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 Whamcloud, Inc.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
39 * lustre/mdd/mdd_lov.c
41 * Lustre Metadata Server (mds) handling of striped file data
43 * Author: Peter Braam <braam@clusterfs.com>
44 * Author: wangdi <wangdi@clusterfs.com>
48 # define EXPORT_SYMTAB
50 #define DEBUG_SUBSYSTEM S_MDS
52 #include <linux/module.h>
54 #include <obd_class.h>
55 #include <lustre_ver.h>
56 #include <obd_support.h>
58 #include <lprocfs_status.h>
59 #include <lustre_mds.h>
60 #include <lustre_fid.h>
61 #include <lustre/lustre_idl.h>
63 #include "mdd_internal.h"
65 static int mdd_notify(struct obd_device *host, struct obd_device *watched,
66 enum obd_notify_event ev, void *owner, void *data)
68 struct mdd_device *mdd = owner;
72 LASSERT(owner != NULL);
75 case OBD_NOTIFY_ACTIVE:
77 case OBD_NOTIFY_SYNC_NONBLOCK:
78 rc = md_do_upcall(NULL, &mdd->mdd_md_dev,
81 case OBD_NOTIFY_CONFIG:
82 rc = md_do_upcall(NULL, &mdd->mdd_md_dev,
85 #ifdef HAVE_QUOTA_SUPPORT
86 case OBD_NOTIFY_QUOTA:
87 rc = md_do_upcall(NULL, &mdd->mdd_md_dev,
92 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
98 /* The obd is created for handling data stack for mdd */
99 int mdd_init_obd(const struct lu_env *env, struct mdd_device *mdd,
100 struct lustre_cfg *cfg)
102 char *dev = lustre_cfg_string(cfg, 0);
103 int rc, name_size, uuid_size;
106 struct lustre_cfg_bufs *bufs;
107 struct lustre_cfg *lcfg;
108 struct obd_device *obd;
111 mds_id = lu_site2md(mdd2lu_dev(mdd)->ld_site)->ms_node_id;
112 name_size = strlen(MDD_OBD_NAME) + 35;
113 uuid_size = strlen(MDD_OBD_UUID) + 35;
115 OBD_ALLOC(name, name_size);
116 OBD_ALLOC(uuid, uuid_size);
117 if (name == NULL || uuid == NULL)
118 GOTO(cleanup_mem, rc = -ENOMEM);
122 GOTO(cleanup_mem, rc = -ENOMEM);
124 snprintf(name, strlen(MDD_OBD_NAME) + 35, "%s-%s",
127 snprintf(uuid, strlen(MDD_OBD_UUID) + 35, "%s-%s",
130 lustre_cfg_bufs_reset(bufs, name);
131 lustre_cfg_bufs_set_string(bufs, 1, MDD_OBD_TYPE);
132 lustre_cfg_bufs_set_string(bufs, 2, uuid);
133 lustre_cfg_bufs_set_string(bufs, 3, (char*)dev/* MDD_OBD_PROFILE */);
134 lustre_cfg_bufs_set_string(bufs, 4, (char*)dev);
136 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
139 GOTO(cleanup_mem, rc = -ENOMEM);
141 rc = class_attach(lcfg);
143 GOTO(lcfg_cleanup, rc);
145 obd = class_name2obd(name);
147 CERROR("Can not find obd %s\n", MDD_OBD_NAME);
151 cfs_spin_lock(&obd->obd_dev_lock);
152 obd->obd_recovering = 1;
153 cfs_spin_unlock(&obd->obd_dev_lock);
154 obd->u.mds.mds_id = mds_id;
155 obd->u.obt.obt_osd_properties.osd_max_ea_size =
156 mdd->mdd_dt_conf.ddp_max_ea_size;
158 rc = class_setup(obd, lcfg);
160 GOTO(class_detach, rc);
163 * Add here for obd notify mechanism, when adding a new ost, the mds
164 * will notify this mdd. The mds will be used for quota also.
166 obd->obd_upcall.onu_upcall = mdd_notify;
167 obd->obd_upcall.onu_owner = mdd;
168 mdd->mdd_obd_dev = obd;
173 class_detach(obd, lcfg);
175 lustre_cfg_free(lcfg);
178 OBD_FREE(name, name_size);
180 OBD_FREE(uuid, uuid_size);
184 int mdd_fini_obd(const struct lu_env *env, struct mdd_device *mdd,
185 struct lustre_cfg *lcfg)
187 struct obd_device *obd;
191 obd = mdd2obd_dev(mdd);
194 rc = class_cleanup(obd, lcfg);
196 GOTO(lcfg_cleanup, rc);
198 obd->obd_upcall.onu_upcall = NULL;
199 obd->obd_upcall.onu_owner = NULL;
200 rc = class_detach(obd, lcfg);
202 GOTO(lcfg_cleanup, rc);
203 mdd->mdd_obd_dev = NULL;
210 int mdd_get_md(const struct lu_env *env, struct mdd_object *obj,
211 void *md, int *md_size, const char *name)
216 rc = mdo_xattr_get(env, obj, mdd_buf_get(env, md, *md_size), name,
217 mdd_object_capa(env, obj));
219 * XXX: Handling of -ENODATA, the right way is to have ->do_md_get()
220 * exported by dt layer.
222 if (rc == 0 || rc == -ENODATA) {
226 CERROR("Error %d reading eadata - %d\n", rc, *md_size);
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 (i.e. all default
285 * values specified) then delete default striping from dir. */
286 if (LOVEA_DELETE_VALUES(lum->lmm_stripe_size, lum->lmm_stripe_count,
287 lum->lmm_stripe_offset) &&
288 lum->lmm_magic != LOV_USER_MAGIC_V3) {
289 rc = mdd_xattr_set_txn(env, obj, &LU_BUF_NULL,
290 XATTR_NAME_LOV, 0, handle);
293 CDEBUG(D_INFO, "delete lov ea of "DFID" rc %d \n",
294 PFID(mdo2fid(obj)), rc);
296 rc = mdd_lov_set_stripe_md(env, obj, buf, handle);
301 int mdd_lsm_sanity_check(const struct lu_env *env, struct mdd_object *obj)
303 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
304 struct md_ucred *uc = md_ucred(env);
308 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
312 if ((uc->mu_fsuid != tmp_la->la_uid) &&
313 !mdd_capable(uc, CFS_CAP_FOWNER))
314 rc = mdd_permission_internal_locked(env, obj, tmp_la,
315 MAY_WRITE, MOR_TGT_CHILD);
320 int mdd_lov_set_md(const struct lu_env *env, struct mdd_object *pobj,
321 struct mdd_object *child, struct lov_mds_md *lmmp,
322 int lmm_size, struct thandle *handle, int set_stripe)
329 buf = mdd_buf_get(env, lmmp, lmm_size);
330 mode = mdd_object_type(child);
331 if (S_ISREG(mode) && lmm_size > 0) {
333 rc = mdd_lov_set_stripe_md(env, child, buf, handle);
335 rc = mdd_xattr_set_txn(env, child, buf,
336 XATTR_NAME_LOV, 0, handle);
338 } else if (S_ISDIR(mode)) {
339 if (lmmp == NULL && lmm_size == 0) {
340 struct mdd_device *mdd = mdd_obj2mdd_dev(child);
341 struct lov_mds_md *lmm = mdd_max_lmm_get(env, mdd);
342 int size = sizeof(struct lov_mds_md_v3);
344 /* Get parent dir stripe and set */
346 rc = mdd_get_md_locked(env, pobj, lmm, &size,
349 buf = mdd_buf_get(env, lmm, size);
350 rc = mdd_xattr_set_txn(env, child, buf,
351 XATTR_NAME_LOV, 0, handle);
353 CERROR("error on copy stripe info: rc "
357 LASSERT(lmmp != NULL && lmm_size > 0);
358 rc = mdd_lov_set_dir_md(env, child, buf, handle);
361 CDEBUG(D_INFO, "Set lov md %p size %d for fid "DFID" rc %d\n",
362 lmmp, lmm_size, PFID(mdo2fid(child)), rc);
366 int mdd_lov_objid_prepare(struct mdd_device *mdd, struct lov_mds_md *lmm)
368 /* copy mds_lov code is using wrong layer */
369 return mds_lov_prepare_objids(mdd->mdd_obd_dev, lmm);
372 int mdd_declare_lov_objid_update(const struct lu_env *env,
373 struct mdd_device *mdd,
374 struct thandle *handle)
376 struct obd_device *obd = mdd2obd_dev(mdd);
379 /* in prepare we create local files */
380 if (unlikely(mdd->mdd_capa == NULL))
383 /* XXX: this is a temporary solution to declare llog changes
384 * will be fixed in 2.3 with new llog implementation */
386 size = obd->u.mds.mds_lov_desc.ld_tgt_count * sizeof(obd_id);
387 return dt_declare_record_write(env, mdd->mdd_capa, size, 0, handle);
390 void mdd_lov_objid_update(struct mdd_device *mdd, struct lov_mds_md *lmm)
392 /* copy mds_lov code is using wrong layer */
393 mds_lov_update_objids(mdd->mdd_obd_dev, lmm);
396 void mdd_lov_create_finish(const struct lu_env *env, struct mdd_device *mdd,
397 struct lov_mds_md *lmm, int lmm_size,
398 const struct md_op_spec *spec)
400 if (lmm && !spec->no_create)
401 OBD_FREE_LARGE(lmm, lmm_size);
404 int mdd_lov_create(const struct lu_env *env, struct mdd_device *mdd,
405 struct mdd_object *parent, struct mdd_object *child,
406 struct lov_mds_md **lmm, int *lmm_size,
407 const struct md_op_spec *spec, struct lu_attr *la)
409 struct obd_device *obd = mdd2obd_dev(mdd);
410 struct obd_export *lov_exp = obd->u.mds.mds_lov_exp;
411 struct lu_site *site = mdd2lu_dev(mdd)->ld_site;
413 struct lov_stripe_md *lsm = NULL;
414 const void *eadata = spec->u.sp_ea.eadata;
415 __u64 create_flags = spec->sp_cr_flags;
416 struct obd_trans_info *oti = &mdd_env_info(env)->mti_oti;
420 if (!md_should_create(create_flags)) {
426 /* replay case, has objects already, only get lov from eadata */
427 if (spec->no_create != 0) {
428 *lmm = (struct lov_mds_md *)spec->u.sp_ea.eadata;
429 *lmm_size = spec->u.sp_ea.eadatalen;
430 if (*lmm_size == lov_mds_md_size((*lmm)->lmm_stripe_count,
431 (*lmm)->lmm_magic)) {
434 CERROR("incorrect lsm received during recovery\n");
439 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
440 GOTO(out_ids, rc = -ENOMEM);
442 LASSERT(lov_exp != NULL);
443 oa = &mdd_env_info(env)->mti_oa;
445 oa->o_uid = 0; /* must have 0 uid / gid on OST */
447 oa->o_seq = mdt_to_obd_objseq(lu_site2md(site)->ms_node_id);
448 oa->o_mode = S_IFREG | 0600;
449 oa->o_id = fid_ver_oid(mdd_object_fid(child));
450 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLFLAGS |
451 OBD_MD_FLMODE | OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLGROUP;
454 if (!(create_flags & MDS_OPEN_HAS_OBJS)) {
455 if (create_flags & MDS_OPEN_HAS_EA) {
456 LASSERT(eadata != NULL);
457 rc = obd_iocontrol(OBD_IOC_LOV_SETSTRIPE, lov_exp,
458 0, &lsm, (void*)eadata);
461 } else if (parent != NULL) {
462 /* get lov ea from parent and set to lov */
463 struct lov_mds_md *_lmm;
466 _lmm_size = mdd_lov_mdsize(env, mdd);
467 _lmm = mdd_max_lmm_get(env, mdd);
470 GOTO(out_oti, rc = -ENOMEM);
472 rc = mdd_get_md_locked(env, parent, _lmm,
476 rc = obd_iocontrol(OBD_IOC_LOV_SETSTRIPE,
484 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_OPEN_WAIT_CREATE, 10);
485 rc = obd_create(lov_exp, oa, &lsm, oti);
488 CERROR("Create error for "DFID": %d\n",
489 PFID(mdo2fid(child)), rc);
494 LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
496 LASSERT(eadata != NULL);
497 rc = obd_iocontrol(OBD_IOC_LOV_SETEA, lov_exp, 0, &lsm,
504 lsm->lsm_object_id = fid_ver_oid(mdd_object_fid(child));
505 lsm->lsm_object_seq = fid_seq(mdd_object_fid(child));
507 * Sometimes, we may truncate some object(without lsm) then open it
508 * (with write flags), so creating lsm above. The Nonzero(truncated)
509 * size should tell ost, since size attr is in charge by OST.
511 if (la->la_size && la->la_valid & LA_SIZE) {
512 struct obd_info *oinfo = &mdd_env_info(env)->mti_oi;
514 memset(oinfo, 0, sizeof(*oinfo));
516 /* When setting attr to ost, FLBKSZ is not needed. */
517 oa->o_valid &= ~OBD_MD_FLBLKSZ;
518 obdo_from_la(oa, la, OBD_MD_FLTYPE | OBD_MD_FLATIME |
519 OBD_MD_FLMTIME | OBD_MD_FLCTIME | OBD_MD_FLSIZE);
522 * XXX: Pack lustre id to OST, in OST, it will be packed by
523 * filter_fid, but can not see what is the usages. So just pack
524 * o_seq o_ver here, maybe fix it after this cycle.
526 obdo_from_inode(oa, NULL,
527 (struct lu_fid *)mdd_object_fid(child), 0);
530 oinfo->oi_capa = NULL;
531 oinfo->oi_policy.l_extent.start = la->la_size;
532 oinfo->oi_policy.l_extent.end = OBD_OBJECT_EOF;
534 rc = obd_punch_rqset(lov_exp, oinfo, oti);
536 CERROR("Error setting attrs for "DFID": rc %d\n",
537 PFID(mdo2fid(child)), rc);
539 CERROR("obd_setattr for "DFID" rc %d\n",
540 PFID(mdo2fid(child)), rc);
546 /* blksize should be changed after create data object */
547 la->la_valid |= LA_BLKSIZE;
548 la->la_blksize = oa->o_blksize;
550 rc = obd_packmd(lov_exp, lmm, lsm);
552 CERROR("Cannot pack lsm, err = %d\n", rc);
555 if (mdd_lov_objid_prepare(mdd, *lmm) != 0) {
556 CERROR("Not have memory for update objid\n");
559 GOTO(out_oti, rc = -ENOMEM);
565 oti_free_cookies(oti);
568 obd_free_memmd(lov_exp, &lsm);
574 * used when destroying orphans and from mds_reint_unlink() when MDS wants to
575 * destroy objects on OSS.
578 int mdd_lovobj_unlink(const struct lu_env *env, struct mdd_device *mdd,
579 struct mdd_object *obj, struct lu_attr *la,
580 struct lov_mds_md *lmm, int lmm_size,
581 struct llog_cookie *logcookies,
584 struct obd_device *obd = mdd2obd_dev(mdd);
585 struct obd_export *lov_exp = obd->u.mds.mds_lov_exp;
586 struct lov_stripe_md *lsm = NULL;
587 struct obd_trans_info *oti = &mdd_env_info(env)->mti_oti;
588 struct obdo *oa = &mdd_env_info(env)->mti_oa;
589 struct lu_site *site = mdd2lu_dev(mdd)->ld_site;
596 rc = obd_unpackmd(lov_exp, &lsm, lmm, lmm_size);
598 CERROR("Error unpack md %p\n", lmm);
601 LASSERT(rc >= sizeof(*lsm));
605 oa->o_id = lsm->lsm_object_id;
606 oa->o_seq = mdt_to_obd_objseq(lu_site2md(site)->ms_node_id);
607 oa->o_mode = la->la_mode & S_IFMT;
608 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
611 if (log_unlink && logcookies) {
612 oa->o_valid |= OBD_MD_FLCOOKIE;
613 oti->oti_logcookies = logcookies;
616 CDEBUG(D_INFO, "destroying OSS object "LPU64":"LPU64"\n", oa->o_seq,
619 rc = obd_destroy(lov_exp, oa, lsm, oti, NULL, NULL);
621 obd_free_memmd(lov_exp, &lsm);
626 * called with obj locked.
628 int mdd_lov_destroy(const struct lu_env *env, struct mdd_device *mdd,
629 struct mdd_object *obj, struct lu_attr *la)
631 struct md_attr *ma = &mdd_env_info(env)->mti_ma;
635 LASSERT(mdd_write_locked(env, obj) != 0);
637 if (unlikely(!S_ISREG(mdd_object_type(obj))))
640 if (unlikely(la->la_nlink != 0)) {
641 CWARN("Attempt to destroy OSS object when nlink == %d\n",
646 ma->ma_lmm_size = mdd_lov_mdsize(env, mdd);
647 ma->ma_lmm = mdd_max_lmm_get(env, mdd);
648 ma->ma_cookie_size = mdd_lov_cookiesize(env, mdd);
649 ma->ma_cookie = mdd_max_cookie_get(env, mdd);
650 if (ma->ma_lmm == NULL || ma->ma_cookie == NULL)
651 RETURN(rc = -ENOMEM);
655 rc = mdd_get_md(env, obj, ma->ma_lmm, &ma->ma_lmm_size,
659 CWARN("Get lov ea failed for "DFID" rc = %d\n",
660 PFID(mdo2fid(obj)), rc);
666 ma->ma_valid = MA_LOV;
668 rc = mdd_unlink_log(env, mdd, obj, ma);
670 CWARN("mds unlink log for "DFID" failed: %d\n",
671 PFID(mdo2fid(obj)), rc);
675 if (ma->ma_valid & MA_COOKIE)
676 rc = mdd_lovobj_unlink(env, mdd, obj, la,
677 ma->ma_lmm, ma->ma_lmm_size,
682 int mdd_declare_unlink_log(const struct lu_env *env, struct mdd_object *obj,
683 struct md_attr *ma, struct thandle *handle)
685 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
691 if (!S_ISREG(lu_object_attr(&obj->mod_obj.mo_lu)))
694 rc = mdd_lmm_get_locked(env, obj, ma);
695 if (rc || !(ma->ma_valid & MA_LOV))
699 if (le32_to_cpu(ma->ma_lmm->lmm_magic) != LOV_MAGIC_V1 &&
700 le32_to_cpu(ma->ma_lmm->lmm_magic) != LOV_MAGIC_V3) {
701 CERROR("%s: invalid LOV_MAGIC %08x on object "DFID"\n",
702 mdd->mdd_obd_dev->obd_name,
703 le32_to_cpu(ma->ma_lmm->lmm_magic),
704 PFID(lu_object_fid(&obj->mod_obj.mo_lu)));
708 if ((int)le32_to_cpu(ma->ma_lmm->lmm_stripe_count) < 0)
709 stripe = mdd2obd_dev(mdd)->u.mds.mds_lov_desc.ld_tgt_count;
711 stripe = le32_to_cpu(ma->ma_lmm->lmm_stripe_count);
713 for (i = 0; i < stripe; i++) {
714 rc = mdd_declare_llog_record(env, mdd,
715 sizeof(struct llog_unlink_rec),
724 int mdd_unlink_log(const struct lu_env *env, struct mdd_device *mdd,
725 struct mdd_object *mdd_cobj, struct md_attr *ma)
727 LASSERT(ma->ma_valid & MA_LOV);
729 if ((ma->ma_cookie_size > 0) &&
730 (mds_log_op_unlink(mdd2obd_dev(mdd), ma->ma_lmm, ma->ma_lmm_size,
731 ma->ma_cookie, ma->ma_cookie_size) > 0)) {
732 CDEBUG(D_HA, "DEBUG: unlink log is added for object "DFID"\n",
733 PFID(mdd_object_fid(mdd_cobj)));
734 ma->ma_valid |= MA_COOKIE;
739 int mdd_log_op_setattr(struct obd_device *obd, __u32 uid, __u32 gid,
740 struct lov_mds_md *lmm, int lmm_size,
741 struct llog_cookie *logcookies, int cookies_size)
743 struct mds_obd *mds = &obd->u.mds;
744 struct lov_stripe_md *lsm = NULL;
745 struct llog_setattr64_rec *lsr;
746 struct llog_ctxt *ctxt;
750 if (IS_ERR(mds->mds_lov_obd))
751 RETURN(PTR_ERR(mds->mds_lov_obd));
753 rc = obd_unpackmd(mds->mds_lov_exp, &lsm, lmm, lmm_size);
757 OBD_ALLOC(lsr, sizeof(*lsr));
759 GOTO(out, rc = -ENOMEM);
761 /* prepare setattr log record */
762 lsr->lsr_hdr.lrh_len = lsr->lsr_tail.lrt_len = sizeof(*lsr);
763 lsr->lsr_hdr.lrh_type = MDS_SETATTR64_REC;
767 /* write setattr log */
768 ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
769 rc = llog_add(ctxt, &lsr->lsr_hdr, lsm, logcookies,
770 cookies_size / sizeof(struct llog_cookie));
774 OBD_FREE(lsr, sizeof(*lsr));
776 obd_free_memmd(mds->mds_lov_exp, &lsm);
780 int mdd_setattr_log(const struct lu_env *env, struct mdd_device *mdd,
781 const struct md_attr *ma,
782 struct lov_mds_md *lmm, int lmm_size,
783 struct llog_cookie *logcookies, int cookies_size)
785 struct obd_device *obd = mdd2obd_dev(mdd);
787 /* journal chown/chgrp in llog, just like unlink */
789 CDEBUG(D_INFO, "setattr llog for uid/gid=%lu/%lu\n",
790 (unsigned long)ma->ma_attr.la_uid,
791 (unsigned long)ma->ma_attr.la_gid);
792 return mdd_log_op_setattr(obd, ma->ma_attr.la_uid,
793 ma->ma_attr.la_gid, lmm,
794 lmm_size, logcookies,
800 static int mdd_osc_setattr_async(struct obd_device *obd, __u32 uid, __u32 gid,
801 struct lov_mds_md *lmm, int lmm_size,
802 struct llog_cookie *logcookies, const struct lu_fid *parent,
805 struct mds_obd *mds = &obd->u.mds;
806 struct obd_trans_info oti = { 0 };
807 struct obd_info oinfo = { { { 0 } } };
811 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OST_SETATTR))
814 /* first get memory EA */
815 OBDO_ALLOC(oinfo.oi_oa);
821 rc = obd_unpackmd(mds->mds_lov_exp, &oinfo.oi_md, lmm, lmm_size);
823 CERROR("Error unpack md %p for obj "DFID"\n", lmm,
829 oinfo.oi_oa->o_uid = uid;
830 oinfo.oi_oa->o_gid = gid;
831 oinfo.oi_oa->o_id = oinfo.oi_md->lsm_object_id;
832 oinfo.oi_oa->o_seq = oinfo.oi_md->lsm_object_seq;
833 oinfo.oi_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP |
834 OBD_MD_FLUID | OBD_MD_FLGID;
836 oinfo.oi_oa->o_valid |= OBD_MD_FLCOOKIE;
837 oti.oti_logcookies = logcookies;
840 obdo_from_inode(oinfo.oi_oa, NULL, (struct lu_fid *)parent, 0);
843 /* do async setattr from mds to ost not waiting for responses. */
844 rc = obd_setattr_async(mds->mds_lov_exp, &oinfo, &oti, NULL);
846 CDEBUG(D_INODE, "mds to ost setattr objid 0x"LPX64
847 " on ost error %d\n", oinfo.oi_md->lsm_object_id, rc);
850 obd_free_memmd(mds->mds_lov_exp, &oinfo.oi_md);
851 OBDO_FREE(oinfo.oi_oa);
855 int mdd_lov_setattr_async(const struct lu_env *env, struct mdd_object *obj,
856 struct lov_mds_md *lmm, int lmm_size,
857 struct llog_cookie *logcookies)
859 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
860 struct obd_device *obd = mdd2obd_dev(mdd);
861 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
862 const struct lu_fid *fid = mdd_object_fid(obj);
866 mdd_read_lock(env, obj, MOR_TGT_CHILD);
867 rc = mdo_attr_get(env, obj, tmp_la, mdd_object_capa(env, obj));
868 mdd_read_unlock(env, obj);
872 rc = mdd_osc_setattr_async(obd, tmp_la->la_uid, tmp_la->la_gid, lmm,
873 lmm_size, logcookies, fid, NULL);
877 static int grouplock_blocking_ast(struct ldlm_lock *lock,
878 struct ldlm_lock_desc *desc,
879 void *data, int flag)
881 struct md_attr *ma = data;
882 struct lustre_handle lockh;
888 case LDLM_CB_BLOCKING :
889 /* lock is canceled */
890 CDEBUG(D_DLMTRACE, "Lock %p is canceled\n", lock);
892 ldlm_lock2handle(lock, &lockh);
893 rc = ldlm_cli_cancel(&lockh);
896 case LDLM_CB_CANCELING :
898 "Lock %p has been canceled, do cleaning\n",
901 if (ma && ma->ma_som)
902 OBD_FREE_PTR(ma->ma_som);
912 static int grouplock_glimpse_ast(struct ldlm_lock *lock, void *data)
914 struct ptlrpc_request *req = data;
920 ma = lock->l_ast_data;
922 req_capsule_extend(&req->rq_pill, &RQF_LDLM_GL_CALLBACK);
923 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
925 rc = req_capsule_server_pack(&req->rq_pill);
927 CERROR("failed pack reply: %d\n", rc);
931 lvb = req_capsule_server_get(&req->rq_pill, &RMF_DLM_LVB);
933 if ((ma) && (ma->ma_valid & MA_SOM)) {
934 lvb->lvb_size = ma->ma_som->msd_size;
935 lvb->lvb_blocks = ma->ma_som->msd_blocks;
936 } else if ((ma) && (ma->ma_valid & MA_INODE)) {
937 lvb->lvb_size = ma->ma_attr.la_size;
938 lvb->lvb_blocks = ma->ma_attr.la_blocks;
941 rc = -ELDLM_NO_LOCK_DATA;
946 if (rc == -ELDLM_NO_LOCK_DATA)
947 lustre_pack_reply(req, 1, NULL, NULL);
953 int mdd_file_lock(const struct lu_env *env, struct md_object *obj,
954 struct lov_mds_md *lmm, struct ldlm_extent *extent,
955 struct lustre_handle *lockh)
957 struct ldlm_enqueue_info einfo = { 0 };
958 struct obd_info oinfo = { { { 0 } } };
959 struct obd_device *obd;
960 struct obd_export *lov_exp;
961 struct lov_stripe_md *lsm = NULL;
962 struct md_attr *ma = NULL;
966 obd = mdo2mdd(obj)->mdd_obd_dev;
967 lov_exp = obd->u.mds.mds_lov_exp;
969 obd_unpackmd(lov_exp, &lsm, lmm,
970 lov_mds_md_size(lmm->lmm_stripe_count, lmm->lmm_magic));
974 GOTO(out, rc = -ENOMEM);
976 OBD_ALLOC_PTR(ma->ma_som);
977 if (ma->ma_som == NULL)
978 GOTO(out, rc = -ENOMEM);
980 ma->ma_need = MA_SOM | MA_INODE;
981 mo_attr_get(env, obj, ma);
983 einfo.ei_type = LDLM_EXTENT;
984 einfo.ei_mode = LCK_GROUP;
985 einfo.ei_cb_bl = grouplock_blocking_ast;
986 einfo.ei_cb_cp = ldlm_completion_ast;
987 einfo.ei_cb_gl = grouplock_glimpse_ast;
989 if (ma->ma_valid & (MA_SOM | MA_INODE))
990 einfo.ei_cbdata = ma;
992 einfo.ei_cbdata = NULL;
994 memset(&oinfo.oi_policy, 0, sizeof(oinfo.oi_policy));
995 oinfo.oi_policy.l_extent = *extent;
996 oinfo.oi_lockh = lockh;
1000 rc = obd_enqueue(lov_exp, &oinfo, &einfo, NULL);
1001 /* ei_cbdata is used as a free flag at exit */
1003 einfo.ei_cbdata = NULL;
1005 obd_unpackmd(lov_exp, &lsm, NULL, 0);
1008 /* ma is freed if not used as callback data */
1009 if ((einfo.ei_cbdata == NULL) && ma && ma->ma_som)
1010 OBD_FREE_PTR(ma->ma_som);
1011 if ((einfo.ei_cbdata == NULL) && ma)
1017 int mdd_file_unlock(const struct lu_env *env, struct md_object *obj,
1018 struct lov_mds_md *lmm, struct lustre_handle *lockh)
1020 struct obd_device *obd;
1021 struct obd_export *lov_exp;
1022 struct lov_stripe_md *lsm = NULL;
1026 LASSERT(lustre_handle_is_used(lockh));
1028 obd = mdo2mdd(obj)->mdd_obd_dev;
1029 lov_exp = obd->u.mds.mds_lov_exp;
1031 obd_unpackmd(lov_exp, &lsm, lmm,
1032 lov_mds_md_size(lmm->lmm_stripe_count, lmm->lmm_magic));
1034 rc = obd_cancel(lov_exp, lsm, LCK_GROUP, lockh);
1036 obd_unpackmd(lov_exp, &lsm, NULL, 0);