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.
33 * 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/mdt/mdt_handler.c
41 * Lustre Metadata Target (mdt) request handler
43 * Author: Peter Braam <braam@clusterfs.com>
44 * Author: Andreas Dilger <adilger@clusterfs.com>
45 * Author: Phil Schwan <phil@clusterfs.com>
46 * Author: Mike Shaver <shaver@clusterfs.com>
47 * Author: Nikita Danilov <nikita@clusterfs.com>
48 * Author: Huang Hua <huanghua@clusterfs.com>
49 * Author: Yury Umanets <umka@clusterfs.com>
53 # define EXPORT_SYMTAB
55 #define DEBUG_SUBSYSTEM S_MDS
57 #include <linux/module.h>
59 * struct OBD_{ALLOC,FREE}*()
61 #include <obd_support.h>
62 /* struct ptlrpc_request */
63 #include <lustre_net.h>
64 /* struct obd_export */
65 #include <lustre_export.h>
66 /* struct obd_device */
69 #include <dt_object.h>
70 #include <lustre_mds.h>
71 #include <lustre_mdt.h>
72 #include "mdt_internal.h"
73 #ifdef HAVE_QUOTA_SUPPORT
74 # include <lustre_quota.h>
76 #include <lustre_acl.h>
77 #include <lustre_param.h>
78 #include <lustre_fsfilt.h>
80 mdl_mode_t mdt_mdl_lock_modes[] = {
81 [LCK_MINMODE] = MDL_MINMODE,
88 [LCK_GROUP] = MDL_GROUP
91 ldlm_mode_t mdt_dlm_lock_modes[] = {
92 [MDL_MINMODE] = LCK_MINMODE,
99 [MDL_GROUP] = LCK_GROUP
103 * Initialized in mdt_mod_init().
105 static unsigned long mdt_num_threads;
106 static unsigned long mdt_min_threads;
107 static unsigned long mdt_max_threads;
109 /* ptlrpc request handler for MDT. All handlers are
110 * grouped into several slices - struct mdt_opc_slice,
111 * and stored in an array - mdt_handlers[].
114 /* The name of this handler. */
116 /* Fail id for this handler, checked at the beginning of this handler*/
118 /* Operation code for this handler */
120 /* flags are listed in enum mdt_handler_flags below. */
122 /* The actual handler function to execute. */
123 int (*mh_act)(struct mdt_thread_info *info);
124 /* Request format for this request. */
125 const struct req_format *mh_fmt;
128 enum mdt_handler_flags {
130 * struct mdt_body is passed in the incoming message, and object
131 * identified by this fid exists on disk.
133 * "habeo corpus" == "I have a body"
135 HABEO_CORPUS = (1 << 0),
137 * struct ldlm_request is passed in the incoming message.
139 * "habeo clavis" == "I have a key"
141 HABEO_CLAVIS = (1 << 1),
143 * this request has fixed reply format, so that reply message can be
144 * packed by generic code.
146 * "habeo refero" == "I have a reply"
148 HABEO_REFERO = (1 << 2),
150 * this request will modify something, so check whether the filesystem
151 * is readonly or not, then return -EROFS to client asap if necessary.
153 * "mutabor" == "I shall modify"
158 struct mdt_opc_slice {
161 struct mdt_handler *mos_hs;
164 static struct mdt_opc_slice mdt_regular_handlers[];
165 static struct mdt_opc_slice mdt_readpage_handlers[];
166 static struct mdt_opc_slice mdt_xmds_handlers[];
167 static struct mdt_opc_slice mdt_seq_handlers[];
168 static struct mdt_opc_slice mdt_fld_handlers[];
170 static struct mdt_device *mdt_dev(struct lu_device *d);
171 static int mdt_regular_handle(struct ptlrpc_request *req);
172 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
173 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
174 struct getinfo_fid2path *fp);
176 static const struct lu_object_operations mdt_obj_ops;
178 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
182 return (rep->lock_policy_res1 & flag);
185 void mdt_clear_disposition(struct mdt_thread_info *info,
186 struct ldlm_reply *rep, int flag)
189 info->mti_opdata &= ~flag;
191 rep->lock_policy_res1 &= ~flag;
194 void mdt_set_disposition(struct mdt_thread_info *info,
195 struct ldlm_reply *rep, int flag)
198 info->mti_opdata |= flag;
200 rep->lock_policy_res1 |= flag;
203 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
205 lh->mlh_pdo_hash = 0;
206 lh->mlh_reg_mode = lm;
207 lh->mlh_type = MDT_REG_LOCK;
210 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
211 const char *name, int namelen)
213 lh->mlh_reg_mode = lm;
214 lh->mlh_type = MDT_PDO_LOCK;
217 LASSERT(namelen > 0);
218 lh->mlh_pdo_hash = full_name_hash(name, namelen);
220 LASSERT(namelen == 0);
221 lh->mlh_pdo_hash = 0ull;
225 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
226 struct mdt_lock_handle *lh)
232 * Any dir access needs couple of locks:
234 * 1) on part of dir we gonna take lookup/modify;
236 * 2) on whole dir to protect it from concurrent splitting and/or to
237 * flush client's cache for readdir().
239 * so, for a given mode and object this routine decides what lock mode
240 * to use for lock #2:
242 * 1) if caller's gonna lookup in dir then we need to protect dir from
243 * being splitted only - LCK_CR
245 * 2) if caller's gonna modify dir then we need to protect dir from
246 * being splitted and to flush cache - LCK_CW
248 * 3) if caller's gonna modify dir and that dir seems ready for
249 * splitting then we need to protect it from any type of access
250 * (lookup/modify/split) - LCK_EX --bzzz
253 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
254 LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
257 * Ask underlaying level its opinion about preferable PDO lock mode
258 * having access type passed as regular lock mode:
260 * - MDL_MINMODE means that lower layer does not want to specify lock
263 * - MDL_NL means that no PDO lock should be taken. This is used in some
264 * cases. Say, for non-splittable directories no need to use PDO locks
267 mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
268 mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
270 if (mode != MDL_MINMODE) {
271 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
274 * Lower layer does not want to specify locking mode. We do it
275 * our selves. No special protection is needed, just flush
276 * client's cache on modification and allow concurrent
279 switch (lh->mlh_reg_mode) {
281 lh->mlh_pdo_mode = LCK_EX;
284 lh->mlh_pdo_mode = LCK_CR;
287 lh->mlh_pdo_mode = LCK_CW;
290 CERROR("Not expected lock type (0x%x)\n",
291 (int)lh->mlh_reg_mode);
296 LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
300 static int mdt_getstatus(struct mdt_thread_info *info)
302 struct mdt_device *mdt = info->mti_mdt;
303 struct md_device *next = mdt->mdt_child;
304 struct mdt_body *repbody;
309 rc = mdt_check_ucred(info);
311 RETURN(err_serious(rc));
313 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
314 RETURN(err_serious(-ENOMEM));
316 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
317 rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
321 repbody->valid |= OBD_MD_FLID;
323 if (mdt->mdt_opts.mo_mds_capa &&
324 info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
325 struct mdt_object *root;
326 struct lustre_capa *capa;
328 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
330 RETURN(PTR_ERR(root));
332 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
334 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
335 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
337 mdt_object_put(info->mti_env, root);
339 repbody->valid |= OBD_MD_FLMDSCAPA;
345 static int mdt_statfs(struct mdt_thread_info *info)
347 struct ptlrpc_request *req = mdt_info_req(info);
348 struct md_device *next = info->mti_mdt->mdt_child;
349 struct ptlrpc_service *svc;
350 struct obd_statfs *osfs;
355 svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
357 /* This will trigger a watchdog timeout */
358 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
359 (MDT_SERVICE_WATCHDOG_FACTOR *
360 at_get(&svc->srv_at_estimate)) + 1);
362 rc = mdt_check_ucred(info);
364 RETURN(err_serious(rc));
366 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
367 rc = err_serious(-ENOMEM);
369 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
370 rc = next->md_ops->mdo_statfs(info->mti_env, next,
372 statfs_pack(osfs, &info->mti_u.ksfs);
376 mdt_counter_incr(req->rq_export, LPROC_MDT_STATFS);
382 * Pack SOM attributes into the reply.
383 * Call under a DLM UPDATE lock.
385 static void mdt_pack_size2body(struct mdt_thread_info *info,
386 struct mdt_object *mo)
389 struct md_attr *ma = &info->mti_attr;
391 LASSERT(ma->ma_attr.la_valid & LA_MODE);
392 b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
394 /* Check if Size-on-MDS is supported, if this is a regular file,
395 * if SOM is enabled on the object and if SOM cache exists and valid.
396 * Otherwise do not pack Size-on-MDS attributes to the reply. */
397 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
398 !S_ISREG(ma->ma_attr.la_mode) ||
399 !mdt_object_is_som_enabled(mo) ||
400 !(ma->ma_valid & MA_SOM))
403 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
404 b->size = ma->ma_som->msd_size;
405 b->blocks = ma->ma_som->msd_blocks;
408 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
409 const struct lu_attr *attr, const struct lu_fid *fid)
411 struct md_attr *ma = &info->mti_attr;
413 LASSERT(ma->ma_valid & MA_INODE);
415 b->atime = attr->la_atime;
416 b->mtime = attr->la_mtime;
417 b->ctime = attr->la_ctime;
418 b->mode = attr->la_mode;
419 b->size = attr->la_size;
420 b->blocks = attr->la_blocks;
421 b->uid = attr->la_uid;
422 b->gid = attr->la_gid;
423 b->flags = attr->la_flags;
424 b->nlink = attr->la_nlink;
425 b->rdev = attr->la_rdev;
427 /*XXX should pack the reply body according to lu_valid*/
428 b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID |
429 OBD_MD_FLGID | OBD_MD_FLTYPE |
430 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
431 OBD_MD_FLATIME | OBD_MD_FLMTIME ;
433 if (!S_ISREG(attr->la_mode)) {
434 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
435 } else if (ma->ma_need & MA_LOV && ma->ma_lmm_size == 0) {
436 /* means no objects are allocated on osts. */
437 LASSERT(!(ma->ma_valid & MA_LOV));
438 /* just ignore blocks occupied by extend attributes on MDS */
440 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
441 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
446 b->valid |= OBD_MD_FLID;
448 /* FIXME: these should be fixed when new igif ready.*/
449 b->ino = fid_oid(fid); /* 1.6 compatibility */
450 b->generation = fid_ver(fid); /* 1.6 compatibility */
451 b->valid |= OBD_MD_FLGENER; /* 1.6 compatibility */
453 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
454 PFID(fid), b->nlink, b->mode, b->size);
458 mdt_body_reverse_idmap(info, b);
460 if (b->valid & OBD_MD_FLSIZE)
461 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
462 PFID(fid), (unsigned long long)b->size);
465 static inline int mdt_body_has_lov(const struct lu_attr *la,
466 const struct mdt_body *body)
468 return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
469 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
472 void mdt_client_compatibility(struct mdt_thread_info *info)
474 struct mdt_body *body;
475 struct ptlrpc_request *req = mdt_info_req(info);
476 struct obd_export *exp = req->rq_export;
477 struct md_attr *ma = &info->mti_attr;
478 struct lu_attr *la = &ma->ma_attr;
481 if (exp->exp_connect_flags & OBD_CONNECT_LAYOUTLOCK)
482 /* the client can deal with 16-bit lmm_stripe_count */
485 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
487 if (!mdt_body_has_lov(la, body))
490 /* now we have a reply with a lov for a client not compatible with the
491 * layout lock so we have to clean the layout generation number */
492 if (S_ISREG(la->la_mode))
493 ma->ma_lmm->lmm_layout_gen = 0;
498 static int mdt_getattr_internal(struct mdt_thread_info *info,
499 struct mdt_object *o, int ma_need)
501 struct md_object *next = mdt_object_child(o);
502 const struct mdt_body *reqbody = info->mti_body;
503 struct ptlrpc_request *req = mdt_info_req(info);
504 struct md_attr *ma = &info->mti_attr;
505 struct lu_attr *la = &ma->ma_attr;
506 struct req_capsule *pill = info->mti_pill;
507 const struct lu_env *env = info->mti_env;
508 struct mdt_body *repbody;
509 struct lu_buf *buffer = &info->mti_buf;
513 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
514 RETURN(err_serious(-ENOMEM));
516 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
520 rc = mdt_object_exists(o);
522 /* This object is located on remote node.*/
523 repbody->fid1 = *mdt_object_fid(o);
524 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
528 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
529 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
531 /* If it is dir object and client require MEA, then we got MEA */
532 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
533 reqbody->valid & OBD_MD_MEA) {
534 /* Assumption: MDT_MD size is enough for lmv size. */
535 ma->ma_lmv = buffer->lb_buf;
536 ma->ma_lmv_size = buffer->lb_len;
537 ma->ma_need = MA_LMV | MA_INODE;
539 ma->ma_lmm = buffer->lb_buf;
540 ma->ma_lmm_size = buffer->lb_len;
541 ma->ma_need = MA_LOV | MA_INODE;
544 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
545 reqbody->valid & OBD_MD_FLDIREA &&
546 lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
547 /* get default stripe info for this dir. */
548 ma->ma_need |= MA_LOV_DEF;
550 ma->ma_need |= ma_need;
551 if (ma->ma_need & MA_SOM)
552 ma->ma_som = &info->mti_u.som.data;
554 rc = mo_attr_get(env, next, ma);
556 CERROR("getattr error for "DFID": %d\n",
557 PFID(mdt_object_fid(o)), rc);
561 if (likely(ma->ma_valid & MA_INODE))
562 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
566 if (mdt_body_has_lov(la, reqbody)) {
567 if (ma->ma_valid & MA_LOV) {
568 LASSERT(ma->ma_lmm_size);
569 mdt_dump_lmm(D_INFO, ma->ma_lmm);
570 repbody->eadatasize = ma->ma_lmm_size;
571 if (S_ISDIR(la->la_mode))
572 repbody->valid |= OBD_MD_FLDIREA;
574 repbody->valid |= OBD_MD_FLEASIZE;
576 if (ma->ma_valid & MA_LMV) {
577 LASSERT(S_ISDIR(la->la_mode));
578 repbody->eadatasize = ma->ma_lmv_size;
579 repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
581 } else if (S_ISLNK(la->la_mode) &&
582 reqbody->valid & OBD_MD_LINKNAME) {
583 buffer->lb_buf = ma->ma_lmm;
584 buffer->lb_len = reqbody->eadatasize;
585 rc = mo_readlink(env, next, buffer);
586 if (unlikely(rc <= 0)) {
587 CERROR("readlink failed: %d\n", rc);
590 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
592 repbody->valid |= OBD_MD_LINKNAME;
593 repbody->eadatasize = rc;
595 ((char*)ma->ma_lmm)[rc - 1] = 0;
596 CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
597 (char*)ma->ma_lmm, rc);
602 if (reqbody->valid & OBD_MD_FLMODEASIZE) {
603 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
604 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
605 repbody->valid |= OBD_MD_FLMODEASIZE;
606 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
607 "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
608 repbody->max_cookiesize);
611 if (exp_connect_rmtclient(info->mti_exp) &&
612 reqbody->valid & OBD_MD_FLRMTPERM) {
613 void *buf = req_capsule_server_get(pill, &RMF_ACL);
615 /* mdt_getattr_lock only */
616 rc = mdt_pack_remote_perm(info, o, buf);
618 repbody->valid &= ~OBD_MD_FLRMTPERM;
619 repbody->aclsize = 0;
622 repbody->valid |= OBD_MD_FLRMTPERM;
623 repbody->aclsize = sizeof(struct mdt_remote_perm);
626 #ifdef CONFIG_FS_POSIX_ACL
627 else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
628 (reqbody->valid & OBD_MD_FLACL)) {
629 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
630 buffer->lb_len = req_capsule_get_size(pill,
631 &RMF_ACL, RCL_SERVER);
632 if (buffer->lb_len > 0) {
633 rc = mo_xattr_get(env, next, buffer,
634 XATTR_NAME_ACL_ACCESS);
636 if (rc == -ENODATA) {
637 repbody->aclsize = 0;
638 repbody->valid |= OBD_MD_FLACL;
640 } else if (rc == -EOPNOTSUPP) {
643 CERROR("got acl size: %d\n", rc);
646 repbody->aclsize = rc;
647 repbody->valid |= OBD_MD_FLACL;
654 if (reqbody->valid & OBD_MD_FLMDSCAPA &&
655 info->mti_mdt->mdt_opts.mo_mds_capa &&
656 info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
657 struct lustre_capa *capa;
659 capa = req_capsule_server_get(pill, &RMF_CAPA1);
661 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
662 rc = mo_capa_get(env, next, capa, 0);
665 repbody->valid |= OBD_MD_FLMDSCAPA;
670 static int mdt_renew_capa(struct mdt_thread_info *info)
672 struct mdt_object *obj = info->mti_object;
673 struct mdt_body *body;
674 struct lustre_capa *capa, *c;
678 /* if object doesn't exist, or server has disabled capability,
679 * return directly, client will find body->valid OBD_MD_FLOSSCAPA
682 if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
683 !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
686 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
687 LASSERT(body != NULL);
689 c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
692 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
696 rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
698 body->valid |= OBD_MD_FLOSSCAPA;
702 static int mdt_getattr(struct mdt_thread_info *info)
704 struct ptlrpc_request *req = mdt_info_req(info);
705 struct mdt_object *obj = info->mti_object;
706 struct req_capsule *pill = info->mti_pill;
707 struct mdt_body *reqbody;
708 struct mdt_body *repbody;
714 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
717 if (reqbody->valid & OBD_MD_FLOSSCAPA) {
718 rc = req_capsule_server_pack(pill);
720 RETURN(err_serious(rc));
721 rc = mdt_renew_capa(info);
722 GOTO(out_shrink, rc);
725 LASSERT(obj != NULL);
726 LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
728 mode = lu_object_attr(&obj->mot_obj.mo_lu);
729 if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
730 (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
731 md_size = reqbody->eadatasize;
733 md_size = info->mti_mdt->mdt_max_mdsize;
735 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
737 rc = req_capsule_server_pack(pill);
738 if (unlikely(rc != 0))
739 RETURN(err_serious(rc));
741 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
742 LASSERT(repbody != NULL);
743 repbody->eadatasize = 0;
744 repbody->aclsize = 0;
746 if (reqbody->valid & OBD_MD_FLRMTPERM)
747 rc = mdt_init_ucred(info, reqbody);
749 rc = mdt_check_ucred(info);
751 GOTO(out_shrink, rc);
753 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
754 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
757 * Don't check capability at all, because rename might getattr for
758 * remote obj, and at that time no capability is available.
760 mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
761 rc = mdt_getattr_internal(info, obj, 0);
762 if (reqbody->valid & OBD_MD_FLRMTPERM)
763 mdt_exit_ucred(info);
767 mdt_counter_incr(req->rq_export, LPROC_MDT_GETATTR);
769 mdt_client_compatibility(info);
770 mdt_shrink_reply(info);
774 static int mdt_is_subdir(struct mdt_thread_info *info)
776 struct mdt_object *o = info->mti_object;
777 struct req_capsule *pill = info->mti_pill;
778 const struct mdt_body *body = info->mti_body;
779 struct mdt_body *repbody;
785 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
788 * We save last checked parent fid to @repbody->fid1 for remote
791 LASSERT(fid_is_sane(&body->fid2));
792 LASSERT(mdt_object_exists(o) > 0);
793 rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
794 &body->fid2, &repbody->fid1);
795 if (rc == 0 || rc == -EREMOTE)
796 repbody->valid |= OBD_MD_FLID;
801 static int mdt_raw_lookup(struct mdt_thread_info *info,
802 struct mdt_object *parent,
803 const struct lu_name *lname,
804 struct ldlm_reply *ldlm_rep)
806 struct md_object *next = mdt_object_child(info->mti_object);
807 const struct mdt_body *reqbody = info->mti_body;
808 struct lu_fid *child_fid = &info->mti_tmp_fid1;
809 struct mdt_body *repbody;
813 if (reqbody->valid != OBD_MD_FLID)
816 LASSERT(!info->mti_cross_ref);
818 /* Only got the fid of this obj by name */
819 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
822 /* XXX is raw_lookup possible as intent operation? */
825 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
828 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
830 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
833 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
834 repbody->fid1 = *child_fid;
835 repbody->valid = OBD_MD_FLID;
841 * UPDATE lock should be taken against parent, and be release before exit;
842 * child_bits lock should be taken against child, and be returned back:
843 * (1)normal request should release the child lock;
844 * (2)intent request will grant the lock to client.
846 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
847 struct mdt_lock_handle *lhc,
849 struct ldlm_reply *ldlm_rep)
851 struct ptlrpc_request *req = mdt_info_req(info);
852 struct mdt_body *reqbody = NULL;
853 struct mdt_object *parent = info->mti_object;
854 struct mdt_object *child;
855 struct md_object *next = mdt_object_child(parent);
856 struct lu_fid *child_fid = &info->mti_tmp_fid1;
857 struct lu_name *lname = NULL;
858 const char *name = NULL;
860 struct mdt_lock_handle *lhp = NULL;
861 struct ldlm_lock *lock;
862 struct ldlm_res_id *res_id;
869 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
870 LASSERT(ergo(is_resent,
871 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
873 LASSERT(parent != NULL);
874 name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
876 RETURN(err_serious(-EFAULT));
878 namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
880 if (!info->mti_cross_ref) {
882 * XXX: Check for "namelen == 0" is for getattr by fid
883 * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
884 * that is the name must contain at least one character and
885 * the terminating '\0'
888 reqbody = req_capsule_client_get(info->mti_pill,
890 if (unlikely(reqbody == NULL))
891 RETURN(err_serious(-EFAULT));
893 if (unlikely(!fid_is_sane(&reqbody->fid2)))
894 RETURN(err_serious(-EINVAL));
897 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
899 PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
902 lname = mdt_name(info->mti_env, (char *)name, namelen);
903 CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
904 "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
908 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
910 rc = mdt_object_exists(parent);
911 if (unlikely(rc == 0)) {
912 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
913 &parent->mot_obj.mo_lu,
914 "Parent doesn't exist!\n");
916 } else if (!info->mti_cross_ref) {
917 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
918 PFID(mdt_object_fid(parent)));
921 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
929 if (info->mti_cross_ref) {
930 /* Only getattr on the child. Parent is on another node. */
931 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
933 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
934 "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
937 /* Do not take lock for resent case. */
938 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
939 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
940 lhc->mlh_reg_lh.cookie);
941 LASSERT(fid_res_name_eq(mdt_object_fid(child),
942 &lock->l_resource->lr_name));
946 mdt_lock_handle_init(lhc);
947 mdt_lock_reg_init(lhc, LCK_PR);
950 * Object's name is on another MDS, no lookup lock is
951 * needed here but update is.
953 child_bits &= ~MDS_INODELOCK_LOOKUP;
954 child_bits |= MDS_INODELOCK_UPDATE;
956 rc = mdt_object_lock(info, child, lhc, child_bits,
960 /* Finally, we can get attr for child. */
961 mdt_set_capainfo(info, 0, mdt_object_fid(child),
963 rc = mdt_getattr_internal(info, child, 0);
964 if (unlikely(rc != 0))
965 mdt_object_unlock(info, child, lhc, 1);
971 /* step 1: lock parent */
972 lhp = &info->mti_lh[MDT_LH_PARENT];
973 mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
974 rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
976 if (unlikely(rc != 0))
979 /* step 2: lookup child's fid by name */
980 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
985 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
986 GOTO(out_parent, rc);
988 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
990 *child_fid = reqbody->fid2;
991 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
995 *step 3: find the child object by fid & lock it.
996 * regardless if it is local or remote.
998 child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
1000 if (unlikely(IS_ERR(child)))
1001 GOTO(out_parent, rc = PTR_ERR(child));
1003 /* Do not take lock for resent case. */
1004 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1005 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1006 lhc->mlh_reg_lh.cookie);
1008 res_id = &lock->l_resource->lr_name;
1009 if (!fid_res_name_eq(mdt_object_fid(child),
1010 &lock->l_resource->lr_name)) {
1011 LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1012 &lock->l_resource->lr_name),
1013 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1014 (unsigned long)res_id->name[0],
1015 (unsigned long)res_id->name[1],
1016 (unsigned long)res_id->name[2],
1017 PFID(mdt_object_fid(parent)));
1018 CWARN("Although resent, but still not get child lock"
1019 "parent:"DFID" child:"DFID"\n",
1020 PFID(mdt_object_fid(parent)),
1021 PFID(mdt_object_fid(child)));
1022 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1023 LDLM_LOCK_PUT(lock);
1026 LDLM_LOCK_PUT(lock);
1030 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1031 mdt_lock_handle_init(lhc);
1032 mdt_lock_reg_init(lhc, LCK_PR);
1034 if (mdt_object_exists(child) == 0) {
1035 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
1036 &child->mot_obj.mo_lu,
1037 "Object doesn't exist!\n");
1038 GOTO(out_child, rc = -ENOENT);
1041 if (!(child_bits & MDS_INODELOCK_UPDATE)) {
1042 struct md_attr *ma = &info->mti_attr;
1045 ma->ma_need = MA_INODE;
1046 rc = mo_attr_get(info->mti_env,
1047 mdt_object_child(child), ma);
1048 if (unlikely(rc != 0))
1049 GOTO(out_child, rc);
1051 /* If the file has not been changed for some time, we
1052 * return not only a LOOKUP lock, but also an UPDATE
1053 * lock and this might save us RPC on later STAT. For
1054 * directories, it also let negative dentry starts
1055 * working for this dir. */
1056 if (ma->ma_valid & MA_INODE &&
1057 ma->ma_attr.la_valid & LA_CTIME &&
1058 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1059 ma->ma_attr.la_ctime < cfs_time_current_sec())
1060 child_bits |= MDS_INODELOCK_UPDATE;
1063 rc = mdt_object_lock(info, child, lhc, child_bits,
1066 if (unlikely(rc != 0))
1067 GOTO(out_child, rc);
1070 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1071 /* Get MA_SOM attributes if update lock is given. */
1073 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1074 S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1077 /* finally, we can get attr for child. */
1078 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1079 rc = mdt_getattr_internal(info, child, ma_need);
1080 if (unlikely(rc != 0)) {
1081 mdt_object_unlock(info, child, lhc, 1);
1083 /* Debugging code. */
1084 res_id = &lock->l_resource->lr_name;
1085 LDLM_DEBUG(lock, "Returning lock to client");
1086 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1087 &lock->l_resource->lr_name),
1088 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1089 (unsigned long)res_id->name[0],
1090 (unsigned long)res_id->name[1],
1091 (unsigned long)res_id->name[2],
1092 PFID(mdt_object_fid(child)));
1093 mdt_pack_size2body(info, child);
1096 LDLM_LOCK_PUT(lock);
1100 mdt_object_put(info->mti_env, child);
1103 mdt_object_unlock(info, parent, lhp, 1);
1107 /* normal handler: should release the child lock */
1108 static int mdt_getattr_name(struct mdt_thread_info *info)
1110 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1111 struct mdt_body *reqbody;
1112 struct mdt_body *repbody;
1116 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1117 LASSERT(reqbody != NULL);
1118 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1119 LASSERT(repbody != NULL);
1121 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1122 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1123 repbody->eadatasize = 0;
1124 repbody->aclsize = 0;
1126 rc = mdt_init_ucred(info, reqbody);
1128 GOTO(out_shrink, rc);
1130 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1131 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1132 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1133 lhc->mlh_reg_lh.cookie = 0;
1135 mdt_exit_ucred(info);
1138 mdt_client_compatibility(info);
1139 mdt_shrink_reply(info);
1143 static const struct lu_device_operations mdt_lu_ops;
1145 static int lu_device_is_mdt(struct lu_device *d)
1147 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1150 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1151 void *karg, void *uarg);
1153 static int mdt_set_info(struct mdt_thread_info *info)
1155 struct ptlrpc_request *req = mdt_info_req(info);
1158 int keylen, vallen, rc = 0;
1161 rc = req_capsule_server_pack(info->mti_pill);
1165 key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1167 DEBUG_REQ(D_HA, req, "no set_info key");
1171 keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1174 val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1176 DEBUG_REQ(D_HA, req, "no set_info val");
1180 vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1183 /* Swab any part of val you need to here */
1184 if (KEY_IS(KEY_READ_ONLY)) {
1186 lustre_msg_set_status(req->rq_repmsg, 0);
1188 cfs_spin_lock(&req->rq_export->exp_lock);
1190 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1192 req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
1193 cfs_spin_unlock(&req->rq_export->exp_lock);
1195 } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1196 struct changelog_setinfo *cs =
1197 (struct changelog_setinfo *)val;
1198 if (vallen != sizeof(*cs)) {
1199 CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1202 if (ptlrpc_req_need_swab(req)) {
1203 __swab64s(&cs->cs_recno);
1204 __swab32s(&cs->cs_id);
1207 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1209 lustre_msg_set_status(req->rq_repmsg, rc);
1217 static int mdt_connect(struct mdt_thread_info *info)
1220 struct ptlrpc_request *req;
1222 req = mdt_info_req(info);
1223 rc = target_handle_connect(req);
1225 LASSERT(req->rq_export != NULL);
1226 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1227 rc = mdt_init_sec_level(info);
1229 rc = mdt_init_idmap(info);
1231 obd_disconnect(class_export_get(req->rq_export));
1233 rc = err_serious(rc);
1238 static int mdt_disconnect(struct mdt_thread_info *info)
1243 rc = target_handle_disconnect(mdt_info_req(info));
1245 rc = err_serious(rc);
1249 static int mdt_sendpage(struct mdt_thread_info *info,
1250 struct lu_rdpg *rdpg, int nob)
1252 struct ptlrpc_request *req = mdt_info_req(info);
1253 struct obd_export *exp = req->rq_export;
1254 struct ptlrpc_bulk_desc *desc;
1255 struct l_wait_info *lwi = &info->mti_u.rdpg.mti_wait_info;
1262 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1267 for (i = 0, tmpcount = nob;
1268 i < rdpg->rp_npages && tmpcount > 0; i++, tmpcount -= tmpsize) {
1269 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1270 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1273 LASSERT(desc->bd_nob == nob);
1274 rc = target_bulk_io(exp, desc, lwi);
1275 ptlrpc_free_bulk(desc);
1279 #ifdef HAVE_SPLIT_SUPPORT
1281 * Retrieve dir entry from the page and insert it to the slave object, actually,
1282 * this should be in osd layer, but since it will not in the final product, so
1283 * just do it here and do not define more moo api anymore for this.
1285 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1288 struct mdt_object *object = info->mti_object;
1289 struct lu_fid *lf = &info->mti_tmp_fid2;
1290 struct md_attr *ma = &info->mti_attr;
1291 struct lu_dirpage *dp;
1292 struct lu_dirent *ent;
1293 int rc = 0, offset = 0;
1296 /* Make sure we have at least one entry. */
1301 * Disable trans for this name insert, since it will include many trans
1304 info->mti_no_need_trans = 1;
1306 * When write_dir_page, no need update parent's ctime,
1307 * and no permission check for name_insert.
1309 ma->ma_attr.la_ctime = 0;
1310 ma->ma_attr.la_valid = LA_MODE;
1311 ma->ma_valid = MA_INODE;
1314 dp = page_address(page);
1315 offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1317 for (ent = lu_dirent_start(dp); ent != NULL;
1318 ent = lu_dirent_next(ent)) {
1319 struct lu_name *lname;
1322 if (le16_to_cpu(ent->lde_namelen) == 0)
1325 fid_le_to_cpu(lf, &ent->lde_fid);
1326 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1327 ma->ma_attr.la_mode = S_IFDIR;
1329 ma->ma_attr.la_mode = 0;
1330 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1332 GOTO(out, rc = -ENOMEM);
1334 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1335 lname = mdt_name(info->mti_env, name,
1336 le16_to_cpu(ent->lde_namelen));
1337 ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE);
1338 rc = mdo_name_insert(info->mti_env,
1339 md_object_next(&object->mot_obj),
1341 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1343 CERROR("Can't insert %*.*s, rc %d\n",
1344 le16_to_cpu(ent->lde_namelen),
1345 le16_to_cpu(ent->lde_namelen),
1350 offset += lu_dirent_size(ent);
1360 static int mdt_bulk_timeout(void *data)
1364 CERROR("mdt bulk transfer timeout \n");
1369 static int mdt_writepage(struct mdt_thread_info *info)
1371 struct ptlrpc_request *req = mdt_info_req(info);
1372 struct mdt_body *reqbody;
1373 struct l_wait_info *lwi;
1374 struct ptlrpc_bulk_desc *desc;
1380 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1381 if (reqbody == NULL)
1382 RETURN(err_serious(-EFAULT));
1384 desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1386 RETURN(err_serious(-ENOMEM));
1388 /* allocate the page for the desc */
1389 page = cfs_alloc_page(CFS_ALLOC_STD);
1391 GOTO(desc_cleanup, rc = -ENOMEM);
1393 CDEBUG(D_INFO, "Received page offset %d size %d \n",
1394 (int)reqbody->size, (int)reqbody->nlink);
1396 ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1397 (int)reqbody->nlink);
1399 rc = sptlrpc_svc_prep_bulk(req, desc);
1401 GOTO(cleanup_page, rc);
1403 * Check if client was evicted while we were doing i/o before touching
1408 GOTO(cleanup_page, rc = -ENOMEM);
1410 if (desc->bd_export->exp_failed)
1413 rc = ptlrpc_start_bulk_transfer (desc);
1415 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * CFS_HZ / 4, CFS_HZ,
1416 mdt_bulk_timeout, desc);
1417 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1418 desc->bd_export->exp_failed, lwi);
1419 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1420 if (rc == -ETIMEDOUT) {
1421 DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1422 ptlrpc_abort_bulk(desc);
1423 } else if (desc->bd_export->exp_failed) {
1424 DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1426 ptlrpc_abort_bulk(desc);
1427 } else if (!desc->bd_success ||
1428 desc->bd_nob_transferred != desc->bd_nob) {
1429 DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1431 "truncated" : "network error on",
1432 desc->bd_nob_transferred, desc->bd_nob);
1433 /* XXX should this be a different errno? */
1437 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1440 GOTO(cleanup_lwi, rc);
1441 rc = mdt_write_dir_page(info, page, reqbody->nlink);
1446 cfs_free_page(page);
1448 ptlrpc_free_bulk(desc);
1453 static int mdt_readpage(struct mdt_thread_info *info)
1455 struct mdt_object *object = info->mti_object;
1456 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1457 struct mdt_body *reqbody;
1458 struct mdt_body *repbody;
1463 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1464 RETURN(err_serious(-ENOMEM));
1466 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1467 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1468 if (reqbody == NULL || repbody == NULL)
1469 RETURN(err_serious(-EFAULT));
1472 * prepare @rdpg before calling lower layers and transfer itself. Here
1473 * reqbody->size contains offset of where to start to read and
1474 * reqbody->nlink contains number bytes to read.
1476 rdpg->rp_hash = reqbody->size;
1477 if (rdpg->rp_hash != reqbody->size) {
1478 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1479 rdpg->rp_hash, reqbody->size);
1483 rdpg->rp_attrs = reqbody->mode;
1484 if (info->mti_exp->exp_connect_flags & OBD_CONNECT_64BITHASH)
1485 rdpg->rp_attrs |= LUDA_64BITHASH;
1486 rdpg->rp_count = min_t(unsigned int, reqbody->nlink,
1487 PTLRPC_MAX_BRW_SIZE);
1488 rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1) >>
1490 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1491 if (rdpg->rp_pages == NULL)
1494 for (i = 0; i < rdpg->rp_npages; ++i) {
1495 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1496 if (rdpg->rp_pages[i] == NULL)
1497 GOTO(free_rdpg, rc = -ENOMEM);
1500 /* call lower layers to fill allocated pages with directory data */
1501 rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1503 GOTO(free_rdpg, rc);
1505 /* send pages to client */
1506 rc = mdt_sendpage(info, rdpg, rc);
1511 for (i = 0; i < rdpg->rp_npages; i++)
1512 if (rdpg->rp_pages[i] != NULL)
1513 cfs_free_page(rdpg->rp_pages[i]);
1514 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1516 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1522 static int mdt_reint_internal(struct mdt_thread_info *info,
1523 struct mdt_lock_handle *lhc,
1526 struct req_capsule *pill = info->mti_pill;
1527 struct mdt_device *mdt = info->mti_mdt;
1528 struct md_quota *mq = md_quota(info->mti_env);
1529 struct mdt_body *repbody;
1534 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1535 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1536 mdt->mdt_max_mdsize);
1537 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1538 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1539 mdt->mdt_max_cookiesize);
1541 rc = req_capsule_server_pack(pill);
1543 CERROR("Can't pack response, rc %d\n", rc);
1544 RETURN(err_serious(rc));
1547 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1548 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1550 repbody->eadatasize = 0;
1551 repbody->aclsize = 0;
1554 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1555 GOTO(out_shrink, rc = err_serious(-EFAULT));
1557 rc = mdt_reint_unpack(info, op);
1559 CERROR("Can't unpack reint, rc %d\n", rc);
1560 GOTO(out_shrink, rc = err_serious(rc));
1563 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1565 /* for replay no cookkie / lmm need, because client have this already */
1566 if (info->mti_spec.no_create == 1) {
1567 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1568 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1570 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1571 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1575 rc = mdt_init_ucred_reint(info);
1577 GOTO(out_shrink, rc);
1579 rc = mdt_fix_attr_ucred(info, op);
1581 GOTO(out_ucred, rc = err_serious(rc));
1583 if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1584 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1585 GOTO(out_ucred, rc);
1587 mq->mq_exp = info->mti_exp;
1588 rc = mdt_reint_rec(info, lhc);
1591 mdt_exit_ucred(info);
1593 mdt_client_compatibility(info);
1594 mdt_shrink_reply(info);
1598 static long mdt_reint_opcode(struct mdt_thread_info *info,
1599 const struct req_format **fmt)
1601 struct mdt_rec_reint *rec;
1604 opc = err_serious(-EFAULT);
1605 rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1607 opc = rec->rr_opcode;
1608 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1609 if (opc < REINT_MAX && fmt[opc] != NULL)
1610 req_capsule_extend(info->mti_pill, fmt[opc]);
1612 CERROR("Unsupported opc: %ld\n", opc);
1613 opc = err_serious(opc);
1619 static int mdt_reint(struct mdt_thread_info *info)
1624 static const struct req_format *reint_fmts[REINT_MAX] = {
1625 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1626 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1627 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1628 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1629 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1630 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1631 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1636 opc = mdt_reint_opcode(info, reint_fmts);
1639 * No lock possible here from client to pass it to reint code
1642 rc = mdt_reint_internal(info, NULL, opc);
1647 info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1651 /* this should sync the whole device */
1652 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1654 struct dt_device *dt = mdt->mdt_bottom;
1658 rc = dt->dd_ops->dt_sync(env, dt);
1662 /* this should sync this object */
1663 static int mdt_object_sync(struct mdt_thread_info *info)
1665 struct md_object *next;
1669 if (!mdt_object_exists(info->mti_object)) {
1670 CWARN("Non existing object "DFID"!\n",
1671 PFID(mdt_object_fid(info->mti_object)));
1674 next = mdt_object_child(info->mti_object);
1675 rc = mo_object_sync(info->mti_env, next);
1680 static int mdt_sync(struct mdt_thread_info *info)
1682 struct ptlrpc_request *req = mdt_info_req(info);
1683 struct req_capsule *pill = info->mti_pill;
1684 struct mdt_body *body;
1688 /* The fid may be zero, so we req_capsule_set manually */
1689 req_capsule_set(pill, &RQF_MDS_SYNC);
1691 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1693 RETURN(err_serious(-EINVAL));
1695 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1696 RETURN(err_serious(-ENOMEM));
1698 if (fid_seq(&body->fid1) == 0) {
1699 /* sync the whole device */
1700 rc = req_capsule_server_pack(pill);
1702 rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1704 rc = err_serious(rc);
1706 /* sync an object */
1707 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1709 rc = mdt_object_sync(info);
1711 struct md_object *next;
1712 const struct lu_fid *fid;
1713 struct lu_attr *la = &info->mti_attr.ma_attr;
1715 next = mdt_object_child(info->mti_object);
1716 info->mti_attr.ma_need = MA_INODE;
1717 info->mti_attr.ma_valid = 0;
1718 rc = mo_attr_get(info->mti_env, next,
1721 body = req_capsule_server_get(pill,
1723 fid = mdt_object_fid(info->mti_object);
1724 mdt_pack_attr2body(info, body, la, fid);
1728 rc = err_serious(rc);
1731 mdt_counter_incr(req->rq_export, LPROC_MDT_SYNC);
1736 #ifdef HAVE_QUOTA_SUPPORT
1737 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1739 struct obd_quotactl *oqctl;
1740 struct req_capsule *pill = info->mti_pill;
1741 struct obd_export *exp = info->mti_exp;
1742 struct md_quota *mq = md_quota(info->mti_env);
1743 struct md_device *next = info->mti_mdt->mdt_child;
1747 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1751 /* remote client has no permission for quotacheck */
1752 if (unlikely(exp_connect_rmtclient(exp)))
1755 rc = req_capsule_server_pack(pill);
1760 rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next,
1765 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1767 struct obd_quotactl *oqctl, *repoqc;
1768 struct req_capsule *pill = info->mti_pill;
1769 struct obd_export *exp = info->mti_exp;
1770 struct md_quota *mq = md_quota(info->mti_env);
1771 struct md_device *next = info->mti_mdt->mdt_child;
1772 const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
1776 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1781 if (exp_connect_rmtclient(exp)) {
1782 struct ptlrpc_request *req = mdt_info_req(info);
1783 struct mdt_export_data *med = mdt_req2med(req);
1784 struct lustre_idmap_table *idmap = med->med_idmap;
1786 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1787 oqctl->qc_cmd != Q_GETINFO))
1791 if (oqctl->qc_type == USRQUOTA)
1792 id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1794 else if (oqctl->qc_type == GRPQUOTA)
1795 id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1800 if (id == CFS_IDMAP_NOTFOUND) {
1801 CDEBUG(D_QUOTA, "no mapping for id %u\n",
1807 rc = req_capsule_server_pack(pill);
1811 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1812 LASSERT(repoqc != NULL);
1815 switch (oqctl->qc_cmd) {
1817 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
1820 rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
1823 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
1827 rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id,
1831 rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id,
1835 rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id,
1839 rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id,
1843 rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id,
1846 case LUSTRE_Q_INVALIDATE:
1847 rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type);
1849 case LUSTRE_Q_FINVALIDATE:
1850 rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type);
1853 CERROR("unsupported mdt_quotactl command: %d\n",
1865 * OBD PING and other handlers.
1867 static int mdt_obd_ping(struct mdt_thread_info *info)
1872 req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1874 rc = target_handle_ping(mdt_info_req(info));
1876 rc = err_serious(rc);
1880 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1882 return err_serious(-EOPNOTSUPP);
1885 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1887 return err_serious(-EOPNOTSUPP);
1895 /** clone llog ctxt from child (mdd)
1896 * This allows remote llog (replicator) access.
1897 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
1898 * context was originally set up, or we can handle them directly.
1899 * I choose the latter, but that means I need any llog
1900 * contexts set up by child to be accessable by the mdt. So we clone the
1901 * context into our context list here.
1903 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
1906 struct md_device *next = mdt->mdt_child;
1907 struct llog_ctxt *ctxt;
1910 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
1913 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
1914 if (rc || ctxt == NULL) {
1915 CERROR("Can't get mdd ctxt %d\n", rc);
1919 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
1921 CERROR("Can't set mdt ctxt %d\n", rc);
1926 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
1927 struct mdt_device *mdt, int idx)
1929 struct llog_ctxt *ctxt;
1931 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
1934 /* Put once for the get we just did, and once for the clone */
1935 llog_ctxt_put(ctxt);
1936 llog_ctxt_put(ctxt);
1940 static int mdt_llog_create(struct mdt_thread_info *info)
1944 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
1945 rc = llog_origin_handle_create(mdt_info_req(info));
1946 return (rc < 0 ? err_serious(rc) : rc);
1949 static int mdt_llog_destroy(struct mdt_thread_info *info)
1953 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
1954 rc = llog_origin_handle_destroy(mdt_info_req(info));
1955 return (rc < 0 ? err_serious(rc) : rc);
1958 static int mdt_llog_read_header(struct mdt_thread_info *info)
1962 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
1963 rc = llog_origin_handle_read_header(mdt_info_req(info));
1964 return (rc < 0 ? err_serious(rc) : rc);
1967 static int mdt_llog_next_block(struct mdt_thread_info *info)
1971 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
1972 rc = llog_origin_handle_next_block(mdt_info_req(info));
1973 return (rc < 0 ? err_serious(rc) : rc);
1976 static int mdt_llog_prev_block(struct mdt_thread_info *info)
1980 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
1981 rc = llog_origin_handle_prev_block(mdt_info_req(info));
1982 return (rc < 0 ? err_serious(rc) : rc);
1989 static struct ldlm_callback_suite cbs = {
1990 .lcs_completion = ldlm_server_completion_ast,
1991 .lcs_blocking = ldlm_server_blocking_ast,
1995 static int mdt_enqueue(struct mdt_thread_info *info)
1997 struct ptlrpc_request *req;
2001 * info->mti_dlm_req already contains swapped and (if necessary)
2002 * converted dlm request.
2004 LASSERT(info->mti_dlm_req != NULL);
2006 req = mdt_info_req(info);
2007 rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
2008 req, info->mti_dlm_req, &cbs);
2009 info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
2010 return rc ? err_serious(rc) : req->rq_status;
2013 static int mdt_convert(struct mdt_thread_info *info)
2016 struct ptlrpc_request *req;
2018 LASSERT(info->mti_dlm_req);
2019 req = mdt_info_req(info);
2020 rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2021 return rc ? err_serious(rc) : req->rq_status;
2024 static int mdt_bl_callback(struct mdt_thread_info *info)
2026 CERROR("bl callbacks should not happen on MDS\n");
2028 return err_serious(-EOPNOTSUPP);
2031 static int mdt_cp_callback(struct mdt_thread_info *info)
2033 CERROR("cp callbacks should not happen on MDS\n");
2035 return err_serious(-EOPNOTSUPP);
2039 * sec context handlers
2041 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2045 rc = mdt_handle_idmap(info);
2048 struct ptlrpc_request *req = mdt_info_req(info);
2051 opc = lustre_msg_get_opc(req->rq_reqmsg);
2052 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2053 sptlrpc_svc_ctx_invalidate(req);
2056 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2061 static struct mdt_object *mdt_obj(struct lu_object *o)
2063 LASSERT(lu_device_is_mdt(o->lo_dev));
2064 return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2067 struct mdt_object *mdt_object_find(const struct lu_env *env,
2068 struct mdt_device *d,
2069 const struct lu_fid *f)
2071 struct lu_object *o;
2072 struct mdt_object *m;
2075 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2076 o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2077 if (unlikely(IS_ERR(o)))
2078 m = (struct mdt_object *)o;
2085 * Asyncronous commit for mdt device.
2087 * Pass asynchonous commit call down the MDS stack.
2089 * \param env environment
2090 * \param mdt the mdt device
2092 static void mdt_device_commit_async(const struct lu_env *env,
2093 struct mdt_device *mdt)
2095 struct dt_device *dt = mdt->mdt_bottom;
2098 rc = dt->dd_ops->dt_commit_async(env, dt);
2099 if (unlikely(rc != 0))
2100 CWARN("async commit start failed with rc = %d", rc);
2104 * Mark the lock as "synchonous".
2106 * Mark the lock to deffer transaction commit to the unlock time.
2108 * \param lock the lock to mark as "synchonous"
2110 * \see mdt_is_lock_sync
2111 * \see mdt_save_lock
2113 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2115 lock->l_ast_data = (void*)1;
2119 * Check whehter the lock "synchonous" or not.
2121 * \param lock the lock to check
2122 * \retval 1 the lock is "synchonous"
2123 * \retval 0 the lock isn't "synchronous"
2125 * \see mdt_set_lock_sync
2126 * \see mdt_save_lock
2128 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2130 return lock->l_ast_data != NULL;
2134 * Blocking AST for mdt locks.
2136 * Starts transaction commit if in case of COS lock conflict or
2137 * deffers such a commit to the mdt_save_lock.
2139 * \param lock the lock which blocks a request or cancelling lock
2140 * \param desc unused
2141 * \param data unused
2142 * \param flag indicates whether this cancelling or blocking callback
2144 * \see ldlm_blocking_ast_nocheck
2146 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2147 void *data, int flag)
2149 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2150 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2154 if (flag == LDLM_CB_CANCELING)
2156 lock_res_and_lock(lock);
2157 if (lock->l_blocking_ast != mdt_blocking_ast) {
2158 unlock_res_and_lock(lock);
2161 if (mdt_cos_is_enabled(mdt) &&
2162 lock->l_req_mode & (LCK_PW | LCK_EX) &&
2163 lock->l_blocking_lock != NULL &&
2164 lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2165 mdt_set_lock_sync(lock);
2167 rc = ldlm_blocking_ast_nocheck(lock);
2169 /* There is no lock conflict if l_blocking_lock == NULL,
2170 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2171 * when the last reference to a local lock was released */
2172 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2175 rc = lu_env_init(&env, LCT_MD_THREAD);
2176 if (unlikely(rc != 0))
2177 CWARN("lu_env initialization failed with rc = %d,"
2178 "cannot start asynchronous commit\n", rc);
2180 mdt_device_commit_async(&env, mdt);
2186 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2187 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2189 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2190 ldlm_policy_data_t *policy = &info->mti_policy;
2191 struct ldlm_res_id *res_id = &info->mti_res_id;
2195 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2196 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2197 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2198 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2200 if (mdt_object_exists(o) < 0) {
2201 if (locality == MDT_CROSS_LOCK) {
2202 /* cross-ref object fix */
2203 ibits &= ~MDS_INODELOCK_UPDATE;
2204 ibits |= MDS_INODELOCK_LOOKUP;
2206 LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2207 LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2209 /* No PDO lock on remote object */
2210 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2213 if (lh->mlh_type == MDT_PDO_LOCK) {
2214 /* check for exists after object is locked */
2215 if (mdt_object_exists(o) == 0) {
2216 /* Non-existent object shouldn't have PDO lock */
2219 /* Non-dir object shouldn't have PDO lock */
2220 if (!S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)))
2225 memset(policy, 0, sizeof(*policy));
2226 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2229 * Take PDO lock on whole directory and build correct @res_id for lock
2230 * on part of directory.
2232 if (lh->mlh_pdo_hash != 0) {
2233 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2234 mdt_lock_pdo_mode(info, o, lh);
2235 if (lh->mlh_pdo_mode != LCK_NL) {
2237 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2238 * is never going to be sent to client and we do not
2239 * want it slowed down due to possible cancels.
2241 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2242 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2243 policy, res_id, LDLM_FL_ATOMIC_CB,
2244 &info->mti_exp->exp_handle.h_cookie);
2250 * Finish res_id initializing by name hash marking part of
2251 * directory which is taking modification.
2253 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2256 policy->l_inodebits.bits = ibits;
2259 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2260 * going to be sent to client. If it is - mdt_intent_policy() path will
2261 * fix it up and turn FL_LOCAL flag off.
2263 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2264 res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
2265 &info->mti_exp->exp_handle.h_cookie);
2267 mdt_object_unlock(info, o, lh, 1);
2268 else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2269 lh->mlh_pdo_hash != 0 &&
2270 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2271 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2278 * Save a lock within request object.
2280 * Keep the lock referenced until whether client ACK or transaction
2281 * commit happens or release the lock immediately depending on input
2282 * parameters. If COS is ON, a write lock is converted to COS lock
2285 * \param info thead info object
2286 * \param h lock handle
2287 * \param mode lock mode
2288 * \param decref force immediate lock releasing
2291 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2292 ldlm_mode_t mode, int decref)
2296 if (lustre_handle_is_used(h)) {
2297 if (decref || !info->mti_has_trans ||
2298 !(mode & (LCK_PW | LCK_EX))){
2299 mdt_fid_unlock(h, mode);
2301 struct mdt_device *mdt = info->mti_mdt;
2302 struct ldlm_lock *lock = ldlm_handle2lock(h);
2303 struct ptlrpc_request *req = mdt_info_req(info);
2306 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2308 CDEBUG(D_HA, "request = %p reply state = %p"
2309 " transno = "LPD64"\n",
2310 req, req->rq_reply_state, req->rq_transno);
2311 if (mdt_cos_is_enabled(mdt)) {
2313 ldlm_lock_downgrade(lock, LCK_COS);
2316 ptlrpc_save_lock(req, h, mode, no_ack);
2317 if (mdt_is_lock_sync(lock)) {
2318 CDEBUG(D_HA, "found sync-lock,"
2319 " async commit started\n");
2320 mdt_device_commit_async(info->mti_env,
2323 LDLM_LOCK_PUT(lock);
2332 * Unlock mdt object.
2334 * Immeditely release the regular lock and the PDO lock or save the
2335 * lock in reqeuest and keep them referenced until client ACK or
2336 * transaction commit.
2338 * \param info thread info object
2339 * \param o mdt object
2340 * \param lh mdt lock handle referencing regular and PDO locks
2341 * \param decref force immediate lock releasing
2343 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2344 struct mdt_lock_handle *lh, int decref)
2348 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2349 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2354 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2355 const struct lu_fid *f,
2356 struct mdt_lock_handle *lh,
2359 struct mdt_object *o;
2361 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2365 rc = mdt_object_lock(info, o, lh, ibits,
2368 mdt_object_put(info->mti_env, o);
2375 void mdt_object_unlock_put(struct mdt_thread_info * info,
2376 struct mdt_object * o,
2377 struct mdt_lock_handle *lh,
2380 mdt_object_unlock(info, o, lh, decref);
2381 mdt_object_put(info->mti_env, o);
2384 static struct mdt_handler *mdt_handler_find(__u32 opc,
2385 struct mdt_opc_slice *supported)
2387 struct mdt_opc_slice *s;
2388 struct mdt_handler *h;
2391 for (s = supported; s->mos_hs != NULL; s++) {
2392 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2393 h = s->mos_hs + (opc - s->mos_opc_start);
2394 if (likely(h->mh_opc != 0))
2395 LASSERTF(h->mh_opc == opc,
2396 "opcode mismatch %d != %d\n",
2399 h = NULL; /* unsupported opc */
2406 static int mdt_lock_resname_compat(struct mdt_device *m,
2407 struct ldlm_request *req)
2409 /* XXX something... later. */
2413 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2415 /* XXX something... later. */
2420 * Generic code handling requests that have struct mdt_body passed in:
2422 * - extract mdt_body from request and save it in @info, if present;
2424 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2427 * - if HABEO_CORPUS flag is set for this request type check whether object
2428 * actually exists on storage (lu_object_exists()).
2431 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2433 const struct mdt_body *body;
2434 struct mdt_object *obj;
2435 const struct lu_env *env;
2436 struct req_capsule *pill;
2440 env = info->mti_env;
2441 pill = info->mti_pill;
2443 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2447 if (!(body->valid & OBD_MD_FLID))
2450 if (!fid_is_sane(&body->fid1)) {
2451 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2456 * Do not get size or any capa fields before we check that request
2457 * contains capa actually. There are some requests which do not, for
2458 * instance MDS_IS_SUBDIR.
2460 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2461 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2462 mdt_set_capainfo(info, 0, &body->fid1,
2463 req_capsule_client_get(pill, &RMF_CAPA1));
2465 obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2467 if ((flags & HABEO_CORPUS) &&
2468 !mdt_object_exists(obj)) {
2469 mdt_object_put(env, obj);
2470 /* for capability renew ENOENT will be handled in
2472 if (body->valid & OBD_MD_FLOSSCAPA)
2477 info->mti_object = obj;
2486 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2488 struct req_capsule *pill = info->mti_pill;
2492 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2493 rc = mdt_body_unpack(info, flags);
2497 if (rc == 0 && (flags & HABEO_REFERO)) {
2498 struct mdt_device *mdt = info->mti_mdt;
2502 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2503 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2504 mdt->mdt_max_mdsize);
2505 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2506 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2507 mdt->mdt_max_cookiesize);
2509 rc = req_capsule_server_pack(pill);
2514 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2516 struct md_device *next = m->mdt_child;
2518 return next->md_ops->mdo_init_capa_ctxt(env, next,
2519 m->mdt_opts.mo_mds_capa,
2520 m->mdt_capa_timeout,
2526 * Invoke handler for this request opc. Also do necessary preprocessing
2527 * (according to handler ->mh_flags), and post-processing (setting of
2528 * ->last_{xid,committed}).
2530 static int mdt_req_handle(struct mdt_thread_info *info,
2531 struct mdt_handler *h, struct ptlrpc_request *req)
2533 int rc, serious = 0;
2538 LASSERT(h->mh_act != NULL);
2539 LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2540 LASSERT(current->journal_info == NULL);
2543 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2544 * to put same checks into handlers like mdt_close(), mdt_reint(),
2545 * etc., without talking to mdt authors first. Checking same thing
2546 * there again is useless and returning 0 error without packing reply
2547 * is buggy! Handlers either pack reply or return error.
2549 * We return 0 here and do not send any reply in order to emulate
2550 * network failure. Do not send any reply in case any of NET related
2551 * fail_id has occured.
2553 if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2557 flags = h->mh_flags;
2558 LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2560 if (h->mh_fmt != NULL) {
2561 req_capsule_set(info->mti_pill, h->mh_fmt);
2562 rc = mdt_unpack_req_pack_rep(info, flags);
2565 if (rc == 0 && flags & MUTABOR &&
2566 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2567 /* should it be rq_status? */
2570 if (rc == 0 && flags & HABEO_CLAVIS) {
2571 struct ldlm_request *dlm_req;
2573 LASSERT(h->mh_fmt != NULL);
2575 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2576 if (dlm_req != NULL) {
2577 if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2579 dlm_req->lock_desc.l_policy_data.\
2580 l_inodebits.bits == 0)) {
2582 * Lock without inodebits makes no sense and
2583 * will oops later in ldlm. If client miss to
2584 * set such bits, do not trigger ASSERTION.
2586 * For liblustre flock case, it maybe zero.
2590 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2591 rc = mdt_lock_resname_compat(
2594 info->mti_dlm_req = dlm_req;
2601 /* capability setting changed via /proc, needs reinitialize ctxt */
2602 if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2603 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2604 info->mti_mdt->mdt_capa_conf = 0;
2607 if (likely(rc == 0)) {
2609 * Process request, there can be two types of rc:
2610 * 1) errors with msg unpack/pack, other failures outside the
2611 * operation itself. This is counted as serious errors;
2612 * 2) errors during fs operation, should be placed in rq_status
2615 rc = h->mh_act(info);
2617 !req->rq_no_reply && req->rq_reply_state == NULL) {
2618 DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2619 "pack reply and returned 0 error\n",
2623 serious = is_serious(rc);
2624 rc = clear_serious(rc);
2628 req->rq_status = rc;
2631 * ELDLM_* codes which > 0 should be in rq_status only as well as
2632 * all non-serious errors.
2634 if (rc > 0 || !serious)
2637 LASSERT(current->journal_info == NULL);
2639 if (rc == 0 && (flags & HABEO_CLAVIS) &&
2640 info->mti_mdt->mdt_opts.mo_compat_resname) {
2641 struct ldlm_reply *dlmrep;
2643 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2645 rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2648 /* If we're DISCONNECTing, the mdt_export_data is already freed */
2649 if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2650 target_committed_to_req(req);
2652 if (unlikely(req_is_replay(req) &&
2653 lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2654 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2658 target_send_reply(req, rc, info->mti_fail_id);
2662 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2664 lh->mlh_type = MDT_NUL_LOCK;
2665 lh->mlh_reg_lh.cookie = 0ull;
2666 lh->mlh_reg_mode = LCK_MINMODE;
2667 lh->mlh_pdo_lh.cookie = 0ull;
2668 lh->mlh_pdo_mode = LCK_MINMODE;
2671 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2673 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2674 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2678 * Initialize fields of struct mdt_thread_info. Other fields are left in
2679 * uninitialized state, because it's too expensive to zero out whole
2680 * mdt_thread_info (> 1K) on each request arrival.
2682 static void mdt_thread_info_init(struct ptlrpc_request *req,
2683 struct mdt_thread_info *info)
2686 struct md_capainfo *ci;
2688 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2689 info->mti_pill = &req->rq_pill;
2692 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2693 mdt_lock_handle_init(&info->mti_lh[i]);
2695 /* mdt device: it can be NULL while CONNECT */
2696 if (req->rq_export) {
2697 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2698 info->mti_exp = req->rq_export;
2700 info->mti_mdt = NULL;
2701 info->mti_env = req->rq_svc_thread->t_env;
2702 ci = md_capainfo(info->mti_env);
2703 memset(ci, 0, sizeof *ci);
2704 if (req->rq_export) {
2705 if (exp_connect_rmtclient(req->rq_export))
2706 ci->mc_auth = LC_ID_CONVERT;
2707 else if (req->rq_export->exp_connect_flags &
2708 OBD_CONNECT_MDS_CAPA)
2709 ci->mc_auth = LC_ID_PLAIN;
2711 ci->mc_auth = LC_ID_NONE;
2714 info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2715 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2716 info->mti_mos = NULL;
2718 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2719 info->mti_body = NULL;
2720 info->mti_object = NULL;
2721 info->mti_dlm_req = NULL;
2722 info->mti_has_trans = 0;
2723 info->mti_no_need_trans = 0;
2724 info->mti_cross_ref = 0;
2725 info->mti_opdata = 0;
2727 /* To not check for split by default. */
2728 info->mti_spec.sp_ck_split = 0;
2729 info->mti_spec.no_create = 0;
2732 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2736 req_capsule_fini(info->mti_pill);
2737 if (info->mti_object != NULL) {
2739 * freeing an object may lead to OSD level transaction, do not
2740 * let it mess with MDT. bz19385.
2742 info->mti_no_need_trans = 1;
2743 mdt_object_put(info->mti_env, info->mti_object);
2744 info->mti_object = NULL;
2746 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2747 mdt_lock_handle_fini(&info->mti_lh[i]);
2748 info->mti_env = NULL;
2751 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2752 struct obd_device *obd, int *process)
2754 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2755 case MDS_CONNECT: /* This will never get here, but for completeness. */
2756 case OST_CONNECT: /* This will never get here, but for completeness. */
2757 case MDS_DISCONNECT:
2758 case OST_DISCONNECT:
2763 case MDS_DONE_WRITING:
2764 case MDS_SYNC: /* used in unmounting */
2770 *process = target_queue_recovery_request(req, obd);
2774 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2781 * Handle recovery. Return:
2782 * +1: continue request processing;
2783 * -ve: abort immediately with the given error code;
2784 * 0: send reply with error code in req->rq_status;
2786 static int mdt_recovery(struct mdt_thread_info *info)
2788 struct ptlrpc_request *req = mdt_info_req(info);
2789 struct obd_device *obd;
2793 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2796 case SEC_CTX_INIT_CONT:
2802 rc = mdt_handle_idmap(info);
2811 if (unlikely(!class_connected_export(req->rq_export))) {
2812 CERROR("operation %d on unconnected MDS from %s\n",
2813 lustre_msg_get_opc(req->rq_reqmsg),
2814 libcfs_id2str(req->rq_peer));
2815 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2816 * mds_A will get -ENOTCONN(especially for ping req),
2817 * which will cause that mds_A deactive timeout, then when
2818 * mds_A cleanup, the cleanup process will be suspended since
2819 * deactive timeout is not zero.
2821 req->rq_status = -ENOTCONN;
2822 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2826 /* sanity check: if the xid matches, the request must be marked as a
2827 * resent or replayed */
2828 if (req_xid_is_last(req)) {
2829 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2830 (MSG_RESENT | MSG_REPLAY))) {
2831 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2832 "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2833 lustre_msg_get_flags(req->rq_reqmsg));
2835 req->rq_status = -ENOTCONN;
2840 /* else: note the opposite is not always true; a RESENT req after a
2841 * failover will usually not match the last_xid, since it was likely
2842 * never committed. A REPLAYed request will almost never match the
2843 * last xid, however it could for a committed, but still retained,
2846 obd = req->rq_export->exp_obd;
2848 /* Check for aborted recovery... */
2849 if (unlikely(obd->obd_recovering)) {
2852 DEBUG_REQ(D_INFO, req, "Got new replay");
2853 rc = mdt_filter_recovery_request(req, obd, &should_process);
2854 if (rc != 0 || !should_process)
2856 else if (should_process < 0) {
2857 req->rq_status = should_process;
2858 rc = ptlrpc_error(req);
2865 static int mdt_msg_check_version(struct lustre_msg *msg)
2869 switch (lustre_msg_get_opc(msg)) {
2871 case MDS_DISCONNECT:
2874 case SEC_CTX_INIT_CONT:
2876 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2878 CERROR("bad opc %u version %08x, expecting %08x\n",
2879 lustre_msg_get_opc(msg),
2880 lustre_msg_get_version(msg),
2881 LUSTRE_OBD_VERSION);
2885 case MDS_GETATTR_NAME:
2892 case MDS_DONE_WRITING:
2899 case MDS_QUOTACHECK:
2905 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2907 CERROR("bad opc %u version %08x, expecting %08x\n",
2908 lustre_msg_get_opc(msg),
2909 lustre_msg_get_version(msg),
2910 LUSTRE_MDS_VERSION);
2914 case LDLM_BL_CALLBACK:
2915 case LDLM_CP_CALLBACK:
2916 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2918 CERROR("bad opc %u version %08x, expecting %08x\n",
2919 lustre_msg_get_opc(msg),
2920 lustre_msg_get_version(msg),
2921 LUSTRE_DLM_VERSION);
2923 case OBD_LOG_CANCEL:
2924 case LLOG_ORIGIN_HANDLE_CREATE:
2925 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2926 case LLOG_ORIGIN_HANDLE_READ_HEADER:
2927 case LLOG_ORIGIN_HANDLE_CLOSE:
2928 case LLOG_ORIGIN_HANDLE_DESTROY:
2929 case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2931 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2933 CERROR("bad opc %u version %08x, expecting %08x\n",
2934 lustre_msg_get_opc(msg),
2935 lustre_msg_get_version(msg),
2936 LUSTRE_LOG_VERSION);
2939 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2945 static int mdt_handle0(struct ptlrpc_request *req,
2946 struct mdt_thread_info *info,
2947 struct mdt_opc_slice *supported)
2949 struct mdt_handler *h;
2950 struct lustre_msg *msg;
2955 if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2958 LASSERT(current->journal_info == NULL);
2960 msg = req->rq_reqmsg;
2961 rc = mdt_msg_check_version(msg);
2962 if (likely(rc == 0)) {
2963 rc = mdt_recovery(info);
2964 if (likely(rc == +1)) {
2965 h = mdt_handler_find(lustre_msg_get_opc(msg),
2967 if (likely(h != NULL)) {
2968 rc = mdt_req_handle(info, h, req);
2970 CERROR("The unsupported opc: 0x%x\n",
2971 lustre_msg_get_opc(msg) );
2972 req->rq_status = -ENOTSUPP;
2973 rc = ptlrpc_error(req);
2978 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2983 * MDT handler function called by ptlrpc service thread when request comes.
2985 * XXX common "target" functionality should be factored into separate module
2986 * shared by mdt, ost and stand-alone services like fld.
2988 static int mdt_handle_common(struct ptlrpc_request *req,
2989 struct mdt_opc_slice *supported)
2992 struct mdt_thread_info *info;
2996 env = req->rq_svc_thread->t_env;
2997 LASSERT(env != NULL);
2998 LASSERT(env->le_ses != NULL);
2999 LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
3000 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3001 LASSERT(info != NULL);
3003 mdt_thread_info_init(req, info);
3005 rc = mdt_handle0(req, info, supported);
3007 mdt_thread_info_fini(info);
3012 * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
3015 int mdt_recovery_handle(struct ptlrpc_request *req)
3020 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3022 rc = mdt_handle_common(req, mdt_fld_handlers);
3025 rc = mdt_handle_common(req, mdt_seq_handlers);
3028 rc = mdt_handle_common(req, mdt_regular_handlers);
3035 static int mdt_regular_handle(struct ptlrpc_request *req)
3037 return mdt_handle_common(req, mdt_regular_handlers);
3040 static int mdt_readpage_handle(struct ptlrpc_request *req)
3042 return mdt_handle_common(req, mdt_readpage_handlers);
3045 static int mdt_xmds_handle(struct ptlrpc_request *req)
3047 return mdt_handle_common(req, mdt_xmds_handlers);
3050 static int mdt_mdsc_handle(struct ptlrpc_request *req)
3052 return mdt_handle_common(req, mdt_seq_handlers);
3055 static int mdt_mdss_handle(struct ptlrpc_request *req)
3057 return mdt_handle_common(req, mdt_seq_handlers);
3060 static int mdt_dtss_handle(struct ptlrpc_request *req)
3062 return mdt_handle_common(req, mdt_seq_handlers);
3065 static int mdt_fld_handle(struct ptlrpc_request *req)
3067 return mdt_handle_common(req, mdt_fld_handlers);
3083 static int mdt_intent_getattr(enum mdt_it_code opcode,
3084 struct mdt_thread_info *info,
3085 struct ldlm_lock **,
3087 static int mdt_intent_reint(enum mdt_it_code opcode,
3088 struct mdt_thread_info *info,
3089 struct ldlm_lock **,
3092 static struct mdt_it_flavor {
3093 const struct req_format *it_fmt;
3095 int (*it_act)(enum mdt_it_code ,
3096 struct mdt_thread_info *,
3097 struct ldlm_lock **,
3100 } mdt_it_flavor[] = {
3102 .it_fmt = &RQF_LDLM_INTENT,
3103 /*.it_flags = HABEO_REFERO,*/
3105 .it_act = mdt_intent_reint,
3106 .it_reint = REINT_OPEN
3109 .it_fmt = &RQF_LDLM_INTENT,
3110 .it_flags = MUTABOR,
3111 .it_act = mdt_intent_reint,
3112 .it_reint = REINT_OPEN
3115 .it_fmt = &RQF_LDLM_INTENT,
3116 .it_flags = MUTABOR,
3117 .it_act = mdt_intent_reint,
3118 .it_reint = REINT_CREATE
3120 [MDT_IT_GETATTR] = {
3121 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3122 .it_flags = HABEO_REFERO,
3123 .it_act = mdt_intent_getattr
3125 [MDT_IT_READDIR] = {
3131 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3132 .it_flags = HABEO_REFERO,
3133 .it_act = mdt_intent_getattr
3136 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
3137 .it_flags = MUTABOR,
3139 .it_reint = REINT_UNLINK
3143 .it_flags = MUTABOR,
3146 [MDT_IT_GETXATTR] = {
3153 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3154 struct ldlm_lock **lockp,
3155 struct ldlm_lock *new_lock,
3156 struct mdt_lock_handle *lh,
3159 struct ptlrpc_request *req = mdt_info_req(info);
3160 struct ldlm_lock *lock = *lockp;
3163 * Get new lock only for cases when possible resent did not find any
3166 if (new_lock == NULL)
3167 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3169 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3170 lh->mlh_reg_lh.cookie = 0;
3174 LASSERTF(new_lock != NULL,
3175 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3178 * If we've already given this lock to a client once, then we should
3179 * have no readers or writers. Otherwise, we should have one reader
3180 * _or_ writer ref (which will be zeroed below) before returning the
3183 if (new_lock->l_export == req->rq_export) {
3184 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3186 LASSERT(new_lock->l_export == NULL);
3187 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3192 if (new_lock->l_export == req->rq_export) {
3194 * Already gave this to the client, which means that we
3195 * reconstructed a reply.
3197 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3199 lh->mlh_reg_lh.cookie = 0;
3200 RETURN(ELDLM_LOCK_REPLACED);
3204 * Fixup the lock to be given to the client.
3206 lock_res_and_lock(new_lock);
3207 /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3208 * possible blocking AST. */
3209 while (new_lock->l_readers > 0) {
3210 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3211 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3212 new_lock->l_readers--;
3214 while (new_lock->l_writers > 0) {
3215 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3216 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3217 new_lock->l_writers--;
3220 new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3221 new_lock->l_blocking_ast = lock->l_blocking_ast;
3222 new_lock->l_completion_ast = lock->l_completion_ast;
3223 new_lock->l_remote_handle = lock->l_remote_handle;
3224 new_lock->l_flags &= ~LDLM_FL_LOCAL;
3226 unlock_res_and_lock(new_lock);
3228 cfs_hash_add(new_lock->l_export->exp_lock_hash,
3229 &new_lock->l_remote_handle,
3230 &new_lock->l_exp_hash);
3232 LDLM_LOCK_RELEASE(new_lock);
3233 lh->mlh_reg_lh.cookie = 0;
3235 RETURN(ELDLM_LOCK_REPLACED);
3238 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3239 struct ldlm_lock *new_lock,
3240 struct ldlm_lock **old_lock,
3241 struct mdt_lock_handle *lh)
3243 struct ptlrpc_request *req = mdt_info_req(info);
3244 struct obd_export *exp = req->rq_export;
3245 struct lustre_handle remote_hdl;
3246 struct ldlm_request *dlmreq;
3247 struct ldlm_lock *lock;
3249 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3252 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3253 remote_hdl = dlmreq->lock_handle[0];
3255 lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3257 if (lock != new_lock) {
3258 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3259 lh->mlh_reg_mode = lock->l_granted_mode;
3261 LDLM_DEBUG(lock, "Restoring lock cookie");
3262 DEBUG_REQ(D_DLMTRACE, req,
3263 "restoring lock cookie "LPX64,
3264 lh->mlh_reg_lh.cookie);
3266 *old_lock = LDLM_LOCK_GET(lock);
3267 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3271 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3275 * If the xid matches, then we know this is a resent request, and allow
3276 * it. (It's probably an OPEN, for which we don't send a lock.
3278 if (req_xid_is_last(req))
3282 * This remote handle isn't enqueued, so we never received or processed
3283 * this request. Clear MSG_RESENT, because it can be handled like any
3284 * normal request now.
3286 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3288 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3292 static int mdt_intent_getattr(enum mdt_it_code opcode,
3293 struct mdt_thread_info *info,
3294 struct ldlm_lock **lockp,
3297 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3298 struct ldlm_lock *new_lock = NULL;
3300 struct ldlm_reply *ldlm_rep;
3301 struct ptlrpc_request *req;
3302 struct mdt_body *reqbody;
3303 struct mdt_body *repbody;
3307 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3310 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3313 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
3314 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3315 repbody->eadatasize = 0;
3316 repbody->aclsize = 0;
3320 child_bits = MDS_INODELOCK_LOOKUP;
3322 case MDT_IT_GETATTR:
3323 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3326 CERROR("Unhandled till now");
3327 GOTO(out_shrink, rc = -EINVAL);
3330 rc = mdt_init_ucred(info, reqbody);
3332 GOTO(out_shrink, rc);
3334 req = info->mti_pill->rc_req;
3335 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3336 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3338 /* Get lock from request for possible resent case. */
3339 mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3341 ldlm_rep->lock_policy_res2 =
3342 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3344 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3345 ldlm_rep->lock_policy_res2 = 0;
3346 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3347 ldlm_rep->lock_policy_res2) {
3348 lhc->mlh_reg_lh.cookie = 0ull;
3349 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3352 rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3355 mdt_exit_ucred(info);
3357 mdt_client_compatibility(info);
3358 mdt_shrink_reply(info);
3362 static int mdt_intent_reint(enum mdt_it_code opcode,
3363 struct mdt_thread_info *info,
3364 struct ldlm_lock **lockp,
3367 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3368 struct ldlm_reply *rep = NULL;
3372 static const struct req_format *intent_fmts[REINT_MAX] = {
3373 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
3374 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
3379 opc = mdt_reint_opcode(info, intent_fmts);
3383 if (mdt_it_flavor[opcode].it_reint != opc) {
3384 CERROR("Reint code %ld doesn't match intent: %d\n",
3386 RETURN(err_serious(-EPROTO));
3389 /* Get lock from request for possible resent case. */
3390 mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3392 rc = mdt_reint_internal(info, lhc, opc);
3394 /* Check whether the reply has been packed successfully. */
3395 if (mdt_info_req(info)->rq_repmsg != NULL)
3396 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3398 RETURN(err_serious(-EFAULT));
3400 /* MDC expects this in any case */
3402 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3404 /* Cross-ref case, the lock should be returned to the client */
3405 if (rc == -EREMOTE) {
3406 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3407 rep->lock_policy_res2 = 0;
3408 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3411 rep->lock_policy_res2 = clear_serious(rc);
3413 if (rep->lock_policy_res2 == -ENOENT &&
3414 mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3415 rep->lock_policy_res2 = 0;
3417 if (rc == -ENOTCONN || rc == -ENODEV ||
3418 rc == -EOVERFLOW) { /**< if VBR failure then return error */
3420 * If it is the disconnect error (ENODEV & ENOCONN), the error
3421 * will be returned by rq_status, and client at ptlrpc layer
3422 * will detect this, then disconnect, reconnect the import
3423 * immediately, instead of impacting the following the rpc.
3425 lhc->mlh_reg_lh.cookie = 0ull;
3429 * For other cases, the error will be returned by intent.
3430 * and client will retrieve the result from intent.
3433 * FIXME: when open lock is finished, that should be
3436 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3437 LASSERTF(rc == 0, "Error occurred but lock handle "
3438 "is still in use\n");
3439 rep->lock_policy_res2 = 0;
3440 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3443 lhc->mlh_reg_lh.cookie = 0ull;
3444 RETURN(ELDLM_LOCK_ABORTED);
3449 static int mdt_intent_code(long itcode)
3457 case IT_OPEN|IT_CREAT:
3464 rc = MDT_IT_READDIR;
3467 rc = MDT_IT_GETATTR;
3479 rc = MDT_IT_GETXATTR;
3482 CERROR("Unknown intent opcode: %ld\n", itcode);
3489 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3490 struct ldlm_lock **lockp, int flags)
3492 struct req_capsule *pill;
3493 struct mdt_it_flavor *flv;
3498 opc = mdt_intent_code(itopc);
3502 pill = info->mti_pill;
3503 flv = &mdt_it_flavor[opc];
3505 if (flv->it_fmt != NULL)
3506 req_capsule_extend(pill, flv->it_fmt);
3508 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3510 struct ptlrpc_request *req = mdt_info_req(info);
3511 if (flv->it_flags & MUTABOR &&
3512 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3515 if (rc == 0 && flv->it_act != NULL) {
3516 /* execute policy */
3517 rc = flv->it_act(opc, info, lockp, flags);
3524 static int mdt_intent_policy(struct ldlm_namespace *ns,
3525 struct ldlm_lock **lockp, void *req_cookie,
3526 ldlm_mode_t mode, int flags, void *data)
3528 struct mdt_thread_info *info;
3529 struct ptlrpc_request *req = req_cookie;
3530 struct ldlm_intent *it;
3531 struct req_capsule *pill;
3536 LASSERT(req != NULL);
3538 info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3540 LASSERT(info != NULL);
3541 pill = info->mti_pill;
3542 LASSERT(pill->rc_req == req);
3544 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3545 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3546 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3548 rc = mdt_intent_opc(it->opc, info, lockp, flags);
3552 /* Lock without inodebits makes no sense and will oops
3553 * later in ldlm. Let's check it now to see if we have
3554 * ibits corrupted somewhere in mdt_intent_opc().
3555 * The case for client miss to set ibits has been
3556 * processed by others. */
3557 LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3558 lr_type == LDLM_IBITS,
3559 info->mti_dlm_req->lock_desc.\
3560 l_policy_data.l_inodebits.bits != 0));
3562 rc = err_serious(-EFAULT);
3564 /* No intent was provided */
3565 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3566 rc = req_capsule_server_pack(pill);
3568 rc = err_serious(rc);
3573 static int mdt_seq_fini(const struct lu_env *env,
3574 struct mdt_device *m)
3576 struct md_site *ms = mdt_md_site(m);
3580 if (ms->ms_server_seq) {
3581 seq_server_fini(ms->ms_server_seq, env);
3582 OBD_FREE_PTR(ms->ms_server_seq);
3583 ms->ms_server_seq = NULL;
3586 if (ms->ms_control_seq) {
3587 seq_server_fini(ms->ms_control_seq, env);
3588 OBD_FREE_PTR(ms->ms_control_seq);
3589 ms->ms_control_seq = NULL;
3592 if (ms->ms_client_seq) {
3593 seq_client_fini(ms->ms_client_seq);
3594 OBD_FREE_PTR(ms->ms_client_seq);
3595 ms->ms_client_seq = NULL;
3602 static int mdt_seq_init(const struct lu_env *env,
3604 struct mdt_device *m)
3611 ms = mdt_md_site(m);
3614 * This is sequence-controller node. Init seq-controller server on local
3617 if (ms->ms_node_id == 0) {
3618 LASSERT(ms->ms_control_seq == NULL);
3620 OBD_ALLOC_PTR(ms->ms_control_seq);
3621 if (ms->ms_control_seq == NULL)
3624 rc = seq_server_init(ms->ms_control_seq,
3625 m->mdt_bottom, uuid,
3626 LUSTRE_SEQ_CONTROLLER,
3631 GOTO(out_seq_fini, rc);
3633 OBD_ALLOC_PTR(ms->ms_client_seq);
3634 if (ms->ms_client_seq == NULL)
3635 GOTO(out_seq_fini, rc = -ENOMEM);
3637 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3638 if (prefix == NULL) {
3639 OBD_FREE_PTR(ms->ms_client_seq);
3640 GOTO(out_seq_fini, rc = -ENOMEM);
3643 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3647 * Init seq-controller client after seq-controller server is
3648 * ready. Pass ms->ms_control_seq to it for direct talking.
3650 rc = seq_client_init(ms->ms_client_seq, NULL,
3651 LUSTRE_SEQ_METADATA, prefix,
3652 ms->ms_control_seq);
3653 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3656 GOTO(out_seq_fini, rc);
3659 /* Init seq-server on local MDT */
3660 LASSERT(ms->ms_server_seq == NULL);
3662 OBD_ALLOC_PTR(ms->ms_server_seq);
3663 if (ms->ms_server_seq == NULL)
3664 GOTO(out_seq_fini, rc = -ENOMEM);
3666 rc = seq_server_init(ms->ms_server_seq,
3667 m->mdt_bottom, uuid,
3672 GOTO(out_seq_fini, rc = -ENOMEM);
3674 /* Assign seq-controller client to local seq-server. */
3675 if (ms->ms_node_id == 0) {
3676 LASSERT(ms->ms_client_seq != NULL);
3678 rc = seq_server_set_cli(ms->ms_server_seq,
3686 mdt_seq_fini(env, m);
3691 * Init client sequence manager which is used by local MDS to talk to sequence
3692 * controller on remote node.
3694 static int mdt_seq_init_cli(const struct lu_env *env,
3695 struct mdt_device *m,
3696 struct lustre_cfg *cfg)
3698 struct md_site *ms = mdt_md_site(m);
3699 struct obd_device *mdc;
3700 struct obd_uuid *uuidp, *mdcuuidp;
3701 char *uuid_str, *mdc_uuid_str;
3704 struct mdt_thread_info *info;
3705 char *p, *index_string = lustre_cfg_string(cfg, 2);
3708 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3709 uuidp = &info->mti_u.uuid[0];
3710 mdcuuidp = &info->mti_u.uuid[1];
3712 LASSERT(index_string);
3714 index = simple_strtol(index_string, &p, 10);
3716 CERROR("Invalid index in lustre_cgf, offset 2\n");
3720 /* check if this is adding the first MDC and controller is not yet
3722 if (index != 0 || ms->ms_client_seq)
3725 uuid_str = lustre_cfg_string(cfg, 1);
3726 mdc_uuid_str = lustre_cfg_string(cfg, 4);
3727 obd_str2uuid(uuidp, uuid_str);
3728 obd_str2uuid(mdcuuidp, mdc_uuid_str);
3730 mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3732 CERROR("can't find controller MDC by uuid %s\n",
3735 } else if (!mdc->obd_set_up) {
3736 CERROR("target %s not set up\n", mdc->obd_name);
3739 LASSERT(ms->ms_control_exp);
3740 OBD_ALLOC_PTR(ms->ms_client_seq);
3741 if (ms->ms_client_seq != NULL) {
3744 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3748 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3751 rc = seq_client_init(ms->ms_client_seq,
3753 LUSTRE_SEQ_METADATA,
3755 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3762 LASSERT(ms->ms_server_seq != NULL);
3763 rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
3770 static void mdt_seq_fini_cli(struct mdt_device *m)
3776 ms = mdt_md_site(m);
3779 if (ms->ms_server_seq)
3780 seq_server_set_cli(ms->ms_server_seq,
3783 if (ms->ms_control_exp) {
3784 class_export_put(ms->ms_control_exp);
3785 ms->ms_control_exp = NULL;
3794 static int mdt_fld_fini(const struct lu_env *env,
3795 struct mdt_device *m)
3797 struct md_site *ms = mdt_md_site(m);
3800 if (ms && ms->ms_server_fld) {
3801 fld_server_fini(ms->ms_server_fld, env);
3802 OBD_FREE_PTR(ms->ms_server_fld);
3803 ms->ms_server_fld = NULL;
3809 static int mdt_fld_init(const struct lu_env *env,
3811 struct mdt_device *m)
3817 ms = mdt_md_site(m);
3819 OBD_ALLOC_PTR(ms->ms_server_fld);
3820 if (ms->ms_server_fld == NULL)
3821 RETURN(rc = -ENOMEM);
3823 rc = fld_server_init(ms->ms_server_fld,
3824 m->mdt_bottom, uuid,
3825 env, ms->ms_node_id);
3827 OBD_FREE_PTR(ms->ms_server_fld);
3828 ms->ms_server_fld = NULL;
3835 /* device init/fini methods */
3836 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3839 if (m->mdt_regular_service != NULL) {
3840 ptlrpc_unregister_service(m->mdt_regular_service);
3841 m->mdt_regular_service = NULL;
3843 if (m->mdt_readpage_service != NULL) {
3844 ptlrpc_unregister_service(m->mdt_readpage_service);
3845 m->mdt_readpage_service = NULL;
3847 if (m->mdt_xmds_service != NULL) {
3848 ptlrpc_unregister_service(m->mdt_xmds_service);
3849 m->mdt_xmds_service = NULL;
3851 if (m->mdt_setattr_service != NULL) {
3852 ptlrpc_unregister_service(m->mdt_setattr_service);
3853 m->mdt_setattr_service = NULL;
3855 if (m->mdt_mdsc_service != NULL) {
3856 ptlrpc_unregister_service(m->mdt_mdsc_service);
3857 m->mdt_mdsc_service = NULL;
3859 if (m->mdt_mdss_service != NULL) {
3860 ptlrpc_unregister_service(m->mdt_mdss_service);
3861 m->mdt_mdss_service = NULL;
3863 if (m->mdt_dtss_service != NULL) {
3864 ptlrpc_unregister_service(m->mdt_dtss_service);
3865 m->mdt_dtss_service = NULL;
3867 if (m->mdt_fld_service != NULL) {
3868 ptlrpc_unregister_service(m->mdt_fld_service);
3869 m->mdt_fld_service = NULL;
3874 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3877 static struct ptlrpc_service_conf conf;
3878 cfs_proc_dir_entry_t *procfs_entry;
3881 procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3883 conf = (typeof(conf)) {
3884 .psc_nbufs = MDS_NBUFS,
3885 .psc_bufsize = MDS_BUFSIZE,
3886 .psc_max_req_size = MDS_MAXREQSIZE,
3887 .psc_max_reply_size = MDS_MAXREPSIZE,
3888 .psc_req_portal = MDS_REQUEST_PORTAL,
3889 .psc_rep_portal = MDC_REPLY_PORTAL,
3890 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3892 * We'd like to have a mechanism to set this on a per-device
3893 * basis, but alas...
3895 .psc_min_threads = mdt_min_threads,
3896 .psc_max_threads = mdt_max_threads,
3897 .psc_ctx_tags = LCT_MD_THREAD
3900 m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3901 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3902 "mdt_ldlm_client", m->mdt_ldlm_client);
3904 m->mdt_regular_service =
3905 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3906 procfs_entry, target_print_req,
3908 if (m->mdt_regular_service == NULL)
3911 rc = ptlrpc_start_threads(m->mdt_regular_service);
3913 GOTO(err_mdt_svc, rc);
3916 * readpage service configuration. Parameters have to be adjusted,
3919 conf = (typeof(conf)) {
3920 .psc_nbufs = MDS_NBUFS,
3921 .psc_bufsize = MDS_BUFSIZE,
3922 .psc_max_req_size = MDS_MAXREQSIZE,
3923 .psc_max_reply_size = MDS_MAXREPSIZE,
3924 .psc_req_portal = MDS_READPAGE_PORTAL,
3925 .psc_rep_portal = MDC_REPLY_PORTAL,
3926 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3927 .psc_min_threads = mdt_min_threads,
3928 .psc_max_threads = mdt_max_threads,
3929 .psc_ctx_tags = LCT_MD_THREAD
3931 m->mdt_readpage_service =
3932 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3933 LUSTRE_MDT_NAME "_readpage",
3934 procfs_entry, target_print_req,"mdt_rdpg");
3936 if (m->mdt_readpage_service == NULL) {
3937 CERROR("failed to start readpage service\n");
3938 GOTO(err_mdt_svc, rc = -ENOMEM);
3941 rc = ptlrpc_start_threads(m->mdt_readpage_service);
3944 * setattr service configuration.
3946 * XXX To keep the compatibility with old client(< 2.2), we need to
3947 * preserve this portal for a certain time, it should be removed
3948 * eventually. LU-617.
3950 conf = (typeof(conf)) {
3951 .psc_nbufs = MDS_NBUFS,
3952 .psc_bufsize = MDS_BUFSIZE,
3953 .psc_max_req_size = MDS_MAXREQSIZE,
3954 .psc_max_reply_size = MDS_MAXREPSIZE,
3955 .psc_req_portal = MDS_SETATTR_PORTAL,
3956 .psc_rep_portal = MDC_REPLY_PORTAL,
3957 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3958 .psc_min_threads = mdt_min_threads,
3959 .psc_max_threads = mdt_max_threads,
3960 .psc_ctx_tags = LCT_MD_THREAD
3963 m->mdt_setattr_service =
3964 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3965 LUSTRE_MDT_NAME "_setattr",
3966 procfs_entry, target_print_req,"mdt_attr");
3968 if (!m->mdt_setattr_service) {
3969 CERROR("failed to start setattr service\n");
3970 GOTO(err_mdt_svc, rc = -ENOMEM);
3973 rc = ptlrpc_start_threads(m->mdt_setattr_service);
3975 GOTO(err_mdt_svc, rc);
3978 * sequence controller service configuration
3980 conf = (typeof(conf)) {
3981 .psc_nbufs = MDS_NBUFS,
3982 .psc_bufsize = MDS_BUFSIZE,
3983 .psc_max_req_size = SEQ_MAXREQSIZE,
3984 .psc_max_reply_size = SEQ_MAXREPSIZE,
3985 .psc_req_portal = SEQ_CONTROLLER_PORTAL,
3986 .psc_rep_portal = MDC_REPLY_PORTAL,
3987 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3988 .psc_min_threads = mdt_min_threads,
3989 .psc_max_threads = mdt_max_threads,
3990 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3993 m->mdt_mdsc_service =
3994 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3995 LUSTRE_MDT_NAME"_mdsc",
3996 procfs_entry, target_print_req,"mdt_mdsc");
3997 if (!m->mdt_mdsc_service) {
3998 CERROR("failed to start seq controller service\n");
3999 GOTO(err_mdt_svc, rc = -ENOMEM);
4002 rc = ptlrpc_start_threads(m->mdt_mdsc_service);
4004 GOTO(err_mdt_svc, rc);
4007 * metadata sequence server service configuration
4009 conf = (typeof(conf)) {
4010 .psc_nbufs = MDS_NBUFS,
4011 .psc_bufsize = MDS_BUFSIZE,
4012 .psc_max_req_size = SEQ_MAXREQSIZE,
4013 .psc_max_reply_size = SEQ_MAXREPSIZE,
4014 .psc_req_portal = SEQ_METADATA_PORTAL,
4015 .psc_rep_portal = MDC_REPLY_PORTAL,
4016 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4017 .psc_min_threads = mdt_min_threads,
4018 .psc_max_threads = mdt_max_threads,
4019 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
4022 m->mdt_mdss_service =
4023 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
4024 LUSTRE_MDT_NAME"_mdss",
4025 procfs_entry, target_print_req,"mdt_mdss");
4026 if (!m->mdt_mdss_service) {
4027 CERROR("failed to start metadata seq server service\n");
4028 GOTO(err_mdt_svc, rc = -ENOMEM);
4031 rc = ptlrpc_start_threads(m->mdt_mdss_service);
4033 GOTO(err_mdt_svc, rc);
4037 * Data sequence server service configuration. We want to have really
4038 * cluster-wide sequences space. This is why we start only one sequence
4039 * controller which manages space.
4041 conf = (typeof(conf)) {
4042 .psc_nbufs = MDS_NBUFS,
4043 .psc_bufsize = MDS_BUFSIZE,
4044 .psc_max_req_size = SEQ_MAXREQSIZE,
4045 .psc_max_reply_size = SEQ_MAXREPSIZE,
4046 .psc_req_portal = SEQ_DATA_PORTAL,
4047 .psc_rep_portal = OSC_REPLY_PORTAL,
4048 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4049 .psc_min_threads = mdt_min_threads,
4050 .psc_max_threads = mdt_max_threads,
4051 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
4054 m->mdt_dtss_service =
4055 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
4056 LUSTRE_MDT_NAME"_dtss",
4057 procfs_entry, target_print_req,"mdt_dtss");
4058 if (!m->mdt_dtss_service) {
4059 CERROR("failed to start data seq server service\n");
4060 GOTO(err_mdt_svc, rc = -ENOMEM);
4063 rc = ptlrpc_start_threads(m->mdt_dtss_service);
4065 GOTO(err_mdt_svc, rc);
4067 /* FLD service start */
4068 conf = (typeof(conf)) {
4069 .psc_nbufs = MDS_NBUFS,
4070 .psc_bufsize = MDS_BUFSIZE,
4071 .psc_max_req_size = FLD_MAXREQSIZE,
4072 .psc_max_reply_size = FLD_MAXREPSIZE,
4073 .psc_req_portal = FLD_REQUEST_PORTAL,
4074 .psc_rep_portal = MDC_REPLY_PORTAL,
4075 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4076 .psc_min_threads = mdt_min_threads,
4077 .psc_max_threads = mdt_max_threads,
4078 .psc_ctx_tags = LCT_DT_THREAD|LCT_MD_THREAD
4081 m->mdt_fld_service =
4082 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
4083 LUSTRE_MDT_NAME"_fld",
4084 procfs_entry, target_print_req, "mdt_fld");
4085 if (!m->mdt_fld_service) {
4086 CERROR("failed to start fld service\n");
4087 GOTO(err_mdt_svc, rc = -ENOMEM);
4090 rc = ptlrpc_start_threads(m->mdt_fld_service);
4092 GOTO(err_mdt_svc, rc);
4095 * mds-mds service configuration. Separate portal is used to allow
4096 * mds-mds requests be not blocked during recovery.
4098 conf = (typeof(conf)) {
4099 .psc_nbufs = MDS_NBUFS,
4100 .psc_bufsize = MDS_BUFSIZE,
4101 .psc_max_req_size = MDS_MAXREQSIZE,
4102 .psc_max_reply_size = MDS_MAXREPSIZE,
4103 .psc_req_portal = MDS_MDS_PORTAL,
4104 .psc_rep_portal = MDC_REPLY_PORTAL,
4105 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4106 .psc_min_threads = mdt_min_threads,
4107 .psc_max_threads = mdt_max_threads,
4108 .psc_ctx_tags = LCT_MD_THREAD
4110 m->mdt_xmds_service =
4111 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
4112 LUSTRE_MDT_NAME "_mds",
4113 procfs_entry, target_print_req,"mdt_xmds");
4115 if (m->mdt_xmds_service == NULL) {
4116 CERROR("failed to start xmds service\n");
4117 GOTO(err_mdt_svc, rc = -ENOMEM);
4120 rc = ptlrpc_start_threads(m->mdt_xmds_service);
4122 GOTO(err_mdt_svc, rc);
4127 mdt_stop_ptlrpc_service(m);
4132 static void mdt_stack_fini(const struct lu_env *env,
4133 struct mdt_device *m, struct lu_device *top)
4135 struct obd_device *obd = mdt2obd_dev(m);
4136 struct lustre_cfg_bufs *bufs;
4137 struct lustre_cfg *lcfg;
4138 struct mdt_thread_info *info;
4142 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4143 LASSERT(info != NULL);
4145 bufs = &info->mti_u.bufs;
4146 /* process cleanup, pass mdt obd name to get obd umount flags */
4147 lustre_cfg_bufs_reset(bufs, obd->obd_name);
4152 lustre_cfg_bufs_set_string(bufs, 1, flags);
4153 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4155 CERROR("Cannot alloc lcfg!\n");
4160 top->ld_ops->ldo_process_config(env, top, lcfg);
4161 lustre_cfg_free(lcfg);
4163 lu_stack_fini(env, top);
4164 m->mdt_child = NULL;
4165 m->mdt_bottom = NULL;
4168 static struct lu_device *mdt_layer_setup(struct lu_env *env,
4169 const char *typename,
4170 struct lu_device *child,
4171 struct lustre_cfg *cfg)
4173 const char *dev = lustre_cfg_string(cfg, 0);
4174 struct obd_type *type;
4175 struct lu_device_type *ldt;
4176 struct lu_device *d;
4181 type = class_get_type(typename);
4183 CERROR("Unknown type: '%s'\n", typename);
4184 GOTO(out, rc = -ENODEV);
4187 rc = lu_env_refill((struct lu_env *)env);
4189 CERROR("Failure to refill session: '%d'\n", rc);
4195 CERROR("type: '%s'\n", typename);
4196 GOTO(out_type, rc = -EINVAL);
4199 ldt->ldt_obd_type = type;
4200 d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
4202 CERROR("Cannot allocate device: '%s'\n", typename);
4203 GOTO(out_type, rc = -ENODEV);
4206 LASSERT(child->ld_site);
4207 d->ld_site = child->ld_site;
4210 rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
4212 CERROR("can't init device '%s', rc %d\n", typename, rc);
4213 GOTO(out_alloc, rc);
4216 lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
4218 cfs_spin_lock(&d->ld_site->ls_ld_lock);
4219 cfs_list_add_tail(&d->ld_linkage, &d->ld_site->ls_ld_linkage);
4220 cfs_spin_unlock(&d->ld_site->ls_ld_lock);
4224 ldt->ldt_ops->ldto_device_free(env, d);
4227 class_put_type(type);
4232 static int mdt_stack_init(struct lu_env *env,
4233 struct mdt_device *m,
4234 struct lustre_cfg *cfg,
4235 struct lustre_mount_info *lmi)
4237 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
4238 struct lu_device *tmp;
4239 struct md_device *md;
4240 struct lu_device *child_lu_dev;
4244 /* init the stack */
4245 tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
4247 RETURN(PTR_ERR(tmp));
4249 m->mdt_bottom = lu2dt_dev(tmp);
4251 tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
4253 GOTO(out, rc = PTR_ERR(tmp));
4258 tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
4260 GOTO(out, rc = PTR_ERR(tmp));
4263 /*set mdd upcall device*/
4264 md_upcall_dev_set(md, lu2md_dev(d));
4267 /*set cmm upcall device*/
4268 md_upcall_dev_set(md, &m->mdt_md_dev);
4270 m->mdt_child = lu2md_dev(d);
4272 /* process setup config */
4273 tmp = &m->mdt_md_dev.md_lu_dev;
4274 rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
4278 /* initialize local objects */
4279 child_lu_dev = &m->mdt_child->md_lu_dev;
4281 rc = child_lu_dev->ld_ops->ldo_prepare(env,
4282 &m->mdt_md_dev.md_lu_dev,
4285 /* fini from last known good lu_device */
4287 mdt_stack_fini(env, m, d);
4293 * setup CONFIG_ORIG context, used to access local config log.
4294 * this may need to be rewrite as part of llog rewrite for lu-api.
4296 static int mdt_obd_llog_setup(struct obd_device *obd,
4297 struct lustre_sb_info *lsi)
4301 LASSERT(obd->obd_fsops == NULL);
4303 obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
4304 if (IS_ERR(obd->obd_fsops))
4305 return PTR_ERR(obd->obd_fsops);
4307 rc = fsfilt_setup(obd, lsi->lsi_srv_mnt->mnt_sb);
4309 fsfilt_put_ops(obd->obd_fsops);
4313 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
4314 obd->obd_lvfs_ctxt.pwdmnt = lsi->lsi_srv_mnt;
4315 obd->obd_lvfs_ctxt.pwd = lsi->lsi_srv_mnt->mnt_root;
4316 obd->obd_lvfs_ctxt.fs = get_ds();
4318 rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd,
4319 0, NULL, &llog_lvfs_ops);
4321 CERROR("llog_setup() failed: %d\n", rc);
4322 fsfilt_put_ops(obd->obd_fsops);
4328 static void mdt_obd_llog_cleanup(struct obd_device *obd)
4330 struct llog_ctxt *ctxt;
4332 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
4336 if (obd->obd_fsops) {
4337 fsfilt_put_ops(obd->obd_fsops);
4338 obd->obd_fsops = NULL;
4342 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4344 struct md_device *next = m->mdt_child;
4345 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
4346 struct lu_site *ls = d->ld_site;
4347 struct obd_device *obd = mdt2obd_dev(m);
4350 target_recovery_fini(obd);
4352 ping_evictor_stop();
4354 mdt_stop_ptlrpc_service(m);
4355 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4356 mdt_obd_llog_cleanup(obd);
4357 obd_exports_barrier(obd);
4358 obd_zombie_barrier();
4362 #ifdef HAVE_QUOTA_SUPPORT
4363 next->md_ops->mdo_quota.mqo_cleanup(env, next);
4365 lut_fini(env, &m->mdt_lut);
4366 mdt_fs_cleanup(env, m);
4367 upcall_cache_cleanup(m->mdt_identity_cache);
4368 m->mdt_identity_cache = NULL;
4370 if (m->mdt_namespace != NULL) {
4371 ldlm_namespace_free(m->mdt_namespace, NULL,
4372 d->ld_obd->obd_force);
4373 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4376 cfs_free_nidlist(&m->mdt_nosquash_nids);
4377 if (m->mdt_nosquash_str) {
4378 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4379 m->mdt_nosquash_str = NULL;
4380 m->mdt_nosquash_strlen = 0;
4383 mdt_seq_fini(env, m);
4384 mdt_seq_fini_cli(m);
4385 mdt_fld_fini(env, m);
4386 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4388 next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4389 cfs_timer_disarm(&m->mdt_ck_timer);
4390 mdt_ck_thread_stop(m);
4395 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4398 struct md_site *mite;
4401 mite = lu_site2md(ls);
4405 LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
4410 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4412 struct mdt_device *m = mdt_dev(obd->obd_lu_dev);
4413 struct sptlrpc_rule_set tmp_rset;
4416 sptlrpc_rule_set_init(&tmp_rset);
4417 rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4419 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4424 sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4426 cfs_write_lock(&m->mdt_sptlrpc_lock);
4427 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4428 m->mdt_sptlrpc_rset = tmp_rset;
4429 cfs_write_unlock(&m->mdt_sptlrpc_lock);
4434 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4436 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4437 struct lu_device_type *ldt, struct lustre_cfg *cfg)
4439 struct mdt_thread_info *info;
4440 struct obd_device *obd;
4441 const char *dev = lustre_cfg_string(cfg, 0);
4442 const char *num = lustre_cfg_string(cfg, 2);
4443 struct lustre_mount_info *lmi = NULL;
4444 struct lustre_sb_info *lsi;
4445 struct lustre_disk_data *ldd;
4447 struct md_site *mite;
4448 const char *identity_upcall = "NONE";
4449 struct md_device *next;
4455 md_device_init(&m->mdt_md_dev, ldt);
4457 * Environment (env) might be missing mdt_thread_key values at that
4458 * point, if device is allocated when mdt_thread_key is in QUIESCENT
4461 * Usually device allocation path doesn't use module key values, but
4462 * mdt has to do a lot of work here, so allocate key value.
4464 rc = lu_env_refill((struct lu_env *)env);
4468 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4469 LASSERT(info != NULL);
4471 obd = class_name2obd(dev);
4472 LASSERT(obd != NULL);
4474 m->mdt_max_mdsize = MAX_MD_SIZE;
4475 m->mdt_max_cookiesize = sizeof(struct llog_cookie);
4476 m->mdt_som_conf = 0;
4478 m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4479 lmi = server_get_mount_2(dev);
4481 CERROR("Cannot get mount info for %s!\n", dev);
4484 lsi = s2lsi(lmi->lmi_sb);
4485 /* CMD is supported only in IAM mode */
4488 node_id = simple_strtol(num, NULL, 10);
4489 if (!(ldd->ldd_flags & LDD_F_IAM_DIR) && node_id) {
4490 CERROR("CMD Operation not allowed in IOP mode\n");
4491 GOTO(err_lmi, rc = -EINVAL);
4494 obd->u.obt.obt_magic = OBT_MAGIC;
4497 cfs_rwlock_init(&m->mdt_sptlrpc_lock);
4498 sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4500 cfs_spin_lock_init(&m->mdt_ioepoch_lock);
4501 m->mdt_opts.mo_compat_resname = 0;
4502 m->mdt_opts.mo_mds_capa = 1;
4503 m->mdt_opts.mo_oss_capa = 1;
4504 m->mdt_capa_timeout = CAPA_TIMEOUT;
4505 m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4506 m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4507 m->mdt_squash_uid = 0;
4508 m->mdt_squash_gid = 0;
4509 CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4510 m->mdt_nosquash_str = NULL;
4511 m->mdt_nosquash_strlen = 0;
4512 cfs_init_rwsem(&m->mdt_squash_sem);
4514 OBD_ALLOC_PTR(mite);
4516 GOTO(err_lmi, rc = -ENOMEM);
4520 m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4521 m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4522 /* set this lu_device to obd, because error handling need it */
4523 obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4525 rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
4527 CERROR("Can't init lu_site, rc %d\n", rc);
4528 GOTO(err_free_site, rc);
4531 /* set server index */
4532 lu_site2md(s)->ms_node_id = node_id;
4534 /* failover is the default
4535 * FIXME: we do not failout mds0/mgs, which may cause some problems.
4536 * assumed whose ms_node_id == 0 XXX
4538 obd->obd_replayable = 1;
4539 /* No connection accepted until configurations will finish */
4540 obd->obd_no_conn = 1;
4542 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4543 char *str = lustre_cfg_string(cfg, 4);
4544 if (strchr(str, 'n')) {
4545 CWARN("%s: recovery disabled\n", obd->obd_name);
4546 obd->obd_replayable = 0;
4550 /* init the stack */
4551 rc = mdt_stack_init((struct lu_env *)env, m, cfg, lmi);
4553 CERROR("Can't init device stack, rc %d\n", rc);
4554 GOTO(err_lu_site, rc);
4557 rc = lut_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4559 GOTO(err_fini_stack, rc);
4561 rc = mdt_fld_init(env, obd->obd_name, m);
4565 rc = mdt_seq_init(env, obd->obd_name, m);
4567 GOTO(err_fini_fld, rc);
4569 snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4570 LUSTRE_MDT_NAME"-%p", m);
4571 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4572 LDLM_NAMESPACE_SERVER,
4573 LDLM_NAMESPACE_GREEDY,
4575 if (m->mdt_namespace == NULL)
4576 GOTO(err_fini_seq, rc = -ENOMEM);
4578 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4579 /* set obd_namespace for compatibility with old code */
4580 obd->obd_namespace = m->mdt_namespace;
4582 cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4584 rc = mdt_ck_thread_start(m);
4586 GOTO(err_free_ns, rc);
4588 rc = mdt_fs_setup(env, m, obd, lsi);
4592 rc = mdt_obd_llog_setup(obd, lsi);
4594 GOTO(err_fs_cleanup, rc);
4596 rc = mdt_llog_ctxt_clone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4598 GOTO(err_llog_cleanup, rc);
4600 mdt_adapt_sptlrpc_conf(obd, 1);
4602 next = m->mdt_child;
4603 #ifdef HAVE_QUOTA_SUPPORT
4604 rc = next->md_ops->mdo_quota.mqo_setup(env, next, lmi->lmi_mnt);
4606 GOTO(err_llog_cleanup, rc);
4609 server_put_mount_2(dev, lmi->lmi_mnt);
4612 rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4615 GOTO(err_quota, rc);
4617 if (mntopts & MNTOPT_USERXATTR)
4618 m->mdt_opts.mo_user_xattr = 1;
4620 m->mdt_opts.mo_user_xattr = 0;
4622 if (mntopts & MNTOPT_ACL)
4623 m->mdt_opts.mo_acl = 1;
4625 m->mdt_opts.mo_acl = 0;
4627 /* XXX: to support suppgid for ACL, we enable identity_upcall
4628 * by default, otherwise, maybe got unexpected -EACCESS. */
4629 if (m->mdt_opts.mo_acl)
4630 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4632 m->mdt_identity_cache = upcall_cache_init(obd->obd_name,identity_upcall,
4633 &mdt_identity_upcall_cache_ops);
4634 if (IS_ERR(m->mdt_identity_cache)) {
4635 rc = PTR_ERR(m->mdt_identity_cache);
4636 m->mdt_identity_cache = NULL;
4637 GOTO(err_quota, rc);
4640 target_recovery_init(&m->mdt_lut, mdt_recovery_handle);
4642 rc = mdt_procfs_init(m, dev);
4644 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4645 GOTO(err_recovery, rc);
4648 rc = mdt_start_ptlrpc_service(m);
4650 GOTO(err_procfs, rc);
4652 ping_evictor_start();
4654 rc = lu_site_init_finish(s);
4656 GOTO(err_stop_service, rc);
4658 if (obd->obd_recovering == 0)
4659 mdt_postrecov(env, m);
4661 mdt_init_capa_ctxt(env, m);
4663 /* Reduce the initial timeout on an MDS because it doesn't need such
4664 * a long timeout as an OST does. Adaptive timeouts will adjust this
4665 * value appropriately. */
4666 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4667 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4672 ping_evictor_stop();
4673 mdt_stop_ptlrpc_service(m);
4677 target_recovery_fini(obd);
4678 upcall_cache_cleanup(m->mdt_identity_cache);
4679 m->mdt_identity_cache = NULL;
4681 #ifdef HAVE_QUOTA_SUPPORT
4682 next->md_ops->mdo_quota.mqo_cleanup(env, next);
4685 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4686 mdt_obd_llog_cleanup(obd);
4688 mdt_fs_cleanup(env, m);
4690 cfs_timer_disarm(&m->mdt_ck_timer);
4691 mdt_ck_thread_stop(m);
4693 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4694 obd->obd_namespace = m->mdt_namespace = NULL;
4696 mdt_seq_fini(env, m);
4698 mdt_fld_fini(env, m);
4700 lut_fini(env, &m->mdt_lut);
4702 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4709 server_put_mount_2(dev, lmi->lmi_mnt);
4713 /* used by MGS to process specific configurations */
4714 static int mdt_process_config(const struct lu_env *env,
4715 struct lu_device *d, struct lustre_cfg *cfg)
4717 struct mdt_device *m = mdt_dev(d);
4718 struct md_device *md_next = m->mdt_child;
4719 struct lu_device *next = md2lu_dev(md_next);
4723 switch (cfg->lcfg_command) {
4725 struct lprocfs_static_vars lvars;
4726 struct obd_device *obd = d->ld_obd;
4729 * For interoperability between 1.8 and 2.0,
4732 /* Skip old "mdt.group_upcall" param. */
4733 char *param = lustre_cfg_string(cfg, 1);
4734 if (param && !strncmp("mdt.group_upcall", param, 16)) {
4735 CWARN("For 1.8 interoperability, skip this"
4736 " mdt.group_upcall. It is obsolete\n");
4739 /* Rename old "mdt.quota_type" to "mdd.quota_type. */
4740 if (param && !strncmp("mdt.quota_type", param, 14)) {
4741 CWARN("Found old param mdt.quota_type, changed"
4742 " it to mdd.quota_type.\n");
4747 lprocfs_mdt_init_vars(&lvars);
4748 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4750 if (rc > 0 || rc == -ENOSYS)
4751 /* we don't understand; pass it on */
4752 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4757 * Add mdc hook to get first MDT uuid and connect it to
4758 * ls->controller to use for seq manager.
4760 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4762 CERROR("Can't add mdc, rc %d\n", rc);
4764 rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4767 /* others are passed further */
4768 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4774 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4775 const struct lu_object_header *hdr,
4776 struct lu_device *d)
4778 struct mdt_object *mo;
4784 struct lu_object *o;
4785 struct lu_object_header *h;
4787 o = &mo->mot_obj.mo_lu;
4788 h = &mo->mot_header;
4789 lu_object_header_init(h);
4790 lu_object_init(o, h, d);
4791 lu_object_add_top(h, o);
4792 o->lo_ops = &mdt_obj_ops;
4793 cfs_sema_init(&mo->mot_ioepoch_sem, 1);
4794 cfs_sema_init(&mo->mot_lov_sem, 1);
4800 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4801 const struct lu_object_conf *unused)
4803 struct mdt_device *d = mdt_dev(o->lo_dev);
4804 struct lu_device *under;
4805 struct lu_object *below;
4809 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4810 PFID(lu_object_fid(o)));
4812 under = &d->mdt_child->md_lu_dev;
4813 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4814 if (below != NULL) {
4815 lu_object_add(o, below);
4822 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4824 struct mdt_object *mo = mdt_obj(o);
4825 struct lu_object_header *h;
4829 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4830 PFID(lu_object_fid(o)));
4833 lu_object_header_fini(h);
4838 static int mdt_object_print(const struct lu_env *env, void *cookie,
4839 lu_printer_t p, const struct lu_object *o)
4841 struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4842 return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4843 "flags="LPX64", epochcount=%d, writecount=%d)",
4844 mdto, mdto->mot_ioepoch, mdto->mot_flags,
4845 mdto->mot_ioepoch_count, mdto->mot_writecount);
4848 static const struct lu_device_operations mdt_lu_ops = {
4849 .ldo_object_alloc = mdt_object_alloc,
4850 .ldo_process_config = mdt_process_config,
4853 static const struct lu_object_operations mdt_obj_ops = {
4854 .loo_object_init = mdt_object_init,
4855 .loo_object_free = mdt_object_free,
4856 .loo_object_print = mdt_object_print
4859 static int mdt_obd_set_info_async(struct obd_export *exp,
4860 __u32 keylen, void *key,
4861 __u32 vallen, void *val,
4862 struct ptlrpc_request_set *set)
4864 struct obd_device *obd = exp->exp_obd;
4870 if (KEY_IS(KEY_SPTLRPC_CONF)) {
4871 rc = mdt_adapt_sptlrpc_conf(obd, 0);
4878 /* mds_connect_internal */
4879 static int mdt_connect_internal(struct obd_export *exp,
4880 struct mdt_device *mdt,
4881 struct obd_connect_data *data)
4884 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4885 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4887 /* If no known bits (which should not happen, probably,
4888 as everybody should support LOOKUP and UPDATE bits at least)
4889 revert to compat mode with plain locks. */
4890 if (!data->ocd_ibits_known &&
4891 data->ocd_connect_flags & OBD_CONNECT_IBITS)
4892 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4894 if (!mdt->mdt_opts.mo_acl)
4895 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4897 if (!mdt->mdt_opts.mo_user_xattr)
4898 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4900 if (!mdt->mdt_som_conf)
4901 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4903 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4904 data->ocd_brw_size = min(data->ocd_brw_size,
4905 (__u32)(PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT));
4906 if (data->ocd_brw_size == 0) {
4907 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4908 " ocd_version: %x ocd_grant: %d "
4909 "ocd_index: %u ocd_brw_size is "
4910 "unexpectedly zero, network data "
4911 "corruption? Refusing connection of this"
4913 exp->exp_obd->obd_name,
4914 exp->exp_client_uuid.uuid,
4915 exp, data->ocd_connect_flags, data->ocd_version,
4916 data->ocd_grant, data->ocd_index);
4921 cfs_spin_lock(&exp->exp_lock);
4922 exp->exp_connect_flags = data->ocd_connect_flags;
4923 cfs_spin_unlock(&exp->exp_lock);
4924 data->ocd_version = LUSTRE_VERSION_CODE;
4925 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4929 if (mdt->mdt_opts.mo_acl &&
4930 ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4931 CWARN("%s: MDS requires ACL support but client does not\n",
4932 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4937 if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4938 CWARN("%s: MDS requires FID support, but client not\n",
4939 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4943 if (mdt->mdt_som_conf && !exp_connect_som(exp) &&
4944 !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS)) {
4945 CWARN("%s: MDS has SOM enabled, but client does not support "
4946 "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4953 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
4954 struct obd_export *exp,
4955 struct ptlrpc_request *req)
4957 struct sptlrpc_flavor flvr;
4960 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4961 cfs_read_lock(&mdt->mdt_sptlrpc_lock);
4962 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
4966 cfs_read_unlock(&mdt->mdt_sptlrpc_lock);
4968 cfs_spin_lock(&exp->exp_lock);
4970 exp->exp_sp_peer = req->rq_sp_from;
4971 exp->exp_flvr = flvr;
4973 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
4974 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4975 CERROR("unauthorized rpc flavor %x from %s, "
4976 "expect %x\n", req->rq_flvr.sf_rpc,
4977 libcfs_nid2str(req->rq_peer.nid),
4978 exp->exp_flvr.sf_rpc);
4982 cfs_spin_unlock(&exp->exp_lock);
4984 if (exp->exp_sp_peer != req->rq_sp_from) {
4985 CERROR("RPC source %s doesn't match %s\n",
4986 sptlrpc_part2name(req->rq_sp_from),
4987 sptlrpc_part2name(exp->exp_sp_peer));
4990 rc = sptlrpc_target_export_check(exp, req);
4997 /* mds_connect copy */
4998 static int mdt_obd_connect(const struct lu_env *env,
4999 struct obd_export **exp, struct obd_device *obd,
5000 struct obd_uuid *cluuid,
5001 struct obd_connect_data *data,
5004 struct mdt_thread_info *info;
5005 struct obd_export *lexp;
5006 struct lustre_handle conn = { 0 };
5007 struct mdt_device *mdt;
5008 struct ptlrpc_request *req;
5012 LASSERT(env != NULL);
5013 if (!exp || !obd || !cluuid)
5016 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5017 req = info->mti_pill->rc_req;
5018 mdt = mdt_dev(obd->obd_lu_dev);
5020 rc = class_connect(&conn, obd, cluuid);
5024 lexp = class_conn2export(&conn);
5025 LASSERT(lexp != NULL);
5027 rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
5031 if (OBD_FAIL_CHECK(OBD_FAIL_TGT_RCVG_FLAG))
5032 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
5034 rc = mdt_connect_internal(lexp, mdt, data);
5036 struct mdt_thread_info *mti;
5037 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5039 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5040 LASSERT(mti != NULL);
5041 mti->mti_exp = lexp;
5042 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5043 rc = mdt_client_new(env, mdt);
5045 mdt_export_stats_init(obd, lexp, localdata);
5050 class_disconnect(lexp);
5059 static int mdt_obd_reconnect(const struct lu_env *env,
5060 struct obd_export *exp, struct obd_device *obd,
5061 struct obd_uuid *cluuid,
5062 struct obd_connect_data *data,
5065 struct mdt_thread_info *info;
5066 struct mdt_device *mdt;
5067 struct ptlrpc_request *req;
5071 if (exp == NULL || obd == NULL || cluuid == NULL)
5074 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5075 req = info->mti_pill->rc_req;
5076 mdt = mdt_dev(obd->obd_lu_dev);
5078 rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5082 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5084 mdt_export_stats_init(obd, exp, localdata);
5088 static int mdt_export_cleanup(struct obd_export *exp)
5090 struct mdt_export_data *med = &exp->exp_mdt_data;
5091 struct obd_device *obd = exp->exp_obd;
5092 struct mdt_device *mdt;
5093 struct mdt_thread_info *info;
5095 CFS_LIST_HEAD(closing_list);
5096 struct mdt_file_data *mfd, *n;
5100 cfs_spin_lock(&med->med_open_lock);
5101 while (!cfs_list_empty(&med->med_open_head)) {
5102 cfs_list_t *tmp = med->med_open_head.next;
5103 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5105 /* Remove mfd handle so it can't be found again.
5106 * We are consuming the mfd_list reference here. */
5107 class_handle_unhash(&mfd->mfd_handle);
5108 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5110 cfs_spin_unlock(&med->med_open_lock);
5111 mdt = mdt_dev(obd->obd_lu_dev);
5112 LASSERT(mdt != NULL);
5114 rc = lu_env_init(&env, LCT_MD_THREAD);
5118 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5119 LASSERT(info != NULL);
5120 memset(info, 0, sizeof *info);
5121 info->mti_env = &env;
5122 info->mti_mdt = mdt;
5123 info->mti_exp = exp;
5125 if (!cfs_list_empty(&closing_list)) {
5126 struct md_attr *ma = &info->mti_attr;
5130 lmm_size = mdt->mdt_max_mdsize;
5131 OBD_ALLOC_LARGE(ma->ma_lmm, lmm_size);
5132 if (ma->ma_lmm == NULL)
5133 GOTO(out_lmm, rc = -ENOMEM);
5135 cookie_size = mdt->mdt_max_cookiesize;
5136 OBD_ALLOC_LARGE(ma->ma_cookie, cookie_size);
5137 if (ma->ma_cookie == NULL)
5138 GOTO(out_cookie, rc = -ENOMEM);
5140 /* Close any open files (which may also cause orphan unlinking). */
5141 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5142 cfs_list_del_init(&mfd->mfd_list);
5143 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
5144 ma->ma_lmm_size = lmm_size;
5145 ma->ma_cookie_size = cookie_size;
5147 /* It is not for setattr, just tell MDD to send
5148 * DESTROY RPC to OSS if needed */
5149 ma->ma_valid = MA_FLAGS;
5150 ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
5151 /* Don't unlink orphan on failover umount, LU-184 */
5152 if (exp->exp_flags & OBD_OPT_FAILOVER)
5153 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5154 mdt_mfd_close(info, mfd);
5156 OBD_FREE_LARGE(ma->ma_cookie, cookie_size);
5157 ma->ma_cookie = NULL;
5159 OBD_FREE_LARGE(ma->ma_lmm, lmm_size);
5163 info->mti_mdt = NULL;
5164 /* cleanup client slot early */
5165 /* Do not erase record for recoverable client. */
5166 if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5167 mdt_client_del(&env, mdt);
5173 static int mdt_obd_disconnect(struct obd_export *exp)
5179 class_export_get(exp);
5181 rc = server_disconnect_export(exp);
5183 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5185 rc = mdt_export_cleanup(exp);
5186 class_export_put(exp);
5190 /* FIXME: Can we avoid using these two interfaces? */
5191 static int mdt_init_export(struct obd_export *exp)
5193 struct mdt_export_data *med = &exp->exp_mdt_data;
5197 CFS_INIT_LIST_HEAD(&med->med_open_head);
5198 cfs_spin_lock_init(&med->med_open_lock);
5199 cfs_sema_init(&med->med_idmap_sem, 1);
5200 med->med_idmap = NULL;
5201 cfs_spin_lock(&exp->exp_lock);
5202 exp->exp_connecting = 1;
5203 cfs_spin_unlock(&exp->exp_lock);
5205 /* self-export doesn't need client data and ldlm initialization */
5206 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5207 &exp->exp_client_uuid)))
5210 rc = lut_client_alloc(exp);
5212 rc = ldlm_init_export(exp);
5214 CERROR("%s: Error %d while initializing export\n",
5215 exp->exp_obd->obd_name, rc);
5219 static int mdt_destroy_export(struct obd_export *exp)
5223 if (exp_connect_rmtclient(exp))
5224 mdt_cleanup_idmap(&exp->exp_mdt_data);
5226 target_destroy_export(exp);
5227 /* destroy can be called from failed obd_setup, so
5228 * checking uuid is safer than obd_self_export */
5229 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5230 &exp->exp_client_uuid)))
5233 ldlm_destroy_export(exp);
5234 lut_client_free(exp);
5236 LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5237 LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5242 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
5244 if (flag & CONFIG_LOG)
5245 cfs_set_bit(MDT_FL_CFGLOG, &m->mdt_state);
5247 /* also notify active event */
5248 if (flag & CONFIG_SYNC)
5249 cfs_set_bit(MDT_FL_SYNCED, &m->mdt_state);
5251 if (cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state) &&
5252 cfs_test_bit(MDT_FL_SYNCED, &m->mdt_state)) {
5253 struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
5255 /* Open for clients */
5256 if (obd->obd_no_conn) {
5257 cfs_spin_lock(&obd->obd_dev_lock);
5258 obd->obd_no_conn = 0;
5259 cfs_spin_unlock(&obd->obd_dev_lock);
5264 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
5265 enum md_upcall_event ev, void *data)
5267 struct mdt_device *m = mdt_dev(&md->md_lu_dev);
5268 struct md_device *next = m->mdt_child;
5269 struct mdt_thread_info *mti;
5275 rc = next->md_ops->mdo_maxsize_get(env, next,
5277 &m->mdt_max_cookiesize);
5278 CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
5279 m->mdt_max_mdsize, m->mdt_max_cookiesize);
5280 mdt_allow_cli(m, CONFIG_SYNC);
5283 m->mdt_lut.lut_obd->u.obt.obt_mount_count;
5286 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5287 mti->mti_no_need_trans = 1;
5288 CDEBUG(D_INFO, "disable mdt trans for this thread\n");
5291 /* Check that MDT is not yet configured */
5292 LASSERT(!cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state));
5294 #ifdef HAVE_QUOTA_SUPPORT
5296 if (md->md_lu_dev.ld_obd->obd_recovering == 0 &&
5297 likely(md->md_lu_dev.ld_obd->obd_stopping == 0))
5298 next->md_ops->mdo_quota.mqo_recovery(env, next);
5302 CERROR("invalid event\n");
5309 static int mdt_obd_notify(struct obd_device *obd,
5310 struct obd_device *watched,
5311 enum obd_notify_event ev, void *data)
5313 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5314 #ifdef HAVE_QUOTA_SUPPORT
5315 struct md_device *next = mdt->mdt_child;
5320 case OBD_NOTIFY_CONFIG:
5321 mdt_allow_cli(mdt, (unsigned long)data);
5323 #ifdef HAVE_QUOTA_SUPPORT
5324 /* quota_type has been processed, we can now handle
5325 * incoming quota requests */
5326 next->md_ops->mdo_quota.mqo_notify(NULL, next);
5330 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
5335 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5336 void *val, int vallen)
5338 struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev);
5339 struct getinfo_fid2path *fpout, *fpin;
5342 fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5345 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5346 lustre_swab_fid2path(fpin);
5348 memcpy(fpout, fpin, sizeof(*fpin));
5349 if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5352 rc = mdt_fid2path(info->mti_env, mdt, fpout);
5356 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
5357 struct getinfo_fid2path *fp)
5359 struct mdt_object *obj;
5363 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5364 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5366 if (!fid_is_sane(&fp->gf_fid))
5369 obj = mdt_object_find(env, mdt, &fp->gf_fid);
5370 if (obj == NULL || IS_ERR(obj)) {
5371 CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid),
5376 rc = lu_object_exists(&obj->mot_obj.mo_lu);
5382 mdt_object_put(env, obj);
5383 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5384 PFID(&fp->gf_fid), rc);
5388 rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path,
5389 fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno);
5390 mdt_object_put(env, obj);
5395 static int mdt_get_info(struct mdt_thread_info *info)
5397 struct ptlrpc_request *req = mdt_info_req(info);
5405 key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5407 CDEBUG(D_IOCTL, "No GETINFO key");
5410 keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5413 vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5414 if (vallen == NULL) {
5415 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5419 req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5421 rc = req_capsule_server_pack(info->mti_pill);
5422 valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5423 if (valout == NULL) {
5424 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5428 if (KEY_IS(KEY_FID2PATH))
5429 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5433 lustre_msg_set_status(req->rq_repmsg, rc);
5438 /* Pass the ioc down */
5439 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5440 unsigned int cmd, int len, void *data)
5442 struct lu_context ioctl_session;
5443 struct md_device *next = mdt->mdt_child;
5447 rc = lu_context_init(&ioctl_session, LCT_SESSION);
5450 ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5451 lu_context_enter(&ioctl_session);
5452 env->le_ses = &ioctl_session;
5454 LASSERT(next->md_ops->mdo_iocontrol);
5455 rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5457 lu_context_exit(&ioctl_session);
5458 lu_context_fini(&ioctl_session);
5462 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5464 struct obd_ioctl_data *data = karg;
5465 struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1;
5467 struct mdt_object *obj;
5468 struct mdt_lock_handle *lh;
5472 CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5473 if (!fid_is_sane(fid))
5476 lh = &mti->mti_lh[MDT_LH_PARENT];
5477 mdt_lock_reg_init(lh, LCK_CR);
5479 obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5481 RETURN(PTR_ERR(obj));
5483 rc = mdt_object_exists(obj);
5487 * before calling version get the correct MDS should be
5488 * fid, this is error to find remote object here
5490 CERROR("nonlocal object "DFID"\n", PFID(fid));
5491 } else if (rc == 0) {
5492 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5495 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5496 *(__u64 *)data->ioc_inlbuf2 = version;
5499 mdt_object_unlock_put(mti, obj, lh, 1);
5503 /* ioctls on obd dev */
5504 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5505 void *karg, void *uarg)
5508 struct obd_device *obd = exp->exp_obd;
5509 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5510 struct dt_device *dt = mdt->mdt_bottom;
5514 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5515 rc = lu_env_init(&env, LCT_MD_THREAD);
5521 rc = mdt_device_sync(&env, mdt);
5523 case OBD_IOC_SET_READONLY:
5524 dt->dd_ops->dt_ro(&env, dt);
5526 case OBD_IOC_ABORT_RECOVERY:
5527 CERROR("Aborting recovery for device %s\n", obd->obd_name);
5528 target_stop_recovery_thread(obd);
5531 case OBD_IOC_CHANGELOG_REG:
5532 case OBD_IOC_CHANGELOG_DEREG:
5533 case OBD_IOC_CHANGELOG_CLEAR:
5534 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5536 case OBD_IOC_GET_OBJ_VERSION: {
5537 struct mdt_thread_info *mti;
5538 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5539 memset(mti, 0, sizeof *mti);
5540 mti->mti_env = &env;
5544 rc = mdt_ioc_version_get(mti, karg);
5548 CERROR("Not supported cmd = %d for device %s\n",
5549 cmd, obd->obd_name);
5557 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5559 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5560 #ifdef HAVE_QUOTA_SUPPORT
5561 struct obd_device *obd = mdt2obd_dev(mdt);
5562 struct md_device *next = mdt->mdt_child;
5567 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5568 #ifdef HAVE_QUOTA_SUPPORT
5569 if (likely(obd->obd_stopping == 0))
5570 next->md_ops->mdo_quota.mqo_recovery(env, next);
5575 int mdt_obd_postrecov(struct obd_device *obd)
5580 rc = lu_env_init(&env, LCT_MD_THREAD);
5583 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5589 * Send a copytool req to a client
5590 * Note this sends a request RPC from a server (MDT) to a client (MDC),
5591 * backwards of normal comms.
5593 int mdt_hsm_copytool_send(struct obd_export *exp)
5596 struct hsm_action_list *hal;
5597 struct hsm_action_item *hai;
5601 CWARN("%s: writing to mdc at %s\n", exp->exp_obd->obd_name,
5602 libcfs_nid2str(exp->exp_connection->c_peer.nid));
5604 len = sizeof(*lh) + sizeof(*hal) + MTI_NAME_MAXLEN +
5605 /* for mockup below */ 2 * cfs_size_round(sizeof(*hai));
5610 lh->kuc_magic = KUC_MAGIC;
5611 lh->kuc_transport = KUC_TRANSPORT_HSM;
5612 lh->kuc_msgtype = HMT_ACTION_LIST;
5613 lh->kuc_msglen = len;
5615 hal = (struct hsm_action_list *)(lh + 1);
5616 hal->hal_version = HAL_VERSION;
5617 hal->hal_archive_num = 1;
5618 obd_uuid2fsname(hal->hal_fsname, exp->exp_obd->obd_name,
5621 /* mock up an action list */
5623 hai = hai_zero(hal);
5624 hai->hai_action = HSMA_ARCHIVE;
5625 hai->hai_fid.f_oid = 0xA00A;
5626 hai->hai_len = sizeof(*hai);
5627 hai = hai_next(hai);
5628 hai->hai_action = HSMA_RESTORE;
5629 hai->hai_fid.f_oid = 0xB00B;
5630 hai->hai_len = sizeof(*hai);
5632 /* Uses the ldlm reverse import; this rpc will be seen by
5633 the ldlm_callback_handler */
5634 rc = do_set_info_async(exp->exp_imp_reverse,
5635 LDLM_SET_INFO, LUSTRE_OBD_VERSION,
5636 sizeof(KEY_HSM_COPYTOOL_SEND),
5637 KEY_HSM_COPYTOOL_SEND,
5645 static struct obd_ops mdt_obd_device_ops = {
5646 .o_owner = THIS_MODULE,
5647 .o_set_info_async = mdt_obd_set_info_async,
5648 .o_connect = mdt_obd_connect,
5649 .o_reconnect = mdt_obd_reconnect,
5650 .o_disconnect = mdt_obd_disconnect,
5651 .o_init_export = mdt_init_export,
5652 .o_destroy_export = mdt_destroy_export,
5653 .o_iocontrol = mdt_iocontrol,
5654 .o_postrecov = mdt_obd_postrecov,
5655 .o_notify = mdt_obd_notify
5658 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5659 struct lu_device *d)
5661 struct mdt_device *m = mdt_dev(d);
5668 static struct lu_device *mdt_device_free(const struct lu_env *env,
5669 struct lu_device *d)
5671 struct mdt_device *m = mdt_dev(d);
5674 md_device_fini(&m->mdt_md_dev);
5679 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5680 struct lu_device_type *t,
5681 struct lustre_cfg *cfg)
5683 struct lu_device *l;
5684 struct mdt_device *m;
5690 l = &m->mdt_md_dev.md_lu_dev;
5691 rc = mdt_init0(env, m, t, cfg);
5693 mdt_device_free(env, l);
5697 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
5699 l = ERR_PTR(-ENOMEM);
5703 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5704 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
5706 /* context key: mdt_thread_key */
5707 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5709 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
5711 return md_ucred(info->mti_env);
5715 * Enable/disable COS (Commit On Sharing).
5717 * Set/Clear the COS flag in mdt options.
5719 * \param mdt mdt device
5720 * \param val 0 disables COS, other values enable COS
5722 void mdt_enable_cos(struct mdt_device *mdt, int val)
5727 mdt->mdt_opts.mo_cos = !!val;
5728 rc = lu_env_init(&env, LCT_MD_THREAD);
5729 if (unlikely(rc != 0)) {
5730 CWARN("lu_env initialization failed with rc = %d,"
5731 "cannot sync\n", rc);
5734 mdt_device_sync(&env, mdt);
5739 * Check COS (Commit On Sharing) status.
5741 * Return COS flag status.
5743 * \param mdt mdt device
5745 int mdt_cos_is_enabled(struct mdt_device *mdt)
5747 return mdt->mdt_opts.mo_cos != 0;
5750 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
5751 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key);
5753 static struct lu_device_type_operations mdt_device_type_ops = {
5754 .ldto_init = mdt_type_init,
5755 .ldto_fini = mdt_type_fini,
5757 .ldto_start = mdt_type_start,
5758 .ldto_stop = mdt_type_stop,
5760 .ldto_device_alloc = mdt_device_alloc,
5761 .ldto_device_free = mdt_device_free,
5762 .ldto_device_fini = mdt_device_fini
5765 static struct lu_device_type mdt_device_type = {
5766 .ldt_tags = LU_DEVICE_MD,
5767 .ldt_name = LUSTRE_MDT_NAME,
5768 .ldt_ops = &mdt_device_type_ops,
5769 .ldt_ctx_tags = LCT_MD_THREAD
5772 static struct lu_local_obj_desc mdt_last_recv = {
5773 .llod_name = LAST_RCVD,
5774 .llod_oid = MDT_LAST_RECV_OID,
5778 static int __init mdt_mod_init(void)
5780 struct lprocfs_static_vars lvars;
5783 llo_local_obj_register(&mdt_last_recv);
5785 if (mdt_num_threads > 0) {
5786 if (mdt_num_threads > MDT_MAX_THREADS)
5787 mdt_num_threads = MDT_MAX_THREADS;
5788 if (mdt_num_threads < MDT_MIN_THREADS)
5789 mdt_num_threads = MDT_MIN_THREADS;
5790 mdt_max_threads = mdt_min_threads = mdt_num_threads;
5792 mdt_max_threads = MDT_MAX_THREADS;
5793 mdt_min_threads = MDT_MIN_THREADS;
5796 lprocfs_mdt_init_vars(&lvars);
5797 rc = class_register_type(&mdt_obd_device_ops, NULL,
5798 lvars.module_vars, LUSTRE_MDT_NAME,
5804 static void __exit mdt_mod_exit(void)
5806 llo_local_obj_unregister(&mdt_last_recv);
5807 class_unregister_type(LUSTRE_MDT_NAME);
5811 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt) \
5812 [prefix ## _ ## opc - prefix ## _ ## base] = { \
5814 .mh_fail_id = OBD_FAIL_ ## prefix ## _ ## opc ## suffix, \
5815 .mh_opc = prefix ## _ ## opc, \
5816 .mh_flags = flags, \
5821 #define DEF_MDT_HNDL(flags, name, fn, fmt) \
5822 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
5824 #define DEF_SEQ_HNDL(flags, name, fn, fmt) \
5825 DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
5827 #define DEF_FLD_HNDL(flags, name, fn, fmt) \
5828 DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
5830 * Request with a format known in advance
5832 #define DEF_MDT_HNDL_F(flags, name, fn) \
5833 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
5835 #define DEF_SEQ_HNDL_F(flags, name, fn) \
5836 DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
5838 #define DEF_FLD_HNDL_F(flags, name, fn) \
5839 DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
5841 * Request with a format we do not yet know
5843 #define DEF_MDT_HNDL_0(flags, name, fn) \
5844 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
5846 static struct mdt_handler mdt_mds_ops[] = {
5847 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5848 DEF_MDT_HNDL_F(0, DISCONNECT, mdt_disconnect),
5849 DEF_MDT_HNDL (0, SET_INFO, mdt_set_info,
5851 DEF_MDT_HNDL_F(0, GET_INFO, mdt_get_info),
5852 DEF_MDT_HNDL_F(0 |HABEO_REFERO, GETSTATUS, mdt_getstatus),
5853 DEF_MDT_HNDL_F(HABEO_CORPUS, GETATTR, mdt_getattr),
5854 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
5855 DEF_MDT_HNDL_F(HABEO_CORPUS, GETXATTR, mdt_getxattr),
5856 DEF_MDT_HNDL_F(0 |HABEO_REFERO, STATFS, mdt_statfs),
5857 DEF_MDT_HNDL_F(0 |MUTABOR, REINT, mdt_reint),
5858 DEF_MDT_HNDL_F(HABEO_CORPUS, CLOSE, mdt_close),
5859 DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing),
5860 DEF_MDT_HNDL_F(0 |HABEO_REFERO, PIN, mdt_pin),
5861 DEF_MDT_HNDL_0(0, SYNC, mdt_sync),
5862 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir),
5863 #ifdef HAVE_QUOTA_SUPPORT
5864 DEF_MDT_HNDL_F(0, QUOTACHECK, mdt_quotacheck_handle),
5865 DEF_MDT_HNDL_F(0, QUOTACTL, mdt_quotactl_handle)
5869 #define DEF_OBD_HNDL(flags, name, fn) \
5870 DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
5873 static struct mdt_handler mdt_obd_ops[] = {
5874 DEF_OBD_HNDL(0, PING, mdt_obd_ping),
5875 DEF_OBD_HNDL(0, LOG_CANCEL, mdt_obd_log_cancel),
5876 DEF_OBD_HNDL(0, QC_CALLBACK, mdt_obd_qc_callback)
5879 #define DEF_DLM_HNDL_0(flags, name, fn) \
5880 DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
5881 #define DEF_DLM_HNDL_F(flags, name, fn) \
5882 DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
5884 static struct mdt_handler mdt_dlm_ops[] = {
5885 DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE, mdt_enqueue),
5886 DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT, mdt_convert),
5887 DEF_DLM_HNDL_0(0, BL_CALLBACK, mdt_bl_callback),
5888 DEF_DLM_HNDL_0(0, CP_CALLBACK, mdt_cp_callback)
5891 #define DEF_LLOG_HNDL(flags, name, fn) \
5892 DEF_HNDL(LLOG, ORIGIN_HANDLE_CREATE, _NET, flags, name, fn, NULL)
5894 static struct mdt_handler mdt_llog_ops[] = {
5895 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CREATE, mdt_llog_create),
5896 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_NEXT_BLOCK, mdt_llog_next_block),
5897 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_READ_HEADER, mdt_llog_read_header),
5898 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_WRITE_REC, NULL),
5899 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CLOSE, NULL),
5900 DEF_LLOG_HNDL(0, ORIGIN_CONNECT, NULL),
5901 DEF_LLOG_HNDL(0, CATINFO, NULL),
5902 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_PREV_BLOCK, mdt_llog_prev_block),
5903 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_DESTROY, mdt_llog_destroy),
5906 #define DEF_SEC_CTX_HNDL(name, fn) \
5907 DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
5909 static struct mdt_handler mdt_sec_ctx_ops[] = {
5910 DEF_SEC_CTX_HNDL(INIT, mdt_sec_ctx_handle),
5911 DEF_SEC_CTX_HNDL(INIT_CONT, mdt_sec_ctx_handle),
5912 DEF_SEC_CTX_HNDL(FINI, mdt_sec_ctx_handle)
5915 static struct mdt_opc_slice mdt_regular_handlers[] = {
5917 .mos_opc_start = MDS_GETATTR,
5918 .mos_opc_end = MDS_LAST_OPC,
5919 .mos_hs = mdt_mds_ops
5922 .mos_opc_start = OBD_PING,
5923 .mos_opc_end = OBD_LAST_OPC,
5924 .mos_hs = mdt_obd_ops
5927 .mos_opc_start = LDLM_ENQUEUE,
5928 .mos_opc_end = LDLM_LAST_OPC,
5929 .mos_hs = mdt_dlm_ops
5932 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
5933 .mos_opc_end = LLOG_LAST_OPC,
5934 .mos_hs = mdt_llog_ops
5937 .mos_opc_start = SEC_CTX_INIT,
5938 .mos_opc_end = SEC_LAST_OPC,
5939 .mos_hs = mdt_sec_ctx_ops
5946 static struct mdt_handler mdt_readpage_ops[] = {
5947 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5948 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
5949 #ifdef HAVE_SPLIT_SUPPORT
5950 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
5954 * XXX: this is ugly and should be fixed one day, see mdc_close() for
5955 * detailed comments. --umka
5957 DEF_MDT_HNDL_F(HABEO_CORPUS, CLOSE, mdt_close),
5958 DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing),
5961 static struct mdt_opc_slice mdt_readpage_handlers[] = {
5963 .mos_opc_start = MDS_GETATTR,
5964 .mos_opc_end = MDS_LAST_OPC,
5965 .mos_hs = mdt_readpage_ops
5972 static struct mdt_handler mdt_xmds_ops[] = {
5973 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5974 DEF_MDT_HNDL_F(HABEO_CORPUS , GETATTR, mdt_getattr),
5975 DEF_MDT_HNDL_F(0 | MUTABOR , REINT, mdt_reint),
5976 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir),
5979 static struct mdt_opc_slice mdt_xmds_handlers[] = {
5981 .mos_opc_start = MDS_GETATTR,
5982 .mos_opc_end = MDS_LAST_OPC,
5983 .mos_hs = mdt_xmds_ops
5986 .mos_opc_start = OBD_PING,
5987 .mos_opc_end = OBD_LAST_OPC,
5988 .mos_hs = mdt_obd_ops
5991 .mos_opc_start = SEC_CTX_INIT,
5992 .mos_opc_end = SEC_LAST_OPC,
5993 .mos_hs = mdt_sec_ctx_ops
6000 static struct mdt_handler mdt_seq_ops[] = {
6001 DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
6004 static struct mdt_opc_slice mdt_seq_handlers[] = {
6006 .mos_opc_start = SEQ_QUERY,
6007 .mos_opc_end = SEQ_LAST_OPC,
6008 .mos_hs = mdt_seq_ops
6015 static struct mdt_handler mdt_fld_ops[] = {
6016 DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
6019 static struct mdt_opc_slice mdt_fld_handlers[] = {
6021 .mos_opc_start = FLD_QUERY,
6022 .mos_opc_end = FLD_LAST_OPC,
6023 .mos_hs = mdt_fld_ops
6030 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6031 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
6032 MODULE_LICENSE("GPL");
6034 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
6035 "number of mdt service threads to start");
6037 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);