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, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdd/mdd_lov.c
38 * Lustre Metadata Server (mds) handling of striped file data
40 * Author: Peter Braam <braam@clusterfs.com>
41 * Author: wangdi <wangdi@clusterfs.com>
44 #define DEBUG_SUBSYSTEM S_MDS
46 #include <linux/module.h>
48 #include <obd_class.h>
49 #include <lustre_ver.h>
50 #include <obd_support.h>
52 #include <lprocfs_status.h>
53 #include <lustre_mds.h>
54 #include <lustre_fid.h>
55 #include <lustre/lustre_idl.h>
57 #include "mdd_internal.h"
59 static int mdd_notify(struct obd_device *host, struct obd_device *watched,
60 enum obd_notify_event ev, void *owner, void *data)
62 struct mdd_device *mdd = owner;
66 LASSERT(owner != NULL);
69 case OBD_NOTIFY_ACTIVE:
71 case OBD_NOTIFY_SYNC_NONBLOCK:
72 rc = md_do_upcall(NULL, &mdd->mdd_md_dev,
75 case OBD_NOTIFY_CONFIG:
76 rc = md_do_upcall(NULL, &mdd->mdd_md_dev,
79 #ifdef HAVE_QUOTA_SUPPORT
80 case OBD_NOTIFY_QUOTA:
81 rc = md_do_upcall(NULL, &mdd->mdd_md_dev,
86 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
92 /* The obd is created for handling data stack for mdd */
93 int mdd_init_obd(const struct lu_env *env, struct mdd_device *mdd,
94 struct lustre_cfg *cfg)
96 char *dev = lustre_cfg_string(cfg, 0);
97 int rc, name_size, uuid_size;
100 struct lustre_cfg_bufs *bufs;
101 struct lustre_cfg *lcfg;
102 struct obd_device *obd;
105 mds_id = lu_site2md(mdd2lu_dev(mdd)->ld_site)->ms_node_id;
106 name_size = strlen(MDD_OBD_NAME) + 35;
107 uuid_size = strlen(MDD_OBD_UUID) + 35;
109 OBD_ALLOC(name, name_size);
110 OBD_ALLOC(uuid, uuid_size);
111 if (name == NULL || uuid == NULL)
112 GOTO(cleanup_mem, rc = -ENOMEM);
116 GOTO(cleanup_mem, rc = -ENOMEM);
118 snprintf(name, strlen(MDD_OBD_NAME) + 35, "%s-%s",
121 snprintf(uuid, strlen(MDD_OBD_UUID) + 35, "%s-%s",
124 lustre_cfg_bufs_reset(bufs, name);
125 lustre_cfg_bufs_set_string(bufs, 1, MDD_OBD_TYPE);
126 lustre_cfg_bufs_set_string(bufs, 2, uuid);
127 lustre_cfg_bufs_set_string(bufs, 3, (char*)dev/* MDD_OBD_PROFILE */);
128 lustre_cfg_bufs_set_string(bufs, 4, (char*)dev);
130 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
133 GOTO(cleanup_mem, rc = -ENOMEM);
135 rc = class_attach(lcfg);
137 GOTO(lcfg_cleanup, rc);
139 obd = class_name2obd(name);
141 CERROR("Can not find obd %s\n", MDD_OBD_NAME);
145 cfs_spin_lock(&obd->obd_dev_lock);
146 obd->obd_recovering = 1;
147 cfs_spin_unlock(&obd->obd_dev_lock);
148 obd->u.mds.mds_id = mds_id;
149 obd->u.obt.obt_osd_properties.osd_max_ea_size =
150 mdd->mdd_dt_conf.ddp_max_ea_size;
152 rc = class_setup(obd, lcfg);
154 GOTO(class_detach, rc);
157 * Add here for obd notify mechanism, when adding a new ost, the mds
158 * will notify this mdd. The mds will be used for quota also.
160 obd->obd_upcall.onu_upcall = mdd_notify;
161 obd->obd_upcall.onu_owner = mdd;
162 mdd->mdd_obd_dev = obd;
167 class_detach(obd, lcfg);
169 lustre_cfg_free(lcfg);
172 OBD_FREE(name, name_size);
174 OBD_FREE(uuid, uuid_size);
178 int mdd_fini_obd(const struct lu_env *env, struct mdd_device *mdd,
179 struct lustre_cfg *lcfg)
181 struct obd_device *obd;
185 obd = mdd2obd_dev(mdd);
188 rc = class_cleanup(obd, lcfg);
190 GOTO(lcfg_cleanup, rc);
192 obd->obd_upcall.onu_upcall = NULL;
193 obd->obd_upcall.onu_owner = NULL;
194 rc = class_detach(obd, lcfg);
196 GOTO(lcfg_cleanup, rc);
197 mdd->mdd_obd_dev = NULL;
204 int mdd_get_md(const struct lu_env *env, struct mdd_object *obj,
205 void *md, int *md_size, const char *name)
210 rc = mdo_xattr_get(env, obj, mdd_buf_get(env, md, *md_size), name,
211 mdd_object_capa(env, obj));
213 * XXX: Handling of -ENODATA, the right way is to have ->do_md_get()
214 * exported by dt layer.
216 if (rc == 0 || rc == -ENODATA) {
220 CDEBUG(D_OTHER, "Error %d reading eadata - %d\n",
223 /* XXX: Convert lov EA but fixed after verification test. */
230 int mdd_get_md_locked(const struct lu_env *env, struct mdd_object *obj,
231 void *md, int *md_size, const char *name)
234 mdd_read_lock(env, obj, MOR_TGT_CHILD);
235 rc = mdd_get_md(env, obj, md, md_size, name);
236 mdd_read_unlock(env, obj);
240 static int mdd_lov_set_stripe_md(const struct lu_env *env,
241 struct mdd_object *obj, struct lu_buf *buf,
242 struct thandle *handle)
244 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
245 struct obd_device *obd = mdd2obd_dev(mdd);
246 struct obd_export *lov_exp = obd->u.mds.mds_lov_exp;
247 struct lov_stripe_md *lsm = NULL;
251 LASSERT(S_ISDIR(mdd_object_type(obj)) || S_ISREG(mdd_object_type(obj)));
252 rc = obd_iocontrol(OBD_IOC_LOV_SETSTRIPE, lov_exp, 0,
256 obd_free_memmd(lov_exp, &lsm);
258 rc = mdd_xattr_set_txn(env, obj, buf, XATTR_NAME_LOV, 0, handle);
260 CDEBUG(D_INFO, "set lov ea of "DFID" rc %d \n", PFID(mdo2fid(obj)), rc);
265 * Permission check is done before call it,
266 * no need check again.
268 static int mdd_lov_set_dir_md(const struct lu_env *env,
269 struct mdd_object *obj, struct lu_buf *buf,
270 struct thandle *handle)
272 struct lov_user_md *lum = NULL;
276 LASSERT(S_ISDIR(mdd_object_type(obj)));
277 lum = (struct lov_user_md*)buf->lb_buf;
279 /* if { size, offset, count } = { 0, -1, 0 } and no pool
280 * (i.e. all default values specified) then delete default
281 * striping from dir. */
282 if (LOVEA_DELETE_VALUES(lum->lmm_stripe_size, lum->lmm_stripe_count,
283 lum->lmm_stripe_offset) &&
284 lum->lmm_magic != LOV_USER_MAGIC_V3) {
285 rc = mdd_xattr_set_txn(env, obj, &LU_BUF_NULL,
286 XATTR_NAME_LOV, 0, handle);
289 CDEBUG(D_INFO, "delete lov ea of "DFID" rc %d \n",
290 PFID(mdo2fid(obj)), rc);
292 rc = mdd_lov_set_stripe_md(env, obj, buf, handle);
297 int mdd_lsm_sanity_check(const struct lu_env *env, struct mdd_object *obj)
299 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
300 struct md_ucred *uc = md_ucred(env);
304 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
308 if ((uc->mu_fsuid != tmp_la->la_uid) &&
309 !mdd_capable(uc, CFS_CAP_FOWNER))
310 rc = mdd_permission_internal_locked(env, obj, tmp_la,
311 MAY_WRITE, MOR_TGT_CHILD);
316 int mdd_lov_set_md(const struct lu_env *env, struct mdd_object *pobj,
317 struct mdd_object *child, struct lov_mds_md *lmmp,
318 int lmm_size, struct thandle *handle, int set_stripe)
325 buf = mdd_buf_get(env, lmmp, lmm_size);
326 mode = mdd_object_type(child);
327 if (S_ISREG(mode) && lmm_size > 0) {
329 rc = mdd_lov_set_stripe_md(env, child, buf, handle);
331 rc = mdd_xattr_set_txn(env, child, buf,
332 XATTR_NAME_LOV, 0, handle);
334 } else if (S_ISDIR(mode)) {
335 if (lmmp == NULL && lmm_size == 0) {
336 struct mdd_device *mdd = mdd_obj2mdd_dev(child);
337 struct lov_mds_md *lmm = mdd_max_lmm_get(env, mdd);
338 int size = sizeof(struct lov_mds_md_v3);
340 /* Get parent dir stripe and set */
342 rc = mdd_get_md_locked(env, pobj, lmm, &size,
345 buf = mdd_buf_get(env, lmm, size);
346 rc = mdd_xattr_set_txn(env, child, buf,
350 CERROR("error on copy stripe info: rc "
354 LASSERT(lmmp != NULL && lmm_size > 0);
355 rc = mdd_lov_set_dir_md(env, child, buf, handle);
358 CDEBUG(D_INFO, "Set lov md %p size %d for fid "DFID" rc %d\n",
359 lmmp, lmm_size, PFID(mdo2fid(child)), rc);
363 int mdd_lov_objid_prepare(struct mdd_device *mdd, struct lov_mds_md *lmm)
365 /* copy mds_lov code is using wrong layer */
366 return mds_lov_prepare_objids(mdd->mdd_obd_dev, lmm);
369 int mdd_declare_lov_objid_update(const struct lu_env *env,
370 struct mdd_device *mdd,
371 struct thandle *handle)
373 struct obd_device *obd = mdd2obd_dev(mdd);
376 /* in prepare we create local files */
377 if (unlikely(mdd->mdd_capa == NULL))
380 /* XXX: this is a temporary solution to declare llog changes
381 * will be fixed in 2.3 with new llog implementation */
383 size = obd->u.mds.mds_lov_desc.ld_tgt_count * sizeof(obd_id);
384 return dt_declare_record_write(env, mdd->mdd_capa, size, 0, handle);
387 void mdd_lov_objid_update(struct mdd_device *mdd, struct lov_mds_md *lmm)
389 /* copy mds_lov code is using wrong layer */
390 mds_lov_update_objids(mdd->mdd_obd_dev, lmm);
393 void mdd_lov_create_finish(const struct lu_env *env, struct mdd_device *mdd,
394 struct lov_mds_md *lmm, int lmm_size,
395 const struct md_op_spec *spec)
397 if (lmm && !spec->no_create)
398 OBD_FREE_LARGE(lmm, lmm_size);
401 int mdd_lov_create(const struct lu_env *env, struct mdd_device *mdd,
402 struct mdd_object *parent, struct mdd_object *child,
403 struct lov_mds_md **lmm, int *lmm_size,
404 const struct md_op_spec *spec, struct md_attr *ma)
406 struct obd_device *obd = mdd2obd_dev(mdd);
407 struct obd_export *lov_exp = obd->u.mds.mds_lov_exp;
408 struct lu_site *site = mdd2lu_dev(mdd)->ld_site;
410 struct lov_stripe_md *lsm = NULL;
411 const void *eadata = spec->u.sp_ea.eadata;
412 __u64 create_flags = spec->sp_cr_flags;
413 struct obd_trans_info *oti = &mdd_env_info(env)->mti_oti;
414 struct lu_attr *la = &ma->ma_attr;
418 if (!md_should_create(create_flags)) {
424 /* replay case, has objects already, only get lov from eadata */
425 if (spec->no_create != 0) {
426 *lmm = (struct lov_mds_md *)spec->u.sp_ea.eadata;
427 *lmm_size = spec->u.sp_ea.eadatalen;
428 if (*lmm_size == lov_mds_md_size((*lmm)->lmm_stripe_count,
429 (*lmm)->lmm_magic)) {
432 CERROR("incorrect lsm received during recovery\n");
437 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
438 GOTO(out_ids, rc = -ENOMEM);
440 LASSERT(lov_exp != NULL);
441 oa = &mdd_env_info(env)->mti_oa;
443 oa->o_uid = 0; /* must have 0 uid / gid on OST */
445 oa->o_seq = mdt_to_obd_objseq(lu_site2md(site)->ms_node_id);
446 oa->o_mode = S_IFREG | 0600;
447 oa->o_id = fid_ver_oid(mdd_object_fid(child));
448 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLFLAGS |
449 OBD_MD_FLMODE | OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLGROUP;
452 if (!(create_flags & MDS_OPEN_HAS_OBJS)) {
453 if (create_flags & MDS_OPEN_HAS_EA) {
454 LASSERT(eadata != NULL);
455 rc = obd_iocontrol(OBD_IOC_LOV_SETSTRIPE, lov_exp,
456 0, &lsm, (void*)eadata);
460 /* get lov ea from parent and set to lov */
461 struct lov_mds_md *_lmm;
462 int _lmm_size = mdd_lov_mdsize(env, mdd);
464 LASSERT(parent != NULL);
466 _lmm = mdd_max_lmm_get(env, mdd);
468 GOTO(out_oti, rc = -ENOMEM);
470 rc = mdd_get_md_locked(env, parent, _lmm,
474 _lmm_size = mdd_lov_mdsize(env, mdd);
475 rc = obd_iocontrol(OBD_IOC_LOV_SETSTRIPE,
483 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_OPEN_WAIT_CREATE, 10);
484 rc = obd_create(env, lov_exp, oa, &lsm, oti);
487 CERROR("Create error for "DFID": %d\n",
488 PFID(mdo2fid(child)), rc);
494 if (ma->ma_valid & MA_LAY_GEN)
495 /* If we already have a lsm, the file is not new and we
496 * are about to change the layout, so we have to bump
497 * the generation. It is worth noting that old versions
498 * will be confused by a non-zero gen, that's why
499 * OBD_INCOMPAT_LMM_VER has been introduced */
500 lsm->lsm_layout_gen = ma->ma_layout_gen + 1;
502 /* Start with a null generation for backward
503 * compatiblity with old versions */
504 lsm->lsm_layout_gen = 0;
506 LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
508 LASSERT(eadata != NULL);
509 rc = obd_iocontrol(OBD_IOC_LOV_SETEA, lov_exp, 0, &lsm,
514 if (ma->ma_valid & MA_LAY_GEN)
515 lsm->lsm_layout_gen = ma->ma_layout_gen;
517 lsm->lsm_layout_gen = 0;
520 lsm->lsm_object_id = fid_ver_oid(mdd_object_fid(child));
521 lsm->lsm_object_seq = fid_seq(mdd_object_fid(child));
523 * Sometimes, we may truncate some object(without lsm) then open it
524 * (with write flags), so creating lsm above. The Nonzero(truncated)
525 * size should tell ost, since size attr is in charge by OST.
527 if (la->la_size && la->la_valid & LA_SIZE) {
528 struct obd_info *oinfo = &mdd_env_info(env)->mti_oi;
530 memset(oinfo, 0, sizeof(*oinfo));
532 /* When setting attr to ost, FLBKSZ is not needed. */
533 oa->o_valid &= ~OBD_MD_FLBLKSZ;
534 obdo_from_la(oa, la, LA_TYPE | LA_ATIME | LA_MTIME |
537 * XXX: Pack lustre id to OST, in OST, it will be packed by
538 * filter_fid, but can not see what is the usages. So just pack
539 * o_seq o_ver here, maybe fix it after this cycle.
541 obdo_set_parent_fid(oa, mdd_object_fid(child));
544 oinfo->oi_capa = NULL;
545 oinfo->oi_policy.l_extent.start = la->la_size;
546 oinfo->oi_policy.l_extent.end = OBD_OBJECT_EOF;
548 rc = obd_punch_rqset(lov_exp, oinfo, oti);
550 CERROR("Error setting attrs for "DFID": rc %d\n",
551 PFID(mdo2fid(child)), rc);
553 CERROR("obd_setattr for "DFID" rc %d\n",
554 PFID(mdo2fid(child)), rc);
560 /* blksize should be changed after create data object */
561 la->la_valid |= LA_BLKSIZE;
562 la->la_blksize = oa->o_blksize;
564 rc = obd_packmd(lov_exp, lmm, lsm);
566 CERROR("Cannot pack lsm, err = %d\n", rc);
569 if (mdd_lov_objid_prepare(mdd, *lmm) != 0) {
570 CERROR("Not have memory for update objid\n");
573 GOTO(out_oti, rc = -ENOMEM);
579 oti_free_cookies(oti);
582 obd_free_memmd(lov_exp, &lsm);
588 * used when destroying orphans and from mds_reint_unlink() when MDS wants to
589 * destroy objects on OSS.
591 int mdd_lovobj_unlink(const struct lu_env *env, struct mdd_device *mdd,
592 struct mdd_object *obj, struct lu_attr *la,
593 struct md_attr *ma, int log_unlink)
595 struct obd_device *obd = mdd2obd_dev(mdd);
596 struct obd_export *lov_exp = obd->u.mds.mds_lov_exp;
597 struct lov_stripe_md *lsm = NULL;
598 struct obd_trans_info *oti = &mdd_env_info(env)->mti_oti;
599 struct obdo *oa = &mdd_env_info(env)->mti_oa;
600 struct lu_site *site = mdd2lu_dev(mdd)->ld_site;
601 struct lov_mds_md *lmm = ma->ma_lmm;
602 int lmm_size = ma->ma_lmm_size;
603 struct llog_cookie *logcookies = ma->ma_cookie;
610 rc = obd_unpackmd(lov_exp, &lsm, lmm, lmm_size);
612 CERROR("Error unpack md %p\n", lmm);
615 LASSERT(rc >= sizeof(*lsm));
619 oa->o_id = lsm->lsm_object_id;
620 oa->o_seq = mdt_to_obd_objseq(lu_site2md(site)->ms_node_id);
621 oa->o_mode = la->la_mode & S_IFMT;
622 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
625 if (log_unlink && logcookies) {
626 oa->o_valid |= OBD_MD_FLCOOKIE;
627 oti->oti_logcookies = logcookies;
630 if (!(ma->ma_attr_flags & MDS_UNLINK_DESTROY))
631 oa->o_flags = OBD_FL_DELORPHAN;
633 CDEBUG(D_INFO, "destroying OSS object "LPU64":"LPU64"\n", oa->o_seq,
636 rc = obd_destroy(env, lov_exp, oa, lsm, oti, NULL, NULL);
638 obd_free_memmd(lov_exp, &lsm);
643 * called with obj locked.
645 int mdd_lov_destroy(const struct lu_env *env, struct mdd_device *mdd,
646 struct mdd_object *obj, struct lu_attr *la)
648 struct md_attr *ma = &mdd_env_info(env)->mti_ma;
652 LASSERT(mdd_write_locked(env, obj) != 0);
654 if (unlikely(!S_ISREG(mdd_object_type(obj))))
657 if (unlikely(la->la_nlink != 0)) {
658 CWARN("Attempt to destroy OSS object when nlink == %d\n",
663 ma->ma_lmm_size = mdd_lov_mdsize(env, mdd);
664 ma->ma_lmm = mdd_max_lmm_get(env, mdd);
665 ma->ma_cookie_size = mdd_lov_cookiesize(env, mdd);
666 ma->ma_cookie = mdd_max_cookie_get(env, mdd);
667 if (ma->ma_lmm == NULL || ma->ma_cookie == NULL)
668 RETURN(rc = -ENOMEM);
672 rc = mdd_get_md(env, obj, ma->ma_lmm, &ma->ma_lmm_size,
676 CWARN("Get lov ea failed for "DFID" rc = %d\n",
677 PFID(mdo2fid(obj)), rc);
683 ma->ma_valid = MA_LOV;
685 rc = mdd_unlink_log(env, mdd, obj, ma);
687 CWARN("mds unlink log for "DFID" failed: %d\n",
688 PFID(mdo2fid(obj)), rc);
692 if (ma->ma_valid & MA_COOKIE)
693 rc = mdd_lovobj_unlink(env, mdd, obj, la, ma, 1);
698 int mdd_declare_unlink_log(const struct lu_env *env, struct mdd_object *obj,
699 struct md_attr *ma, struct thandle *handle)
701 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
708 if (!S_ISREG(lu_object_attr(&obj->mod_obj.mo_lu)))
711 rc = mdd_lmm_get_locked(env, obj, ma);
712 if (rc || !(ma->ma_valid & MA_LOV))
716 if (le32_to_cpu(ma->ma_lmm->lmm_magic) != LOV_MAGIC_V1 &&
717 le32_to_cpu(ma->ma_lmm->lmm_magic) != LOV_MAGIC_V3) {
718 CERROR("%s: invalid LOV_MAGIC %08x on object "DFID"\n",
719 mdd->mdd_obd_dev->obd_name,
720 le32_to_cpu(ma->ma_lmm->lmm_magic),
721 PFID(lu_object_fid(&obj->mod_obj.mo_lu)));
725 stripe = le16_to_cpu(ma->ma_lmm->lmm_stripe_count);
726 if (stripe == LOV_ALL_STRIPES);
727 stripe = mdd2obd_dev(mdd)->u.mds.mds_lov_desc.ld_tgt_count;
729 for (i = 0; i < stripe; i++) {
730 rc = mdd_declare_llog_record(env, mdd,
731 sizeof(struct llog_unlink_rec),
740 int mdd_unlink_log(const struct lu_env *env, struct mdd_device *mdd,
741 struct mdd_object *mdd_cobj, struct md_attr *ma)
743 LASSERT(ma->ma_valid & MA_LOV);
745 if ((ma->ma_cookie_size > 0) &&
746 (mds_log_op_unlink(mdd2obd_dev(mdd), ma->ma_lmm, ma->ma_lmm_size,
747 ma->ma_cookie, ma->ma_cookie_size) > 0)) {
748 CDEBUG(D_HA, "DEBUG: unlink log is added for object "DFID"\n",
749 PFID(mdd_object_fid(mdd_cobj)));
750 ma->ma_valid |= MA_COOKIE;
755 int mdd_log_op_setattr(struct obd_device *obd, __u32 uid, __u32 gid,
756 struct lov_mds_md *lmm, int lmm_size,
757 struct llog_cookie *logcookies, int cookies_size)
759 struct mds_obd *mds = &obd->u.mds;
760 struct lov_stripe_md *lsm = NULL;
761 struct llog_setattr64_rec *lsr;
762 struct llog_ctxt *ctxt;
766 if (IS_ERR(mds->mds_lov_obd))
767 RETURN(PTR_ERR(mds->mds_lov_obd));
769 rc = obd_unpackmd(mds->mds_lov_exp, &lsm, lmm, lmm_size);
773 OBD_ALLOC(lsr, sizeof(*lsr));
775 GOTO(out, rc = -ENOMEM);
777 /* prepare setattr log record */
778 lsr->lsr_hdr.lrh_len = lsr->lsr_tail.lrt_len = sizeof(*lsr);
779 lsr->lsr_hdr.lrh_type = MDS_SETATTR64_REC;
783 /* write setattr log */
784 ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
785 rc = llog_add(ctxt, &lsr->lsr_hdr, lsm, logcookies,
786 cookies_size / sizeof(struct llog_cookie));
790 OBD_FREE(lsr, sizeof(*lsr));
792 obd_free_memmd(mds->mds_lov_exp, &lsm);
796 int mdd_setattr_log(const struct lu_env *env, struct mdd_device *mdd,
797 const struct md_attr *ma,
798 struct lov_mds_md *lmm, int lmm_size,
799 struct llog_cookie *logcookies, int cookies_size)
801 struct obd_device *obd = mdd2obd_dev(mdd);
803 /* journal chown/chgrp in llog, just like unlink */
805 CDEBUG(D_INFO, "setattr llog for uid/gid=%lu/%lu\n",
806 (unsigned long)ma->ma_attr.la_uid,
807 (unsigned long)ma->ma_attr.la_gid);
808 return mdd_log_op_setattr(obd, ma->ma_attr.la_uid,
809 ma->ma_attr.la_gid, lmm,
810 lmm_size, logcookies,
816 static int mdd_osc_setattr_async(struct obd_device *obd, __u32 uid, __u32 gid,
817 struct lov_mds_md *lmm, int lmm_size,
818 struct llog_cookie *logcookies, const struct lu_fid *parent,
821 struct mds_obd *mds = &obd->u.mds;
822 struct obd_trans_info oti = { 0 };
823 struct obd_info oinfo = { { { 0 } } };
827 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OST_SETATTR))
830 /* first get memory EA */
831 OBDO_ALLOC(oinfo.oi_oa);
837 rc = obd_unpackmd(mds->mds_lov_exp, &oinfo.oi_md, lmm, lmm_size);
839 CERROR("Error unpack md %p for obj "DFID"\n", lmm,
845 oinfo.oi_oa->o_uid = uid;
846 oinfo.oi_oa->o_gid = gid;
847 oinfo.oi_oa->o_id = oinfo.oi_md->lsm_object_id;
848 oinfo.oi_oa->o_seq = oinfo.oi_md->lsm_object_seq;
849 oinfo.oi_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP |
850 OBD_MD_FLUID | OBD_MD_FLGID;
852 oinfo.oi_oa->o_valid |= OBD_MD_FLCOOKIE;
853 oti.oti_logcookies = logcookies;
856 obdo_set_parent_fid(oinfo.oi_oa, parent);
859 /* do async setattr from mds to ost not waiting for responses. */
860 rc = obd_setattr_async(mds->mds_lov_exp, &oinfo, &oti, NULL);
862 CDEBUG(D_INODE, "mds to ost setattr objid 0x"LPX64
863 " on ost error %d\n", oinfo.oi_md->lsm_object_id, rc);
866 obd_free_memmd(mds->mds_lov_exp, &oinfo.oi_md);
867 OBDO_FREE(oinfo.oi_oa);
871 int mdd_lov_setattr_async(const struct lu_env *env, struct mdd_object *obj,
872 struct lov_mds_md *lmm, int lmm_size,
873 struct llog_cookie *logcookies)
875 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
876 struct obd_device *obd = mdd2obd_dev(mdd);
877 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
878 const struct lu_fid *fid = mdd_object_fid(obj);
882 mdd_read_lock(env, obj, MOR_TGT_CHILD);
883 rc = mdo_attr_get(env, obj, tmp_la, mdd_object_capa(env, obj));
884 mdd_read_unlock(env, obj);
888 rc = mdd_osc_setattr_async(obd, tmp_la->la_uid, tmp_la->la_gid, lmm,
889 lmm_size, logcookies, fid, NULL);
893 static int grouplock_blocking_ast(struct ldlm_lock *lock,
894 struct ldlm_lock_desc *desc,
895 void *data, int flag)
897 struct md_attr *ma = data;
898 struct lustre_handle lockh;
904 case LDLM_CB_BLOCKING :
905 /* lock is canceled */
906 CDEBUG(D_DLMTRACE, "Lock %p is canceled\n", lock);
908 ldlm_lock2handle(lock, &lockh);
909 rc = ldlm_cli_cancel(&lockh);
912 case LDLM_CB_CANCELING :
914 "Lock %p has been canceled, do cleaning\n",
917 if (ma && ma->ma_som)
918 OBD_FREE_PTR(ma->ma_som);
928 static int grouplock_glimpse_ast(struct ldlm_lock *lock, void *data)
930 struct ptlrpc_request *req = data;
936 ma = lock->l_ast_data;
938 req_capsule_extend(&req->rq_pill, &RQF_LDLM_GL_CALLBACK);
939 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
941 rc = req_capsule_server_pack(&req->rq_pill);
943 CERROR("failed pack reply: %d\n", rc);
947 lvb = req_capsule_server_get(&req->rq_pill, &RMF_DLM_LVB);
949 if ((ma) && (ma->ma_valid & MA_SOM)) {
950 lvb->lvb_size = ma->ma_som->msd_size;
951 lvb->lvb_blocks = ma->ma_som->msd_blocks;
952 } else if ((ma) && (ma->ma_valid & MA_INODE)) {
953 lvb->lvb_size = ma->ma_attr.la_size;
954 lvb->lvb_blocks = ma->ma_attr.la_blocks;
957 rc = -ELDLM_NO_LOCK_DATA;
962 if (rc == -ELDLM_NO_LOCK_DATA)
963 lustre_pack_reply(req, 1, NULL, NULL);
969 int mdd_file_lock(const struct lu_env *env, struct md_object *obj,
970 struct lov_mds_md *lmm, struct ldlm_extent *extent,
971 struct lustre_handle *lockh)
973 struct ldlm_enqueue_info einfo = { 0 };
974 struct obd_info oinfo = { { { 0 } } };
975 struct obd_device *obd;
976 struct obd_export *lov_exp;
977 struct lov_stripe_md *lsm = NULL;
978 struct md_attr *ma = NULL;
982 obd = mdo2mdd(obj)->mdd_obd_dev;
983 lov_exp = obd->u.mds.mds_lov_exp;
985 obd_unpackmd(lov_exp, &lsm, lmm,
986 lov_mds_md_size(lmm->lmm_stripe_count, lmm->lmm_magic));
990 GOTO(out, rc = -ENOMEM);
992 OBD_ALLOC_PTR(ma->ma_som);
993 if (ma->ma_som == NULL)
994 GOTO(out, rc = -ENOMEM);
996 ma->ma_need = MA_SOM | MA_INODE;
997 mo_attr_get(env, obj, ma);
999 einfo.ei_type = LDLM_EXTENT;
1000 einfo.ei_mode = LCK_GROUP;
1001 einfo.ei_cb_bl = grouplock_blocking_ast;
1002 einfo.ei_cb_cp = ldlm_completion_ast;
1003 einfo.ei_cb_gl = grouplock_glimpse_ast;
1005 if (ma->ma_valid & (MA_SOM | MA_INODE))
1006 einfo.ei_cbdata = ma;
1008 einfo.ei_cbdata = NULL;
1010 memset(&oinfo.oi_policy, 0, sizeof(oinfo.oi_policy));
1011 oinfo.oi_policy.l_extent = *extent;
1012 oinfo.oi_lockh = lockh;
1016 rc = obd_enqueue(lov_exp, &oinfo, &einfo, NULL);
1017 /* ei_cbdata is used as a free flag at exit */
1019 einfo.ei_cbdata = NULL;
1021 obd_unpackmd(lov_exp, &lsm, NULL, 0);
1024 /* ma is freed if not used as callback data */
1025 if ((einfo.ei_cbdata == NULL) && ma && ma->ma_som)
1026 OBD_FREE_PTR(ma->ma_som);
1027 if ((einfo.ei_cbdata == NULL) && ma)
1033 int mdd_file_unlock(const struct lu_env *env, struct md_object *obj,
1034 struct lov_mds_md *lmm, struct lustre_handle *lockh)
1036 struct obd_device *obd;
1037 struct obd_export *lov_exp;
1038 struct lov_stripe_md *lsm = NULL;
1042 LASSERT(lustre_handle_is_used(lockh));
1044 obd = mdo2mdd(obj)->mdd_obd_dev;
1045 lov_exp = obd->u.mds.mds_lov_exp;
1047 obd_unpackmd(lov_exp, &lsm, lmm,
1048 lov_mds_md_size(lmm->lmm_stripe_count, lmm->lmm_magic));
1050 rc = obd_cancel(lov_exp, lsm, LCK_GROUP, lockh);
1052 obd_unpackmd(lov_exp, &lsm, NULL, 0);
1057 /* file lov is in ma->ma_lmm */
1058 /* requested lov is in info->mti_spec.u.sp_ea.eadata */
1059 int mdd_lum_lmm_cmp(const struct lu_env *env, struct md_object *cobj,
1060 const struct md_op_spec *spec, struct md_attr *ma)
1062 struct obd_export *lov_exp =
1063 mdd2obd_dev(mdo2mdd(cobj))->u.mds.mds_lov_exp;
1064 struct lov_mds_md *lmm = ma->ma_lmm;
1065 struct lov_user_md_v3 *lum =
1066 (struct lov_user_md_v3 *)(spec->u.sp_ea.eadata);
1067 struct lov_stripe_md *lsm = NULL;
1071 rc = obd_unpackmd(lov_exp, &lsm, lmm,
1072 lov_mds_md_size(lmm->lmm_stripe_count,
1074 ma->ma_layout_gen = lsm->lsm_layout_gen;
1075 ma->ma_valid |= MA_LAY_GEN;
1077 rc = lov_lum_swab_if_needed(lum, &lmm_magic, NULL);
1081 rc = lov_lum_lsm_cmp((struct lov_user_md *)lum, lsm);
1083 GOTO(out, rc); /* keep GOTO to for traces */
1087 obd_unpackmd(lov_exp, &lsm, NULL, 0);