1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/mdd/mdd_lov.c
40 * Lustre Metadata Server (mds) handling of striped file data
42 * Author: Peter Braam <braam@clusterfs.com>
43 * Author: wangdi <wangdi@clusterfs.com>
47 # define EXPORT_SYMTAB
49 #define DEBUG_SUBSYSTEM S_MDS
51 #include <linux/module.h>
53 #include <obd_class.h>
54 #include <lustre_ver.h>
55 #include <obd_support.h>
57 #include <lprocfs_status.h>
58 #include <lustre_mds.h>
59 #include <lustre_fid.h>
60 #include <lustre/lustre_idl.h>
62 #include "mdd_internal.h"
64 static int mdd_notify(struct obd_device *host, struct obd_device *watched,
65 enum obd_notify_event ev, void *owner, void *data)
67 struct mdd_device *mdd = owner;
71 LASSERT(owner != NULL);
74 case OBD_NOTIFY_ACTIVE:
76 case OBD_NOTIFY_SYNC_NONBLOCK:
77 rc = md_do_upcall(NULL, &mdd->mdd_md_dev,
80 case OBD_NOTIFY_CONFIG:
81 rc = md_do_upcall(NULL, &mdd->mdd_md_dev,
84 #ifdef HAVE_QUOTA_SUPPORT
85 case OBD_NOTIFY_QUOTA:
86 rc = md_do_upcall(NULL, &mdd->mdd_md_dev,
91 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
97 /* The obd is created for handling data stack for mdd */
98 int mdd_init_obd(const struct lu_env *env, struct mdd_device *mdd,
99 struct lustre_cfg *cfg)
101 char *dev = lustre_cfg_string(cfg, 0);
102 int rc, name_size, uuid_size;
105 struct lustre_cfg_bufs *bufs;
106 struct lustre_cfg *lcfg;
107 struct obd_device *obd;
110 mds_id = lu_site2md(mdd2lu_dev(mdd)->ld_site)->ms_node_id;
111 name_size = strlen(MDD_OBD_NAME) + 35;
112 uuid_size = strlen(MDD_OBD_UUID) + 35;
114 OBD_ALLOC(name, name_size);
115 OBD_ALLOC(uuid, uuid_size);
116 if (name == NULL || uuid == NULL)
117 GOTO(cleanup_mem, rc = -ENOMEM);
121 GOTO(cleanup_mem, rc = -ENOMEM);
123 snprintf(name, strlen(MDD_OBD_NAME) + 35, "%s-%s",
126 snprintf(uuid, strlen(MDD_OBD_UUID) + 35, "%s-%s",
129 lustre_cfg_bufs_reset(bufs, name);
130 lustre_cfg_bufs_set_string(bufs, 1, MDD_OBD_TYPE);
131 lustre_cfg_bufs_set_string(bufs, 2, uuid);
132 lustre_cfg_bufs_set_string(bufs, 3, (char*)dev/* MDD_OBD_PROFILE */);
133 lustre_cfg_bufs_set_string(bufs, 4, (char*)dev);
135 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
138 GOTO(cleanup_mem, rc = -ENOMEM);
140 rc = class_attach(lcfg);
142 GOTO(lcfg_cleanup, rc);
144 obd = class_name2obd(name);
146 CERROR("Can not find obd %s\n", MDD_OBD_NAME);
150 cfs_spin_lock(&obd->obd_dev_lock);
151 obd->obd_recovering = 1;
152 cfs_spin_unlock(&obd->obd_dev_lock);
153 obd->u.mds.mds_id = mds_id;
154 obd->u.obt.obt_osd_properties.osd_max_ea_size =
155 mdd->mdd_dt_conf.ddp_max_ea_size;
157 rc = class_setup(obd, lcfg);
159 GOTO(class_detach, rc);
162 * Add here for obd notify mechanism, when adding a new ost, the mds
163 * will notify this mdd. The mds will be used for quota also.
165 obd->obd_upcall.onu_upcall = mdd_notify;
166 obd->obd_upcall.onu_owner = mdd;
167 mdd->mdd_obd_dev = obd;
172 class_detach(obd, lcfg);
174 lustre_cfg_free(lcfg);
177 OBD_FREE(name, name_size);
179 OBD_FREE(uuid, uuid_size);
183 int mdd_fini_obd(const struct lu_env *env, struct mdd_device *mdd,
184 struct lustre_cfg *lcfg)
186 struct obd_device *obd;
190 obd = mdd2obd_dev(mdd);
193 rc = class_cleanup(obd, lcfg);
195 GOTO(lcfg_cleanup, rc);
197 obd->obd_upcall.onu_upcall = NULL;
198 obd->obd_upcall.onu_owner = NULL;
199 rc = class_detach(obd, lcfg);
201 GOTO(lcfg_cleanup, rc);
202 mdd->mdd_obd_dev = NULL;
209 int mdd_get_md(const struct lu_env *env, struct mdd_object *obj,
210 void *md, int *md_size, const char *name)
215 rc = mdo_xattr_get(env, obj, mdd_buf_get(env, md, *md_size), name,
216 mdd_object_capa(env, obj));
218 * XXX: Handling of -ENODATA, the right way is to have ->do_md_get()
219 * exported by dt layer.
221 if (rc == 0 || rc == -ENODATA) {
225 CDEBUG(D_OTHER, "Error %d reading eadata - %d\n",
228 /* XXX: Convert lov EA but fixed after verification test. */
235 int mdd_get_md_locked(const struct lu_env *env, struct mdd_object *obj,
236 void *md, int *md_size, const char *name)
239 mdd_read_lock(env, obj, MOR_TGT_CHILD);
240 rc = mdd_get_md(env, obj, md, md_size, name);
241 mdd_read_unlock(env, obj);
245 static int mdd_lov_set_stripe_md(const struct lu_env *env,
246 struct mdd_object *obj, struct lu_buf *buf,
247 struct thandle *handle)
249 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
250 struct obd_device *obd = mdd2obd_dev(mdd);
251 struct obd_export *lov_exp = obd->u.mds.mds_lov_exp;
252 struct lov_stripe_md *lsm = NULL;
256 LASSERT(S_ISDIR(mdd_object_type(obj)) || S_ISREG(mdd_object_type(obj)));
257 rc = obd_iocontrol(OBD_IOC_LOV_SETSTRIPE, lov_exp, 0,
261 obd_free_memmd(lov_exp, &lsm);
263 rc = mdd_xattr_set_txn(env, obj, buf, XATTR_NAME_LOV, 0, handle);
265 CDEBUG(D_INFO, "set lov ea of "DFID" rc %d \n", PFID(mdo2fid(obj)), rc);
270 * Permission check is done before call it,
271 * no need check again.
273 static int mdd_lov_set_dir_md(const struct lu_env *env,
274 struct mdd_object *obj, struct lu_buf *buf,
275 struct thandle *handle)
277 struct lov_user_md *lum = NULL;
281 LASSERT(S_ISDIR(mdd_object_type(obj)));
282 lum = (struct lov_user_md*)buf->lb_buf;
284 /* if { size, offset, count } = { 0, -1, 0 } and no pool
285 * (i.e. all default values specified) then delete default
286 * striping from dir. */
287 if (LOVEA_DELETE_VALUES(lum->lmm_stripe_size, lum->lmm_stripe_count,
288 lum->lmm_stripe_offset) &&
289 lum->lmm_magic != LOV_USER_MAGIC_V3) {
290 rc = mdd_xattr_set_txn(env, obj, &LU_BUF_NULL,
291 XATTR_NAME_LOV, 0, handle);
294 CDEBUG(D_INFO, "delete lov ea of "DFID" rc %d \n",
295 PFID(mdo2fid(obj)), rc);
297 rc = mdd_lov_set_stripe_md(env, obj, buf, handle);
302 int mdd_lsm_sanity_check(const struct lu_env *env, struct mdd_object *obj)
304 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
305 struct md_ucred *uc = md_ucred(env);
309 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
313 if ((uc->mu_fsuid != tmp_la->la_uid) &&
314 !mdd_capable(uc, CFS_CAP_FOWNER))
315 rc = mdd_permission_internal_locked(env, obj, tmp_la,
316 MAY_WRITE, MOR_TGT_CHILD);
321 int mdd_lov_set_md(const struct lu_env *env, struct mdd_object *pobj,
322 struct mdd_object *child, struct lov_mds_md *lmmp,
323 int lmm_size, struct thandle *handle, int set_stripe)
330 buf = mdd_buf_get(env, lmmp, lmm_size);
331 mode = mdd_object_type(child);
332 if (S_ISREG(mode) && lmm_size > 0) {
334 rc = mdd_lov_set_stripe_md(env, child, buf, handle);
336 rc = mdd_xattr_set_txn(env, child, buf,
337 XATTR_NAME_LOV, 0, handle);
339 } else if (S_ISDIR(mode)) {
340 if (lmmp == NULL && lmm_size == 0) {
341 struct mdd_device *mdd = mdd_obj2mdd_dev(child);
342 struct lov_mds_md *lmm = mdd_max_lmm_get(env, mdd);
343 int size = sizeof(struct lov_mds_md_v3);
345 /* Get parent dir stripe and set */
347 rc = mdd_get_md_locked(env, pobj, lmm, &size,
350 buf = mdd_buf_get(env, lmm, size);
351 rc = mdd_xattr_set_txn(env, child, buf,
355 CERROR("error on copy stripe info: rc "
359 LASSERT(lmmp != NULL && lmm_size > 0);
360 rc = mdd_lov_set_dir_md(env, child, buf, handle);
363 CDEBUG(D_INFO, "Set lov md %p size %d for fid "DFID" rc %d\n",
364 lmmp, lmm_size, PFID(mdo2fid(child)), rc);
368 int mdd_lov_objid_prepare(struct mdd_device *mdd, struct lov_mds_md *lmm)
370 /* copy mds_lov code is using wrong layer */
371 return mds_lov_prepare_objids(mdd->mdd_obd_dev, lmm);
374 int mdd_declare_lov_objid_update(const struct lu_env *env,
375 struct mdd_device *mdd,
376 struct thandle *handle)
378 struct obd_device *obd = mdd2obd_dev(mdd);
381 /* in prepare we create local files */
382 if (unlikely(mdd->mdd_capa == NULL))
385 /* XXX: this is a temporary solution to declare llog changes
386 * will be fixed in 2.3 with new llog implementation */
388 size = obd->u.mds.mds_lov_desc.ld_tgt_count * sizeof(obd_id);
389 return dt_declare_record_write(env, mdd->mdd_capa, size, 0, handle);
392 void mdd_lov_objid_update(struct mdd_device *mdd, struct lov_mds_md *lmm)
394 /* copy mds_lov code is using wrong layer */
395 mds_lov_update_objids(mdd->mdd_obd_dev, lmm);
398 void mdd_lov_create_finish(const struct lu_env *env, struct mdd_device *mdd,
399 struct lov_mds_md *lmm, int lmm_size,
400 const struct md_op_spec *spec)
402 if (lmm && !spec->no_create)
403 OBD_FREE_LARGE(lmm, lmm_size);
406 int mdd_lov_create(const struct lu_env *env, struct mdd_device *mdd,
407 struct mdd_object *parent, struct mdd_object *child,
408 struct lov_mds_md **lmm, int *lmm_size,
409 const struct md_op_spec *spec, struct md_attr *ma)
411 struct obd_device *obd = mdd2obd_dev(mdd);
412 struct obd_export *lov_exp = obd->u.mds.mds_lov_exp;
413 struct lu_site *site = mdd2lu_dev(mdd)->ld_site;
415 struct lov_stripe_md *lsm = NULL;
416 const void *eadata = spec->u.sp_ea.eadata;
417 __u64 create_flags = spec->sp_cr_flags;
418 struct obd_trans_info *oti = &mdd_env_info(env)->mti_oti;
419 struct lu_attr *la = &ma->ma_attr;
423 if (!md_should_create(create_flags)) {
429 /* replay case, has objects already, only get lov from eadata */
430 if (spec->no_create != 0) {
431 *lmm = (struct lov_mds_md *)spec->u.sp_ea.eadata;
432 *lmm_size = spec->u.sp_ea.eadatalen;
433 if (*lmm_size == lov_mds_md_size((*lmm)->lmm_stripe_count,
434 (*lmm)->lmm_magic)) {
437 CERROR("incorrect lsm received during recovery\n");
442 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
443 GOTO(out_ids, rc = -ENOMEM);
445 LASSERT(lov_exp != NULL);
446 oa = &mdd_env_info(env)->mti_oa;
448 oa->o_uid = 0; /* must have 0 uid / gid on OST */
450 oa->o_seq = mdt_to_obd_objseq(lu_site2md(site)->ms_node_id);
451 oa->o_mode = S_IFREG | 0600;
452 oa->o_id = fid_ver_oid(mdd_object_fid(child));
453 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLFLAGS |
454 OBD_MD_FLMODE | OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLGROUP;
457 if (!(create_flags & MDS_OPEN_HAS_OBJS)) {
458 if (create_flags & MDS_OPEN_HAS_EA) {
459 LASSERT(eadata != NULL);
460 rc = obd_iocontrol(OBD_IOC_LOV_SETSTRIPE, lov_exp,
461 0, &lsm, (void*)eadata);
465 /* get lov ea from parent and set to lov */
466 struct lov_mds_md *_lmm;
467 int _lmm_size = mdd_lov_mdsize(env, mdd);
469 LASSERT(parent != NULL);
471 _lmm = mdd_max_lmm_get(env, mdd);
473 GOTO(out_oti, rc = -ENOMEM);
475 rc = mdd_get_md_locked(env, parent, _lmm,
479 _lmm_size = mdd_lov_mdsize(env, mdd);
480 rc = obd_iocontrol(OBD_IOC_LOV_SETSTRIPE,
488 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_OPEN_WAIT_CREATE, 10);
489 rc = obd_create(lov_exp, oa, &lsm, oti);
492 CERROR("Create error for "DFID": %d\n",
493 PFID(mdo2fid(child)), rc);
499 if (ma->ma_valid & MA_LAY_GEN)
500 /* If we already have a lsm, the file is not new and we
501 * are about to change the layout, so we have to bump
502 * the generation. It is worth noting that old versions
503 * will be confused by a non-zero gen, that's why
504 * OBD_INCOMPAT_LMM_VER has been introduced */
505 lsm->lsm_layout_gen = ma->ma_layout_gen + 1;
507 /* Start with a null generation for backward
508 * compatiblity with old versions */
509 lsm->lsm_layout_gen = 0;
511 LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
513 LASSERT(eadata != NULL);
514 rc = obd_iocontrol(OBD_IOC_LOV_SETEA, lov_exp, 0, &lsm,
519 if (ma->ma_valid & MA_LAY_GEN)
520 lsm->lsm_layout_gen = ma->ma_layout_gen;
522 lsm->lsm_layout_gen = 0;
525 lsm->lsm_object_id = fid_ver_oid(mdd_object_fid(child));
526 lsm->lsm_object_seq = fid_seq(mdd_object_fid(child));
528 * Sometimes, we may truncate some object(without lsm) then open it
529 * (with write flags), so creating lsm above. The Nonzero(truncated)
530 * size should tell ost, since size attr is in charge by OST.
532 if (la->la_size && la->la_valid & LA_SIZE) {
533 struct obd_info *oinfo = &mdd_env_info(env)->mti_oi;
535 memset(oinfo, 0, sizeof(*oinfo));
537 /* When setting attr to ost, FLBKSZ is not needed. */
538 oa->o_valid &= ~OBD_MD_FLBLKSZ;
539 obdo_from_la(oa, la, LA_TYPE | LA_ATIME | LA_MTIME |
542 * XXX: Pack lustre id to OST, in OST, it will be packed by
543 * filter_fid, but can not see what is the usages. So just pack
544 * o_seq o_ver here, maybe fix it after this cycle.
546 obdo_set_parent_fid(oa, mdd_object_fid(child));
549 oinfo->oi_capa = NULL;
550 oinfo->oi_policy.l_extent.start = la->la_size;
551 oinfo->oi_policy.l_extent.end = OBD_OBJECT_EOF;
553 rc = obd_punch_rqset(lov_exp, oinfo, oti);
555 CERROR("Error setting attrs for "DFID": rc %d\n",
556 PFID(mdo2fid(child)), rc);
558 CERROR("obd_setattr for "DFID" rc %d\n",
559 PFID(mdo2fid(child)), rc);
565 /* blksize should be changed after create data object */
566 la->la_valid |= LA_BLKSIZE;
567 la->la_blksize = oa->o_blksize;
569 rc = obd_packmd(lov_exp, lmm, lsm);
571 CERROR("Cannot pack lsm, err = %d\n", rc);
574 if (mdd_lov_objid_prepare(mdd, *lmm) != 0) {
575 CERROR("Not have memory for update objid\n");
578 GOTO(out_oti, rc = -ENOMEM);
584 oti_free_cookies(oti);
587 obd_free_memmd(lov_exp, &lsm);
593 * used when destroying orphans and from mds_reint_unlink() when MDS wants to
594 * destroy objects on OSS.
596 int mdd_lovobj_unlink(const struct lu_env *env, struct mdd_device *mdd,
597 struct mdd_object *obj, struct lu_attr *la,
598 struct lov_mds_md *lmm, int lmm_size,
599 struct llog_cookie *logcookies,
602 struct obd_device *obd = mdd2obd_dev(mdd);
603 struct obd_export *lov_exp = obd->u.mds.mds_lov_exp;
604 struct lov_stripe_md *lsm = NULL;
605 struct obd_trans_info *oti = &mdd_env_info(env)->mti_oti;
606 struct obdo *oa = &mdd_env_info(env)->mti_oa;
607 struct lu_site *site = mdd2lu_dev(mdd)->ld_site;
614 rc = obd_unpackmd(lov_exp, &lsm, lmm, lmm_size);
616 CERROR("Error unpack md %p\n", lmm);
619 LASSERT(rc >= sizeof(*lsm));
623 oa->o_id = lsm->lsm_object_id;
624 oa->o_seq = mdt_to_obd_objseq(lu_site2md(site)->ms_node_id);
625 oa->o_mode = la->la_mode & S_IFMT;
626 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
629 if (log_unlink && logcookies) {
630 oa->o_valid |= OBD_MD_FLCOOKIE;
631 oti->oti_logcookies = logcookies;
634 CDEBUG(D_INFO, "destroying OSS object "LPU64":"LPU64"\n", oa->o_seq,
637 rc = obd_destroy(lov_exp, oa, lsm, oti, NULL, NULL);
639 obd_free_memmd(lov_exp, &lsm);
644 * called with obj locked.
646 int mdd_lov_destroy(const struct lu_env *env, struct mdd_device *mdd,
647 struct mdd_object *obj, struct lu_attr *la)
649 struct md_attr *ma = &mdd_env_info(env)->mti_ma;
653 LASSERT(mdd_write_locked(env, obj) != 0);
655 if (unlikely(!S_ISREG(mdd_object_type(obj))))
658 if (unlikely(la->la_nlink != 0)) {
659 CWARN("Attempt to destroy OSS object when nlink == %d\n",
664 ma->ma_lmm_size = mdd_lov_mdsize(env, mdd);
665 ma->ma_lmm = mdd_max_lmm_get(env, mdd);
666 ma->ma_cookie_size = mdd_lov_cookiesize(env, mdd);
667 ma->ma_cookie = mdd_max_cookie_get(env, mdd);
668 if (ma->ma_lmm == NULL || ma->ma_cookie == NULL)
669 RETURN(rc = -ENOMEM);
673 rc = mdd_get_md(env, obj, ma->ma_lmm, &ma->ma_lmm_size,
677 CWARN("Get lov ea failed for "DFID" rc = %d\n",
678 PFID(mdo2fid(obj)), rc);
684 ma->ma_valid = MA_LOV;
686 rc = mdd_unlink_log(env, mdd, obj, ma);
688 CWARN("mds unlink log for "DFID" failed: %d\n",
689 PFID(mdo2fid(obj)), rc);
693 if (ma->ma_valid & MA_COOKIE)
694 rc = mdd_lovobj_unlink(env, mdd, obj, la,
695 ma->ma_lmm, ma->ma_lmm_size,
700 int mdd_declare_unlink_log(const struct lu_env *env, struct mdd_object *obj,
701 struct md_attr *ma, struct thandle *handle)
703 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
710 if (!S_ISREG(lu_object_attr(&obj->mod_obj.mo_lu)))
713 rc = mdd_lmm_get_locked(env, obj, ma);
714 if (rc || !(ma->ma_valid & MA_LOV))
718 if (le32_to_cpu(ma->ma_lmm->lmm_magic) != LOV_MAGIC_V1 &&
719 le32_to_cpu(ma->ma_lmm->lmm_magic) != LOV_MAGIC_V3) {
720 CERROR("%s: invalid LOV_MAGIC %08x on object "DFID"\n",
721 mdd->mdd_obd_dev->obd_name,
722 le32_to_cpu(ma->ma_lmm->lmm_magic),
723 PFID(lu_object_fid(&obj->mod_obj.mo_lu)));
727 stripe = le16_to_cpu(ma->ma_lmm->lmm_stripe_count);
728 if (stripe == LOV_ALL_STRIPES);
729 stripe = mdd2obd_dev(mdd)->u.mds.mds_lov_desc.ld_tgt_count;
731 for (i = 0; i < stripe; i++) {
732 rc = mdd_declare_llog_record(env, mdd,
733 sizeof(struct llog_unlink_rec),
742 int mdd_unlink_log(const struct lu_env *env, struct mdd_device *mdd,
743 struct mdd_object *mdd_cobj, struct md_attr *ma)
745 LASSERT(ma->ma_valid & MA_LOV);
747 if ((ma->ma_cookie_size > 0) &&
748 (mds_log_op_unlink(mdd2obd_dev(mdd), ma->ma_lmm, ma->ma_lmm_size,
749 ma->ma_cookie, ma->ma_cookie_size) > 0)) {
750 CDEBUG(D_HA, "DEBUG: unlink log is added for object "DFID"\n",
751 PFID(mdd_object_fid(mdd_cobj)));
752 ma->ma_valid |= MA_COOKIE;
757 int mdd_log_op_setattr(struct obd_device *obd, __u32 uid, __u32 gid,
758 struct lov_mds_md *lmm, int lmm_size,
759 struct llog_cookie *logcookies, int cookies_size)
761 struct mds_obd *mds = &obd->u.mds;
762 struct lov_stripe_md *lsm = NULL;
763 struct llog_setattr64_rec *lsr;
764 struct llog_ctxt *ctxt;
768 if (IS_ERR(mds->mds_lov_obd))
769 RETURN(PTR_ERR(mds->mds_lov_obd));
771 rc = obd_unpackmd(mds->mds_lov_exp, &lsm, lmm, lmm_size);
775 OBD_ALLOC(lsr, sizeof(*lsr));
777 GOTO(out, rc = -ENOMEM);
779 /* prepare setattr log record */
780 lsr->lsr_hdr.lrh_len = lsr->lsr_tail.lrt_len = sizeof(*lsr);
781 lsr->lsr_hdr.lrh_type = MDS_SETATTR64_REC;
785 /* write setattr log */
786 ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
787 rc = llog_add(ctxt, &lsr->lsr_hdr, lsm, logcookies,
788 cookies_size / sizeof(struct llog_cookie));
792 OBD_FREE(lsr, sizeof(*lsr));
794 obd_free_memmd(mds->mds_lov_exp, &lsm);
798 int mdd_setattr_log(const struct lu_env *env, struct mdd_device *mdd,
799 const struct md_attr *ma,
800 struct lov_mds_md *lmm, int lmm_size,
801 struct llog_cookie *logcookies, int cookies_size)
803 struct obd_device *obd = mdd2obd_dev(mdd);
805 /* journal chown/chgrp in llog, just like unlink */
807 CDEBUG(D_INFO, "setattr llog for uid/gid=%lu/%lu\n",
808 (unsigned long)ma->ma_attr.la_uid,
809 (unsigned long)ma->ma_attr.la_gid);
810 return mdd_log_op_setattr(obd, ma->ma_attr.la_uid,
811 ma->ma_attr.la_gid, lmm,
812 lmm_size, logcookies,
818 static int mdd_osc_setattr_async(struct obd_device *obd, __u32 uid, __u32 gid,
819 struct lov_mds_md *lmm, int lmm_size,
820 struct llog_cookie *logcookies, const struct lu_fid *parent,
823 struct mds_obd *mds = &obd->u.mds;
824 struct obd_trans_info oti = { 0 };
825 struct obd_info oinfo = { { { 0 } } };
829 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OST_SETATTR))
832 /* first get memory EA */
833 OBDO_ALLOC(oinfo.oi_oa);
839 rc = obd_unpackmd(mds->mds_lov_exp, &oinfo.oi_md, lmm, lmm_size);
841 CERROR("Error unpack md %p for obj "DFID"\n", lmm,
847 oinfo.oi_oa->o_uid = uid;
848 oinfo.oi_oa->o_gid = gid;
849 oinfo.oi_oa->o_id = oinfo.oi_md->lsm_object_id;
850 oinfo.oi_oa->o_seq = oinfo.oi_md->lsm_object_seq;
851 oinfo.oi_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP |
852 OBD_MD_FLUID | OBD_MD_FLGID;
854 oinfo.oi_oa->o_valid |= OBD_MD_FLCOOKIE;
855 oti.oti_logcookies = logcookies;
858 obdo_set_parent_fid(oinfo.oi_oa, parent);
861 /* do async setattr from mds to ost not waiting for responses. */
862 rc = obd_setattr_async(mds->mds_lov_exp, &oinfo, &oti, NULL);
864 CDEBUG(D_INODE, "mds to ost setattr objid 0x"LPX64
865 " on ost error %d\n", oinfo.oi_md->lsm_object_id, rc);
868 obd_free_memmd(mds->mds_lov_exp, &oinfo.oi_md);
869 OBDO_FREE(oinfo.oi_oa);
873 int mdd_lov_setattr_async(const struct lu_env *env, struct mdd_object *obj,
874 struct lov_mds_md *lmm, int lmm_size,
875 struct llog_cookie *logcookies)
877 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
878 struct obd_device *obd = mdd2obd_dev(mdd);
879 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
880 const struct lu_fid *fid = mdd_object_fid(obj);
884 mdd_read_lock(env, obj, MOR_TGT_CHILD);
885 rc = mdo_attr_get(env, obj, tmp_la, mdd_object_capa(env, obj));
886 mdd_read_unlock(env, obj);
890 rc = mdd_osc_setattr_async(obd, tmp_la->la_uid, tmp_la->la_gid, lmm,
891 lmm_size, logcookies, fid, NULL);
895 static int grouplock_blocking_ast(struct ldlm_lock *lock,
896 struct ldlm_lock_desc *desc,
897 void *data, int flag)
899 struct md_attr *ma = data;
900 struct lustre_handle lockh;
906 case LDLM_CB_BLOCKING :
907 /* lock is canceled */
908 CDEBUG(D_DLMTRACE, "Lock %p is canceled\n", lock);
910 ldlm_lock2handle(lock, &lockh);
911 rc = ldlm_cli_cancel(&lockh);
914 case LDLM_CB_CANCELING :
916 "Lock %p has been canceled, do cleaning\n",
919 if (ma && ma->ma_som)
920 OBD_FREE_PTR(ma->ma_som);
930 static int grouplock_glimpse_ast(struct ldlm_lock *lock, void *data)
932 struct ptlrpc_request *req = data;
938 ma = lock->l_ast_data;
940 req_capsule_extend(&req->rq_pill, &RQF_LDLM_GL_CALLBACK);
941 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
943 rc = req_capsule_server_pack(&req->rq_pill);
945 CERROR("failed pack reply: %d\n", rc);
949 lvb = req_capsule_server_get(&req->rq_pill, &RMF_DLM_LVB);
951 if ((ma) && (ma->ma_valid & MA_SOM)) {
952 lvb->lvb_size = ma->ma_som->msd_size;
953 lvb->lvb_blocks = ma->ma_som->msd_blocks;
954 } else if ((ma) && (ma->ma_valid & MA_INODE)) {
955 lvb->lvb_size = ma->ma_attr.la_size;
956 lvb->lvb_blocks = ma->ma_attr.la_blocks;
959 rc = -ELDLM_NO_LOCK_DATA;
964 if (rc == -ELDLM_NO_LOCK_DATA)
965 lustre_pack_reply(req, 1, NULL, NULL);
971 int mdd_file_lock(const struct lu_env *env, struct md_object *obj,
972 struct lov_mds_md *lmm, struct ldlm_extent *extent,
973 struct lustre_handle *lockh)
975 struct ldlm_enqueue_info einfo = { 0 };
976 struct obd_info oinfo = { { { 0 } } };
977 struct obd_device *obd;
978 struct obd_export *lov_exp;
979 struct lov_stripe_md *lsm = NULL;
980 struct md_attr *ma = NULL;
984 obd = mdo2mdd(obj)->mdd_obd_dev;
985 lov_exp = obd->u.mds.mds_lov_exp;
987 obd_unpackmd(lov_exp, &lsm, lmm,
988 lov_mds_md_size(lmm->lmm_stripe_count, lmm->lmm_magic));
992 GOTO(out, rc = -ENOMEM);
994 OBD_ALLOC_PTR(ma->ma_som);
995 if (ma->ma_som == NULL)
996 GOTO(out, rc = -ENOMEM);
998 ma->ma_need = MA_SOM | MA_INODE;
999 mo_attr_get(env, obj, ma);
1001 einfo.ei_type = LDLM_EXTENT;
1002 einfo.ei_mode = LCK_GROUP;
1003 einfo.ei_cb_bl = grouplock_blocking_ast;
1004 einfo.ei_cb_cp = ldlm_completion_ast;
1005 einfo.ei_cb_gl = grouplock_glimpse_ast;
1007 if (ma->ma_valid & (MA_SOM | MA_INODE))
1008 einfo.ei_cbdata = ma;
1010 einfo.ei_cbdata = NULL;
1012 memset(&oinfo.oi_policy, 0, sizeof(oinfo.oi_policy));
1013 oinfo.oi_policy.l_extent = *extent;
1014 oinfo.oi_lockh = lockh;
1018 rc = obd_enqueue(lov_exp, &oinfo, &einfo, NULL);
1019 /* ei_cbdata is used as a free flag at exit */
1021 einfo.ei_cbdata = NULL;
1023 obd_unpackmd(lov_exp, &lsm, NULL, 0);
1026 /* ma is freed if not used as callback data */
1027 if ((einfo.ei_cbdata == NULL) && ma && ma->ma_som)
1028 OBD_FREE_PTR(ma->ma_som);
1029 if ((einfo.ei_cbdata == NULL) && ma)
1035 int mdd_file_unlock(const struct lu_env *env, struct md_object *obj,
1036 struct lov_mds_md *lmm, struct lustre_handle *lockh)
1038 struct obd_device *obd;
1039 struct obd_export *lov_exp;
1040 struct lov_stripe_md *lsm = NULL;
1044 LASSERT(lustre_handle_is_used(lockh));
1046 obd = mdo2mdd(obj)->mdd_obd_dev;
1047 lov_exp = obd->u.mds.mds_lov_exp;
1049 obd_unpackmd(lov_exp, &lsm, lmm,
1050 lov_mds_md_size(lmm->lmm_stripe_count, lmm->lmm_magic));
1052 rc = obd_cancel(lov_exp, lsm, LCK_GROUP, lockh);
1054 obd_unpackmd(lov_exp, &lsm, NULL, 0);
1059 /* file lov is in ma->ma_lmm */
1060 /* requested lov is in info->mti_spec.u.sp_ea.eadata */
1061 int mdd_lum_lmm_cmp(const struct lu_env *env, struct md_object *cobj,
1062 const struct md_op_spec *spec, struct md_attr *ma)
1064 struct obd_export *lov_exp =
1065 mdd2obd_dev(mdo2mdd(cobj))->u.mds.mds_lov_exp;
1066 struct lov_mds_md *lmm = ma->ma_lmm;
1067 struct lov_user_md_v3 *lum =
1068 (struct lov_user_md_v3 *)(spec->u.sp_ea.eadata);
1069 struct lov_stripe_md *lsm = NULL;
1073 rc = obd_unpackmd(lov_exp, &lsm, lmm,
1074 lov_mds_md_size(lmm->lmm_stripe_count,
1076 ma->ma_layout_gen = lsm->lsm_layout_gen;
1077 ma->ma_valid |= MA_LAY_GEN;
1079 rc = lov_lum_swab_if_needed(lum, &lmm_magic, NULL);
1083 rc = lov_lum_lsm_cmp((struct lov_user_md *)lum, lsm);
1085 GOTO(out, rc); /* keep GOTO to for traces */
1089 obd_unpackmd(lov_exp, &lsm, NULL, 0);