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, 2014, 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>
52 #include <lustre_idmap.h>
53 #include <lustre_param.h>
54 #include <lustre_mds.h>
55 #include <lustre/lustre_idl.h>
57 #include "mdd_internal.h"
59 static const struct lu_object_operations mdd_lu_obj_ops;
61 static int mdd_xattr_get(const struct lu_env *env,
62 struct md_object *obj, struct lu_buf *buf,
65 int mdd_la_get(const struct lu_env *env, struct mdd_object *obj,
68 if (mdd_object_exists(obj) == 0) {
69 CERROR("%s: object "DFID" not found: rc = -2\n",
70 mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj)));
74 return mdo_attr_get(env, obj, la);
77 struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
79 struct mdd_thread_info *info;
81 lu_env_refill((struct lu_env *)env);
82 info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
83 LASSERT(info != NULL);
87 struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
91 buf = &mdd_env_info(env)->mti_buf[0];
97 const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
98 const void *area, ssize_t len)
102 buf = &mdd_env_info(env)->mti_buf[0];
103 buf->lb_buf = (void *)area;
108 struct lu_object *mdd_object_alloc(const struct lu_env *env,
109 const struct lu_object_header *hdr,
112 struct mdd_object *mdd_obj;
114 OBD_SLAB_ALLOC_PTR_GFP(mdd_obj, mdd_object_kmem, GFP_NOFS);
115 if (mdd_obj != NULL) {
118 o = mdd2lu_obj(mdd_obj);
119 lu_object_init(o, NULL, d);
120 mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
121 mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
122 mdd_obj->mod_count = 0;
123 o->lo_ops = &mdd_lu_obj_ops;
130 static int mdd_object_init(const struct lu_env *env, struct lu_object *o,
131 const struct lu_object_conf *unused)
133 struct mdd_device *d = lu2mdd_dev(o->lo_dev);
134 struct mdd_object *mdd_obj = lu2mdd_obj(o);
135 struct lu_object *below;
136 struct lu_device *under;
139 mdd_obj->mod_cltime = 0;
140 under = &d->mdd_child->dd_lu_dev;
141 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
143 RETURN(PTR_ERR(below));
145 lu_object_add(o, below);
150 static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
154 if (lu_object_exists(o)) {
155 struct mdd_object *mdd_obj = lu2mdd_obj(o);
156 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
158 rc = mdd_la_get(env, mdd_obj, attr);
164 static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
166 struct mdd_object *mdd = lu2mdd_obj(o);
169 OBD_SLAB_FREE_PTR(mdd, mdd_object_kmem);
172 static int mdd_object_print(const struct lu_env *env, void *cookie,
173 lu_printer_t p, const struct lu_object *o)
175 struct mdd_object *mdd = lu2mdd_obj((struct lu_object *)o);
176 return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p(open_count=%d, "
177 "valid=%x, cltime="LPU64", flags=%lx)",
178 mdd, mdd->mod_count, mdd->mod_valid,
179 mdd->mod_cltime, mdd->mod_flags);
182 static const struct lu_object_operations mdd_lu_obj_ops = {
183 .loo_object_init = mdd_object_init,
184 .loo_object_start = mdd_object_start,
185 .loo_object_free = mdd_object_free,
186 .loo_object_print = mdd_object_print,
189 struct mdd_object *mdd_object_find(const struct lu_env *env,
190 struct mdd_device *d,
191 const struct lu_fid *f)
193 return md2mdd_obj(md_object_find_slice(env, &d->mdd_md_dev, f));
197 * No permission check is needed.
199 int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
202 struct mdd_object *mdd_obj = md2mdd_obj(obj);
207 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr);
208 if ((ma->ma_need & MA_INODE) != 0 && mdd_is_dead_obj(mdd_obj))
209 ma->ma_attr.la_nlink = 0;
215 * No permission check is needed.
217 static int mdd_xattr_get(const struct lu_env *env,
218 struct md_object *obj, struct lu_buf *buf,
221 struct mdd_object *mdd_obj = md2mdd_obj(obj);
226 if (mdd_object_exists(mdd_obj) == 0) {
227 CERROR("%s: object "DFID" not found: rc = -2\n",
228 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
232 /* If the object has been destroyed, then do not get LMVEA, because
233 * it needs to load stripes from the iteration of the master object,
234 * and it will cause problem if master object has been destroyed, see
236 if (unlikely((mdd_obj->mod_flags & DEAD_OBJ) &&
237 !(mdd_obj->mod_flags & ORPHAN_OBJ) &&
238 strcmp(name, XATTR_NAME_LMV) == 0))
241 /* If the object has been delete from the namespace, then
242 * get linkEA should return -ENOENT as well */
243 if (unlikely((mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ)) &&
244 strcmp(name, XATTR_NAME_LINK) == 0))
247 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
248 rc = mdo_xattr_get(env, mdd_obj, buf, name);
249 mdd_read_unlock(env, mdd_obj);
255 * Permission check is done when open,
256 * no need check again.
258 int mdd_readlink(const struct lu_env *env, struct md_object *obj,
261 struct mdd_object *mdd_obj = md2mdd_obj(obj);
262 struct dt_object *next;
267 if (mdd_object_exists(mdd_obj) == 0) {
268 CERROR("%s: object "DFID" not found: rc = -2\n",
269 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
273 next = mdd_object_child(mdd_obj);
274 LASSERT(next != NULL);
275 LASSERT(next->do_body_ops != NULL);
276 LASSERT(next->do_body_ops->dbo_read != NULL);
277 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
278 rc = dt_read(env, next, buf, &pos);
279 mdd_read_unlock(env, mdd_obj);
284 * No permission check is needed.
286 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
289 struct mdd_object *mdd_obj = md2mdd_obj(obj);
294 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
295 rc = mdo_xattr_list(env, mdd_obj, buf);
296 mdd_read_unlock(env, mdd_obj);
302 * Filter out XATTR_NAME_LINK if this is an orphan object. See
305 if (unlikely(mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ))) {
306 char *end = (char *)buf->lb_buf + rc;
307 char *p = buf->lb_buf;
310 char *next = p + strlen(p) + 1;
312 if (strcmp(p, XATTR_NAME_LINK) == 0) {
314 memmove(p, next, end - next);
316 CDEBUG(D_INFO, "Filtered out "XATTR_NAME_LINK
317 " of orphan "DFID"\n",
318 PFID(mdd_object_fid(mdd_obj)));
329 int mdd_declare_object_create_internal(const struct lu_env *env,
330 struct mdd_object *p,
331 struct mdd_object *c,
332 struct lu_attr *attr,
333 struct thandle *handle,
334 const struct md_op_spec *spec,
335 struct dt_allocation_hint *hint)
337 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
338 const struct dt_index_features *feat = spec->sp_feat;
342 if (feat != &dt_directory_features && feat != NULL) {
343 dof->dof_type = DFT_INDEX;
344 dof->u.dof_idx.di_feat = feat;
347 dof->dof_type = dt_mode_to_dft(attr->la_mode);
348 if (dof->dof_type == DFT_REGULAR) {
349 dof->u.dof_reg.striped =
350 md_should_create(spec->sp_cr_flags);
351 if (spec->sp_cr_flags & MDS_OPEN_HAS_EA)
352 dof->u.dof_reg.striped = 0;
353 /* is this replay? */
355 dof->u.dof_reg.striped = 0;
359 rc = mdo_declare_create_obj(env, c, attr, hint, dof, handle);
364 int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p,
365 struct mdd_object *c, struct lu_attr *attr,
366 struct thandle *handle,
367 const struct md_op_spec *spec,
368 struct dt_allocation_hint *hint)
370 struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
374 LASSERT(!mdd_object_exists(c));
376 rc = mdo_create_obj(env, c, attr, hint, dof, handle);
381 int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj,
382 const struct lu_attr *attr, struct thandle *handle,
388 rc = mdo_attr_set(env, obj, attr, handle);
389 #ifdef CONFIG_FS_POSIX_ACL
390 if (!rc && (attr->la_valid & LA_MODE) && needacl)
391 rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
396 int mdd_update_time(const struct lu_env *env, struct mdd_object *obj,
397 const struct lu_attr *oattr, struct lu_attr *attr,
398 struct thandle *handle)
403 LASSERT(attr->la_valid & LA_CTIME);
404 LASSERT(oattr != NULL);
406 /* Make sure the ctime is increased only, however, it's not strictly
407 * reliable at here because there is not guarantee to hold lock on
408 * object, so we just bypass some unnecessary cmtime setting first
409 * and OSD has to check it again. */
410 if (attr->la_ctime < oattr->la_ctime)
411 attr->la_valid &= ~(LA_MTIME | LA_CTIME);
412 else if (attr->la_valid == LA_CTIME &&
413 attr->la_ctime == oattr->la_ctime)
414 attr->la_valid &= ~LA_CTIME;
416 if (attr->la_valid != 0)
417 rc = mdd_attr_set_internal(env, obj, attr, handle, 0);
422 * This gives the same functionality as the code between
423 * sys_chmod and inode_setattr
424 * chown_common and inode_setattr
425 * utimes and inode_setattr
426 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
428 static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
429 const struct lu_attr *oattr, struct lu_attr *la,
430 const unsigned long flags)
439 /* Do not permit change file type */
440 if (la->la_valid & LA_TYPE)
443 /* They should not be processed by setattr */
444 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
447 LASSERT(oattr != NULL);
449 uc = lu_ucred_check(env);
453 if (la->la_valid == LA_CTIME) {
454 if (!(flags & MDS_PERM_BYPASS))
455 /* This is only for set ctime when rename's source is
457 rc = mdd_may_delete(env, NULL, NULL, obj, oattr, NULL,
459 if (rc == 0 && la->la_ctime <= oattr->la_ctime)
460 la->la_valid &= ~LA_CTIME;
464 if (la->la_valid == LA_ATIME) {
465 /* This is atime only set for read atime update on close. */
466 if (la->la_atime >= oattr->la_atime &&
467 la->la_atime < (oattr->la_atime +
468 mdd_obj2mdd_dev(obj)->mdd_atime_diff))
469 la->la_valid &= ~LA_ATIME;
473 /* Check if flags change. */
474 if (la->la_valid & LA_FLAGS) {
475 unsigned int oldflags = oattr->la_flags &
476 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
477 unsigned int newflags = la->la_flags &
478 (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
480 if ((uc->uc_fsuid != oattr->la_uid) &&
481 !md_capable(uc, CFS_CAP_FOWNER))
484 /* The IMMUTABLE and APPEND_ONLY flags can
485 * only be changed by the relevant capability. */
486 if ((oldflags ^ newflags) &&
487 !md_capable(uc, CFS_CAP_LINUX_IMMUTABLE))
490 if (!S_ISDIR(oattr->la_mode))
491 la->la_flags &= ~LUSTRE_DIRSYNC_FL;
494 if (oattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL) &&
495 (la->la_valid & ~LA_FLAGS) &&
496 !(flags & MDS_PERM_BYPASS))
499 /* Check for setting the obj time. */
500 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
501 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
502 if ((uc->uc_fsuid != oattr->la_uid) &&
503 !md_capable(uc, CFS_CAP_FOWNER)) {
504 rc = mdd_permission_internal(env, obj, oattr,
511 if (la->la_valid & LA_KILL_SUID) {
512 la->la_valid &= ~LA_KILL_SUID;
513 if ((oattr->la_mode & S_ISUID) &&
514 !(la->la_valid & LA_MODE)) {
515 la->la_mode = oattr->la_mode;
516 la->la_valid |= LA_MODE;
518 la->la_mode &= ~S_ISUID;
521 if (la->la_valid & LA_KILL_SGID) {
522 la->la_valid &= ~LA_KILL_SGID;
523 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
524 (S_ISGID | S_IXGRP)) &&
525 !(la->la_valid & LA_MODE)) {
526 la->la_mode = oattr->la_mode;
527 la->la_valid |= LA_MODE;
529 la->la_mode &= ~S_ISGID;
532 /* Make sure a caller can chmod. */
533 if (la->la_valid & LA_MODE) {
534 if (!(flags & MDS_PERM_BYPASS) &&
535 (uc->uc_fsuid != oattr->la_uid) &&
536 !md_capable(uc, CFS_CAP_FOWNER))
539 if (la->la_mode == (umode_t) -1)
540 la->la_mode = oattr->la_mode;
542 la->la_mode = (la->la_mode & S_IALLUGO) |
543 (oattr->la_mode & ~S_IALLUGO);
545 /* Also check the setgid bit! */
546 if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ?
547 la->la_gid : oattr->la_gid) &&
548 !md_capable(uc, CFS_CAP_FSETID))
549 la->la_mode &= ~S_ISGID;
551 la->la_mode = oattr->la_mode;
554 /* Make sure a caller can chown. */
555 if (la->la_valid & LA_UID) {
556 if (la->la_uid == (uid_t) -1)
557 la->la_uid = oattr->la_uid;
558 if (((uc->uc_fsuid != oattr->la_uid) ||
559 (la->la_uid != oattr->la_uid)) &&
560 !md_capable(uc, CFS_CAP_CHOWN))
563 /* If the user or group of a non-directory has been
564 * changed by a non-root user, remove the setuid bit.
565 * 19981026 David C Niemi <niemi@tux.org>
567 * Changed this to apply to all users, including root,
568 * to avoid some races. This is the behavior we had in
569 * 2.0. The check for non-root was definitely wrong
570 * for 2.2 anyway, as it should have been using
571 * CAP_FSETID rather than fsuid -- 19990830 SD. */
572 if (((oattr->la_mode & S_ISUID) == S_ISUID) &&
573 !S_ISDIR(oattr->la_mode)) {
574 la->la_mode &= ~S_ISUID;
575 la->la_valid |= LA_MODE;
579 /* Make sure caller can chgrp. */
580 if (la->la_valid & LA_GID) {
581 if (la->la_gid == (gid_t) -1)
582 la->la_gid = oattr->la_gid;
583 if (((uc->uc_fsuid != oattr->la_uid) ||
584 ((la->la_gid != oattr->la_gid) &&
585 !lustre_in_group_p(uc, la->la_gid))) &&
586 !md_capable(uc, CFS_CAP_CHOWN))
589 /* Likewise, if the user or group of a non-directory
590 * has been changed by a non-root user, remove the
591 * setgid bit UNLESS there is no group execute bit
592 * (this would be a file marked for mandatory
593 * locking). 19981026 David C Niemi <niemi@tux.org>
595 * Removed the fsuid check (see the comment above) --
597 if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
598 (S_ISGID | S_IXGRP)) && !S_ISDIR(oattr->la_mode)) {
599 la->la_mode &= ~S_ISGID;
600 la->la_valid |= LA_MODE;
604 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
605 if (!((flags & MDS_OWNEROVERRIDE) &&
606 (uc->uc_fsuid == oattr->la_uid)) &&
607 !(flags & MDS_PERM_BYPASS)) {
608 rc = mdd_permission_internal(env, obj, oattr,
615 if (la->la_valid & LA_CTIME) {
616 /* The pure setattr, it has the priority over what is
617 * already set, do not drop it if ctime is equal. */
618 if (la->la_ctime < oattr->la_ctime)
619 la->la_valid &= ~(LA_ATIME | LA_MTIME | LA_CTIME);
625 /** Store a data change changelog record
626 * If this fails, we must fail the whole transaction; we don't
627 * want the change to commit without the log entry.
628 * \param mdd_obj - mdd_object of change
629 * \param handle - transaction handle
631 int mdd_changelog_data_store(const struct lu_env *env, struct mdd_device *mdd,
632 enum changelog_rec_type type, int flags,
633 struct mdd_object *mdd_obj, struct thandle *handle)
635 const struct lu_ucred *uc = lu_ucred(env);
636 const struct lu_fid *tfid;
637 struct llog_changelog_rec *rec;
643 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
645 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
648 LASSERT(mdd_obj != NULL);
649 LASSERT(handle != NULL);
651 tfid = mdo2fid(mdd_obj);
653 if ((type >= CL_MTIME) && (type <= CL_ATIME) &&
654 cfs_time_before_64(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) {
655 /* Don't need multiple updates in this log */
656 /* Don't check under lock - no big deal if we get an extra
661 flags = (flags & CLF_FLAGMASK) | CLF_VERSION;
662 if (uc != NULL && uc->uc_jobid[0] != '\0')
665 reclen = llog_data_len(changelog_rec_offset(flags & CLF_SUPPORTED));
666 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
667 if (buf->lb_buf == NULL)
671 rec->cr.cr_flags = flags;
672 rec->cr.cr_type = (__u32)type;
673 rec->cr.cr_tfid = *tfid;
674 rec->cr.cr_namelen = 0;
675 mdd_obj->mod_cltime = cfs_time_current_64();
677 if (flags & CLF_JOBID)
678 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
680 rc = mdd_changelog_store(env, mdd, rec, handle);
685 static int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type,
686 int flags, struct md_object *obj)
688 struct thandle *handle;
689 struct mdd_object *mdd_obj = md2mdd_obj(obj);
690 struct mdd_device *mdd = mdo2mdd(obj);
694 handle = mdd_trans_create(env, mdd);
696 RETURN(PTR_ERR(handle));
698 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
702 rc = mdd_trans_start(env, mdd, handle);
706 rc = mdd_changelog_data_store(env, mdd, type, flags, mdd_obj,
710 mdd_trans_stop(env, mdd, rc, handle);
716 * Save LMA extended attributes with data from \a ma.
718 * HSM and Size-On-MDS data will be extracted from \ma if they are valid, if
719 * not, LMA EA will be first read from disk, modified and write back.
722 /* Precedence for choosing record type when multiple
723 * attributes change: setattr > mtime > ctime > atime
724 * (ctime changes when mtime does, plus chmod/chown.
725 * atime and ctime are independent.) */
726 static int mdd_attr_set_changelog(const struct lu_env *env,
727 struct md_object *obj, struct thandle *handle,
730 struct mdd_device *mdd = mdo2mdd(obj);
733 bits = (valid & LA_SIZE) ? 1 << CL_TRUNC : 0;
734 bits |= (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0;
735 bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0;
736 bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0;
737 bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0;
738 bits = bits & mdd->mdd_cl.mc_mask;
739 /* This is an implementation limit rather than a protocol limit */
740 CLASSERT(CL_LAST <= sizeof(int) * 8);
744 /* The record type is the lowest non-masked set bit */
747 /* FYI we only store the first CLF_FLAGMASK bits of la_valid */
748 return mdd_changelog_data_store(env, mdd, type, (int)valid,
749 md2mdd_obj(obj), handle);
752 static int mdd_declare_attr_set(const struct lu_env *env,
753 struct mdd_device *mdd,
754 struct mdd_object *obj,
755 const struct lu_attr *attr,
756 struct thandle *handle)
760 rc = mdo_declare_attr_set(env, obj, attr, handle);
764 #ifdef CONFIG_FS_POSIX_ACL
765 if (attr->la_valid & LA_MODE) {
766 mdd_read_lock(env, obj, MOR_TGT_CHILD);
767 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL,
768 XATTR_NAME_ACL_ACCESS);
769 mdd_read_unlock(env, obj);
770 if (rc == -EOPNOTSUPP || rc == -ENODATA)
776 struct lu_buf *buf = mdd_buf_get(env, NULL, rc);
777 rc = mdo_declare_xattr_set(env, obj, buf,
778 XATTR_NAME_ACL_ACCESS, 0,
786 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
793 * permission changes may require sync operation, to mitigate performance
794 * impact, only do this for dir and when permission is reduced.
796 * For regular files, version is updated with permission change (see VBR), async
797 * permission won't cause any issue, while missing permission change on
798 * directory may affect accessibility of other objects after recovery.
800 static inline bool permission_needs_sync(const struct lu_attr *old,
801 const struct lu_attr *new)
803 if (!S_ISDIR(old->la_mode))
806 if (new->la_valid & (LA_UID | LA_GID))
809 if (new->la_valid & LA_MODE &&
810 new->la_mode & (S_ISUID | S_ISGID | S_ISVTX))
813 if ((new->la_valid & LA_MODE) &&
814 ((new->la_mode & old->la_mode) & S_IRWXUGO) !=
815 (old->la_mode & S_IRWXUGO))
821 /* set attr and LOV EA at once, return updated attr */
822 int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
823 const struct md_attr *ma)
825 struct mdd_object *mdd_obj = md2mdd_obj(obj);
826 struct mdd_device *mdd = mdo2mdd(obj);
827 struct thandle *handle;
828 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
829 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
830 const struct lu_attr *la = &ma->ma_attr;
834 /* we do not use ->attr_set() for LOV/HSM EA any more */
835 LASSERT((ma->ma_valid & MA_LOV) == 0);
836 LASSERT((ma->ma_valid & MA_HSM) == 0);
838 rc = mdd_la_get(env, mdd_obj, attr);
842 *la_copy = ma->ma_attr;
843 rc = mdd_fix_attr(env, mdd_obj, attr, la_copy, ma->ma_attr_flags);
847 /* no need to setattr anymore */
848 if (la_copy->la_valid == 0) {
849 CDEBUG(D_INODE, "%s: no valid attribute on "DFID", previous"
850 "valid is "LPX64"\n", mdd2obd_dev(mdd)->obd_name,
851 PFID(mdo2fid(mdd_obj)), la->la_valid);
856 handle = mdd_trans_create(env, mdd);
858 RETURN(PTR_ERR(handle));
860 rc = mdd_declare_attr_set(env, mdd, mdd_obj, la_copy, handle);
864 rc = mdd_trans_start(env, mdd, handle);
868 if (mdd->mdd_sync_permission && permission_needs_sync(attr, la))
871 if (la->la_valid & (LA_MTIME | LA_CTIME))
872 CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n",
873 la->la_mtime, la->la_ctime);
875 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
876 if (la_copy->la_valid & LA_FLAGS)
877 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
878 else if (la_copy->la_valid) /* setattr */
879 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
880 mdd_write_unlock(env, mdd_obj);
883 rc = mdd_attr_set_changelog(env, obj, handle,
889 mdd_trans_stop(env, mdd, rc, handle);
893 static int mdd_xattr_sanity_check(const struct lu_env *env,
894 struct mdd_object *obj,
895 const struct lu_attr *attr)
897 struct lu_ucred *uc = lu_ucred_assert(env);
900 if (attr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
903 if ((uc->uc_fsuid != attr->la_uid) && !md_capable(uc, CFS_CAP_FOWNER))
909 static int mdd_declare_xattr_set(const struct lu_env *env,
910 struct mdd_device *mdd,
911 struct mdd_object *obj,
912 const struct lu_buf *buf,
914 int fl, struct thandle *handle)
918 rc = mdo_declare_xattr_set(env, obj, buf, name, fl, handle);
922 /* Only record user and layout xattr changes */
923 if (strncmp(XATTR_USER_PREFIX, name,
924 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
925 strcmp(XATTR_NAME_LOV, name) == 0) {
926 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
931 /* If HSM data is modified, this could add a changelog */
932 if (strcmp(XATTR_NAME_HSM, name) == 0) {
933 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
938 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
943 * Compare current and future data of HSM EA and add a changelog if needed.
945 * Caller should have write-locked \param obj.
947 * \param buf - Future HSM EA content.
948 * \retval 0 if no changelog is needed or changelog was added properly.
949 * \retval -ve errno if there was a problem
951 static int mdd_hsm_update_locked(const struct lu_env *env,
952 struct md_object *obj,
953 const struct lu_buf *buf,
954 struct thandle *handle)
956 struct mdd_thread_info *info = mdd_env_info(env);
957 struct mdd_device *mdd = mdo2mdd(obj);
958 struct mdd_object *mdd_obj = md2mdd_obj(obj);
959 struct lu_buf *current_buf;
960 struct md_hsm *current_mh;
961 struct md_hsm *new_mh;
965 OBD_ALLOC_PTR(current_mh);
966 if (current_mh == NULL)
969 /* Read HSM attrs from disk */
970 CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
971 current_buf = mdd_buf_get(env, info->mti_xattr_buf,
972 sizeof(info->mti_xattr_buf));
973 rc = mdo_xattr_get(env, mdd_obj, current_buf, XATTR_NAME_HSM);
974 rc = lustre_buf2hsm(current_buf->lb_buf, rc, current_mh);
975 if (rc < 0 && rc != -ENODATA)
977 else if (rc == -ENODATA)
978 current_mh->mh_flags = 0;
980 /* Map future HSM xattr */
981 OBD_ALLOC_PTR(new_mh);
983 GOTO(free, rc = -ENOMEM);
984 lustre_buf2hsm(buf->lb_buf, buf->lb_len, new_mh);
986 /* If HSM flags are different, add a changelog */
988 if (current_mh->mh_flags != new_mh->mh_flags) {
990 hsm_set_cl_event(&flags, HE_STATE);
991 if (new_mh->mh_flags & HS_DIRTY)
992 hsm_set_cl_flags(&flags, CLF_HSM_DIRTY);
994 rc = mdd_changelog_data_store(env, mdd, CL_HSM, flags, mdd_obj,
998 OBD_FREE_PTR(new_mh);
1000 OBD_FREE_PTR(current_mh);
1004 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1008 * The caller should guarantee to update the object ctime
1009 * after xattr_set if needed.
1011 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
1012 const struct lu_buf *buf, const char *name,
1015 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1016 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1017 struct mdd_device *mdd = mdo2mdd(obj);
1018 struct thandle *handle;
1022 rc = mdd_la_get(env, mdd_obj, attr);
1026 rc = mdd_xattr_sanity_check(env, mdd_obj, attr);
1030 if (strcmp(name, XATTR_NAME_ACL_ACCESS) == 0 ||
1031 strcmp(name, XATTR_NAME_ACL_DEFAULT) == 0) {
1032 struct posix_acl *acl;
1034 /* user may set empty ACL, which should be treated as removing
1036 acl = posix_acl_from_xattr(&init_user_ns, buf->lb_buf,
1039 rc = mdd_xattr_del(env, obj, name);
1042 posix_acl_release(acl);
1045 if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) {
1046 rc = mdd_acl_set(env, mdd_obj, attr, buf, fl);
1050 handle = mdd_trans_create(env, mdd);
1052 RETURN(PTR_ERR(handle));
1054 rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, fl, handle);
1058 rc = mdd_trans_start(env, mdd, handle);
1062 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1064 if (strcmp(XATTR_NAME_HSM, name) == 0) {
1065 rc = mdd_hsm_update_locked(env, obj, buf, handle);
1067 mdd_write_unlock(env, mdd_obj);
1072 rc = mdo_xattr_set(env, mdd_obj, buf, name, fl, handle);
1073 mdd_write_unlock(env, mdd_obj);
1077 if (strcmp(XATTR_NAME_LOV, name) == 0)
1078 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, mdd_obj,
1080 else if (strncmp(XATTR_USER_PREFIX, name,
1081 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1082 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1083 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1084 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1085 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1086 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1090 mdd_trans_stop(env, mdd, rc, handle);
1095 static int mdd_declare_xattr_del(const struct lu_env *env,
1096 struct mdd_device *mdd,
1097 struct mdd_object *obj,
1099 struct thandle *handle)
1103 rc = mdo_declare_xattr_del(env, obj, name, handle);
1107 /* Only record user xattr changes */
1108 if ((strncmp(XATTR_USER_PREFIX, name,
1109 sizeof(XATTR_USER_PREFIX) - 1) == 0))
1110 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1116 * The caller should guarantee to update the object ctime
1117 * after xattr_set if needed.
1119 static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
1122 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1123 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1124 struct mdd_device *mdd = mdo2mdd(obj);
1125 struct thandle *handle;
1129 rc = mdd_la_get(env, mdd_obj, attr);
1133 rc = mdd_xattr_sanity_check(env, mdd_obj, attr);
1137 handle = mdd_trans_create(env, mdd);
1139 RETURN(PTR_ERR(handle));
1141 rc = mdd_declare_xattr_del(env, mdd, mdd_obj, name, handle);
1145 rc = mdd_trans_start(env, mdd, handle);
1149 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1150 rc = mdo_xattr_del(env, mdd_obj, name, handle);
1151 mdd_write_unlock(env, mdd_obj);
1155 /* Only record system & user xattr changes */
1156 if (strncmp(XATTR_USER_PREFIX, name,
1157 sizeof(XATTR_USER_PREFIX) - 1) == 0 ||
1158 strncmp(POSIX_ACL_XATTR_ACCESS, name,
1159 sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 ||
1160 strncmp(POSIX_ACL_XATTR_DEFAULT, name,
1161 sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0)
1162 rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj,
1166 mdd_trans_stop(env, mdd, rc, handle);
1172 * read lov EA of an object
1173 * return the lov EA in an allocated lu_buf
1175 int mdd_get_lov_ea(const struct lu_env *env, struct mdd_object *obj,
1176 struct lu_buf *lmm_buf)
1178 struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf;
1183 rc = mdo_xattr_get(env, obj, buf, XATTR_NAME_LOV);
1185 if (rc == -ERANGE) {
1186 /* mti_big_buf is allocated but is too small
1187 * we need to increase it */
1188 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1190 if (buf->lb_buf == NULL)
1191 GOTO(out, rc = -ENOMEM);
1202 if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) {
1203 /* mti_big_buf was not allocated, so we have to
1204 * allocate it based on the ea size */
1205 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
1207 if (buf->lb_buf == NULL)
1208 GOTO(out, rc = -ENOMEM);
1212 lu_buf_alloc(lmm_buf, bufsize);
1213 if (lmm_buf->lb_buf == NULL)
1214 GOTO(out, rc = -ENOMEM);
1216 memcpy(lmm_buf->lb_buf, buf->lb_buf, bufsize);
1222 lu_buf_free(lmm_buf);
1226 static int mdd_xattr_hsm_replace(const struct lu_env *env,
1227 struct mdd_object *o, struct lu_buf *buf,
1228 struct thandle *handle)
1230 struct hsm_attrs *attrs;
1236 rc = mdo_xattr_set(env, o, buf, XATTR_NAME_HSM, LU_XATTR_REPLACE,
1241 attrs = buf->lb_buf;
1242 hsm_flags = le32_to_cpu(attrs->hsm_flags);
1243 if (!(hsm_flags & HS_RELEASED) || mdd_is_dead_obj(o))
1246 /* Add a changelog record for release. */
1247 hsm_set_cl_event(&flags, HE_RELEASE);
1248 rc = mdd_changelog_data_store(env, mdo2mdd(&o->mod_obj), CL_HSM,
1254 * check if layout swapping between 2 objects is allowed
1256 * - only normal FIDs or non-system IGIFs
1257 * - same type of objects
1258 * - same owner/group (so quotas are still valid)
1260 static int mdd_layout_swap_allowed(const struct lu_env *env,
1261 struct mdd_object *o1,
1262 const struct lu_attr *attr1,
1263 struct mdd_object *o2,
1264 const struct lu_attr *attr2)
1266 const struct lu_fid *fid1, *fid2;
1272 if (!fid_is_norm(fid1) &&
1273 (!fid_is_igif(fid1) || IS_ERR(mdd_links_get(env, o1))))
1276 if (!fid_is_norm(fid2) &&
1277 (!fid_is_igif(fid2) || IS_ERR(mdd_links_get(env, o2))))
1280 if (mdd_object_type(o1) != mdd_object_type(o2)) {
1281 if (S_ISDIR(mdd_object_type(o1)))
1283 if (S_ISREG(mdd_object_type(o1)))
1288 if ((attr1->la_uid != attr2->la_uid) ||
1289 (attr1->la_gid != attr2->la_gid))
1296 * swap layouts between 2 lustre objects
1298 static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1,
1299 struct md_object *obj2, __u64 flags)
1301 struct mdd_thread_info *info = mdd_env_info(env);
1302 struct mdd_object *fst_o = md2mdd_obj(obj1);
1303 struct mdd_object *snd_o = md2mdd_obj(obj2);
1304 struct lu_attr *fst_la = MDD_ENV_VAR(env, cattr);
1305 struct lu_attr *snd_la = MDD_ENV_VAR(env, tattr);
1306 struct mdd_device *mdd = mdo2mdd(obj1);
1307 struct lov_mds_md *fst_lmm, *snd_lmm;
1308 struct lu_buf *fst_buf = &info->mti_buf[0];
1309 struct lu_buf *snd_buf = &info->mti_buf[1];
1310 struct lu_buf *fst_hsm_buf = &info->mti_buf[2];
1311 struct lu_buf *snd_hsm_buf = &info->mti_buf[3];
1312 struct ost_id *saved_oi = NULL;
1313 struct thandle *handle;
1314 __u16 fst_gen, snd_gen;
1320 CLASSERT(ARRAY_SIZE(info->mti_buf) >= 4);
1321 memset(info->mti_buf, 0, sizeof(info->mti_buf));
1323 /* we have to sort the 2 obj, so locking will always
1324 * be in the same order, even in case of 2 concurrent swaps */
1325 rc = lu_fid_cmp(mdo2fid(fst_o), mdo2fid(snd_o));
1326 if (rc == 0) /* same fid ? */
1332 rc = mdd_la_get(env, fst_o, fst_la);
1336 rc = mdd_la_get(env, snd_o, snd_la);
1340 /* check if layout swapping is allowed */
1341 rc = mdd_layout_swap_allowed(env, fst_o, fst_la, snd_o, snd_la);
1345 handle = mdd_trans_create(env, mdd);
1347 RETURN(PTR_ERR(handle));
1349 /* objects are already sorted */
1350 mdd_write_lock(env, fst_o, MOR_TGT_CHILD);
1351 mdd_write_lock(env, snd_o, MOR_TGT_CHILD);
1353 rc = mdd_get_lov_ea(env, fst_o, fst_buf);
1354 if (rc < 0 && rc != -ENODATA)
1357 rc = mdd_get_lov_ea(env, snd_o, snd_buf);
1358 if (rc < 0 && rc != -ENODATA)
1361 /* swapping 2 non existant layouts is a success */
1362 if (fst_buf->lb_buf == NULL && snd_buf->lb_buf == NULL)
1365 /* to help inode migration between MDT, it is better to
1366 * start by the no layout file (if one), so we order the swap */
1367 if (snd_buf->lb_buf == NULL) {
1369 swap(fst_buf, snd_buf);
1372 /* lmm and generation layout initialization */
1373 if (fst_buf->lb_buf != NULL) {
1374 fst_lmm = fst_buf->lb_buf;
1375 fst_gen = le16_to_cpu(fst_lmm->lmm_layout_gen);
1376 fst_fl = LU_XATTR_REPLACE;
1380 fst_fl = LU_XATTR_CREATE;
1383 snd_lmm = snd_buf->lb_buf;
1384 snd_gen = le16_to_cpu(snd_lmm->lmm_layout_gen);
1386 /* increase the generation layout numbers */
1390 /* set the file specific informations in lmm */
1391 if (fst_lmm != NULL) {
1392 saved_oi = &info->mti_oa.o_oi;
1394 *saved_oi = fst_lmm->lmm_oi;
1395 fst_lmm->lmm_layout_gen = cpu_to_le16(snd_gen);
1396 fst_lmm->lmm_oi = snd_lmm->lmm_oi;
1397 snd_lmm->lmm_oi = *saved_oi;
1399 if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1))
1400 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEF);
1401 else if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))
1402 snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V3_DEF);
1404 GOTO(stop, rc = -EPROTO);
1406 snd_lmm->lmm_layout_gen = cpu_to_le16(fst_gen);
1408 /* Prepare HSM attribute if it's required */
1409 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1410 const int buflen = sizeof(struct hsm_attrs);
1412 lu_buf_alloc(fst_hsm_buf, buflen);
1413 lu_buf_alloc(snd_hsm_buf, buflen);
1414 if (fst_hsm_buf->lb_buf == NULL || snd_hsm_buf->lb_buf == NULL)
1415 GOTO(stop, rc = -ENOMEM);
1417 /* Read HSM attribute */
1418 rc = mdo_xattr_get(env, fst_o, fst_hsm_buf, XATTR_NAME_HSM);
1422 rc = mdo_xattr_get(env, snd_o, snd_hsm_buf, XATTR_NAME_HSM);
1426 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_hsm_buf,
1427 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1432 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_hsm_buf,
1433 XATTR_NAME_HSM, LU_XATTR_REPLACE,
1439 /* prepare transaction */
1440 rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_buf, XATTR_NAME_LOV,
1445 if (fst_buf->lb_buf != NULL)
1446 rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_buf,
1447 XATTR_NAME_LOV, LU_XATTR_REPLACE,
1450 rc = mdd_declare_xattr_del(env, mdd, snd_o, XATTR_NAME_LOV,
1455 rc = mdd_trans_start(env, mdd, handle);
1459 if (flags & SWAP_LAYOUTS_MDS_HSM) {
1460 rc = mdd_xattr_hsm_replace(env, fst_o, snd_hsm_buf, handle);
1464 rc = mdd_xattr_hsm_replace(env, snd_o, fst_hsm_buf, handle);
1466 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf,
1469 CERROR("%s: restore "DFID" HSM error: %d/%d\n",
1470 mdd_obj_dev_name(fst_o),
1471 PFID(mdo2fid(fst_o)), rc, rc2);
1476 rc = mdo_xattr_set(env, fst_o, snd_buf, XATTR_NAME_LOV, fst_fl, handle);
1480 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_MDS_HSM_SWAP_LAYOUTS))) {
1483 if (fst_buf->lb_buf != NULL)
1484 rc = mdo_xattr_set(env, snd_o, fst_buf, XATTR_NAME_LOV,
1485 LU_XATTR_REPLACE, handle);
1487 rc = mdo_xattr_del(env, snd_o, XATTR_NAME_LOV, handle);
1493 /* failure on second file, but first was done, so we have
1494 * to roll back first. */
1495 if (fst_buf->lb_buf != NULL) {
1496 fst_lmm->lmm_oi = *saved_oi;
1497 fst_lmm->lmm_layout_gen = cpu_to_le16(fst_gen - 1);
1498 rc2 = mdo_xattr_set(env, fst_o, fst_buf, XATTR_NAME_LOV,
1499 LU_XATTR_REPLACE, handle);
1501 rc2 = mdo_xattr_del(env, fst_o, XATTR_NAME_LOV, handle);
1507 rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, handle);
1512 rc2 = mdd_xattr_hsm_replace(env, snd_o, snd_hsm_buf, handle);
1517 CERROR("%s: unable to roll back layout swap. FIDs: "
1518 DFID" and "DFID "error: %d/%d, steps: %d\n",
1519 mdd_obj_dev_name(fst_o),
1520 PFID(mdo2fid(snd_o)), PFID(mdo2fid(fst_o)),
1522 /* a solution to avoid journal commit is to panic,
1523 * but it has strong consequences so we use LBUG to
1524 * allow sysdamin to choose to panic or not
1531 /* Issue one changelog record per file */
1532 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle);
1536 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle);
1542 mdd_trans_stop(env, mdd, rc, handle);
1543 mdd_write_unlock(env, snd_o);
1544 mdd_write_unlock(env, fst_o);
1546 lu_buf_free(fst_buf);
1547 lu_buf_free(snd_buf);
1548 lu_buf_free(fst_hsm_buf);
1549 lu_buf_free(snd_hsm_buf);
1553 void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent,
1554 struct mdd_object *child, const struct lu_attr *attr,
1555 const struct md_op_spec *spec,
1556 struct dt_allocation_hint *hint)
1558 struct dt_object *np = parent ? mdd_object_child(parent) : NULL;
1559 struct dt_object *nc = mdd_object_child(child);
1561 memset(hint, 0, sizeof(*hint));
1563 /* For striped directory, give striping EA to lod_ah_init, which will
1564 * decide the stripe_offset and stripe count by it. */
1565 if (S_ISDIR(attr->la_mode) &&
1566 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1567 hint->dah_eadata = spec->u.sp_ea.eadata;
1568 hint->dah_eadata_len = spec->u.sp_ea.eadatalen;
1570 hint->dah_eadata = NULL;
1571 hint->dah_eadata_len = 0;
1574 CDEBUG(D_INFO, DFID" eadata %p len %d\n", PFID(mdd_object_fid(child)),
1575 hint->dah_eadata, hint->dah_eadata_len);
1576 /* @hint will be initialized by underlying device. */
1577 nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT);
1581 * do NOT or the MAY_*'s, you'll get the weakest
1583 int accmode(const struct lu_env *env, const struct lu_attr *la, int flags)
1587 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1588 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1589 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1590 * owner can write to a file even if it is marked readonly to hide
1591 * its brokenness. (bug 5781) */
1592 if (flags & MDS_OPEN_OWNEROVERRIDE) {
1593 struct lu_ucred *uc = lu_ucred_check(env);
1595 if ((uc == NULL) || (la->la_uid == uc->uc_fsuid))
1599 if (flags & FMODE_READ)
1601 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1603 if (flags & MDS_FMODE_EXEC)
1608 static int mdd_open_sanity_check(const struct lu_env *env,
1609 struct mdd_object *obj,
1610 const struct lu_attr *attr, int flag)
1616 if (mdd_is_dead_obj(obj))
1619 if (S_ISLNK(attr->la_mode))
1622 mode = accmode(env, attr, flag);
1624 if (S_ISDIR(attr->la_mode) && (mode & MAY_WRITE))
1627 if (!(flag & MDS_OPEN_CREATED)) {
1628 rc = mdd_permission_internal(env, obj, attr, mode);
1633 if (S_ISFIFO(attr->la_mode) || S_ISSOCK(attr->la_mode) ||
1634 S_ISBLK(attr->la_mode) || S_ISCHR(attr->la_mode))
1635 flag &= ~MDS_OPEN_TRUNC;
1637 /* For writing append-only file must open it with append mode. */
1638 if (attr->la_flags & LUSTRE_APPEND_FL) {
1639 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1641 if (flag & MDS_OPEN_TRUNC)
1647 * Now, flag -- O_NOATIME does not be packed by client.
1649 if (flag & O_NOATIME) {
1650 struct lu_ucred *uc = lu_ucred(env);
1652 if (uc && ((uc->uc_valid == UCRED_OLD) ||
1653 (uc->uc_valid == UCRED_NEW)) &&
1654 (uc->uc_fsuid != attr->la_uid) &&
1655 !md_capable(uc, CFS_CAP_FOWNER))
1663 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1666 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1667 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1670 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1672 rc = mdd_la_get(env, mdd_obj, attr);
1676 rc = mdd_open_sanity_check(env, mdd_obj, attr, flags);
1680 mdd_obj->mod_count++;
1683 mdd_write_unlock(env, mdd_obj);
1687 static int mdd_declare_close(const struct lu_env *env,
1688 struct mdd_object *obj,
1690 struct thandle *handle)
1694 rc = orph_declare_index_delete(env, obj, handle);
1698 return mdo_declare_destroy(env, obj, handle);
1702 * No permission check is needed.
1704 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1705 struct md_attr *ma, int mode)
1707 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1708 struct mdd_device *mdd = mdo2mdd(obj);
1709 struct thandle *handle = NULL;
1710 int rc, is_orphan = 0;
1713 if (ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_KEEP_ORPHAN) {
1714 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1715 mdd_obj->mod_count--;
1716 mdd_write_unlock(env, mdd_obj);
1718 if (mdd_obj->mod_flags & ORPHAN_OBJ && !mdd_obj->mod_count)
1719 CDEBUG(D_HA, "Object "DFID" is retained in orphan "
1720 "list\n", PFID(mdd_object_fid(mdd_obj)));
1724 /* mdd_finish_unlink() will always set orphan object as DEAD_OBJ, but
1725 * it might fail to add the object to orphan list (w/o ORPHAN_OBJ). */
1726 /* check without any lock */
1727 is_orphan = mdd_obj->mod_count == 1 &&
1728 (mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0;
1732 handle = mdd_trans_create(env, mdo2mdd(obj));
1734 RETURN(PTR_ERR(handle));
1736 rc = mdd_declare_close(env, mdd_obj, ma, handle);
1740 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
1744 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1749 mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
1750 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr);
1752 CERROR("Failed to get lu_attr of "DFID": %d\n",
1753 PFID(mdd_object_fid(mdd_obj)), rc);
1757 /* check again with lock */
1758 is_orphan = (mdd_obj->mod_count == 1) &&
1759 ((mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0 ||
1760 ma->ma_attr.la_nlink == 0);
1762 if (is_orphan && handle == NULL) {
1763 mdd_write_unlock(env, mdd_obj);
1767 mdd_obj->mod_count--; /*release open count */
1773 /* NB: Object maybe not in orphan list originally, it is rare case for
1774 * mdd_finish_unlink() failure, in that case, the object doesn't have
1775 * ORPHAN_OBJ flag */
1776 if ((mdd_obj->mod_flags & ORPHAN_OBJ) != 0) {
1777 /* remove link to object from orphan index */
1778 LASSERT(handle != NULL);
1779 rc = __mdd_orphan_del(env, mdd_obj, handle);
1781 CERROR("%s: unable to delete "DFID" from orphan list: "
1782 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1783 PFID(mdd_object_fid(mdd_obj)), rc);
1784 /* If object was not deleted from orphan list, do not
1785 * destroy OSS objects, which will be done when next
1790 CDEBUG(D_HA, "Object "DFID" is deleted from orphan "
1791 "list, OSS objects to be destroyed.\n",
1792 PFID(mdd_object_fid(mdd_obj)));
1795 rc = mdo_destroy(env, mdd_obj, handle);
1798 CERROR("%s: unable to delete "DFID" from orphan list: "
1799 "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
1800 PFID(mdd_object_fid(mdd_obj)), rc);
1805 mdd_write_unlock(env, mdd_obj);
1808 (mode & (FMODE_WRITE | MDS_OPEN_APPEND | MDS_OPEN_TRUNC)) &&
1809 !(ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_RECOV_OPEN)) {
1810 if (handle == NULL) {
1811 handle = mdd_trans_create(env, mdo2mdd(obj));
1813 GOTO(stop, rc = PTR_ERR(handle));
1815 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL,
1820 rc = mdd_trans_start(env, mdo2mdd(obj), handle);
1825 mdd_changelog_data_store(env, mdd, CL_CLOSE, mode,
1830 if (handle != NULL && !IS_ERR(handle))
1831 mdd_trans_stop(env, mdd, rc, handle);
1837 * Permission check is done when open,
1838 * no need check again.
1840 static int mdd_readpage_sanity_check(const struct lu_env *env,
1841 struct mdd_object *obj)
1843 struct dt_object *next = mdd_object_child(obj);
1847 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
1855 static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp,
1856 size_t nob, const struct dt_it_ops *iops,
1857 struct dt_it *it, __u32 attr, void *arg)
1859 struct lu_dirpage *dp = &lp->lp_dir;
1863 struct lu_dirent *ent;
1864 struct lu_dirent *last = NULL;
1868 if (nob < sizeof(*dp))
1871 memset(area, 0, sizeof (*dp));
1872 area += sizeof (*dp);
1873 nob -= sizeof (*dp);
1880 len = iops->key_size(env, it);
1882 /* IAM iterator can return record with zero len. */
1886 hash = iops->store(env, it);
1887 if (unlikely(first)) {
1889 dp->ldp_hash_start = cpu_to_le64(hash);
1892 /* calculate max space required for lu_dirent */
1893 recsize = lu_dirent_calc_size(len, attr);
1895 if (nob >= recsize) {
1896 result = iops->rec(env, it, (struct dt_rec *)ent, attr);
1897 if (result == -ESTALE)
1902 /* osd might not able to pack all attributes,
1903 * so recheck rec length */
1904 recsize = le16_to_cpu(ent->lde_reclen);
1906 if (le32_to_cpu(ent->lde_attrs) & LUDA_FID) {
1907 fid_le_to_cpu(&fid, &ent->lde_fid);
1908 if (fid_is_dot_lustre(&fid))
1912 result = (last != NULL) ? 0 :-EINVAL;
1916 ent = (void *)ent + recsize;
1920 result = iops->next(env, it);
1921 if (result == -ESTALE)
1923 } while (result == 0);
1926 dp->ldp_hash_end = cpu_to_le64(hash);
1928 if (last->lde_hash == dp->ldp_hash_end)
1929 dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE);
1930 last->lde_reclen = 0; /* end mark */
1933 /* end of directory */
1934 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1935 else if (result < 0)
1936 CWARN("build page failed: %d!\n", result);
1940 int mdd_readpage(const struct lu_env *env, struct md_object *obj,
1941 const struct lu_rdpg *rdpg)
1943 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1947 if (mdd_object_exists(mdd_obj) == 0) {
1948 CERROR("%s: object "DFID" not found: rc = -2\n",
1949 mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
1953 mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
1954 rc = mdd_readpage_sanity_check(env, mdd_obj);
1956 GOTO(out_unlock, rc);
1958 if (mdd_is_dead_obj(mdd_obj)) {
1960 struct lu_dirpage *dp;
1963 * According to POSIX, please do not return any entry to client:
1964 * even dot and dotdot should not be returned.
1966 CDEBUG(D_INODE, "readdir from dead object: "DFID"\n",
1967 PFID(mdd_object_fid(mdd_obj)));
1969 if (rdpg->rp_count <= 0)
1970 GOTO(out_unlock, rc = -EFAULT);
1971 LASSERT(rdpg->rp_pages != NULL);
1973 pg = rdpg->rp_pages[0];
1974 dp = (struct lu_dirpage *)kmap(pg);
1975 memset(dp, 0 , sizeof(struct lu_dirpage));
1976 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
1977 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1978 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
1980 GOTO(out_unlock, rc = LU_PAGE_SIZE);
1983 rc = dt_index_walk(env, mdd_object_child(mdd_obj), rdpg,
1984 mdd_dir_page_build, NULL);
1986 struct lu_dirpage *dp;
1988 dp = kmap(rdpg->rp_pages[0]);
1989 dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
1992 * No pages were processed, mark this for first page
1995 dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
1996 dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
1997 rc = min_t(unsigned int, LU_PAGE_SIZE, rdpg->rp_count);
1999 kunmap(rdpg->rp_pages[0]);
2002 GOTO(out_unlock, rc);
2004 mdd_read_unlock(env, mdd_obj);
2008 static int mdd_object_sync(const struct lu_env *env, struct md_object *obj)
2010 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2012 if (mdd_object_exists(mdd_obj) == 0) {
2015 CERROR("%s: object "DFID" not found: rc = %d\n",
2016 mdd_obj_dev_name(mdd_obj),
2017 PFID(mdd_object_fid(mdd_obj)), rc);
2020 return dt_object_sync(env, mdd_object_child(mdd_obj),
2024 static int mdd_object_lock(const struct lu_env *env,
2025 struct md_object *obj,
2026 struct lustre_handle *lh,
2027 struct ldlm_enqueue_info *einfo,
2028 ldlm_policy_data_t *policy)
2030 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2031 return dt_object_lock(env, mdd_object_child(mdd_obj), lh,
2035 static int mdd_object_unlock(const struct lu_env *env,
2036 struct md_object *obj,
2037 struct ldlm_enqueue_info *einfo,
2038 ldlm_policy_data_t *policy)
2040 struct mdd_object *mdd_obj = md2mdd_obj(obj);
2041 return dt_object_unlock(env, mdd_object_child(mdd_obj), einfo, policy);
2044 const struct md_object_operations mdd_obj_ops = {
2045 .moo_permission = mdd_permission,
2046 .moo_attr_get = mdd_attr_get,
2047 .moo_attr_set = mdd_attr_set,
2048 .moo_xattr_get = mdd_xattr_get,
2049 .moo_xattr_set = mdd_xattr_set,
2050 .moo_xattr_list = mdd_xattr_list,
2051 .moo_xattr_del = mdd_xattr_del,
2052 .moo_swap_layouts = mdd_swap_layouts,
2053 .moo_open = mdd_open,
2054 .moo_close = mdd_close,
2055 .moo_readpage = mdd_readpage,
2056 .moo_readlink = mdd_readlink,
2057 .moo_changelog = mdd_changelog,
2058 .moo_object_sync = mdd_object_sync,
2059 .moo_object_lock = mdd_object_lock,
2060 .moo_object_unlock = mdd_object_unlock,