {
144 OBD_FREE_LARGE(buf->lb_buf, buf->lb_len);
147 if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) {
149 OBD_ALLOC_LARGE(buf->lb_buf, buf->lb_len);
150 if (buf->lb_buf == NULL)
156 /** Increase the size of the \a mti_big_buf.
157 * preserves old data in buffer
158 * old buffer remains unchanged on error
159 * \retval 0 or -ENOMEM
161 int mdd_buf_grow(const struct lu_env *env, ssize_t len)
163 struct lu_buf *oldbuf = &mdd_env_info(env)->mti_big_buf;
166 LASSERT(len >= oldbuf->lb_len);
167 OBD_ALLOC_LARGE(buf.lb_buf, len);
169 if (buf.lb_buf == NULL)
173 memcpy(buf.lb_buf, oldbuf->lb_buf, oldbuf->lb_len);
175 OBD_FREE_LARGE(oldbuf->lb_buf, oldbuf->lb_len);
177 memcpy(oldbuf, &buf, sizeof(buf));
182 struct lu_object *mdd_object_alloc(const struct lu_env *env,
183 const struct lu_object_header *hdr,
186 struct mdd_object *mdd_obj;
188 OBD_SLAB_ALLOC_PTR_GFP(mdd_obj, mdd_object_kmem, CFS_ALLOC_IO);
189 if (mdd_obj != NULL) {
192 o = mdd2lu_obj(mdd_obj);
193 lu_object_init(o, NULL, d);
194 mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
195 mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
196 mdd_obj->mod_count = 0;
197 o->lo_ops = &mdd_lu_obj_ops;
204 static int mdd_object_init(const struct lu_env *env, struct lu_object *o,
205 const struct lu_object_conf *unused)
207 struct mdd_device *d = lu2mdd_dev(o->lo_dev);
208 struct mdd_object *mdd_obj = lu2mdd_obj(o);
209 struct lu_object *below;
210 struct lu_device *under;
213 mdd_obj->mod_cltime = 0;
214 under = &d->mdd_child->dd_lu_dev;
215 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
216 mdd_pdlock_init(mdd_obj);
220 lu_object_add(o, below);
225 static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
227 if (lu_object_exists(o))
228 return mdd_get_flags(env, lu2mdd_obj(o));
233 static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
235 struct mdd_object *mdd = lu2mdd_obj(o);
238 OBD_SLAB_FREE_PTR(mdd, mdd_object_kmem);
241 static int mdd_object_print(const struct lu_env *env, void *cookie,
242 lu_printer_t p, const struct lu_object *o)
244 struct mdd_object *mdd = lu2mdd_obj((struct lu_object *)o);
245 return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p(open_count=%d, "
246 "valid=%x, cltime="LPU64", flags=%lx)",
247 mdd, mdd->mod_count, mdd->mod_valid,
248 mdd->mod_cltime, mdd->mod_flags);
251 static const struct lu_object_operations mdd_lu_obj_ops = {
252 .loo_object_init = mdd_object_init,
253 .loo_object_start = mdd_object_start,
254 .loo_object_free = mdd_object_free,
255 .loo_object_print = mdd_object_print,
258 struct mdd_object *mdd_object_find(const struct lu_env *env,
259 struct mdd_device *d,
260 const struct lu_fid *f)
262 return md2mdd_obj(md_object_find_slice(env, &d->mdd_md_dev, f));
265 static int mdd_path2fid(const struct lu_env *env, struct mdd_device *mdd,
266 const char *path, struct lu_fid *fid)
269 struct lu_fid *f = &mdd_env_info(env)->mti_fid;
270 struct mdd_object *obj;
271 struct lu_name *lname = &mdd_env_info(env)->mti_name;
276 /* temp buffer for path element */
277 buf = mdd_buf_alloc(env, PATH_MAX);
278 if (buf->lb_buf == NULL)
281 lname->ln_name = name = buf->lb_buf;
282 lname->ln_namelen = 0;
283 *f = mdd->mdd_root_fid;
290 while (*path != '/' && *path != '\0') {
298 /* find obj corresponding to fid */
299 obj = mdd_object_find(env, mdd, f);
301 GOTO(out, rc = -EREMOTE);
303 GOTO(out, rc = PTR_ERR(obj));
304 /* get child fid from parent and name */
305 rc = mdd_lookup(env, &obj->mod_obj, lname, f, NULL);
306 mdd_object_put(env, obj);
311 lname->ln_namelen = 0;
320 /** The maximum depth that fid2path() will search.
321 * This is limited only because we want to store the fids for
322 * historical path lookup purposes.
324 #define MAX_PATH_DEPTH 100
326 /** mdd_path() lookup structure. */
327 struct path_lookup_info {
328 __u64 pli_recno; /**< history point */
329 __u64 pli_currec; /**< current record */
330 struct lu_fid pli_fid;
331 struct lu_fid pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
332 struct mdd_object *pli_mdd_obj;
333 char *pli_path; /**< full path */
335 int pli_linkno; /**< which hardlink to follow */
336 int pli_fidcount; /**< number of \a pli_fids */
339 static int mdd_path_current(const struct lu_env *env,
340 struct path_lookup_info *pli)
342 struct mdd_device *mdd = mdo2mdd(&pli->pli_mdd_obj->mod_obj);
343 struct mdd_object *mdd_obj;
344 struct lu_buf *buf = NULL;
345 struct link_ea_header *leh;
346 struct link_ea_entry *lee;
347 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
348 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
354 ptr = pli->pli_path + pli->pli_pathlen - 1;
357 pli->pli_fidcount = 0;
358 pli->pli_fids[0] = *(struct lu_fid *)mdd_object_fid(pli->pli_mdd_obj);
360 while (!mdd_is_root(mdd, &pli->pli_fids[pli->pli_fidcount])) {
361 mdd_obj = mdd_object_find(env, mdd,
362 &pli->pli_fids[pli->pli_fidcount]);
364 GOTO(out, rc = -EREMOTE);
366 GOTO(out, rc = PTR_ERR(mdd_obj));
367 rc = lu_object_exists(&mdd_obj->mod_obj.mo_lu);
369 mdd_object_put(env, mdd_obj);
373 /* Do I need to error out here? */
378 /* Get parent fid and object name */
379 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
380 buf = mdd_links_get(env, mdd_obj);
381 mdd_read_unlock(env, mdd_obj);
382 mdd_object_put(env, mdd_obj);
384 GOTO(out, rc = PTR_ERR(buf));
387 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
388 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
390 /* If set, use link #linkno for path lookup, otherwise use
391 link #0. Only do this for the final path element. */
392 if ((pli->pli_fidcount == 0) &&
393 (pli->pli_linkno < leh->leh_reccount)) {
395 for (count = 0; count < pli->pli_linkno; count++) {
396 lee = (struct link_ea_entry *)
397 ((char *)lee + reclen);
398 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
400 if (pli->pli_linkno < leh->leh_reccount - 1)
401 /* indicate to user there are more links */
405 /* Pack the name in the end of the buffer */
406 ptr -= tmpname->ln_namelen;
407 if (ptr - 1 <= pli->pli_path)
408 GOTO(out, rc = -EOVERFLOW);
409 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
412 /* Store the parent fid for historic lookup */
413 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
414 GOTO(out, rc = -EOVERFLOW);
415 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
418 /* Verify that our path hasn't changed since we started the lookup.
419 Record the current index, and verify the path resolves to the
420 same fid. If it does, then the path is correct as of this index. */
421 spin_lock(&mdd->mdd_cl.mc_lock);
422 pli->pli_currec = mdd->mdd_cl.mc_index;
423 spin_unlock(&mdd->mdd_cl.mc_lock);
424 rc = mdd_path2fid(env, mdd, ptr, &pli->pli_fid);
426 CDEBUG(D_INFO, "mdd_path2fid(%s) failed %d\n", ptr, rc);
427 GOTO (out, rc = -EAGAIN);
429 if (!lu_fid_eq(&pli->pli_fids[0], &pli->pli_fid)) {
430 CDEBUG(D_INFO, "mdd_path2fid(%s) found another FID o="DFID
431 " n="DFID"\n", ptr, PFID(&pli->pli_fids[0]),
432 PFID(&pli->pli_fid));
433 GOTO(out, rc = -EAGAIN);
435 ptr++; /* skip leading / */
436 memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
440 if (buf && !IS_ERR(buf) && buf->lb_len > OBD_ALLOC_BIG)
441 /* if we vmalloced a large buffer drop it */
447 static int mdd_path_historic(const struct lu_env *env,
448 struct path_lookup_info *pli)
453 /* Returns the full path to this fid, as of changelog record recno. */
454 static int mdd_path(const struct lu_env *env, struct md_object *obj,
455 char *path, int pathlen, __u64 *recno, int *linkno)
457 struct path_lookup_info *pli;
465 if (mdd_is_root(mdo2mdd(obj), mdd_object_fid(md2mdd_obj(obj)))) {
474 pli->pli_mdd_obj = md2mdd_obj(obj);
475 pli->pli_recno = *recno;
476 pli->pli_path = path;
477 pli->pli_pathlen = pathlen;
478 pli->pli_linkno = *linkno;
480 /* Retry multiple times in case file is being moved */
481 while (tries-- && rc == -EAGAIN)
482 rc = mdd_path_current(env, pli);
484 /* For historical path lookup, the current links may not have existed
485 * at "recno" time. We must switch over to earlier links/parents
486 * by using the changelog records. If the earlier parent doesn't
487 * exist, we must search back through the changelog to reconstruct
488 * its parents, then check if it exists, etc.
489 * We may ignore this problem for the initial implementation and
490 * state that an "original" hardlink must still exist for us to find
491 * historic path name. */
492 if (pli->pli_recno != -1) {
493 rc = mdd_path_historic(env, pli);
495 *recno = pli->pli_currec;
496 /* Return next link index to caller */
497 *linkno = pli->pli_linkno;
505 int mdd_get_flags(const struct lu_env *env, struct mdd_object *obj)
507 struct lu_attr *la = &mdd_env_info(env)->mti_la;
511 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
513 mdd_flags_xlate(obj, la->la_flags);
519 * No permission check is needed.
521 int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
527 return mdd_la_get(env, md2mdd_obj(obj), &ma->ma_attr,
528 mdd_object_capa(env, md2mdd_obj(obj)));
533 * No permission check is needed.
535 static int mdd_xattr_get(const struct lu_env *env,
536 struct md_object *obj, struct lu_buf *buf,
539 struct mdd_object *mdd_obj = md2mdd_obj(obj);
544 if (mdd_object_exists(mdd_obj) == 0) {
545 CERROR("%s: object "DFID" not found: rc = -2\n",
546 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
550 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
551 rc = mdo_xattr_get(env, mdd_obj, buf, name,
552 mdd_object_capa(env, mdd_obj));
553 mdd_read_unlock(env, mdd_obj);
559 * Permission check is done when open,
560 * no need check again.
562 static int mdd_readlink(const struct lu_env *env, struct md_object *obj,
565 struct mdd_object *mdd_obj = md2mdd_obj(obj);
566 struct dt_object *next;
571 if (mdd_object_exists(mdd_obj) == 0) {
572 CERROR("%s: object "DFID" not found: rc = -2\n",
573 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
577 next = mdd_object_child(mdd_obj);
578 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
579 rc = next->do_body_ops->dbo_read(env, next, buf, &pos,
580 mdd_object_capa(env, mdd_obj));
581 mdd_read_unlock(env, mdd_obj);
586 * No permission check is needed.
588 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
591 struct mdd_object *mdd_obj = md2mdd_obj(obj);
596 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
597 rc = mdo_xattr_list(env, mdd_obj, buf, mdd_object_capa(env, mdd_obj));
598 mdd_read_unlock(env, mdd_obj);
603 int mdd_declare_object_create_internal(const struct lu_env *env,
604 struct mdd_object *p,
605 struct mdd_object *c,
606 struct lu_attr *attr,
607 struct thandle *handle,
608 const struct md_op_spec *spec)
610 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
611 const struct dt_index_features *feat = spec->sp_feat;
615 if (feat != &dt_directory_features && feat != NULL) {
616 dof->dof_type = DFT_INDEX;
617 dof->u.dof_idx.di_feat = feat;
620 dof->dof_type = dt_mode_to_dft(attr->la_mode);
621 if (dof->dof_type == DFT_REGULAR) {
622 dof->u.dof_reg.striped =
623 md_should_create(spec->sp_cr_flags);
624 if (spec->sp_cr_flags & MDS_OPEN_HAS_EA)
625 dof->u.dof_reg.striped = 0;
626 /* is this replay? */
628 dof->u.dof_reg.striped = 0;
632 rc = mdo_declare_create_obj(env, c, attr, NULL, dof, handle);
637 int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p,
638 struct mdd_object *c, struct lu_attr *attr,
639 struct thandle *handle,
640 const struct md_op_spec *spec)
642 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
643 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
647 LASSERT(!mdd_object_exists(c));
649 rc = mdo_create_obj(env, c, attr, hint, dof, handle);
651 LASSERT(ergo(rc == 0, mdd_object_exists(c)));
657 * Make sure the ctime is increased only.
659 static inline int mdd_attr_check(const struct lu_env *env,
660 struct mdd_object *obj,
661 struct lu_attr *attr)
663 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
667 if (attr->la_valid & LA_CTIME) {
668 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
672 if (attr->la_ctime < tmp_la->la_ctime)
673 attr->la_valid &= ~(LA_MTIME | LA_CTIME);
674 else if (attr->la_valid == LA_CTIME &&
675 attr->la_ctime == tmp_la->la_ctime)
676 attr->la_valid &= ~LA_CTIME;
681 int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj,
682 struct lu_attr *attr, struct thandle *handle,
688 rc = mdo_attr_set(env, obj, attr, handle, mdd_object_capa(env, obj));
689 #ifdef CONFIG_FS_POSIX_ACL
690 if (!rc && (attr->la_valid & LA_MODE) && needacl)
691 rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
696 int mdd_attr_check_set_internal(const struct lu_env *env,
697 struct mdd_object *obj, struct lu_attr *attr,
698 struct thandle *handle, int needacl)
703 rc = mdd_attr_check(env, obj, attr);
708 rc = mdd_attr_set_internal(env, obj, attr, handle, needacl);
713 * This gives the same functionality as the code between
714 * sys_chmod and inode_setattr
715 * chown_common and inode_setattr
716 * utimes and inode_setattr
717 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
719 static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
720 struct lu_attr *la, const unsigned long flags)
722 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
730 /* Do not permit change file type */
731 if (la->la_valid & LA_TYPE)
734 /* They should not be processed by setattr */
735 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
738 /* export destroy does not have ->le_ses, but we may want
739 * to drop LUSTRE_SOM_FL. */
740 uc = lu_ucred_check(env);
744 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
748 if (la->la_valid == LA_CTIME) {
749 if (!(flags & MDS_PERM_BYPASS))
750 /* This is only for set ctime when rename's source is
752 rc = mdd_may_delete(env, NULL, obj, tmp_la, NULL, 1, 0);
753 if (rc == 0 && la->la_ctime <= tmp_la->la_ctime)
754 la->la_valid &= ~LA_CTIME;
758 if (la->la_valid == LA_ATIME) {
759 /* This is atime only set for read atime update on close. */
760 if (la->la_atime >= tmp_la->la_atime &&
761 la->la_atime < (tmp_la->la_atime +
762 mdd_obj2mdd_dev(obj)->mdd_atime_diff))
763 la->la_valid &= ~LA_ATIME;
767 /* Check if flags change. */
768 if (la->la_valid & LA_FLAGS) {
769 unsigned int oldflags = 0;
770 unsigned int newflags = la->la_flags &
771 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
773 if ((uc->uc_fsuid != tmp_la->la_uid) &&
774 !mdd_capable(uc, CFS_CAP_FOWNER))
777 /* XXX: the IMMUTABLE and APPEND_ONLY flags can
778 * only be changed by the relevant capability. */
779 if (mdd_is_immutable(obj))
780 oldflags |= LUSTRE_IMMUTABLE_FL;
781 if (mdd_is_append(obj))
782 oldflags |= LUSTRE_APPEND_FL;
783 if ((oldflags ^ newflags) &&
784 !mdd_capable(uc, CFS_CAP_LINUX_IMMUTABLE))
787 if (!S_ISDIR(tmp_la->la_mode))
788 la->la_flags &= ~LUSTRE_DIRSYNC_FL;
791 if ((mdd_is_immutable(obj) || mdd_is_append(obj)) &&
792 (la->la_valid & ~LA_FLAGS) &&
793 !(flags & MDS_PERM_BYPASS))
796 /* Check for setting the obj time. */
797 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
798 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
799 if ((uc->uc_fsuid != tmp_la->la_uid) &&
800 !mdd_capable(uc, CFS_CAP_FOWNER)) {
801 rc = mdd_permission_internal(env, obj, tmp_la,
808 if (la->la_valid & LA_KILL_SUID) {
809 la->la_valid &= ~LA_KILL_SUID;
810 if ((tmp_la->la_mode & S_ISUID) &&
811 !(la->la_valid & LA_MODE)) {
812 la->la_mode = tmp_la->la_mode;
813 la->la_valid |= LA_MODE;
815 la->la_mode &= ~S_ISUID;
818 if (la->la_valid & LA_KILL_SGID) {
819 la->la_valid &= ~LA_KILL_SGID;
820 if (((tmp_la->la_mode & (S_ISGID | S_IXGRP)) ==
821 (S_ISGID | S_IXGRP)) &&
822 !(la->la_valid & LA_MODE)) {
823 la->la_mode = tmp_la->la_mode;
824 la->la_valid |= LA_MODE;
826 la->la_mode &= ~S_ISGID;
829 /* Make sure a caller can chmod. */
830 if (la->la_valid & LA_MODE) {
831 if (!(flags & MDS_PERM_BYPASS) &&
832 (uc->uc_fsuid != tmp_la->la_uid) &&
833 !mdd_capable(uc, CFS_CAP_FOWNER))
836 if (la->la_mode == (cfs_umode_t) -1)
837 la->la_mode = tmp_la->la_mode;
839 la->la_mode = (la->la_mode & S_IALLUGO) |
840 (tmp_la->la_mode & ~S_IALLUGO);
842 /* Also check the setgid bit! */
843 if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ?
844 la->la_gid : tmp_la->la_gid) &&
845 !mdd_capable(uc, CFS_CAP_FSETID))
846 la->la_mode &= ~S_ISGID;
848 la->la_mode = tmp_la->la_mode;
851 /* Make sure a caller can chown. */
852 if (la->la_valid & LA_UID) {
853 if (la->la_uid == (uid_t) -1)
854 la->la_uid = tmp_la->la_uid;
855 if (((uc->uc_fsuid != tmp_la->la_uid) ||
856 (la->la_uid != tmp_la->la_uid)) &&
857 !mdd_capable(uc, CFS_CAP_CHOWN))
860 /* If the user or group of a non-directory has been
861 * changed by a non-root user, remove the setuid bit.
862 * 19981026 David C Niemi <niemi@tux.org>
864 * Changed this to apply to all users, including root,
865 * to avoid some races. This is the behavior we had in
866 * 2.0. The check for non-root was definitely wrong
867 * for 2.2 anyway, as it should have been using
868 * CAP_FSETID rather than fsuid -- 19990830 SD. */
869 if (((tmp_la->la_mode & S_ISUID) == S_ISUID) &&
870 !S_ISDIR(tmp_la->la_mode)) {
871 la->la_mode &= ~S_ISUID;
872 la->la_valid |= LA_MODE;
876 /* Make sure caller can chgrp. */
877 if (la->la_valid & LA_GID) {
878 if (la->la_gid == (gid_t) -1)
879 la->la_gid = tmp_la->la_gid;
880 if (((uc->uc_fsuid != tmp_la->la_uid) ||
881 ((la->la_gid != tmp_la->la_gid) &&
882 !lustre_in_group_p(uc, la->la_gid))) &&
883 !mdd_capable(uc, CFS_CAP_CHOWN))
886 /* Likewise, if the user or group of a non-directory
887 * has been changed by a non-root user, remove the
888 * setgid bit UNLESS there is no group execute bit
889 * (this would be a file marked for mandatory
890 * locking). 19981026 David C Niemi <niemi@tux.org>
892 * Removed the fsuid check (see the comment above) --
894 if (((tmp_la->la_mode & (S_ISGID | S_IXGRP)) ==
895 (S_ISGID | S_IXGRP)) && !S_ISDIR(tmp_la->la_mode)) {
896 la->la_mode &= ~S_ISGID;
897 la->la_valid |= LA_MODE;
901 /* For both Size-on-MDS case and truncate case,
902 * "la->la_valid & (LA_SIZE | LA_BLOCKS)" are ture.
903 * We distinguish them by "flags & MDS_SOM".
904 * For SOM case, it is true, the MAY_WRITE perm has been checked
905 * when open, no need check again. For truncate case, it is false,
906 * the MAY_WRITE perm should be checked here. */
907 if (flags & MDS_SOM) {
908 /* For the "Size-on-MDS" setattr update, merge coming
909 * attributes with the set in the inode. BUG 10641 */
910 if ((la->la_valid & LA_ATIME) &&
911 (la->la_atime <= tmp_la->la_atime))
912 la->la_valid &= ~LA_ATIME;
914 /* OST attributes do not have a priority over MDS attributes,
915 * so drop times if ctime is equal. */
916 if ((la->la_valid & LA_CTIME) &&
917 (la->la_ctime <= tmp_la->la_ctime))
918 la->la_valid &= ~(LA_MTIME | LA_CTIME);
920 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
921 if (!((flags & MDS_OPEN_OWNEROVERRIDE) &&
922 (uc->uc_fsuid == tmp_la->la_uid)) &&
923 !(flags & MDS_PERM_BYPASS)) {
924 rc = mdd_permission_internal(env, obj,
930 if (la->la_valid & LA_CTIME) {
931 /* The pure setattr, it has the priority over what is
932 * already set, do not drop it if ctime is equal. */
933 if (la->la_ctime < tmp_la->la_ctime)
934 la->la_valid &= ~(LA_ATIME | LA_MTIME |
942 /** Store a data change changelog record
943 * If this fails, we must fail the whole transaction; we don't
944 * want the change to commit without the log entry.
945 * \param mdd_obj - mdd_object of change
946 * \param handle - transacion handle
948 static int mdd_changelog_data_store(const struct lu_env *env,
949 struct mdd_device *mdd,
950 enum changelog_rec_type type,
951 int flags, struct mdd_object *mdd_obj,
952 struct thandle *handle)
954 const struct lu_fid *tfid = mdo2fid(mdd_obj);
955 struct llog_changelog_rec *rec;
961 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
963 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
966 LASSERT(mdd_obj != NULL);
967 LASSERT(handle != NULL);
969 if ((type >= CL_MTIME) && (type <= CL_ATIME) &&
970 cfs_time_before_64(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) {
971 /* Don't need multiple updates in this log */
972 /* Don't check under lock - no big deal if we get an extra
977 reclen = llog_data_len(sizeof(*rec));
978 buf = mdd_buf_alloc(env, reclen);
979 if (buf->lb_buf == NULL)
983 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
984 rec->cr.cr_type = (__u32)type;
985 rec->cr.cr_tfid = *tfid;
986 rec->cr.cr_namelen = 0;
987 mdd_obj->mod_cltime = cfs_time_current_64();
989 rc = mdd_changelog_store(env, mdd, rec, handle);
994 int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type,
995 int flags, struct md_object *obj)
997 struct thandle *handle;
998 struct mdd_object *mdd_obj = md2mdd_obj(obj);
999 struct mdd_device *mdd = mdo2mdd(obj);
1003 handle = mdd_trans_create(env, mdd);
1005 RETURN(PTR_ERR(handle));
1007 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1011 rc = mdd_trans_start(env, mdd, handle);
1015 rc = mdd_changelog_data_store(env, mdd, type, flags, mdd_obj,
1019 mdd_trans_stop(env, mdd, rc, handle);
1025 * Save LMA extended attributes with data from \a ma.
1027 * HSM and Size-On-MDS data will be extracted from \ma if they are valid, if
1028 * not, LMA EA will be first read from disk, modified and write back.
1031 /* Precedence for choosing record type when multiple
1032 * attributes change: setattr > mtime > ctime > atime
1033 * (ctime changes when mtime does, plus chmod/chown.
1034 * atime and ctime are independent.) */
1035 static int mdd_attr_set_changelog(const struct lu_env *env,
1036 struct md_object *obj, struct thandle *handle,
1039 struct mdd_device *mdd = mdo2mdd(obj);
1042 bits = (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0;
1043 bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0;
1044 bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0;
1045 bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0;
1046 bits = bits & mdd->mdd_cl.mc_mask;
1050 /* The record type is the lowest non-masked set bit */
1051 while (bits && ((bits & 1) == 0)) {
1056 /* FYI we only store the first CLF_FLAGMASK bits of la_valid */
1057 return mdd_changelog_data_store(env, mdd, type, (int)valid,
1058 md2mdd_obj(obj), handle);
1061 static int mdd_declare_attr_set(const struct lu_env *env,
1062 struct mdd_device *mdd,
1063 struct mdd_object *obj,
1064 const struct lu_attr *attr,
1065 struct thandle *handle)
1069 rc = mdo_declare_attr_set(env, obj, attr, handle);
1073 #ifdef CONFIG_FS_POSIX_ACL
1074 if (attr->la_valid & LA_MODE) {
1075 mdd_read_lock(env, obj, MOR_TGT_CHILD);
1076 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL,
1077 XATTR_NAME_ACL_ACCESS, BYPASS_CAPA);
1078 mdd_read_unlock(env, obj);
1079 if (rc == -EOPNOTSUPP || rc == -ENODATA)
1085 struct lu_buf *buf = mdd_buf_get(env, NULL, rc);
1086 rc = mdo_declare_xattr_set(env, obj, buf,
1087 XATTR_NAME_ACL_ACCESS, 0,
1095 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1099 /* set attr and LOV EA at once, return updated attr */
1100 int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
1101 const struct md_attr *ma)
1103 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1104 struct mdd_device *mdd = mdo2mdd(obj);
1105 struct thandle *handle;
1106 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
1107 const struct lu_attr *la = &ma->ma_attr;
1111 /* we do not use ->attr_set() for LOV/SOM/HSM EA any more */
1112 LASSERT((ma->ma_valid & MA_LOV) == 0);
1113 LASSERT((ma->ma_valid & MA_HSM) == 0);
1114 LASSERT((ma->ma_valid & MA_SOM) == 0);
1116 *la_copy = ma->ma_attr;
1117 rc = mdd_fix_attr(env, mdd_obj, la_copy, ma->ma_attr_flags);
1121 /* setattr on "close" only change atime, or do nothing */
1122 if (la->la_valid == LA_ATIME && la_copy->la_valid == 0)
1125 handle = mdd_trans_create(env, mdd);
1127 RETURN(PTR_ERR(handle));
1129 rc = mdd_declare_attr_set(env, mdd, mdd_obj, la, handle);
1133 rc = mdd_trans_start(env, mdd, handle);
1137 /* permission changes may require sync operation */
1138 if (ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID))
1139 handle->th_sync |= !!mdd->mdd_sync_permission;
1141 if (la->la_valid & (LA_MTIME | LA_CTIME))
1142 CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n",
1143 la->la_mtime, la->la_ctime);
1145 if (la_copy->la_valid & LA_FLAGS) {
1146 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
1148 mdd_flags_xlate(mdd_obj, la_copy->la_flags);
1149 } else if (la_copy->la_valid) { /* setattr */
1150 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
1154 rc = mdd_attr_set_changelog(env, obj, handle,
1157 mdd_trans_stop(env, mdd, rc, handle);
1161 static int mdd_xattr_sanity_check(const struct lu_env *env,
1162 struct mdd_object *obj)
1164 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
1165 struct lu_ucred *uc = lu_ucred_assert(env);
1169 if (mdd_is_immutable(obj) || mdd_is_append(obj))
1172 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
1176 if ((uc->uc_fsuid != tmp_la->la_uid) &&
1177 !mdd_capable(uc, CFS_CAP_FOWNER))
1183 static int mdd_declare_xattr_set(const struct lu_env *env,
1184 struct mdd_device *mdd,
1185 struct mdd_object *obj,
1186 const struct lu_buf *buf,
1188 int fl, struct thandle *handle)
1192 rc = mdo_declare_xattr_set(env, obj, buf, name, fl, handle);
1196 /* Only record user xattr changes */
1197 if ((strncmp("user.", name, 5) == 0)) {
1198 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1203 /* If HSM data is modified, this could add a changelog */
1204 if (strncmp(XATTR_NAME_HSM, name, sizeof(XATTR_NAME_HSM) - 1) == 0)
1205 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1207 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1212 * Compare current and future data of HSM EA and add a changelog if needed.
1214 * Caller should have write-locked \param obj.
1216 * \param buf - Future HSM EA content.
1217 * \retval 0 if no changelog is needed or changelog was added properly.
1218 * \retval -ve errno if there was a problem
1220 static int mdd_hsm_update_locked(const struct lu_env *env,
1221 struct md_object *obj,
1222 const struct lu_buf *buf,
1223 struct thandle *handle)
1225 struct mdd_thread_info *info = mdd_env_info(env);
1226 struct mdd_device *mdd = mdo2mdd(obj);
1227 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1228 struct lu_buf *current_buf = &info->mti_buf;
1229 struct md_hsm *current_mh;
1230 struct md_hsm *new_mh;
1234 OBD_ALLOC_PTR(current_mh);
1235 if (current_mh == NULL)
1238 /* Read HSM attrs from disk */
1239 current_buf->lb_buf = info->mti_xattr_buf;
1240 current_buf->lb_len = sizeof(info->mti_xattr_buf);
1241 CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
1242 rc = mdo_xattr_get(env, mdd_obj, current_buf, XATTR_NAME_HSM,
1243 mdd_object_capa(env, mdd_obj));
1244 rc = lustre_buf2hsm(info->mti_xattr_buf, rc, current_mh);
1245 if (rc < 0 && rc != -ENODATA)
1247 else if (rc == -ENODATA)
1248 current_mh->mh_flags = 0;
1250 /* Map future HSM xattr */
1251 OBD_ALLOC_PTR(new_mh);
1253 GOTO(free, rc = -ENOMEM);
1254 lustre_buf2hsm(buf->lb_buf, buf->lb_len, new_mh);
1256 /* If HSM flags are different, add a changelog */
1258 if (current_mh->mh_flags != new_mh->mh_flags) {
1260 hsm_set_cl_event(&flags, HE_STATE);
1261 if (new_mh->mh_flags & HS_DIRTY)
1262 hsm_set_cl_flags(&flags, CLF_HSM_DIRTY);
1264 rc = mdd_changelog_data_store(env, mdd, CL_HSM, flags, mdd_obj,
1268 OBD_FREE_PTR(new_mh);
1270 OBD_FREE_PTR(current_mh);
1276 * The caller should guarantee to update the object ctime
1277 * after xattr_set if needed.
1279 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
1280 const struct lu_buf *buf, const char *name,
1283 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1284 struct mdd_device *mdd = mdo2mdd(obj);
1285 struct thandle *handle;
1289 if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) {
1290 rc = mdd_acl_set(env, mdd_obj, buf, fl);
1294 rc = mdd_xattr_sanity_check(env, mdd_obj);
1298 handle = mdd_trans_create(env, mdd);
1300 RETURN(PTR_ERR(handle));
1302 rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, 0, handle);
1306 rc = mdd_trans_start(env, mdd, handle);
1310 /* security-replated changes may require sync */
1311 if (!strcmp(name, XATTR_NAME_ACL_ACCESS))
1312 handle->th_sync |= !!mdd->mdd_sync_permission;
1314 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1316 if (strncmp(XATTR_NAME_HSM, name, sizeof(XATTR_NAME_HSM) - 1) == 0) {
1317 rc = mdd_hsm_update_locked(env, obj, buf, handle);
1319 mdd_write_unlock(env, mdd_obj);
1324 rc = mdo_xattr_set(env, mdd_obj, buf, name, fl, handle,
1325 mdd_object_capa(env, mdd_obj));
1326 mdd_write_unlock(env, mdd_obj);
1330 /* Only record system & user xattr changes */
1331 if (strncmp(XATTR_USER_PREFIX, name,
1332 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1333 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1334 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1335 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1336 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1337 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1341 mdd_trans_stop(env, mdd, rc, handle);
1346 static int mdd_declare_xattr_del(const struct lu_env *env,
1347 struct mdd_device *mdd,
1348 struct mdd_object *obj,
1350 struct thandle *handle)
1354 rc = mdo_declare_xattr_del(env, obj, name, handle);
1358 /* Only record user xattr changes */
1359 if ((strncmp("user.", name, 5) == 0))
1360 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1366 * The caller should guarantee to update the object ctime
1367 * after xattr_set if needed.
1369 int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1372 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1373 struct mdd_device *mdd = mdo2mdd(obj);
1374 struct thandle *handle;
1378 rc = mdd_xattr_sanity_check(env, mdd_obj);
1382 handle = mdd_trans_create(env, mdd);
1384 RETURN(PTR_ERR(handle));
1386 rc = mdd_declare_xattr_del(env, mdd, mdd_obj, name, handle);
1390 rc = mdd_trans_start(env, mdd, handle);
1394 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1395 rc = mdo_xattr_del(env, mdd_obj, name, handle,
1396 mdd_object_capa(env, mdd_obj));
1397 mdd_write_unlock(env, mdd_obj);
1401 /* Only record system & user xattr changes */
1402 if (strncmp(XATTR_USER_PREFIX, name,
1403 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1404 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1405 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1406 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1407 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1408 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1412 mdd_trans_stop(env, mdd, rc, handle);
1418 * read lov EA of an object
1419 * return the lov EA in an allocated lu_buf
1421 static struct lu_buf *mdd_get_lov_ea(const struct lu_env *env,
1422 struct mdd_object *obj)
1424 struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf;
1425 struct lu_buf *lmm_buf = NULL;
1430 rc = mdo_xattr_get(env, obj, buf, XATTR_NAME_LOV,
1431 mdd_object_capa(env, obj));
1436 GOTO(out, rc = -ENODATA);
1439 if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) {
1440 /* mti_big_buf was not allocated, so we have to
1441 * allocate it based on the ea size */
1442 buf = mdd_buf_alloc(env, sz);
1443 if (buf->lb_buf == NULL)
1444 GOTO(out, rc = -ENOMEM);
1448 OBD_ALLOC_PTR(lmm_buf);
1450 GOTO(out, rc = -ENOMEM);
1452 OBD_ALLOC(lmm_buf->lb_buf, sz);
1453 if (!lmm_buf->lb_buf)
1454 GOTO(free, rc = -ENOMEM);
1456 memcpy(lmm_buf->lb_buf, buf->lb_buf, sz);
1457 lmm_buf->lb_len = sz;
1463 OBD_FREE_PTR(lmm_buf);
1472 * check if layout swapping between 2 objects is allowed
1474 * - same type of objects
1475 * - same owner/group (so quotas are still valid)
1477 static int mdd_layout_swap_allowed(const struct lu_env *env,
1478 struct mdd_object *o1,
1479 struct mdd_object *o2)
1481 const struct lu_fid *fid1, *fid2;
1483 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
1490 if (!fid_is_norm(fid1) || !fid_is_norm(fid2) ||
1491 (mdd_object_type(o1) != mdd_object_type(o2)))
1494 tmp_la->la_valid = 0;
1495 rc = mdd_la_get(env, o1, tmp_la, BYPASS_CAPA);
1498 uid = tmp_la->la_uid;
1499 gid = tmp_la->la_gid;
1501 tmp_la->la_valid = 0;
1502 rc = mdd_la_get(env, o2, tmp_la, BYPASS_CAPA);
1506 if ((uid != tmp_la->la_uid) || (gid != tmp_la->la_gid))
1513 * swap layouts between 2 lustre objects
1515 static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1,
1516 struct md_object *obj2, __u64 flags)
1518 struct mdd_object *o1, *o2, *fst_o, *snd_o;
1519 struct lu_buf *lmm1_buf = NULL, *lmm2_buf = NULL;
1520 struct lu_buf *fst_buf, *snd_buf;
1521 struct lov_mds_md *fst_lmm, *snd_lmm, *old_fst_lmm = NULL;
1522 struct thandle *handle;
1523 struct mdd_device *mdd = mdo2mdd(obj1);
1525 __u16 fst_gen, snd_gen;
1528 /* we have to sort the 2 obj, so locking will always
1529 * be in the same order, even in case of 2 concurrent swaps */
1530 rc = lu_fid_cmp(mdo2fid(md2mdd_obj(obj1)),
1531 mdo2fid(md2mdd_obj(obj2)));
1537 o1 = md2mdd_obj(obj1);
1538 o2 = md2mdd_obj(obj2);
1540 o1 = md2mdd_obj(obj2);
1541 o2 = md2mdd_obj(obj1);
1544 /* check if layout swapping is allowed */
1545 rc = mdd_layout_swap_allowed(env, o1, o2);
1549 handle = mdd_trans_create(env, mdd);
1551 RETURN(PTR_ERR(handle));
1553 /* objects are already sorted */
1554 mdd_write_lock(env, o1, MOR_TGT_CHILD);
1555 mdd_write_lock(env, o2, MOR_TGT_CHILD);
1557 lmm1_buf = mdd_get_lov_ea(env, o1);
1558 if (IS_ERR(lmm1_buf)) {
1559 rc = PTR_ERR(lmm1_buf);
1565 lmm2_buf = mdd_get_lov_ea(env, o2);
1566 if (IS_ERR(lmm2_buf)) {
1567 rc = PTR_ERR(lmm2_buf);
1573 /* swapping 2 non existant layouts is a success */
1574 if ((lmm1_buf == NULL) && (lmm2_buf == NULL))
1575 GOTO(unlock, rc = 0);
1577 /* to help inode migration between MDT, it is better to
1578 * start by the no layout file (if one), so we order the swap */
1579 if (lmm1_buf == NULL) {
1591 /* lmm and generation layout initialization */
1593 fst_lmm = fst_buf->lb_buf;
1594 fst_gen = le16_to_cpu(fst_lmm->lmm_layout_gen);
1601 snd_lmm = snd_buf->lb_buf;
1602 snd_gen = le16_to_cpu(snd_lmm->lmm_layout_gen);
1608 /* save the orignal lmm common header of first file
1609 * to be able to roll back */
1610 OBD_ALLOC_PTR(old_fst_lmm);
1611 if (old_fst_lmm == NULL)
1612 GOTO(unlock, rc = -ENOMEM);
1614 memcpy(old_fst_lmm, fst_lmm, sizeof(*old_fst_lmm));
1616 /* increase the generation layout numbers */
1620 /* set the file specific informations in lmm */
1622 fst_lmm->lmm_layout_gen = cpu_to_le16(snd_gen);
1623 fst_lmm->lmm_object_seq = snd_lmm->lmm_object_seq;
1624 fst_lmm->lmm_object_id = snd_lmm->lmm_object_id;
1628 snd_lmm->lmm_layout_gen = cpu_to_le16(fst_gen);
1629 snd_lmm->lmm_object_seq = old_fst_lmm->lmm_object_seq;
1630 snd_lmm->lmm_object_id = old_fst_lmm->lmm_object_id;
1633 /* prepare transaction */
1634 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_buf, XATTR_NAME_LOV,
1635 LU_XATTR_REPLACE, handle);
1639 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_buf, XATTR_NAME_LOV,
1640 LU_XATTR_REPLACE, handle);
1644 rc = mdd_trans_start(env, mdd, handle);
1648 rc = mdo_xattr_set(env, fst_o, snd_buf, XATTR_NAME_LOV,
1649 LU_XATTR_REPLACE, handle,
1650 mdd_object_capa(env, fst_o));
1654 rc = mdo_xattr_set(env, snd_o, fst_buf, XATTR_NAME_LOV,
1655 LU_XATTR_REPLACE, handle,
1656 mdd_object_capa(env, snd_o));
1660 /* failure on second file, but first was done, so we have
1661 * to roll back first */
1662 /* restore object_id, object_seq and generation number
1665 fst_lmm->lmm_object_id = old_fst_lmm->lmm_object_id;
1666 fst_lmm->lmm_object_seq = old_fst_lmm->lmm_object_seq;
1667 fst_lmm->lmm_layout_gen = old_fst_lmm->lmm_layout_gen;
1670 rc2 = mdo_xattr_set(env, fst_o, fst_buf, XATTR_NAME_LOV,
1671 LU_XATTR_REPLACE, handle,
1672 mdd_object_capa(env, fst_o));
1675 CERROR("%s: unable to roll back after swap layouts"
1676 " failure between "DFID" and "DFID
1677 " rc2 = %d rc = %d)\n",
1678 mdd2obd_dev(mdd)->obd_name,
1679 PFID(mdo2fid(snd_o)), PFID(mdo2fid(fst_o)),
1681 /* a solution to avoid journal commit is to panic,
1682 * but it has strong consequences so we use LBUG to
1683 * allow sysdamin to choose to panic or not
1692 mdd_trans_stop(env, mdd, rc, handle);
1694 mdd_write_unlock(env, o2);
1695 mdd_write_unlock(env, o1);
1697 if (lmm1_buf && lmm1_buf->lb_buf)
1698 OBD_FREE(lmm1_buf->lb_buf, lmm1_buf->lb_len);
1700 OBD_FREE_PTR(lmm1_buf);
1702 if (lmm2_buf && lmm2_buf->lb_buf)
1703 OBD_FREE(lmm2_buf->lb_buf, lmm2_buf->lb_len);
1705 OBD_FREE_PTR(lmm2_buf);
1708 OBD_FREE_PTR(old_fst_lmm);
1713 void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent,
1714 struct mdd_object *child, struct lu_attr *attr)
1716 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1717 struct dt_object *np = parent ? mdd_object_child(parent) : NULL;
1718 struct dt_object *nc = mdd_object_child(child);
1720 /* @hint will be initialized by underlying device. */
1721 nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT);
1725 * do NOT or the MAY_*'s, you'll get the weakest
1727 int accmode(const struct lu_env *env, struct lu_attr *la, int flags)
1731 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1732 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1733 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1734 * owner can write to a file even if it is marked readonly to hide
1735 * its brokenness. (bug 5781) */
1736 if (flags & MDS_OPEN_OWNEROVERRIDE) {
1737 struct lu_ucred *uc = lu_ucred_check(env);
1739 if ((uc == NULL) || (la->la_uid == uc->uc_fsuid))
1743 if (flags & FMODE_READ)
1745 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1747 if (flags & MDS_FMODE_EXEC)
1752 static int mdd_open_sanity_check(const struct lu_env *env,
1753 struct mdd_object *obj, int flag)
1755 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
1760 if (mdd_is_dead_obj(obj))
1763 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
1767 if (S_ISLNK(tmp_la->la_mode))
1770 mode = accmode(env, tmp_la, flag);
1772 if (S_ISDIR(tmp_la->la_mode) && (mode & MAY_WRITE))
1775 if (!(flag & MDS_OPEN_CREATED)) {
1776 rc = mdd_permission_internal(env, obj, tmp_la, mode);
1781 if (S_ISFIFO(tmp_la->la_mode) || S_ISSOCK(tmp_la->la_mode) ||
1782 S_ISBLK(tmp_la->la_mode) || S_ISCHR(tmp_la->la_mode))
1783 flag &= ~MDS_OPEN_TRUNC;
1785 /* For writing append-only file must open it with append mode. */
1786 if (mdd_is_append(obj)) {
1787 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1789 if (flag & MDS_OPEN_TRUNC)
1795 * Now, flag -- O_NOATIME does not be packed by client.
1797 if (flag & O_NOATIME) {
1798 struct lu_ucred *uc = lu_ucred(env);
1800 if (uc && ((uc->uc_valid == UCRED_OLD) ||
1801 (uc->uc_valid == UCRED_NEW)) &&
1802 (uc->uc_fsuid != tmp_la->la_uid) &&
1803 !mdd_capable(uc, CFS_CAP_FOWNER))
1811 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1814 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1817 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1819 rc = mdd_open_sanity_check(env, mdd_obj, flags);
1821 mdd_obj->mod_count++;
1823 mdd_write_unlock(env, mdd_obj);
1827 int mdd_declare_object_kill(const struct lu_env *env, struct mdd_object *obj,
1828 struct md_attr *ma, struct thandle *handle)
1830 return mdo_declare_destroy(env, obj, handle);
1833 /* return md_attr back,
1834 * if it is last unlink then return lov ea + llog cookie*/
1835 int mdd_object_kill(const struct lu_env *env, struct mdd_object *obj,
1836 struct md_attr *ma, struct thandle *handle)
1841 rc = mdo_destroy(env, obj, handle);
1846 static int mdd_declare_close(const struct lu_env *env,
1847 struct mdd_object *obj,
1849 struct thandle *handle)
1853 rc = orph_declare_index_delete(env, obj, handle);
1857 return mdo_declare_destroy(env, obj, handle);
1861 * No permission check is needed.
1863 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1864 struct md_attr *ma, int mode)
1866 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1867 struct mdd_device *mdd = mdo2mdd(obj);
1868 struct thandle *handle = NULL;
1869 int rc, is_orphan = 0;
1872 if (ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_KEEP_ORPHAN) {
1873 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1874 mdd_obj->mod_count--;
1875 mdd_write_unlock(env, mdd_obj);
1877 if (mdd_obj->mod_flags & ORPHAN_OBJ && !mdd_obj->mod_count)
1878 CDEBUG(D_HA, "Object "DFID" is retained in orphan "
1879 "list\n", PFID(mdd_object_fid(mdd_obj)));
1883 /* mdd_finish_unlink() will always set orphan object as DEAD_OBJ, but
1884 * it might fail to add the object to orphan list (w/o ORPHAN_OBJ). */
1885 /* check without any lock */
1886 is_orphan = mdd_obj->mod_count == 1 &&
1887 (mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0;
1891 handle = mdd_trans_create(env, mdo2mdd(obj));
1893 RETURN(PTR_ERR(handle));
1895 rc = mdd_declare_close(env, mdd_obj, ma, handle);
1899 rc = mdd_declare_changelog_store(env, mdd, NULL, handle);
1903 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1908 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1909 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr,
1910 mdd_object_capa(env, mdd_obj));
1912 CERROR("Failed to get lu_attr of "DFID": %d\n",
1913 PFID(mdd_object_fid(mdd_obj)), rc);
1917 /* check again with lock */
1918 is_orphan = (mdd_obj->mod_count == 1) &&
1919 ((mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0 ||
1920 ma->ma_attr.la_nlink == 0);
1922 if (is_orphan && handle == NULL) {
1923 mdd_write_unlock(env, mdd_obj);
1927 mdd_obj->mod_count--; /*release open count */
1933 /* NB: Object maybe not in orphan list originally, it is rare case for
1934 * mdd_finish_unlink() failure, in that case, the object doesn't have
1935 * ORPHAN_OBJ flag */
1936 if ((mdd_obj->mod_flags & ORPHAN_OBJ) != 0) {
1937 /* remove link to object from orphan index */
1938 LASSERT(handle != NULL);
1939 rc = __mdd_orphan_del(env, mdd_obj, handle);
1941 CERROR("%s: unable to delete "DFID" from orphan list: "
1942 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1943 PFID(mdd_object_fid(mdd_obj)), rc);
1944 /* If object was not deleted from orphan list, do not
1945 * destroy OSS objects, which will be done when next
1950 CDEBUG(D_HA, "Object "DFID" is deleted from orphan "
1951 "list, OSS objects to be destroyed.\n",
1952 PFID(mdd_object_fid(mdd_obj)));
1955 rc = mdo_destroy(env, mdd_obj, handle);
1958 CERROR("%s: unable to delete "DFID" from orphan list: "
1959 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1960 PFID(mdd_object_fid(mdd_obj)), rc);
1965 mdd_write_unlock(env, mdd_obj);
1968 (mode & (FMODE_WRITE | MDS_OPEN_APPEND | MDS_OPEN_TRUNC)) &&
1969 !(ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_RECOV_OPEN)) {
1970 if (handle == NULL) {
1971 handle = mdd_trans_create(env, mdo2mdd(obj));
1973 GOTO(stop, rc = IS_ERR(handle));
1975 rc = mdd_declare_changelog_store(env, mdd, NULL,
1980 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1985 mdd_changelog_data_store(env, mdd, CL_CLOSE, mode,
1991 mdd_trans_stop(env, mdd, rc, handle);
1996 * Permission check is done when open,
1997 * no need check again.
1999 static int mdd_readpage_sanity_check(const struct lu_env *env,
2000 struct mdd_object *obj)
2002 struct dt_object *next = mdd_object_child(obj);
2006 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
2014 static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp,
2015 int nob, const struct dt_it_ops *iops,
2016 struct dt_it *it, __u32 attr, void *arg)
2018 struct lu_dirpage *dp = &lp->lp_dir;
2022 struct lu_dirent *ent;
2023 struct lu_dirent *last = NULL;
2026 memset(area, 0, sizeof (*dp));
2027 area += sizeof (*dp);
2028 nob -= sizeof (*dp);
2035 len = iops->key_size(env, it);
2037 /* IAM iterator can return record with zero len. */
2041 hash = iops->store(env, it);
2042 if (unlikely(first)) {
2044 dp->ldp_hash_start = cpu_to_le64(hash);
2047 /* calculate max space required for lu_dirent */
2048 recsize = lu_dirent_calc_size(len, attr);
2050 if (nob >= recsize) {
2051 result = iops->rec(env, it, (struct dt_rec *)ent, attr);
2052 if (result == -ESTALE)
2057 /* osd might not able to pack all attributes,
2058 * so recheck rec length */
2059 recsize = le16_to_cpu(ent->lde_reclen);
2061 result = (last != NULL) ? 0 :-EINVAL;
2065 ent = (void *)ent + recsize;
2069 result = iops->next(env, it);
2070 if (result == -ESTALE)
2072 } while (result == 0);
2075 dp->ldp_hash_end = cpu_to_le64(hash);
2077 if (last->lde_hash == dp->ldp_hash_end)
2078 dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE);
2079 last->lde_reclen = 0; /* end mark */
2082 /* end of directory */
2083 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
2085 CWARN("build page failed: %d!\n", result);
2089 int mdd_readpage(const struct lu_env *env, struct md_object *obj,
2090 const struct lu_rdpg *rdpg)
2092 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2096 if (mdd_object_exists(mdd_obj) == 0) {
2097 CERROR("%s: object "DFID" not found: rc = -2\n",
2098 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
2102 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
2103 rc = mdd_readpage_sanity_check(env, mdd_obj);
2105 GOTO(out_unlock, rc);
2107 if (mdd_is_dead_obj(mdd_obj)) {
2109 struct lu_dirpage *dp;
2112 * According to POSIX, please do not return any entry to client:
2113 * even dot and dotdot should not be returned.
2115 CDEBUG(D_INODE, "readdir from dead object: "DFID"\n",
2116 PFID(mdd_object_fid(mdd_obj)));
2118 if (rdpg->rp_count <= 0)
2119 GOTO(out_unlock, rc = -EFAULT);
2120 LASSERT(rdpg->rp_pages != NULL);
2122 pg = rdpg->rp_pages[0];
2123 dp = (struct lu_dirpage*)cfs_kmap(pg);
2124 memset(dp, 0 , sizeof(struct lu_dirpage));
2125 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
2126 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
2127 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
2129 GOTO(out_unlock, rc = LU_PAGE_SIZE);
2132 rc = dt_index_walk(env, mdd_object_child(mdd_obj), rdpg,
2133 mdd_dir_page_build, NULL);
2135 struct lu_dirpage *dp;
2137 dp = cfs_kmap(rdpg->rp_pages[0]);
2138 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
2141 * No pages were processed, mark this for first page
2144 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
2145 rc = min_t(unsigned int, LU_PAGE_SIZE, rdpg->rp_count);
2147 cfs_kunmap(rdpg->rp_pages[0]);
2150 GOTO(out_unlock, rc);
2152 mdd_read_unlock(env, mdd_obj);
2156 static int mdd_object_sync(const struct lu_env *env, struct md_object *obj)
2158 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2160 if (mdd_object_exists(mdd_obj) == 0) {
2161 CERROR("%s: object "DFID" not found: rc = -2\n",
2162 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
2165 return dt_object_sync(env, mdd_object_child(mdd_obj));
2168 const struct md_object_operations mdd_obj_ops = {
2169 .moo_permission = mdd_permission,
2170 .moo_attr_get = mdd_attr_get,
2171 .moo_attr_set = mdd_attr_set,
2172 .moo_xattr_get = mdd_xattr_get,
2173 .moo_xattr_set = mdd_xattr_set,
2174 .moo_xattr_list = mdd_xattr_list,
2175 .moo_xattr_del = mdd_xattr_del,
2176 .moo_swap_layouts = mdd_swap_layouts,
2177 .moo_open = mdd_open,
2178 .moo_close = mdd_close,
2179 .moo_readpage = mdd_readpage,
2180 .moo_readlink = mdd_readlink,
2181 .moo_changelog = mdd_changelog,
2182 .moo_capa_get = mdd_capa_get,
2183 .moo_object_sync = mdd_object_sync,
2184 .moo_path = mdd_path,