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 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdt/mdt_handler.c
38 * Lustre Metadata Target (mdt) request handler
40 * Author: Peter Braam <braam@clusterfs.com>
41 * Author: Andreas Dilger <adilger@clusterfs.com>
42 * Author: Phil Schwan <phil@clusterfs.com>
43 * Author: Mike Shaver <shaver@clusterfs.com>
44 * Author: Nikita Danilov <nikita@clusterfs.com>
45 * Author: Huang Hua <huanghua@clusterfs.com>
46 * Author: Yury Umanets <umka@clusterfs.com>
50 # define EXPORT_SYMTAB
52 #define DEBUG_SUBSYSTEM S_MDS
54 #include <linux/module.h>
56 * struct OBD_{ALLOC,FREE}*()
58 #include <obd_support.h>
59 /* struct ptlrpc_request */
60 #include <lustre_net.h>
61 /* struct obd_export */
62 #include <lustre_export.h>
63 /* struct obd_device */
66 #include <dt_object.h>
67 #include <lustre_mds.h>
68 #include <lustre_mdt.h>
69 #include "mdt_internal.h"
70 #ifdef HAVE_QUOTA_SUPPORT
71 # include <lustre_quota.h>
73 #include <lustre_acl.h>
74 #include <lustre_param.h>
75 #include <lustre_fsfilt.h>
77 mdl_mode_t mdt_mdl_lock_modes[] = {
78 [LCK_MINMODE] = MDL_MINMODE,
85 [LCK_GROUP] = MDL_GROUP
88 ldlm_mode_t mdt_dlm_lock_modes[] = {
89 [MDL_MINMODE] = LCK_MINMODE,
96 [MDL_GROUP] = LCK_GROUP
100 * Initialized in mdt_mod_init().
102 unsigned long mdt_num_threads;
104 /* ptlrpc request handler for MDT. All handlers are
105 * grouped into several slices - struct mdt_opc_slice,
106 * and stored in an array - mdt_handlers[].
109 /* The name of this handler. */
111 /* Fail id for this handler, checked at the beginning of this handler*/
113 /* Operation code for this handler */
115 /* flags are listed in enum mdt_handler_flags below. */
117 /* The actual handler function to execute. */
118 int (*mh_act)(struct mdt_thread_info *info);
119 /* Request format for this request. */
120 const struct req_format *mh_fmt;
123 enum mdt_handler_flags {
125 * struct mdt_body is passed in the incoming message, and object
126 * identified by this fid exists on disk.
128 * "habeo corpus" == "I have a body"
130 HABEO_CORPUS = (1 << 0),
132 * struct ldlm_request is passed in the incoming message.
134 * "habeo clavis" == "I have a key"
136 HABEO_CLAVIS = (1 << 1),
138 * this request has fixed reply format, so that reply message can be
139 * packed by generic code.
141 * "habeo refero" == "I have a reply"
143 HABEO_REFERO = (1 << 2),
145 * this request will modify something, so check whether the filesystem
146 * is readonly or not, then return -EROFS to client asap if necessary.
148 * "mutabor" == "I shall modify"
153 struct mdt_opc_slice {
156 struct mdt_handler *mos_hs;
159 static struct mdt_opc_slice mdt_regular_handlers[];
160 static struct mdt_opc_slice mdt_readpage_handlers[];
161 static struct mdt_opc_slice mdt_xmds_handlers[];
162 static struct mdt_opc_slice mdt_seq_handlers[];
163 static struct mdt_opc_slice mdt_fld_handlers[];
165 static struct mdt_device *mdt_dev(struct lu_device *d);
166 static int mdt_regular_handle(struct ptlrpc_request *req);
167 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
168 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
169 struct getinfo_fid2path *fp);
171 static const struct lu_object_operations mdt_obj_ops;
173 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
177 return (rep->lock_policy_res1 & flag);
180 void mdt_clear_disposition(struct mdt_thread_info *info,
181 struct ldlm_reply *rep, int flag)
184 info->mti_opdata &= ~flag;
186 rep->lock_policy_res1 &= ~flag;
189 void mdt_set_disposition(struct mdt_thread_info *info,
190 struct ldlm_reply *rep, int flag)
193 info->mti_opdata |= flag;
195 rep->lock_policy_res1 |= flag;
198 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
200 lh->mlh_pdo_hash = 0;
201 lh->mlh_reg_mode = lm;
202 lh->mlh_type = MDT_REG_LOCK;
205 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
206 const char *name, int namelen)
208 lh->mlh_reg_mode = lm;
209 lh->mlh_type = MDT_PDO_LOCK;
212 LASSERT(namelen > 0);
213 lh->mlh_pdo_hash = full_name_hash(name, namelen);
215 LASSERT(namelen == 0);
216 lh->mlh_pdo_hash = 0ull;
220 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
221 struct mdt_lock_handle *lh)
227 * Any dir access needs couple of locks:
229 * 1) on part of dir we gonna take lookup/modify;
231 * 2) on whole dir to protect it from concurrent splitting and/or to
232 * flush client's cache for readdir().
234 * so, for a given mode and object this routine decides what lock mode
235 * to use for lock #2:
237 * 1) if caller's gonna lookup in dir then we need to protect dir from
238 * being splitted only - LCK_CR
240 * 2) if caller's gonna modify dir then we need to protect dir from
241 * being splitted and to flush cache - LCK_CW
243 * 3) if caller's gonna modify dir and that dir seems ready for
244 * splitting then we need to protect it from any type of access
245 * (lookup/modify/split) - LCK_EX --bzzz
248 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
249 LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
252 * Ask underlaying level its opinion about preferable PDO lock mode
253 * having access type passed as regular lock mode:
255 * - MDL_MINMODE means that lower layer does not want to specify lock
258 * - MDL_NL means that no PDO lock should be taken. This is used in some
259 * cases. Say, for non-splittable directories no need to use PDO locks
262 mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
263 mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
265 if (mode != MDL_MINMODE) {
266 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
269 * Lower layer does not want to specify locking mode. We do it
270 * our selves. No special protection is needed, just flush
271 * client's cache on modification and allow concurrent
274 switch (lh->mlh_reg_mode) {
276 lh->mlh_pdo_mode = LCK_EX;
279 lh->mlh_pdo_mode = LCK_CR;
282 lh->mlh_pdo_mode = LCK_CW;
285 CERROR("Not expected lock type (0x%x)\n",
286 (int)lh->mlh_reg_mode);
291 LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
295 static int mdt_getstatus(struct mdt_thread_info *info)
297 struct mdt_device *mdt = info->mti_mdt;
298 struct md_device *next = mdt->mdt_child;
299 struct mdt_body *repbody;
304 rc = mdt_check_ucred(info);
306 RETURN(err_serious(rc));
308 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
309 RETURN(err_serious(-ENOMEM));
311 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
312 rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
316 repbody->valid |= OBD_MD_FLID;
318 if (mdt->mdt_opts.mo_mds_capa &&
319 info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
320 struct mdt_object *root;
321 struct lustre_capa *capa;
323 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
325 RETURN(PTR_ERR(root));
327 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
329 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
330 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
332 mdt_object_put(info->mti_env, root);
334 repbody->valid |= OBD_MD_FLMDSCAPA;
340 static int mdt_statfs(struct mdt_thread_info *info)
342 struct md_device *next = info->mti_mdt->mdt_child;
343 struct ptlrpc_service *svc;
344 struct obd_statfs *osfs;
349 svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
351 /* This will trigger a watchdog timeout */
352 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
353 (MDT_SERVICE_WATCHDOG_FACTOR *
354 at_get(&svc->srv_at_estimate)) + 1);
356 rc = mdt_check_ucred(info);
358 RETURN(err_serious(rc));
360 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
361 rc = err_serious(-ENOMEM);
363 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
364 rc = next->md_ops->mdo_statfs(info->mti_env, next,
366 statfs_pack(osfs, &info->mti_u.ksfs);
371 void mdt_pack_size2body(struct mdt_thread_info *info, struct mdt_object *o)
374 struct lu_attr *attr = &info->mti_attr.ma_attr;
376 b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
378 /* Check if Size-on-MDS is enabled. */
379 if ((mdt_conn_flags(info) & OBD_CONNECT_SOM) &&
380 S_ISREG(attr->la_mode) && mdt_sizeonmds_enabled(o)) {
381 b->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
382 b->size = attr->la_size;
383 b->blocks = attr->la_blocks;
387 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
388 const struct lu_attr *attr, const struct lu_fid *fid)
390 /*XXX should pack the reply body according to lu_valid*/
391 b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID |
392 OBD_MD_FLGID | OBD_MD_FLTYPE |
393 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
394 OBD_MD_FLATIME | OBD_MD_FLMTIME ;
396 if (!S_ISREG(attr->la_mode))
397 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
399 b->atime = attr->la_atime;
400 b->mtime = attr->la_mtime;
401 b->ctime = attr->la_ctime;
402 b->mode = attr->la_mode;
403 b->size = attr->la_size;
404 b->blocks = attr->la_blocks;
405 b->uid = attr->la_uid;
406 b->gid = attr->la_gid;
407 b->flags = attr->la_flags;
408 b->nlink = attr->la_nlink;
409 b->rdev = attr->la_rdev;
413 b->valid |= OBD_MD_FLID;
415 /* FIXME: these should be fixed when new igif ready.*/
416 b->ino = fid_oid(fid); /* 1.6 compatibility */
417 b->generation = fid_ver(fid); /* 1.6 compatibility */
418 b->valid |= OBD_MD_FLGENER; /* 1.6 compatibility */
420 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
421 PFID(fid), b->nlink, b->mode, b->size);
425 mdt_body_reverse_idmap(info, b);
428 static inline int mdt_body_has_lov(const struct lu_attr *la,
429 const struct mdt_body *body)
431 return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
432 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
435 static int mdt_getattr_internal(struct mdt_thread_info *info,
436 struct mdt_object *o)
438 struct md_object *next = mdt_object_child(o);
439 const struct mdt_body *reqbody = info->mti_body;
440 struct ptlrpc_request *req = mdt_info_req(info);
441 struct md_attr *ma = &info->mti_attr;
442 struct lu_attr *la = &ma->ma_attr;
443 struct req_capsule *pill = info->mti_pill;
444 const struct lu_env *env = info->mti_env;
445 struct mdt_body *repbody;
446 struct lu_buf *buffer = &info->mti_buf;
450 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
451 RETURN(err_serious(-ENOMEM));
453 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
457 rc = mdt_object_exists(o);
459 /* This object is located on remote node.*/
460 repbody->fid1 = *mdt_object_fid(o);
461 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
465 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
466 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
468 /* If it is dir object and client require MEA, then we got MEA */
469 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
470 reqbody->valid & OBD_MD_MEA) {
471 /* Assumption: MDT_MD size is enough for lmv size. */
472 ma->ma_lmv = buffer->lb_buf;
473 ma->ma_lmv_size = buffer->lb_len;
474 ma->ma_need = MA_LMV | MA_INODE;
476 ma->ma_lmm = buffer->lb_buf;
477 ma->ma_lmm_size = buffer->lb_len;
478 ma->ma_need = MA_LOV | MA_INODE;
481 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
482 reqbody->valid & OBD_MD_FLDIREA &&
483 lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
484 /* get default stripe info for this dir. */
485 ma->ma_need |= MA_LOV_DEF;
487 rc = mo_attr_get(env, next, ma);
489 CERROR("getattr error for "DFID": %d\n",
490 PFID(mdt_object_fid(o)), rc);
494 if (likely(ma->ma_valid & MA_INODE))
495 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
499 if (mdt_body_has_lov(la, reqbody)) {
500 if (ma->ma_valid & MA_LOV) {
501 LASSERT(ma->ma_lmm_size);
502 mdt_dump_lmm(D_INFO, ma->ma_lmm);
503 repbody->eadatasize = ma->ma_lmm_size;
504 if (S_ISDIR(la->la_mode))
505 repbody->valid |= OBD_MD_FLDIREA;
507 repbody->valid |= OBD_MD_FLEASIZE;
509 if (ma->ma_valid & MA_LMV) {
510 LASSERT(S_ISDIR(la->la_mode));
511 repbody->eadatasize = ma->ma_lmv_size;
512 repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
514 if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
515 repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
517 } else if (S_ISLNK(la->la_mode) &&
518 reqbody->valid & OBD_MD_LINKNAME) {
519 buffer->lb_buf = ma->ma_lmm;
520 buffer->lb_len = reqbody->eadatasize;
521 rc = mo_readlink(env, next, buffer);
522 if (unlikely(rc <= 0)) {
523 CERROR("readlink failed: %d\n", rc);
526 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
528 repbody->valid |= OBD_MD_LINKNAME;
529 repbody->eadatasize = rc;
531 ((char*)ma->ma_lmm)[rc - 1] = 0;
532 CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
533 (char*)ma->ma_lmm, rc);
538 if (reqbody->valid & OBD_MD_FLMODEASIZE) {
539 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
540 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
541 repbody->valid |= OBD_MD_FLMODEASIZE;
542 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
543 "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
544 repbody->max_cookiesize);
547 if (exp_connect_rmtclient(info->mti_exp) &&
548 reqbody->valid & OBD_MD_FLRMTPERM) {
549 void *buf = req_capsule_server_get(pill, &RMF_ACL);
551 /* mdt_getattr_lock only */
552 rc = mdt_pack_remote_perm(info, o, buf);
554 repbody->valid &= ~OBD_MD_FLRMTPERM;
555 repbody->aclsize = 0;
558 repbody->valid |= OBD_MD_FLRMTPERM;
559 repbody->aclsize = sizeof(struct mdt_remote_perm);
562 #ifdef CONFIG_FS_POSIX_ACL
563 else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
564 (reqbody->valid & OBD_MD_FLACL)) {
565 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
566 buffer->lb_len = req_capsule_get_size(pill,
567 &RMF_ACL, RCL_SERVER);
568 if (buffer->lb_len > 0) {
569 rc = mo_xattr_get(env, next, buffer,
570 XATTR_NAME_ACL_ACCESS);
572 if (rc == -ENODATA) {
573 repbody->aclsize = 0;
574 repbody->valid |= OBD_MD_FLACL;
576 } else if (rc == -EOPNOTSUPP) {
579 CERROR("got acl size: %d\n", rc);
582 repbody->aclsize = rc;
583 repbody->valid |= OBD_MD_FLACL;
590 if (reqbody->valid & OBD_MD_FLMDSCAPA &&
591 info->mti_mdt->mdt_opts.mo_mds_capa &&
592 info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
593 struct lustre_capa *capa;
595 capa = req_capsule_server_get(pill, &RMF_CAPA1);
597 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
598 rc = mo_capa_get(env, next, capa, 0);
601 repbody->valid |= OBD_MD_FLMDSCAPA;
606 static int mdt_renew_capa(struct mdt_thread_info *info)
608 struct mdt_object *obj = info->mti_object;
609 struct mdt_body *body;
610 struct lustre_capa *capa, *c;
614 /* if object doesn't exist, or server has disabled capability,
615 * return directly, client will find body->valid OBD_MD_FLOSSCAPA
618 if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
619 !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
622 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
623 LASSERT(body != NULL);
625 c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
628 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
632 rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
634 body->valid |= OBD_MD_FLOSSCAPA;
638 static int mdt_getattr(struct mdt_thread_info *info)
640 struct mdt_object *obj = info->mti_object;
641 struct req_capsule *pill = info->mti_pill;
642 struct mdt_body *reqbody;
643 struct mdt_body *repbody;
649 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
652 if (reqbody->valid & OBD_MD_FLOSSCAPA) {
653 rc = req_capsule_server_pack(pill);
655 RETURN(err_serious(rc));
656 rc = mdt_renew_capa(info);
657 GOTO(out_shrink, rc);
660 LASSERT(obj != NULL);
661 LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
663 mode = lu_object_attr(&obj->mot_obj.mo_lu);
664 if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
665 (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
666 md_size = reqbody->eadatasize;
668 md_size = info->mti_mdt->mdt_max_mdsize;
670 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
672 rc = req_capsule_server_pack(pill);
673 if (unlikely(rc != 0))
674 RETURN(err_serious(rc));
676 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
677 LASSERT(repbody != NULL);
678 repbody->eadatasize = 0;
679 repbody->aclsize = 0;
681 if (reqbody->valid & OBD_MD_FLRMTPERM)
682 rc = mdt_init_ucred(info, reqbody);
684 rc = mdt_check_ucred(info);
686 GOTO(out_shrink, rc);
688 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
689 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
692 * Don't check capability at all, because rename might getattr for
693 * remote obj, and at that time no capability is available.
695 mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
696 rc = mdt_getattr_internal(info, obj);
697 if (reqbody->valid & OBD_MD_FLRMTPERM)
698 mdt_exit_ucred(info);
701 mdt_shrink_reply(info);
705 static int mdt_is_subdir(struct mdt_thread_info *info)
707 struct mdt_object *o = info->mti_object;
708 struct req_capsule *pill = info->mti_pill;
709 const struct mdt_body *body = info->mti_body;
710 struct mdt_body *repbody;
716 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
719 * We save last checked parent fid to @repbody->fid1 for remote
722 LASSERT(fid_is_sane(&body->fid2));
723 LASSERT(mdt_object_exists(o) > 0);
724 rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
725 &body->fid2, &repbody->fid1);
726 if (rc == 0 || rc == -EREMOTE)
727 repbody->valid |= OBD_MD_FLID;
732 static int mdt_raw_lookup(struct mdt_thread_info *info,
733 struct mdt_object *parent,
734 const struct lu_name *lname,
735 struct ldlm_reply *ldlm_rep)
737 struct md_object *next = mdt_object_child(info->mti_object);
738 const struct mdt_body *reqbody = info->mti_body;
739 struct lu_fid *child_fid = &info->mti_tmp_fid1;
740 struct mdt_body *repbody;
744 if (reqbody->valid != OBD_MD_FLID)
747 LASSERT(!info->mti_cross_ref);
749 /* Only got the fid of this obj by name */
750 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
753 /* XXX is raw_lookup possible as intent operation? */
756 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
759 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
761 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
764 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
765 repbody->fid1 = *child_fid;
766 repbody->valid = OBD_MD_FLID;
772 * UPDATE lock should be taken against parent, and be release before exit;
773 * child_bits lock should be taken against child, and be returned back:
774 * (1)normal request should release the child lock;
775 * (2)intent request will grant the lock to client.
777 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
778 struct mdt_lock_handle *lhc,
780 struct ldlm_reply *ldlm_rep)
782 struct ptlrpc_request *req = mdt_info_req(info);
783 struct mdt_body *reqbody = NULL;
784 struct mdt_object *parent = info->mti_object;
785 struct mdt_object *child;
786 struct md_object *next = mdt_object_child(parent);
787 struct lu_fid *child_fid = &info->mti_tmp_fid1;
788 struct lu_name *lname = NULL;
789 const char *name = NULL;
791 struct mdt_lock_handle *lhp;
792 struct ldlm_lock *lock;
793 struct ldlm_res_id *res_id;
799 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
800 LASSERT(ergo(is_resent,
801 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
803 LASSERT(parent != NULL);
804 name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
806 RETURN(err_serious(-EFAULT));
808 namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
810 if (!info->mti_cross_ref) {
812 * XXX: Check for "namelen == 0" is for getattr by fid
813 * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
814 * that is the name must contain at least one character and
815 * the terminating '\0'
818 reqbody = req_capsule_client_get(info->mti_pill,
820 LASSERT(fid_is_sane(&reqbody->fid2));
823 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
825 PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
828 lname = mdt_name(info->mti_env, (char *)name, namelen);
829 CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
830 "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
834 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
836 rc = mdt_object_exists(parent);
837 if (unlikely(rc == 0)) {
838 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
839 &parent->mot_obj.mo_lu,
840 "Parent doesn't exist!\n");
842 } else if (!info->mti_cross_ref) {
843 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
844 PFID(mdt_object_fid(parent)));
847 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
855 if (info->mti_cross_ref) {
856 /* Only getattr on the child. Parent is on another node. */
857 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
859 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
860 "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
863 /* Do not take lock for resent case. */
864 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
865 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
866 lhc->mlh_reg_lh.cookie);
867 LASSERT(fid_res_name_eq(mdt_object_fid(child),
868 &lock->l_resource->lr_name));
872 mdt_lock_handle_init(lhc);
873 mdt_lock_reg_init(lhc, LCK_PR);
876 * Object's name is on another MDS, no lookup lock is
877 * needed here but update is.
879 child_bits &= ~MDS_INODELOCK_LOOKUP;
880 child_bits |= MDS_INODELOCK_UPDATE;
882 rc = mdt_object_lock(info, child, lhc, child_bits,
886 /* Finally, we can get attr for child. */
887 mdt_set_capainfo(info, 0, mdt_object_fid(child),
889 rc = mdt_getattr_internal(info, child);
890 if (unlikely(rc != 0))
891 mdt_object_unlock(info, child, lhc, 1);
896 /* step 1: lock parent */
897 lhp = &info->mti_lh[MDT_LH_PARENT];
898 mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
899 rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
902 if (unlikely(rc != 0))
906 /* step 2: lookup child's fid by name */
907 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
912 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
913 GOTO(out_parent, rc);
915 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
917 *child_fid = reqbody->fid2;
918 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
922 *step 3: find the child object by fid & lock it.
923 * regardless if it is local or remote.
925 child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
927 if (unlikely(IS_ERR(child)))
928 GOTO(out_parent, rc = PTR_ERR(child));
930 /* Do not take lock for resent case. */
931 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
932 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
933 lhc->mlh_reg_lh.cookie);
935 res_id = &lock->l_resource->lr_name;
936 if (!fid_res_name_eq(mdt_object_fid(child),
937 &lock->l_resource->lr_name)) {
938 LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
939 &lock->l_resource->lr_name),
940 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
941 (unsigned long)res_id->name[0],
942 (unsigned long)res_id->name[1],
943 (unsigned long)res_id->name[2],
944 PFID(mdt_object_fid(parent)));
945 CWARN("Although resent, but still not get child lock"
946 "parent:"DFID" child:"DFID"\n",
947 PFID(mdt_object_fid(parent)),
948 PFID(mdt_object_fid(child)));
949 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
958 ma = &info->mti_attr;
960 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
961 mdt_lock_handle_init(lhc);
962 mdt_lock_reg_init(lhc, LCK_PR);
964 if (mdt_object_exists(child) == 0) {
965 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
966 &child->mot_obj.mo_lu,
967 "Object doesn't exist!\n");
968 GOTO(out_child, rc = -ESTALE);
972 ma->ma_need = MA_INODE;
973 rc = mo_attr_get(info->mti_env, next, ma);
974 if (unlikely(rc != 0))
977 /* If the file has not been changed for some time, we return
978 * not only a LOOKUP lock, but also an UPDATE lock and this
979 * might save us RPC on later STAT. For directories, it also
980 * let negative dentry starts working for this dir. */
981 if (ma->ma_valid & MA_INODE &&
982 ma->ma_attr.la_valid & LA_CTIME &&
983 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
984 ma->ma_attr.la_ctime < cfs_time_current_sec())
985 child_bits |= MDS_INODELOCK_UPDATE;
987 rc = mdt_object_lock(info, child, lhc, child_bits,
990 if (unlikely(rc != 0))
994 /* finally, we can get attr for child. */
995 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
996 rc = mdt_getattr_internal(info, child);
997 if (unlikely(rc != 0)) {
998 mdt_object_unlock(info, child, lhc, 1);
1000 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1002 struct mdt_body *repbody;
1004 /* Debugging code. */
1005 res_id = &lock->l_resource->lr_name;
1006 LDLM_DEBUG(lock, "Returning lock to client");
1007 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1008 &lock->l_resource->lr_name),
1009 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1010 (unsigned long)res_id->name[0],
1011 (unsigned long)res_id->name[1],
1012 (unsigned long)res_id->name[2],
1013 PFID(mdt_object_fid(child)));
1015 * Pack Size-on-MDS inode attributes to the body if
1016 * update lock is given.
1018 repbody = req_capsule_server_get(info->mti_pill,
1020 if (lock->l_policy_data.l_inodebits.bits &
1021 MDS_INODELOCK_UPDATE)
1022 mdt_pack_size2body(info, child);
1023 LDLM_LOCK_PUT(lock);
1028 mdt_object_put(info->mti_env, child);
1030 mdt_object_unlock(info, parent, lhp, 1);
1034 /* normal handler: should release the child lock */
1035 static int mdt_getattr_name(struct mdt_thread_info *info)
1037 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1038 struct mdt_body *reqbody;
1039 struct mdt_body *repbody;
1043 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1044 LASSERT(reqbody != NULL);
1045 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1046 LASSERT(repbody != NULL);
1048 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1049 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1050 repbody->eadatasize = 0;
1051 repbody->aclsize = 0;
1053 rc = mdt_init_ucred(info, reqbody);
1055 GOTO(out_shrink, rc);
1057 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1058 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1059 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1060 lhc->mlh_reg_lh.cookie = 0;
1062 mdt_exit_ucred(info);
1065 mdt_shrink_reply(info);
1069 static const struct lu_device_operations mdt_lu_ops;
1071 static int lu_device_is_mdt(struct lu_device *d)
1073 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1076 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1077 void *karg, void *uarg);
1079 static int mdt_set_info(struct mdt_thread_info *info)
1081 struct ptlrpc_request *req = mdt_info_req(info);
1084 int keylen, vallen, rc = 0;
1087 rc = req_capsule_server_pack(info->mti_pill);
1091 key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1093 DEBUG_REQ(D_HA, req, "no set_info key");
1097 keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1100 val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1102 DEBUG_REQ(D_HA, req, "no set_info val");
1106 vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1109 /* Swab any part of val you need to here */
1110 if (KEY_IS(KEY_READ_ONLY)) {
1112 lustre_msg_set_status(req->rq_repmsg, 0);
1114 spin_lock(&req->rq_export->exp_lock);
1116 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1118 req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
1119 spin_unlock(&req->rq_export->exp_lock);
1121 } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1122 struct changelog_setinfo *cs =
1123 (struct changelog_setinfo *)val;
1124 if (vallen != sizeof(*cs)) {
1125 CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1128 if (ptlrpc_req_need_swab(req)) {
1129 __swab64s(&cs->cs_recno);
1130 __swab32s(&cs->cs_id);
1133 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1135 lustre_msg_set_status(req->rq_repmsg, rc);
1143 static int mdt_connect(struct mdt_thread_info *info)
1146 struct ptlrpc_request *req;
1148 req = mdt_info_req(info);
1149 rc = target_handle_connect(req);
1151 LASSERT(req->rq_export != NULL);
1152 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1153 rc = mdt_init_sec_level(info);
1155 rc = mdt_init_idmap(info);
1157 obd_disconnect(class_export_get(req->rq_export));
1159 rc = err_serious(rc);
1164 static int mdt_disconnect(struct mdt_thread_info *info)
1169 rc = target_handle_disconnect(mdt_info_req(info));
1171 rc = err_serious(rc);
1175 static int mdt_sendpage(struct mdt_thread_info *info,
1176 struct lu_rdpg *rdpg)
1178 struct ptlrpc_request *req = mdt_info_req(info);
1179 struct ptlrpc_bulk_desc *desc;
1180 struct l_wait_info *lwi = &info->mti_u.rdpg.mti_wait_info;
1188 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1193 for (i = 0, tmpcount = rdpg->rp_count;
1194 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1195 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1196 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1199 LASSERT(desc->bd_nob == rdpg->rp_count);
1200 rc = sptlrpc_svc_wrap_bulk(req, desc);
1202 GOTO(free_desc, rc);
1204 rc = ptlrpc_start_bulk_transfer(desc);
1206 GOTO(free_desc, rc);
1208 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1209 GOTO(abort_bulk, rc = 0);
1211 timeout = (int) req->rq_deadline - cfs_time_current_sec();
1213 CERROR("Req deadline already passed %lu (now: %lu)\n",
1214 req->rq_deadline, cfs_time_current_sec());
1215 *lwi = LWI_TIMEOUT(cfs_time_seconds(max(timeout, 1)), NULL, NULL);
1216 rc = l_wait_event(desc->bd_waitq, !ptlrpc_server_bulk_active(desc), lwi);
1217 LASSERT (rc == 0 || rc == -ETIMEDOUT);
1220 if (desc->bd_success &&
1221 desc->bd_nob_transferred == rdpg->rp_count)
1222 GOTO(free_desc, rc);
1224 rc = -ETIMEDOUT; /* XXX should this be a different errno? */
1227 DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1228 (rc == -ETIMEDOUT) ? "timeout" : "network error",
1229 desc->bd_nob_transferred, rdpg->rp_count,
1230 req->rq_export->exp_client_uuid.uuid,
1231 req->rq_export->exp_connection->c_remote_uuid.uuid);
1233 class_fail_export(req->rq_export);
1237 ptlrpc_abort_bulk(desc);
1239 ptlrpc_free_bulk(desc);
1243 #ifdef HAVE_SPLIT_SUPPORT
1245 * Retrieve dir entry from the page and insert it to the slave object, actually,
1246 * this should be in osd layer, but since it will not in the final product, so
1247 * just do it here and do not define more moo api anymore for this.
1249 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1252 struct mdt_object *object = info->mti_object;
1253 struct lu_fid *lf = &info->mti_tmp_fid2;
1254 struct md_attr *ma = &info->mti_attr;
1255 struct lu_dirpage *dp;
1256 struct lu_dirent *ent;
1257 int rc = 0, offset = 0;
1260 /* Make sure we have at least one entry. */
1265 * Disable trans for this name insert, since it will include many trans
1268 info->mti_no_need_trans = 1;
1270 * When write_dir_page, no need update parent's ctime,
1271 * and no permission check for name_insert.
1273 ma->ma_attr.la_ctime = 0;
1274 ma->ma_attr.la_valid = LA_MODE;
1275 ma->ma_valid = MA_INODE;
1278 dp = page_address(page);
1279 offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1281 for (ent = lu_dirent_start(dp); ent != NULL;
1282 ent = lu_dirent_next(ent)) {
1283 struct lu_name *lname;
1286 if (le16_to_cpu(ent->lde_namelen) == 0)
1289 fid_le_to_cpu(lf, &ent->lde_fid);
1290 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1291 ma->ma_attr.la_mode = S_IFDIR;
1293 ma->ma_attr.la_mode = 0;
1294 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1296 GOTO(out, rc = -ENOMEM);
1298 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1299 lname = mdt_name(info->mti_env, name,
1300 le16_to_cpu(ent->lde_namelen));
1301 ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE);
1302 rc = mdo_name_insert(info->mti_env,
1303 md_object_next(&object->mot_obj),
1305 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1307 CERROR("Can't insert %*.*s, rc %d\n",
1308 le16_to_cpu(ent->lde_namelen),
1309 le16_to_cpu(ent->lde_namelen),
1314 offset += lu_dirent_size(ent);
1324 static int mdt_bulk_timeout(void *data)
1328 CERROR("mdt bulk transfer timeout \n");
1333 static int mdt_writepage(struct mdt_thread_info *info)
1335 struct ptlrpc_request *req = mdt_info_req(info);
1336 struct mdt_body *reqbody;
1337 struct l_wait_info *lwi;
1338 struct ptlrpc_bulk_desc *desc;
1344 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1345 if (reqbody == NULL)
1346 RETURN(err_serious(-EFAULT));
1348 desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1350 RETURN(err_serious(-ENOMEM));
1352 /* allocate the page for the desc */
1353 page = cfs_alloc_page(CFS_ALLOC_STD);
1355 GOTO(desc_cleanup, rc = -ENOMEM);
1357 CDEBUG(D_INFO, "Received page offset %d size %d \n",
1358 (int)reqbody->size, (int)reqbody->nlink);
1360 ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1361 (int)reqbody->nlink);
1363 rc = sptlrpc_svc_prep_bulk(req, desc);
1365 GOTO(cleanup_page, rc);
1367 * Check if client was evicted while we were doing i/o before touching
1372 GOTO(cleanup_page, rc = -ENOMEM);
1374 if (desc->bd_export->exp_failed)
1377 rc = ptlrpc_start_bulk_transfer (desc);
1379 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
1380 mdt_bulk_timeout, desc);
1381 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1382 desc->bd_export->exp_failed, lwi);
1383 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1384 if (rc == -ETIMEDOUT) {
1385 DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1386 ptlrpc_abort_bulk(desc);
1387 } else if (desc->bd_export->exp_failed) {
1388 DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1390 ptlrpc_abort_bulk(desc);
1391 } else if (!desc->bd_success ||
1392 desc->bd_nob_transferred != desc->bd_nob) {
1393 DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1395 "truncated" : "network error on",
1396 desc->bd_nob_transferred, desc->bd_nob);
1397 /* XXX should this be a different errno? */
1401 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1404 GOTO(cleanup_lwi, rc);
1405 rc = mdt_write_dir_page(info, page, reqbody->nlink);
1410 cfs_free_page(page);
1412 ptlrpc_free_bulk(desc);
1417 static int mdt_readpage(struct mdt_thread_info *info)
1419 struct mdt_object *object = info->mti_object;
1420 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1421 struct mdt_body *reqbody;
1422 struct mdt_body *repbody;
1427 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1428 RETURN(err_serious(-ENOMEM));
1430 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1431 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1432 if (reqbody == NULL || repbody == NULL)
1433 RETURN(err_serious(-EFAULT));
1436 * prepare @rdpg before calling lower layers and transfer itself. Here
1437 * reqbody->size contains offset of where to start to read and
1438 * reqbody->nlink contains number bytes to read.
1440 rdpg->rp_hash = reqbody->size;
1441 if (rdpg->rp_hash != reqbody->size) {
1442 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1443 rdpg->rp_hash, reqbody->size);
1447 rdpg->rp_attrs = reqbody->mode;
1448 rdpg->rp_count = reqbody->nlink;
1449 rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1450 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1451 if (rdpg->rp_pages == NULL)
1454 for (i = 0; i < rdpg->rp_npages; ++i) {
1455 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1456 if (rdpg->rp_pages[i] == NULL)
1457 GOTO(free_rdpg, rc = -ENOMEM);
1460 /* call lower layers to fill allocated pages with directory data */
1461 rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1463 GOTO(free_rdpg, rc);
1465 /* send pages to client */
1466 rc = mdt_sendpage(info, rdpg);
1471 for (i = 0; i < rdpg->rp_npages; i++)
1472 if (rdpg->rp_pages[i] != NULL)
1473 cfs_free_page(rdpg->rp_pages[i]);
1474 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1476 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1482 static int mdt_reint_internal(struct mdt_thread_info *info,
1483 struct mdt_lock_handle *lhc,
1486 struct req_capsule *pill = info->mti_pill;
1487 struct mdt_device *mdt = info->mti_mdt;
1488 struct mdt_body *repbody;
1493 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1494 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1495 mdt->mdt_max_mdsize);
1496 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1497 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1498 mdt->mdt_max_cookiesize);
1500 rc = req_capsule_server_pack(pill);
1502 CERROR("Can't pack response, rc %d\n", rc);
1503 RETURN(err_serious(rc));
1506 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1507 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1509 repbody->eadatasize = 0;
1510 repbody->aclsize = 0;
1513 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1514 GOTO(out_shrink, rc = err_serious(-EFAULT));
1516 rc = mdt_reint_unpack(info, op);
1518 CERROR("Can't unpack reint, rc %d\n", rc);
1519 GOTO(out_shrink, rc = err_serious(rc));
1522 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1524 /* for replay no cookkie / lmm need, because client have this already */
1525 if (info->mti_spec.no_create == 1) {
1526 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1527 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1529 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1530 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1534 rc = mdt_init_ucred_reint(info);
1536 GOTO(out_shrink, rc);
1538 rc = mdt_fix_attr_ucred(info, op);
1540 GOTO(out_ucred, rc = err_serious(rc));
1542 if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1543 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1544 GOTO(out_ucred, rc);
1546 rc = mdt_reint_rec(info, lhc);
1549 mdt_exit_ucred(info);
1551 mdt_shrink_reply(info);
1555 static long mdt_reint_opcode(struct mdt_thread_info *info,
1556 const struct req_format **fmt)
1558 struct mdt_rec_reint *rec;
1561 opc = err_serious(-EFAULT);
1562 rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1564 opc = rec->rr_opcode;
1565 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1566 if (opc < REINT_MAX && fmt[opc] != NULL)
1567 req_capsule_extend(info->mti_pill, fmt[opc]);
1569 CERROR("Unsupported opc: %ld\n", opc);
1570 opc = err_serious(opc);
1576 static int mdt_reint(struct mdt_thread_info *info)
1581 static const struct req_format *reint_fmts[REINT_MAX] = {
1582 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1583 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1584 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1585 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1586 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1587 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1588 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1593 opc = mdt_reint_opcode(info, reint_fmts);
1596 * No lock possible here from client to pass it to reint code
1599 rc = mdt_reint_internal(info, NULL, opc);
1604 info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1608 /* this should sync the whole device */
1609 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1611 struct dt_device *dt = mdt->mdt_bottom;
1615 rc = dt->dd_ops->dt_sync(env, dt);
1619 /* this should sync this object */
1620 static int mdt_object_sync(struct mdt_thread_info *info)
1622 struct md_object *next;
1626 if (!mdt_object_exists(info->mti_object)) {
1627 CWARN("Non existing object "DFID"!\n",
1628 PFID(mdt_object_fid(info->mti_object)));
1631 next = mdt_object_child(info->mti_object);
1632 rc = mo_object_sync(info->mti_env, next);
1637 static int mdt_sync(struct mdt_thread_info *info)
1639 struct req_capsule *pill = info->mti_pill;
1640 struct mdt_body *body;
1644 /* The fid may be zero, so we req_capsule_set manually */
1645 req_capsule_set(pill, &RQF_MDS_SYNC);
1647 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1649 RETURN(err_serious(-EINVAL));
1651 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1652 RETURN(err_serious(-ENOMEM));
1654 if (fid_seq(&body->fid1) == 0) {
1655 /* sync the whole device */
1656 rc = req_capsule_server_pack(pill);
1658 rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1660 rc = err_serious(rc);
1662 /* sync an object */
1663 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1665 rc = mdt_object_sync(info);
1667 struct md_object *next;
1668 const struct lu_fid *fid;
1669 struct lu_attr *la = &info->mti_attr.ma_attr;
1671 next = mdt_object_child(info->mti_object);
1672 info->mti_attr.ma_need = MA_INODE;
1673 info->mti_attr.ma_valid = 0;
1674 rc = mo_attr_get(info->mti_env, next,
1677 body = req_capsule_server_get(pill,
1679 fid = mdt_object_fid(info->mti_object);
1680 mdt_pack_attr2body(info, body, la, fid);
1684 rc = err_serious(rc);
1689 #ifdef HAVE_QUOTA_SUPPORT
1690 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1692 struct obd_quotactl *oqctl;
1693 struct req_capsule *pill = info->mti_pill;
1694 struct obd_export *exp = info->mti_exp;
1695 struct md_device *next = info->mti_mdt->mdt_child;
1699 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1703 /* remote client has no permission for quotacheck */
1704 if (unlikely(exp_connect_rmtclient(exp)))
1707 rc = req_capsule_server_pack(pill);
1711 rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next, exp,
1716 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1718 struct obd_quotactl *oqctl, *repoqc;
1719 struct req_capsule *pill = info->mti_pill;
1720 struct obd_export *exp = info->mti_exp;
1721 struct md_device *next = info->mti_mdt->mdt_child;
1722 const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
1726 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1731 if (exp_connect_rmtclient(exp)) {
1732 struct ptlrpc_request *req = mdt_info_req(info);
1733 struct mdt_export_data *med = mdt_req2med(req);
1734 struct lustre_idmap_table *idmap = med->med_idmap;
1736 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1737 oqctl->qc_cmd != Q_GETINFO))
1741 if (oqctl->qc_type == USRQUOTA)
1742 id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1744 else if (oqctl->qc_type == GRPQUOTA)
1745 id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1750 if (id == CFS_IDMAP_NOTFOUND) {
1751 CDEBUG(D_QUOTA, "no mapping for id %u\n",
1757 rc = req_capsule_server_pack(pill);
1761 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1762 LASSERT(repoqc != NULL);
1764 switch (oqctl->qc_cmd) {
1766 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
1769 rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
1772 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
1776 rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id,
1780 rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id,
1784 rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id,
1788 rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id,
1792 rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id,
1795 case LUSTRE_Q_INVALIDATE:
1796 rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type);
1798 case LUSTRE_Q_FINVALIDATE:
1799 rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type);
1802 CERROR("unsupported mdt_quotactl command: %d\n",
1814 * OBD PING and other handlers.
1816 static int mdt_obd_ping(struct mdt_thread_info *info)
1821 req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1823 rc = target_handle_ping(mdt_info_req(info));
1825 rc = err_serious(rc);
1829 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1831 return err_serious(-EOPNOTSUPP);
1834 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1836 return err_serious(-EOPNOTSUPP);
1844 /** clone llog ctxt from child (mdd)
1845 * This allows remote llog (replicator) access.
1846 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
1847 * context was originally set up, or we can handle them directly.
1848 * I choose the latter, but that means I need any llog
1849 * contexts set up by child to be accessable by the mdt. So we clone the
1850 * context into our context list here.
1852 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
1855 struct md_device *next = mdt->mdt_child;
1856 struct llog_ctxt *ctxt;
1859 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
1862 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
1863 if (rc || ctxt == NULL) {
1864 CERROR("Can't get mdd ctxt %d\n", rc);
1868 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
1870 CERROR("Can't set mdt ctxt %d\n", rc);
1875 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
1876 struct mdt_device *mdt, int idx)
1878 struct llog_ctxt *ctxt;
1880 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
1883 /* Put once for the get we just did, and once for the clone */
1884 llog_ctxt_put(ctxt);
1885 llog_ctxt_put(ctxt);
1889 static int mdt_llog_create(struct mdt_thread_info *info)
1893 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
1894 rc = llog_origin_handle_create(mdt_info_req(info));
1895 return (rc < 0 ? err_serious(rc) : rc);
1898 static int mdt_llog_destroy(struct mdt_thread_info *info)
1902 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
1903 rc = llog_origin_handle_destroy(mdt_info_req(info));
1904 return (rc < 0 ? err_serious(rc) : rc);
1907 static int mdt_llog_read_header(struct mdt_thread_info *info)
1911 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
1912 rc = llog_origin_handle_read_header(mdt_info_req(info));
1913 return (rc < 0 ? err_serious(rc) : rc);
1916 static int mdt_llog_next_block(struct mdt_thread_info *info)
1920 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
1921 rc = llog_origin_handle_next_block(mdt_info_req(info));
1922 return (rc < 0 ? err_serious(rc) : rc);
1925 static int mdt_llog_prev_block(struct mdt_thread_info *info)
1929 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
1930 rc = llog_origin_handle_prev_block(mdt_info_req(info));
1931 return (rc < 0 ? err_serious(rc) : rc);
1938 static struct ldlm_callback_suite cbs = {
1939 .lcs_completion = ldlm_server_completion_ast,
1940 .lcs_blocking = ldlm_server_blocking_ast,
1944 static int mdt_enqueue(struct mdt_thread_info *info)
1946 struct ptlrpc_request *req;
1951 * info->mti_dlm_req already contains swapped and (if necessary)
1952 * converted dlm request.
1954 LASSERT(info->mti_dlm_req != NULL);
1956 req = mdt_info_req(info);
1959 * Lock without inodebits makes no sense and will oops later in
1960 * ldlm. Let's check it now to see if we have wrong lock from client or
1961 * bits get corrupted somewhere in mdt_intent_policy().
1963 req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1964 /* This is disabled because we need to support liblustre flock.
1965 * LASSERT(req_bits != 0);
1968 rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1969 req, info->mti_dlm_req, &cbs);
1970 info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1971 return rc ? err_serious(rc) : req->rq_status;
1974 static int mdt_convert(struct mdt_thread_info *info)
1977 struct ptlrpc_request *req;
1979 LASSERT(info->mti_dlm_req);
1980 req = mdt_info_req(info);
1981 rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1982 return rc ? err_serious(rc) : req->rq_status;
1985 static int mdt_bl_callback(struct mdt_thread_info *info)
1987 CERROR("bl callbacks should not happen on MDS\n");
1989 return err_serious(-EOPNOTSUPP);
1992 static int mdt_cp_callback(struct mdt_thread_info *info)
1994 CERROR("cp callbacks should not happen on MDS\n");
1996 return err_serious(-EOPNOTSUPP);
2000 * sec context handlers
2002 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2006 rc = mdt_handle_idmap(info);
2009 struct ptlrpc_request *req = mdt_info_req(info);
2012 opc = lustre_msg_get_opc(req->rq_reqmsg);
2013 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2014 sptlrpc_svc_ctx_invalidate(req);
2017 OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
2022 static struct mdt_object *mdt_obj(struct lu_object *o)
2024 LASSERT(lu_device_is_mdt(o->lo_dev));
2025 return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2028 struct mdt_object *mdt_object_find(const struct lu_env *env,
2029 struct mdt_device *d,
2030 const struct lu_fid *f)
2032 struct lu_object *o;
2033 struct mdt_object *m;
2036 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2037 o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2038 if (unlikely(IS_ERR(o)))
2039 m = (struct mdt_object *)o;
2046 * Asyncronous commit for mdt device.
2048 * Pass asynchonous commit call down the MDS stack.
2050 * \param env environment
2051 * \param mdt the mdt device
2053 static void mdt_device_commit_async(const struct lu_env *env,
2054 struct mdt_device *mdt)
2056 struct dt_device *dt = mdt->mdt_bottom;
2059 rc = dt->dd_ops->dt_commit_async(env, dt);
2060 if (unlikely(rc != 0))
2061 CWARN("async commit start failed with rc = %d", rc);
2065 * Mark the lock as "synchonous".
2067 * Mark the lock to deffer transaction commit to the unlock time.
2069 * \param lock the lock to mark as "synchonous"
2071 * \see mdt_is_lock_sync
2072 * \see mdt_save_lock
2074 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2076 lock->l_ast_data = (void*)1;
2080 * Check whehter the lock "synchonous" or not.
2082 * \param lock the lock to check
2083 * \retval 1 the lock is "synchonous"
2084 * \retval 0 the lock isn't "synchronous"
2086 * \see mdt_set_lock_sync
2087 * \see mdt_save_lock
2089 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2091 return lock->l_ast_data != NULL;
2095 * Blocking AST for mdt locks.
2097 * Starts transaction commit if in case of COS lock conflict or
2098 * deffers such a commit to the mdt_save_lock.
2100 * \param lock the lock which blocks a request or cancelling lock
2101 * \param desc unused
2102 * \param data unused
2103 * \param flag indicates whether this cancelling or blocking callback
2105 * \see ldlm_blocking_ast_nocheck
2107 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2108 void *data, int flag)
2110 struct obd_device *obd = lock->l_resource->lr_namespace->ns_obd;
2111 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2115 if (flag == LDLM_CB_CANCELING)
2117 lock_res_and_lock(lock);
2118 if (lock->l_blocking_ast != mdt_blocking_ast) {
2119 unlock_res_and_lock(lock);
2122 if (mdt_cos_is_enabled(mdt) &&
2123 lock->l_req_mode & (LCK_PW | LCK_EX) &&
2124 lock->l_blocking_lock != NULL &&
2125 lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2126 mdt_set_lock_sync(lock);
2128 rc = ldlm_blocking_ast_nocheck(lock);
2130 /* There is no lock conflict if l_blocking_lock == NULL,
2131 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2132 * when the last reference to a local lock was released */
2133 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2136 rc = lu_env_init(&env, LCT_MD_THREAD);
2137 if (unlikely(rc != 0))
2138 CWARN("lu_env initialization failed with rc = %d,"
2139 "cannot start asynchronous commit\n", rc);
2141 mdt_device_commit_async(&env, mdt);
2147 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2148 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2150 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2151 ldlm_policy_data_t *policy = &info->mti_policy;
2152 struct ldlm_res_id *res_id = &info->mti_res_id;
2156 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2157 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2158 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2159 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2161 if (mdt_object_exists(o) < 0) {
2162 if (locality == MDT_CROSS_LOCK) {
2163 /* cross-ref object fix */
2164 ibits &= ~MDS_INODELOCK_UPDATE;
2165 ibits |= MDS_INODELOCK_LOOKUP;
2167 LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2168 LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2170 /* No PDO lock on remote object */
2171 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2174 if (lh->mlh_type == MDT_PDO_LOCK) {
2175 /* check for exists after object is locked */
2176 if (mdt_object_exists(o) == 0) {
2177 /* Non-existent object shouldn't have PDO lock */
2180 /* Non-dir object shouldn't have PDO lock */
2181 LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
2185 memset(policy, 0, sizeof(*policy));
2186 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2189 * Take PDO lock on whole directory and build correct @res_id for lock
2190 * on part of directory.
2192 if (lh->mlh_pdo_hash != 0) {
2193 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2194 mdt_lock_pdo_mode(info, o, lh);
2195 if (lh->mlh_pdo_mode != LCK_NL) {
2197 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2198 * is never going to be sent to client and we do not
2199 * want it slowed down due to possible cancels.
2201 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2202 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2203 policy, res_id, LDLM_FL_ATOMIC_CB,
2204 &info->mti_exp->exp_handle.h_cookie);
2210 * Finish res_id initializing by name hash marking part of
2211 * directory which is taking modification.
2213 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2216 policy->l_inodebits.bits = ibits;
2219 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2220 * going to be sent to client. If it is - mdt_intent_policy() path will
2221 * fix it up and turn FL_LOCAL flag off.
2223 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2224 res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
2225 &info->mti_exp->exp_handle.h_cookie);
2231 mdt_object_unlock(info, o, lh, 1);
2238 * Save a lock within request object.
2240 * Keep the lock referenced until whether client ACK or transaction
2241 * commit happens or release the lock immediately depending on input
2242 * parameters. If COS is ON, a write lock is converted to COS lock
2245 * \param info thead info object
2246 * \param h lock handle
2247 * \param mode lock mode
2248 * \param decref force immediate lock releasing
2251 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2252 ldlm_mode_t mode, int decref)
2256 if (lustre_handle_is_used(h)) {
2257 if (decref || !info->mti_has_trans ||
2258 !(mode & (LCK_PW | LCK_EX))){
2259 mdt_fid_unlock(h, mode);
2261 struct mdt_device *mdt = info->mti_mdt;
2262 struct ldlm_lock *lock = ldlm_handle2lock(h);
2263 struct ptlrpc_request *req = mdt_info_req(info);
2266 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2268 CDEBUG(D_HA, "request = %p reply state = %p"
2269 " transno = "LPD64"\n",
2270 req, req->rq_reply_state, req->rq_transno);
2271 if (mdt_cos_is_enabled(mdt)) {
2273 ldlm_lock_downgrade(lock, LCK_COS);
2276 ptlrpc_save_lock(req, h, mode, no_ack);
2277 if (mdt_is_lock_sync(lock)) {
2278 CDEBUG(D_HA, "found sync-lock,"
2279 " async commit started\n");
2280 mdt_device_commit_async(info->mti_env,
2283 LDLM_LOCK_PUT(lock);
2292 * Unlock mdt object.
2294 * Immeditely release the regular lock and the PDO lock or save the
2295 * lock in reqeuest and keep them referenced until client ACK or
2296 * transaction commit.
2298 * \param info thread info object
2299 * \param o mdt object
2300 * \param lh mdt lock handle referencing regular and PDO locks
2301 * \param decref force immediate lock releasing
2303 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2304 struct mdt_lock_handle *lh, int decref)
2308 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2309 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2314 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2315 const struct lu_fid *f,
2316 struct mdt_lock_handle *lh,
2319 struct mdt_object *o;
2321 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2325 rc = mdt_object_lock(info, o, lh, ibits,
2328 mdt_object_put(info->mti_env, o);
2335 void mdt_object_unlock_put(struct mdt_thread_info * info,
2336 struct mdt_object * o,
2337 struct mdt_lock_handle *lh,
2340 mdt_object_unlock(info, o, lh, decref);
2341 mdt_object_put(info->mti_env, o);
2344 static struct mdt_handler *mdt_handler_find(__u32 opc,
2345 struct mdt_opc_slice *supported)
2347 struct mdt_opc_slice *s;
2348 struct mdt_handler *h;
2351 for (s = supported; s->mos_hs != NULL; s++) {
2352 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2353 h = s->mos_hs + (opc - s->mos_opc_start);
2354 if (likely(h->mh_opc != 0))
2355 LASSERTF(h->mh_opc == opc,
2356 "opcode mismatch %d != %d\n",
2359 h = NULL; /* unsupported opc */
2366 static int mdt_lock_resname_compat(struct mdt_device *m,
2367 struct ldlm_request *req)
2369 /* XXX something... later. */
2373 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2375 /* XXX something... later. */
2380 * Generic code handling requests that have struct mdt_body passed in:
2382 * - extract mdt_body from request and save it in @info, if present;
2384 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2387 * - if HABEO_CORPUS flag is set for this request type check whether object
2388 * actually exists on storage (lu_object_exists()).
2391 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2393 const struct mdt_body *body;
2394 struct mdt_object *obj;
2395 const struct lu_env *env;
2396 struct req_capsule *pill;
2400 env = info->mti_env;
2401 pill = info->mti_pill;
2403 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2407 if (!(body->valid & OBD_MD_FLID))
2410 if (!fid_is_sane(&body->fid1)) {
2411 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2416 * Do not get size or any capa fields before we check that request
2417 * contains capa actually. There are some requests which do not, for
2418 * instance MDS_IS_SUBDIR.
2420 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2421 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2422 mdt_set_capainfo(info, 0, &body->fid1,
2423 req_capsule_client_get(pill, &RMF_CAPA1));
2425 obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2427 if ((flags & HABEO_CORPUS) &&
2428 !mdt_object_exists(obj)) {
2429 mdt_object_put(env, obj);
2430 /* for capability renew ENOENT will be handled in
2432 if (body->valid & OBD_MD_FLOSSCAPA)
2437 info->mti_object = obj;
2446 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2448 struct req_capsule *pill = info->mti_pill;
2452 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2453 rc = mdt_body_unpack(info, flags);
2457 if (rc == 0 && (flags & HABEO_REFERO)) {
2458 struct mdt_device *mdt = info->mti_mdt;
2462 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2463 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2464 mdt->mdt_max_mdsize);
2465 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2466 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2467 mdt->mdt_max_cookiesize);
2469 rc = req_capsule_server_pack(pill);
2474 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2476 struct md_device *next = m->mdt_child;
2478 return next->md_ops->mdo_init_capa_ctxt(env, next,
2479 m->mdt_opts.mo_mds_capa,
2480 m->mdt_capa_timeout,
2486 * Invoke handler for this request opc. Also do necessary preprocessing
2487 * (according to handler ->mh_flags), and post-processing (setting of
2488 * ->last_{xid,committed}).
2490 static int mdt_req_handle(struct mdt_thread_info *info,
2491 struct mdt_handler *h, struct ptlrpc_request *req)
2493 int rc, serious = 0;
2498 LASSERT(h->mh_act != NULL);
2499 LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2500 LASSERT(current->journal_info == NULL);
2503 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2504 * to put same checks into handlers like mdt_close(), mdt_reint(),
2505 * etc., without talking to mdt authors first. Checking same thing
2506 * there again is useless and returning 0 error without packing reply
2507 * is buggy! Handlers either pack reply or return error.
2509 * We return 0 here and do not send any reply in order to emulate
2510 * network failure. Do not send any reply in case any of NET related
2511 * fail_id has occured.
2513 if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2517 flags = h->mh_flags;
2518 LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2520 if (h->mh_fmt != NULL) {
2521 req_capsule_set(info->mti_pill, h->mh_fmt);
2522 rc = mdt_unpack_req_pack_rep(info, flags);
2525 if (rc == 0 && flags & MUTABOR &&
2526 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2527 /* should it be rq_status? */
2530 if (rc == 0 && flags & HABEO_CLAVIS) {
2531 struct ldlm_request *dlm_req;
2533 LASSERT(h->mh_fmt != NULL);
2535 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2536 if (dlm_req != NULL) {
2537 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2538 rc = mdt_lock_resname_compat(info->mti_mdt,
2540 info->mti_dlm_req = dlm_req;
2546 /* capability setting changed via /proc, needs reinitialize ctxt */
2547 if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2548 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2549 info->mti_mdt->mdt_capa_conf = 0;
2552 if (likely(rc == 0)) {
2554 * Process request, there can be two types of rc:
2555 * 1) errors with msg unpack/pack, other failures outside the
2556 * operation itself. This is counted as serious errors;
2557 * 2) errors during fs operation, should be placed in rq_status
2560 rc = h->mh_act(info);
2562 !req->rq_no_reply && req->rq_reply_state == NULL) {
2563 DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2564 "pack reply and returned 0 error\n",
2568 serious = is_serious(rc);
2569 rc = clear_serious(rc);
2573 req->rq_status = rc;
2576 * ELDLM_* codes which > 0 should be in rq_status only as well as
2577 * all non-serious errors.
2579 if (rc > 0 || !serious)
2582 LASSERT(current->journal_info == NULL);
2584 if (rc == 0 && (flags & HABEO_CLAVIS) &&
2585 info->mti_mdt->mdt_opts.mo_compat_resname) {
2586 struct ldlm_reply *dlmrep;
2588 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2590 rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2593 /* If we're DISCONNECTing, the mdt_export_data is already freed */
2594 if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2595 target_committed_to_req(req);
2597 if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2598 lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2599 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2603 target_send_reply(req, rc, info->mti_fail_id);
2607 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2609 lh->mlh_type = MDT_NUL_LOCK;
2610 lh->mlh_reg_lh.cookie = 0ull;
2611 lh->mlh_reg_mode = LCK_MINMODE;
2612 lh->mlh_pdo_lh.cookie = 0ull;
2613 lh->mlh_pdo_mode = LCK_MINMODE;
2616 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2618 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2619 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2623 * Initialize fields of struct mdt_thread_info. Other fields are left in
2624 * uninitialized state, because it's too expensive to zero out whole
2625 * mdt_thread_info (> 1K) on each request arrival.
2627 static void mdt_thread_info_init(struct ptlrpc_request *req,
2628 struct mdt_thread_info *info)
2631 struct md_capainfo *ci;
2633 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2634 info->mti_pill = &req->rq_pill;
2637 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2638 mdt_lock_handle_init(&info->mti_lh[i]);
2640 /* mdt device: it can be NULL while CONNECT */
2641 if (req->rq_export) {
2642 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2643 info->mti_exp = req->rq_export;
2645 info->mti_mdt = NULL;
2646 info->mti_env = req->rq_svc_thread->t_env;
2647 ci = md_capainfo(info->mti_env);
2648 memset(ci, 0, sizeof *ci);
2649 if (req->rq_export) {
2650 if (exp_connect_rmtclient(req->rq_export))
2651 ci->mc_auth = LC_ID_CONVERT;
2652 else if (req->rq_export->exp_connect_flags &
2653 OBD_CONNECT_MDS_CAPA)
2654 ci->mc_auth = LC_ID_PLAIN;
2656 ci->mc_auth = LC_ID_NONE;
2659 info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2660 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2661 info->mti_mos[0] = NULL;
2662 info->mti_mos[1] = NULL;
2663 info->mti_mos[2] = NULL;
2664 info->mti_mos[3] = NULL;
2666 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2667 info->mti_body = NULL;
2668 info->mti_object = NULL;
2669 info->mti_dlm_req = NULL;
2670 info->mti_has_trans = 0;
2671 info->mti_no_need_trans = 0;
2672 info->mti_cross_ref = 0;
2673 info->mti_opdata = 0;
2675 /* To not check for split by default. */
2676 info->mti_spec.sp_ck_split = 0;
2677 info->mti_spec.no_create = 0;
2680 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2684 req_capsule_fini(info->mti_pill);
2685 if (info->mti_object != NULL) {
2687 * freeing an object may lead to OSD level transaction, do not
2688 * let it mess with MDT. bz19385.
2690 info->mti_no_need_trans = 1;
2691 mdt_object_put(info->mti_env, info->mti_object);
2692 info->mti_object = NULL;
2694 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2695 mdt_lock_handle_fini(&info->mti_lh[i]);
2696 info->mti_env = NULL;
2699 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2700 struct obd_device *obd, int *process)
2702 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2703 case MDS_CONNECT: /* This will never get here, but for completeness. */
2704 case OST_CONNECT: /* This will never get here, but for completeness. */
2705 case MDS_DISCONNECT:
2706 case OST_DISCONNECT:
2711 case MDS_DONE_WRITING:
2712 case MDS_SYNC: /* used in unmounting */
2718 *process = target_queue_recovery_request(req, obd);
2722 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2729 * Handle recovery. Return:
2730 * +1: continue request processing;
2731 * -ve: abort immediately with the given error code;
2732 * 0: send reply with error code in req->rq_status;
2734 static int mdt_recovery(struct mdt_thread_info *info)
2736 struct ptlrpc_request *req = mdt_info_req(info);
2738 struct obd_device *obd;
2742 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2745 case SEC_CTX_INIT_CONT:
2751 rc = mdt_handle_idmap(info);
2760 if (unlikely(req->rq_export == NULL)) {
2761 CERROR("operation %d on unconnected MDS from %s\n",
2762 lustre_msg_get_opc(req->rq_reqmsg),
2763 libcfs_id2str(req->rq_peer));
2764 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2765 * mds_A will get -ENOTCONN(especially for ping req),
2766 * which will cause that mds_A deactive timeout, then when
2767 * mds_A cleanup, the cleanup process will be suspended since
2768 * deactive timeout is not zero.
2770 req->rq_status = -ENOTCONN;
2771 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2775 /* sanity check: if the xid matches, the request must be marked as a
2776 * resent or replayed */
2777 if (req_xid_is_last(req)) {
2778 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2779 (MSG_RESENT | MSG_REPLAY))) {
2780 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2781 "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2782 lustre_msg_get_flags(req->rq_reqmsg));
2784 req->rq_status = -ENOTCONN;
2789 /* else: note the opposite is not always true; a RESENT req after a
2790 * failover will usually not match the last_xid, since it was likely
2791 * never committed. A REPLAYed request will almost never match the
2792 * last xid, however it could for a committed, but still retained,
2795 obd = req->rq_export->exp_obd;
2797 /* Check for aborted recovery... */
2798 spin_lock_bh(&obd->obd_processing_task_lock);
2799 recovering = obd->obd_recovering;
2800 spin_unlock_bh(&obd->obd_processing_task_lock);
2801 if (unlikely(recovering)) {
2804 DEBUG_REQ(D_INFO, req, "Got new replay");
2805 rc = mdt_filter_recovery_request(req, obd, &should_process);
2806 if (rc != 0 || !should_process)
2808 else if (should_process < 0) {
2809 req->rq_status = should_process;
2810 rc = ptlrpc_error(req);
2817 static int mdt_msg_check_version(struct lustre_msg *msg)
2821 switch (lustre_msg_get_opc(msg)) {
2823 case MDS_DISCONNECT:
2827 case SEC_CTX_INIT_CONT:
2829 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2831 CERROR("bad opc %u version %08x, expecting %08x\n",
2832 lustre_msg_get_opc(msg),
2833 lustre_msg_get_version(msg),
2834 LUSTRE_OBD_VERSION);
2838 case MDS_GETATTR_NAME:
2845 case MDS_DONE_WRITING:
2851 case MDS_QUOTACHECK:
2857 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2859 CERROR("bad opc %u version %08x, expecting %08x\n",
2860 lustre_msg_get_opc(msg),
2861 lustre_msg_get_version(msg),
2862 LUSTRE_MDS_VERSION);
2866 case LDLM_BL_CALLBACK:
2867 case LDLM_CP_CALLBACK:
2868 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2870 CERROR("bad opc %u version %08x, expecting %08x\n",
2871 lustre_msg_get_opc(msg),
2872 lustre_msg_get_version(msg),
2873 LUSTRE_DLM_VERSION);
2875 case OBD_LOG_CANCEL:
2876 case LLOG_ORIGIN_HANDLE_CREATE:
2877 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2878 case LLOG_ORIGIN_HANDLE_READ_HEADER:
2879 case LLOG_ORIGIN_HANDLE_CLOSE:
2880 case LLOG_ORIGIN_HANDLE_DESTROY:
2881 case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2883 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2885 CERROR("bad opc %u version %08x, expecting %08x\n",
2886 lustre_msg_get_opc(msg),
2887 lustre_msg_get_version(msg),
2888 LUSTRE_LOG_VERSION);
2891 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2897 static int mdt_handle0(struct ptlrpc_request *req,
2898 struct mdt_thread_info *info,
2899 struct mdt_opc_slice *supported)
2901 struct mdt_handler *h;
2902 struct lustre_msg *msg;
2907 if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2910 LASSERT(current->journal_info == NULL);
2912 msg = req->rq_reqmsg;
2913 rc = mdt_msg_check_version(msg);
2914 if (likely(rc == 0)) {
2915 rc = mdt_recovery(info);
2916 if (likely(rc == +1)) {
2917 h = mdt_handler_find(lustre_msg_get_opc(msg),
2919 if (likely(h != NULL)) {
2920 rc = mdt_req_handle(info, h, req);
2922 CERROR("The unsupported opc: 0x%x\n",
2923 lustre_msg_get_opc(msg) );
2924 req->rq_status = -ENOTSUPP;
2925 rc = ptlrpc_error(req);
2930 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2935 * MDT handler function called by ptlrpc service thread when request comes.
2937 * XXX common "target" functionality should be factored into separate module
2938 * shared by mdt, ost and stand-alone services like fld.
2940 static int mdt_handle_common(struct ptlrpc_request *req,
2941 struct mdt_opc_slice *supported)
2944 struct mdt_thread_info *info;
2948 env = req->rq_svc_thread->t_env;
2949 LASSERT(env != NULL);
2950 LASSERT(env->le_ses != NULL);
2951 LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2952 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2953 LASSERT(info != NULL);
2955 mdt_thread_info_init(req, info);
2957 rc = mdt_handle0(req, info, supported);
2959 mdt_thread_info_fini(info);
2964 * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2967 int mdt_recovery_handle(struct ptlrpc_request *req)
2972 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2974 rc = mdt_handle_common(req, mdt_fld_handlers);
2977 rc = mdt_handle_common(req, mdt_seq_handlers);
2980 rc = mdt_handle_common(req, mdt_regular_handlers);
2987 static int mdt_regular_handle(struct ptlrpc_request *req)
2989 return mdt_handle_common(req, mdt_regular_handlers);
2992 static int mdt_readpage_handle(struct ptlrpc_request *req)
2994 return mdt_handle_common(req, mdt_readpage_handlers);
2997 static int mdt_xmds_handle(struct ptlrpc_request *req)
2999 return mdt_handle_common(req, mdt_xmds_handlers);
3002 static int mdt_mdsc_handle(struct ptlrpc_request *req)
3004 return mdt_handle_common(req, mdt_seq_handlers);
3007 static int mdt_mdss_handle(struct ptlrpc_request *req)
3009 return mdt_handle_common(req, mdt_seq_handlers);
3012 static int mdt_dtss_handle(struct ptlrpc_request *req)
3014 return mdt_handle_common(req, mdt_seq_handlers);
3017 static int mdt_fld_handle(struct ptlrpc_request *req)
3019 return mdt_handle_common(req, mdt_fld_handlers);
3035 static int mdt_intent_getattr(enum mdt_it_code opcode,
3036 struct mdt_thread_info *info,
3037 struct ldlm_lock **,
3039 static int mdt_intent_reint(enum mdt_it_code opcode,
3040 struct mdt_thread_info *info,
3041 struct ldlm_lock **,
3044 static struct mdt_it_flavor {
3045 const struct req_format *it_fmt;
3047 int (*it_act)(enum mdt_it_code ,
3048 struct mdt_thread_info *,
3049 struct ldlm_lock **,
3052 } mdt_it_flavor[] = {
3054 .it_fmt = &RQF_LDLM_INTENT,
3055 /*.it_flags = HABEO_REFERO,*/
3057 .it_act = mdt_intent_reint,
3058 .it_reint = REINT_OPEN
3061 .it_fmt = &RQF_LDLM_INTENT,
3062 .it_flags = MUTABOR,
3063 .it_act = mdt_intent_reint,
3064 .it_reint = REINT_OPEN
3067 .it_fmt = &RQF_LDLM_INTENT,
3068 .it_flags = MUTABOR,
3069 .it_act = mdt_intent_reint,
3070 .it_reint = REINT_CREATE
3072 [MDT_IT_GETATTR] = {
3073 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3074 .it_flags = HABEO_REFERO,
3075 .it_act = mdt_intent_getattr
3077 [MDT_IT_READDIR] = {
3083 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3084 .it_flags = HABEO_REFERO,
3085 .it_act = mdt_intent_getattr
3088 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
3089 .it_flags = MUTABOR,
3091 .it_reint = REINT_UNLINK
3095 .it_flags = MUTABOR,
3098 [MDT_IT_GETXATTR] = {
3105 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3106 struct ldlm_lock **lockp,
3107 struct ldlm_lock *new_lock,
3108 struct mdt_lock_handle *lh,
3111 struct ptlrpc_request *req = mdt_info_req(info);
3112 struct ldlm_lock *lock = *lockp;
3115 * Get new lock only for cases when possible resent did not find any
3118 if (new_lock == NULL)
3119 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3121 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3122 lh->mlh_reg_lh.cookie = 0;
3126 LASSERTF(new_lock != NULL,
3127 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3130 * If we've already given this lock to a client once, then we should
3131 * have no readers or writers. Otherwise, we should have one reader
3132 * _or_ writer ref (which will be zeroed below) before returning the
3135 if (new_lock->l_export == req->rq_export) {
3136 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3138 LASSERT(new_lock->l_export == NULL);
3139 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3144 if (new_lock->l_export == req->rq_export) {
3146 * Already gave this to the client, which means that we
3147 * reconstructed a reply.
3149 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3151 lh->mlh_reg_lh.cookie = 0;
3152 RETURN(ELDLM_LOCK_REPLACED);
3156 * Fixup the lock to be given to the client.
3158 lock_res_and_lock(new_lock);
3159 /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3160 * possible blocking AST. */
3161 while (new_lock->l_readers > 0) {
3162 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3163 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3164 new_lock->l_readers--;
3166 while (new_lock->l_writers > 0) {
3167 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3168 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3169 new_lock->l_writers--;
3172 new_lock->l_export = class_export_lock_get(req->rq_export);
3173 new_lock->l_blocking_ast = lock->l_blocking_ast;
3174 new_lock->l_completion_ast = lock->l_completion_ast;
3175 new_lock->l_remote_handle = lock->l_remote_handle;
3176 new_lock->l_flags &= ~LDLM_FL_LOCAL;
3178 unlock_res_and_lock(new_lock);
3180 lustre_hash_add(new_lock->l_export->exp_lock_hash,
3181 &new_lock->l_remote_handle,
3182 &new_lock->l_exp_hash);
3184 LDLM_LOCK_RELEASE(new_lock);
3185 lh->mlh_reg_lh.cookie = 0;
3187 RETURN(ELDLM_LOCK_REPLACED);
3190 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3191 struct ldlm_lock *new_lock,
3192 struct ldlm_lock **old_lock,
3193 struct mdt_lock_handle *lh)
3195 struct ptlrpc_request *req = mdt_info_req(info);
3196 struct obd_export *exp = req->rq_export;
3197 struct lustre_handle remote_hdl;
3198 struct ldlm_request *dlmreq;
3199 struct ldlm_lock *lock;
3201 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3204 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3205 remote_hdl = dlmreq->lock_handle[0];
3207 lock = lustre_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3209 if (lock != new_lock) {
3210 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3211 lh->mlh_reg_mode = lock->l_granted_mode;
3213 LDLM_DEBUG(lock, "Restoring lock cookie");
3214 DEBUG_REQ(D_DLMTRACE, req,
3215 "restoring lock cookie "LPX64,
3216 lh->mlh_reg_lh.cookie);
3218 *old_lock = LDLM_LOCK_GET(lock);
3219 lh_put(exp->exp_lock_hash, &lock->l_exp_hash);
3223 lh_put(exp->exp_lock_hash, &lock->l_exp_hash);
3227 * If the xid matches, then we know this is a resent request, and allow
3228 * it. (It's probably an OPEN, for which we don't send a lock.
3230 if (req_xid_is_last(req))
3234 * This remote handle isn't enqueued, so we never received or processed
3235 * this request. Clear MSG_RESENT, because it can be handled like any
3236 * normal request now.
3238 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3240 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3244 static int mdt_intent_getattr(enum mdt_it_code opcode,
3245 struct mdt_thread_info *info,
3246 struct ldlm_lock **lockp,
3249 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3250 struct ldlm_lock *new_lock = NULL;
3252 struct ldlm_reply *ldlm_rep;
3253 struct ptlrpc_request *req;
3254 struct mdt_body *reqbody;
3255 struct mdt_body *repbody;
3259 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3262 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3265 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
3266 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3267 repbody->eadatasize = 0;
3268 repbody->aclsize = 0;
3272 child_bits = MDS_INODELOCK_LOOKUP;
3274 case MDT_IT_GETATTR:
3275 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3278 CERROR("Unhandled till now");
3279 GOTO(out_shrink, rc = -EINVAL);
3282 rc = mdt_init_ucred(info, reqbody);
3284 GOTO(out_shrink, rc);
3286 req = info->mti_pill->rc_req;
3287 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3288 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3290 /* Get lock from request for possible resent case. */
3291 mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3293 ldlm_rep->lock_policy_res2 =
3294 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3296 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3297 ldlm_rep->lock_policy_res2 = 0;
3298 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3299 ldlm_rep->lock_policy_res2) {
3300 lhc->mlh_reg_lh.cookie = 0ull;
3301 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3304 rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3307 mdt_exit_ucred(info);
3309 mdt_shrink_reply(info);
3313 static int mdt_intent_reint(enum mdt_it_code opcode,
3314 struct mdt_thread_info *info,
3315 struct ldlm_lock **lockp,
3318 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3319 struct ldlm_reply *rep = NULL;
3323 static const struct req_format *intent_fmts[REINT_MAX] = {
3324 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
3325 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
3330 opc = mdt_reint_opcode(info, intent_fmts);
3334 if (mdt_it_flavor[opcode].it_reint != opc) {
3335 CERROR("Reint code %ld doesn't match intent: %d\n",
3337 RETURN(err_serious(-EPROTO));
3340 /* Get lock from request for possible resent case. */
3341 mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3343 rc = mdt_reint_internal(info, lhc, opc);
3345 /* Check whether the reply has been packed successfully. */
3346 if (mdt_info_req(info)->rq_repmsg != NULL)
3347 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3349 RETURN(err_serious(-EFAULT));
3351 /* MDC expects this in any case */
3353 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3355 /* Cross-ref case, the lock should be returned to the client */
3356 if (rc == -EREMOTE) {
3357 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3358 rep->lock_policy_res2 = 0;
3359 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3362 rep->lock_policy_res2 = clear_serious(rc);
3364 if (rc == -ENOTCONN || rc == -ENODEV ||
3365 rc == -EOVERFLOW) { /**< if VBR failure then return error */
3367 * If it is the disconnect error (ENODEV & ENOCONN), the error
3368 * will be returned by rq_status, and client at ptlrpc layer
3369 * will detect this, then disconnect, reconnect the import
3370 * immediately, instead of impacting the following the rpc.
3372 lhc->mlh_reg_lh.cookie = 0ull;
3376 * For other cases, the error will be returned by intent.
3377 * and client will retrieve the result from intent.
3380 * FIXME: when open lock is finished, that should be
3383 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3384 rep->lock_policy_res2 = 0;
3385 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3388 lhc->mlh_reg_lh.cookie = 0ull;
3389 RETURN(ELDLM_LOCK_ABORTED);
3394 static int mdt_intent_code(long itcode)
3402 case IT_OPEN|IT_CREAT:
3409 rc = MDT_IT_READDIR;
3412 rc = MDT_IT_GETATTR;
3424 rc = MDT_IT_GETXATTR;
3427 CERROR("Unknown intent opcode: %ld\n", itcode);
3434 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3435 struct ldlm_lock **lockp, int flags)
3437 struct req_capsule *pill;
3438 struct mdt_it_flavor *flv;
3443 opc = mdt_intent_code(itopc);
3447 pill = info->mti_pill;
3448 flv = &mdt_it_flavor[opc];
3450 if (flv->it_fmt != NULL)
3451 req_capsule_extend(pill, flv->it_fmt);
3453 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3455 struct ptlrpc_request *req = mdt_info_req(info);
3456 if (flv->it_flags & MUTABOR &&
3457 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3460 if (rc == 0 && flv->it_act != NULL) {
3461 /* execute policy */
3462 rc = flv->it_act(opc, info, lockp, flags);
3469 static int mdt_intent_policy(struct ldlm_namespace *ns,
3470 struct ldlm_lock **lockp, void *req_cookie,
3471 ldlm_mode_t mode, int flags, void *data)
3473 struct mdt_thread_info *info;
3474 struct ptlrpc_request *req = req_cookie;
3475 struct ldlm_intent *it;
3476 struct req_capsule *pill;
3481 LASSERT(req != NULL);
3483 info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3485 LASSERT(info != NULL);
3486 pill = info->mti_pill;
3487 LASSERT(pill->rc_req == req);
3489 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3490 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3491 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3493 const struct ldlm_request *dlmreq;
3496 rc = mdt_intent_opc(it->opc, info, lockp, flags);
3501 * Lock without inodebits makes no sense and will oops
3502 * later in ldlm. Let's check it now to see if we have
3503 * wrong lock from client or bits get corrupted
3504 * somewhere in mdt_intent_opc().
3506 dlmreq = info->mti_dlm_req;
3507 req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
3508 LASSERT(req_bits != 0);
3511 rc = err_serious(-EFAULT);
3513 /* No intent was provided */
3514 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3515 rc = req_capsule_server_pack(pill);
3517 rc = err_serious(rc);
3525 static void mdt_seq_adjust(const struct lu_env *env,
3526 struct mdt_device *m, int lost)
3528 struct md_site *ms = mdt_md_site(m);
3529 struct lu_seq_range out;
3532 LASSERT(ms && ms->ms_server_seq);
3534 /* get extra seq from seq_server, moving it's range up */
3535 while (lost-- > 0) {
3536 seq_server_alloc_meta(ms->ms_server_seq, NULL, &out, env);
3541 static int mdt_seq_fini(const struct lu_env *env,
3542 struct mdt_device *m)
3544 struct md_site *ms = mdt_md_site(m);
3548 if (ms->ms_server_seq) {
3549 seq_server_fini(ms->ms_server_seq, env);
3550 OBD_FREE_PTR(ms->ms_server_seq);
3551 ms->ms_server_seq = NULL;
3554 if (ms->ms_control_seq) {
3555 seq_server_fini(ms->ms_control_seq, env);
3556 OBD_FREE_PTR(ms->ms_control_seq);
3557 ms->ms_control_seq = NULL;
3560 if (ms->ms_client_seq) {
3561 seq_client_fini(ms->ms_client_seq);
3562 OBD_FREE_PTR(ms->ms_client_seq);
3563 ms->ms_client_seq = NULL;
3570 static int mdt_seq_init(const struct lu_env *env,
3572 struct mdt_device *m)
3579 ms = mdt_md_site(m);
3582 * This is sequence-controller node. Init seq-controller server on local
3585 if (ms->ms_node_id == 0) {
3586 LASSERT(ms->ms_control_seq == NULL);
3588 OBD_ALLOC_PTR(ms->ms_control_seq);
3589 if (ms->ms_control_seq == NULL)
3592 rc = seq_server_init(ms->ms_control_seq,
3593 m->mdt_bottom, uuid,
3594 LUSTRE_SEQ_CONTROLLER,
3599 GOTO(out_seq_fini, rc);
3601 OBD_ALLOC_PTR(ms->ms_client_seq);
3602 if (ms->ms_client_seq == NULL)
3603 GOTO(out_seq_fini, rc = -ENOMEM);
3605 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3606 if (prefix == NULL) {
3607 OBD_FREE_PTR(ms->ms_client_seq);
3608 GOTO(out_seq_fini, rc = -ENOMEM);
3611 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3615 * Init seq-controller client after seq-controller server is
3616 * ready. Pass ms->ms_control_seq to it for direct talking.
3618 rc = seq_client_init(ms->ms_client_seq, NULL,
3619 LUSTRE_SEQ_METADATA, prefix,
3620 ms->ms_control_seq);
3621 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3624 GOTO(out_seq_fini, rc);
3627 /* Init seq-server on local MDT */
3628 LASSERT(ms->ms_server_seq == NULL);
3630 OBD_ALLOC_PTR(ms->ms_server_seq);
3631 if (ms->ms_server_seq == NULL)
3632 GOTO(out_seq_fini, rc = -ENOMEM);
3634 rc = seq_server_init(ms->ms_server_seq,
3635 m->mdt_bottom, uuid,
3640 GOTO(out_seq_fini, rc = -ENOMEM);
3642 /* Assign seq-controller client to local seq-server. */
3643 if (ms->ms_node_id == 0) {
3644 LASSERT(ms->ms_client_seq != NULL);
3646 rc = seq_server_set_cli(ms->ms_server_seq,
3654 mdt_seq_fini(env, m);
3659 * Init client sequence manager which is used by local MDS to talk to sequence
3660 * controller on remote node.
3662 static int mdt_seq_init_cli(const struct lu_env *env,
3663 struct mdt_device *m,
3664 struct lustre_cfg *cfg)
3666 struct md_site *ms = mdt_md_site(m);
3667 struct obd_device *mdc;
3668 struct obd_uuid *uuidp, *mdcuuidp;
3669 char *uuid_str, *mdc_uuid_str;
3672 struct mdt_thread_info *info;
3673 char *p, *index_string = lustre_cfg_string(cfg, 2);
3676 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3677 uuidp = &info->mti_u.uuid[0];
3678 mdcuuidp = &info->mti_u.uuid[1];
3680 LASSERT(index_string);
3682 index = simple_strtol(index_string, &p, 10);
3684 CERROR("Invalid index in lustre_cgf, offset 2\n");
3688 /* check if this is adding the first MDC and controller is not yet
3690 if (index != 0 || ms->ms_client_seq)
3693 uuid_str = lustre_cfg_string(cfg, 1);
3694 mdc_uuid_str = lustre_cfg_string(cfg, 4);
3695 obd_str2uuid(uuidp, uuid_str);
3696 obd_str2uuid(mdcuuidp, mdc_uuid_str);
3698 mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3700 CERROR("can't find controller MDC by uuid %s\n",
3703 } else if (!mdc->obd_set_up) {
3704 CERROR("target %s not set up\n", mdc->obd_name);
3707 LASSERT(ms->ms_control_exp);
3708 OBD_ALLOC_PTR(ms->ms_client_seq);
3709 if (ms->ms_client_seq != NULL) {
3712 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3716 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3719 rc = seq_client_init(ms->ms_client_seq,
3721 LUSTRE_SEQ_METADATA,
3723 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3730 LASSERT(ms->ms_server_seq != NULL);
3731 rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
3738 static void mdt_seq_fini_cli(struct mdt_device *m)
3744 ms = mdt_md_site(m);
3747 if (ms->ms_server_seq)
3748 seq_server_set_cli(ms->ms_server_seq,
3751 if (ms->ms_control_exp) {
3752 class_export_put(ms->ms_control_exp);
3753 ms->ms_control_exp = NULL;
3762 static int mdt_fld_fini(const struct lu_env *env,
3763 struct mdt_device *m)
3765 struct md_site *ms = mdt_md_site(m);
3768 if (ms && ms->ms_server_fld) {
3769 fld_server_fini(ms->ms_server_fld, env);
3770 OBD_FREE_PTR(ms->ms_server_fld);
3771 ms->ms_server_fld = NULL;
3777 static int mdt_fld_init(const struct lu_env *env,
3779 struct mdt_device *m)
3785 ms = mdt_md_site(m);
3787 OBD_ALLOC_PTR(ms->ms_server_fld);
3788 if (ms->ms_server_fld == NULL)
3789 RETURN(rc = -ENOMEM);
3791 rc = fld_server_init(ms->ms_server_fld,
3792 m->mdt_bottom, uuid,
3793 env, ms->ms_node_id);
3795 OBD_FREE_PTR(ms->ms_server_fld);
3796 ms->ms_server_fld = NULL;
3803 /* device init/fini methods */
3804 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3807 if (m->mdt_regular_service != NULL) {
3808 ptlrpc_unregister_service(m->mdt_regular_service);
3809 m->mdt_regular_service = NULL;
3811 if (m->mdt_readpage_service != NULL) {
3812 ptlrpc_unregister_service(m->mdt_readpage_service);
3813 m->mdt_readpage_service = NULL;
3815 if (m->mdt_xmds_service != NULL) {
3816 ptlrpc_unregister_service(m->mdt_xmds_service);
3817 m->mdt_xmds_service = NULL;
3819 if (m->mdt_setattr_service != NULL) {
3820 ptlrpc_unregister_service(m->mdt_setattr_service);
3821 m->mdt_setattr_service = NULL;
3823 if (m->mdt_mdsc_service != NULL) {
3824 ptlrpc_unregister_service(m->mdt_mdsc_service);
3825 m->mdt_mdsc_service = NULL;
3827 if (m->mdt_mdss_service != NULL) {
3828 ptlrpc_unregister_service(m->mdt_mdss_service);
3829 m->mdt_mdss_service = NULL;
3831 if (m->mdt_dtss_service != NULL) {
3832 ptlrpc_unregister_service(m->mdt_dtss_service);
3833 m->mdt_dtss_service = NULL;
3835 if (m->mdt_fld_service != NULL) {
3836 ptlrpc_unregister_service(m->mdt_fld_service);
3837 m->mdt_fld_service = NULL;
3842 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3845 static struct ptlrpc_service_conf conf;
3846 cfs_proc_dir_entry_t *procfs_entry;
3849 procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3851 conf = (typeof(conf)) {
3852 .psc_nbufs = MDS_NBUFS,
3853 .psc_bufsize = MDS_BUFSIZE,
3854 .psc_max_req_size = MDS_MAXREQSIZE,
3855 .psc_max_reply_size = MDS_MAXREPSIZE,
3856 .psc_req_portal = MDS_REQUEST_PORTAL,
3857 .psc_rep_portal = MDC_REPLY_PORTAL,
3858 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3860 * We'd like to have a mechanism to set this on a per-device
3861 * basis, but alas...
3863 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3865 .psc_max_threads = MDT_MAX_THREADS,
3866 .psc_ctx_tags = LCT_MD_THREAD
3869 m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3870 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3871 "mdt_ldlm_client", m->mdt_ldlm_client);
3873 m->mdt_regular_service =
3874 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3875 procfs_entry, target_print_req,
3877 if (m->mdt_regular_service == NULL)
3880 rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3882 GOTO(err_mdt_svc, rc);
3885 * readpage service configuration. Parameters have to be adjusted,
3888 conf = (typeof(conf)) {
3889 .psc_nbufs = MDS_NBUFS,
3890 .psc_bufsize = MDS_BUFSIZE,
3891 .psc_max_req_size = MDS_MAXREQSIZE,
3892 .psc_max_reply_size = MDS_MAXREPSIZE,
3893 .psc_req_portal = MDS_READPAGE_PORTAL,
3894 .psc_rep_portal = MDC_REPLY_PORTAL,
3895 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3896 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3898 .psc_max_threads = MDT_MAX_THREADS,
3899 .psc_ctx_tags = LCT_MD_THREAD
3901 m->mdt_readpage_service =
3902 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3903 LUSTRE_MDT_NAME "_readpage",
3904 procfs_entry, target_print_req,"mdt_rdpg");
3906 if (m->mdt_readpage_service == NULL) {
3907 CERROR("failed to start readpage service\n");
3908 GOTO(err_mdt_svc, rc = -ENOMEM);
3911 rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3914 * setattr service configuration.
3916 conf = (typeof(conf)) {
3917 .psc_nbufs = MDS_NBUFS,
3918 .psc_bufsize = MDS_BUFSIZE,
3919 .psc_max_req_size = MDS_MAXREQSIZE,
3920 .psc_max_reply_size = MDS_MAXREPSIZE,
3921 .psc_req_portal = MDS_SETATTR_PORTAL,
3922 .psc_rep_portal = MDC_REPLY_PORTAL,
3923 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3924 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3926 .psc_max_threads = MDT_MAX_THREADS,
3927 .psc_ctx_tags = LCT_MD_THREAD
3930 m->mdt_setattr_service =
3931 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3932 LUSTRE_MDT_NAME "_setattr",
3933 procfs_entry, target_print_req,"mdt_attr");
3935 if (!m->mdt_setattr_service) {
3936 CERROR("failed to start setattr service\n");
3937 GOTO(err_mdt_svc, rc = -ENOMEM);
3940 rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3942 GOTO(err_mdt_svc, rc);
3945 * sequence controller service configuration
3947 conf = (typeof(conf)) {
3948 .psc_nbufs = MDS_NBUFS,
3949 .psc_bufsize = MDS_BUFSIZE,
3950 .psc_max_req_size = SEQ_MAXREQSIZE,
3951 .psc_max_reply_size = SEQ_MAXREPSIZE,
3952 .psc_req_portal = SEQ_CONTROLLER_PORTAL,
3953 .psc_rep_portal = MDC_REPLY_PORTAL,
3954 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3955 .psc_min_threads = SEQ_NUM_THREADS,
3956 .psc_max_threads = SEQ_NUM_THREADS,
3957 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3960 m->mdt_mdsc_service =
3961 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3962 LUSTRE_MDT_NAME"_mdsc",
3963 procfs_entry, target_print_req,"mdt_mdsc");
3964 if (!m->mdt_mdsc_service) {
3965 CERROR("failed to start seq controller service\n");
3966 GOTO(err_mdt_svc, rc = -ENOMEM);
3969 rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3971 GOTO(err_mdt_svc, rc);
3974 * metadata sequence server service configuration
3976 conf = (typeof(conf)) {
3977 .psc_nbufs = MDS_NBUFS,
3978 .psc_bufsize = MDS_BUFSIZE,
3979 .psc_max_req_size = SEQ_MAXREQSIZE,
3980 .psc_max_reply_size = SEQ_MAXREPSIZE,
3981 .psc_req_portal = SEQ_METADATA_PORTAL,
3982 .psc_rep_portal = MDC_REPLY_PORTAL,
3983 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3984 .psc_min_threads = SEQ_NUM_THREADS,
3985 .psc_max_threads = SEQ_NUM_THREADS,
3986 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3989 m->mdt_mdss_service =
3990 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
3991 LUSTRE_MDT_NAME"_mdss",
3992 procfs_entry, target_print_req,"mdt_mdss");
3993 if (!m->mdt_mdss_service) {
3994 CERROR("failed to start metadata seq server service\n");
3995 GOTO(err_mdt_svc, rc = -ENOMEM);
3998 rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
4000 GOTO(err_mdt_svc, rc);
4004 * Data sequence server service configuration. We want to have really
4005 * cluster-wide sequences space. This is why we start only one sequence
4006 * controller which manages space.
4008 conf = (typeof(conf)) {
4009 .psc_nbufs = MDS_NBUFS,
4010 .psc_bufsize = MDS_BUFSIZE,
4011 .psc_max_req_size = SEQ_MAXREQSIZE,
4012 .psc_max_reply_size = SEQ_MAXREPSIZE,
4013 .psc_req_portal = SEQ_DATA_PORTAL,
4014 .psc_rep_portal = OSC_REPLY_PORTAL,
4015 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4016 .psc_min_threads = SEQ_NUM_THREADS,
4017 .psc_max_threads = SEQ_NUM_THREADS,
4018 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
4021 m->mdt_dtss_service =
4022 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
4023 LUSTRE_MDT_NAME"_dtss",
4024 procfs_entry, target_print_req,"mdt_dtss");
4025 if (!m->mdt_dtss_service) {
4026 CERROR("failed to start data seq server service\n");
4027 GOTO(err_mdt_svc, rc = -ENOMEM);
4030 rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
4032 GOTO(err_mdt_svc, rc);
4034 /* FLD service start */
4035 conf = (typeof(conf)) {
4036 .psc_nbufs = MDS_NBUFS,
4037 .psc_bufsize = MDS_BUFSIZE,
4038 .psc_max_req_size = FLD_MAXREQSIZE,
4039 .psc_max_reply_size = FLD_MAXREPSIZE,
4040 .psc_req_portal = FLD_REQUEST_PORTAL,
4041 .psc_rep_portal = MDC_REPLY_PORTAL,
4042 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4043 .psc_min_threads = FLD_NUM_THREADS,
4044 .psc_max_threads = FLD_NUM_THREADS,
4045 .psc_ctx_tags = LCT_DT_THREAD|LCT_MD_THREAD
4048 m->mdt_fld_service =
4049 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
4050 LUSTRE_MDT_NAME"_fld",
4051 procfs_entry, target_print_req, "mdt_fld");
4052 if (!m->mdt_fld_service) {
4053 CERROR("failed to start fld service\n");
4054 GOTO(err_mdt_svc, rc = -ENOMEM);
4057 rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
4059 GOTO(err_mdt_svc, rc);
4062 * mds-mds service configuration. Separate portal is used to allow
4063 * mds-mds requests be not blocked during recovery.
4065 conf = (typeof(conf)) {
4066 .psc_nbufs = MDS_NBUFS,
4067 .psc_bufsize = MDS_BUFSIZE,
4068 .psc_max_req_size = MDS_MAXREQSIZE,
4069 .psc_max_reply_size = MDS_MAXREPSIZE,
4070 .psc_req_portal = MDS_MDS_PORTAL,
4071 .psc_rep_portal = MDC_REPLY_PORTAL,
4072 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4073 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
4075 .psc_max_threads = MDT_MAX_THREADS,
4076 .psc_ctx_tags = LCT_MD_THREAD
4078 m->mdt_xmds_service =
4079 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
4080 LUSTRE_MDT_NAME "_mds",
4081 procfs_entry, target_print_req,"mdt_xmds");
4083 if (m->mdt_xmds_service == NULL) {
4084 CERROR("failed to start readpage service\n");
4085 GOTO(err_mdt_svc, rc = -ENOMEM);
4088 rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
4090 GOTO(err_mdt_svc, rc);
4095 mdt_stop_ptlrpc_service(m);
4100 static void mdt_stack_fini(const struct lu_env *env,
4101 struct mdt_device *m, struct lu_device *top)
4103 struct obd_device *obd = mdt2obd_dev(m);
4104 struct lustre_cfg_bufs *bufs;
4105 struct lustre_cfg *lcfg;
4106 struct mdt_thread_info *info;
4110 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4111 LASSERT(info != NULL);
4113 bufs = &info->mti_u.bufs;
4114 /* process cleanup, pass mdt obd name to get obd umount flags */
4115 lustre_cfg_bufs_reset(bufs, obd->obd_name);
4120 lustre_cfg_bufs_set_string(bufs, 1, flags);
4121 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4123 CERROR("Cannot alloc lcfg!\n");
4128 top->ld_ops->ldo_process_config(env, top, lcfg);
4129 lustre_cfg_free(lcfg);
4131 lu_stack_fini(env, top);
4132 m->mdt_child = NULL;
4133 m->mdt_bottom = NULL;
4136 static struct lu_device *mdt_layer_setup(struct lu_env *env,
4137 const char *typename,
4138 struct lu_device *child,
4139 struct lustre_cfg *cfg)
4141 const char *dev = lustre_cfg_string(cfg, 0);
4142 struct obd_type *type;
4143 struct lu_device_type *ldt;
4144 struct lu_device *d;
4149 type = class_get_type(typename);
4151 CERROR("Unknown type: '%s'\n", typename);
4152 GOTO(out, rc = -ENODEV);
4155 rc = lu_env_refill((struct lu_env *)env);
4157 CERROR("Failure to refill session: '%d'\n", rc);
4163 CERROR("type: '%s'\n", typename);
4164 GOTO(out_type, rc = -EINVAL);
4167 ldt->ldt_obd_type = type;
4168 d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
4170 CERROR("Cannot allocate device: '%s'\n", typename);
4171 GOTO(out_type, rc = -ENODEV);
4174 LASSERT(child->ld_site);
4175 d->ld_site = child->ld_site;
4178 rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
4180 CERROR("can't init device '%s', rc %d\n", typename, rc);
4181 GOTO(out_alloc, rc);
4184 lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
4189 ldt->ldt_ops->ldto_device_free(env, d);
4192 class_put_type(type);
4197 static int mdt_stack_init(struct lu_env *env,
4198 struct mdt_device *m,
4199 struct lustre_cfg *cfg,
4200 struct lustre_mount_info *lmi)
4202 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
4203 struct lu_device *tmp;
4204 struct md_device *md;
4205 struct lu_device *child_lu_dev;
4209 /* init the stack */
4210 tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
4212 RETURN(PTR_ERR(tmp));
4214 m->mdt_bottom = lu2dt_dev(tmp);
4216 tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
4218 GOTO(out, rc = PTR_ERR(tmp));
4223 tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
4225 GOTO(out, rc = PTR_ERR(tmp));
4228 /*set mdd upcall device*/
4229 md_upcall_dev_set(md, lu2md_dev(d));
4232 /*set cmm upcall device*/
4233 md_upcall_dev_set(md, &m->mdt_md_dev);
4235 m->mdt_child = lu2md_dev(d);
4237 /* process setup config */
4238 tmp = &m->mdt_md_dev.md_lu_dev;
4239 rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
4243 /* initialize local objects */
4244 child_lu_dev = &m->mdt_child->md_lu_dev;
4246 rc = child_lu_dev->ld_ops->ldo_prepare(env,
4247 &m->mdt_md_dev.md_lu_dev,
4250 /* fini from last known good lu_device */
4252 mdt_stack_fini(env, m, d);
4258 * setup CONFIG_ORIG context, used to access local config log.
4259 * this may need to be rewrite as part of llog rewrite for lu-api.
4261 static int mdt_obd_llog_setup(struct obd_device *obd,
4262 struct lustre_sb_info *lsi)
4266 LASSERT(obd->obd_fsops == NULL);
4268 obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
4269 if (IS_ERR(obd->obd_fsops))
4270 return PTR_ERR(obd->obd_fsops);
4272 rc = fsfilt_setup(obd, lsi->lsi_srv_mnt->mnt_sb);
4274 fsfilt_put_ops(obd->obd_fsops);
4278 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
4279 obd->obd_lvfs_ctxt.pwdmnt = lsi->lsi_srv_mnt;
4280 obd->obd_lvfs_ctxt.pwd = lsi->lsi_srv_mnt->mnt_root;
4281 obd->obd_lvfs_ctxt.fs = get_ds();
4283 rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd,
4284 0, NULL, &llog_lvfs_ops);
4286 CERROR("llog_setup() failed: %d\n", rc);
4287 fsfilt_put_ops(obd->obd_fsops);
4293 static void mdt_obd_llog_cleanup(struct obd_device *obd)
4295 struct llog_ctxt *ctxt;
4297 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
4301 if (obd->obd_fsops) {
4302 fsfilt_put_ops(obd->obd_fsops);
4303 obd->obd_fsops = NULL;
4307 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4309 struct md_device *next = m->mdt_child;
4310 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
4311 struct lu_site *ls = d->ld_site;
4312 struct obd_device *obd = mdt2obd_dev(m);
4315 target_recovery_fini(obd);
4317 ping_evictor_stop();
4319 mdt_stop_ptlrpc_service(m);
4320 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4321 mdt_obd_llog_cleanup(obd);
4322 obd_exports_barrier(obd);
4323 obd_zombie_barrier();
4324 #ifdef HAVE_QUOTA_SUPPORT
4325 next->md_ops->mdo_quota.mqo_cleanup(env, next);
4327 lut_fini(env, &m->mdt_lut);
4328 mdt_fs_cleanup(env, m);
4329 upcall_cache_cleanup(m->mdt_identity_cache);
4330 m->mdt_identity_cache = NULL;
4332 if (m->mdt_namespace != NULL) {
4333 ldlm_namespace_free(m->mdt_namespace, NULL,
4334 d->ld_obd->obd_force);
4335 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4338 cfs_free_nidlist(&m->mdt_nosquash_nids);
4339 if (m->mdt_nosquash_str) {
4340 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4341 m->mdt_nosquash_str = NULL;
4342 m->mdt_nosquash_strlen = 0;
4345 mdt_seq_fini(env, m);
4346 mdt_seq_fini_cli(m);
4347 mdt_fld_fini(env, m);
4348 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4350 next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4351 cfs_timer_disarm(&m->mdt_ck_timer);
4352 mdt_ck_thread_stop(m);
4357 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4360 if (obd->obd_proc_exports_entry) {
4361 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
4362 obd->obd_proc_exports_entry = NULL;
4364 lprocfs_free_per_client_stats(obd);
4365 lprocfs_free_obd_stats(obd);
4366 ptlrpc_lprocfs_unregister_obd(obd);
4367 lprocfs_obd_cleanup(obd);
4370 struct md_site *mite;
4373 mite = lu_site2md(ls);
4377 LASSERT(atomic_read(&d->ld_ref) == 0);
4382 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4384 struct mdt_device *m = mdt_dev(obd->obd_lu_dev);
4385 struct sptlrpc_rule_set tmp_rset;
4388 sptlrpc_rule_set_init(&tmp_rset);
4389 rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4391 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4396 sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4398 write_lock(&m->mdt_sptlrpc_lock);
4399 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4400 m->mdt_sptlrpc_rset = tmp_rset;
4401 write_unlock(&m->mdt_sptlrpc_lock);
4406 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
4410 m->mdt_opts.mo_mds_capa = 1;
4411 m->mdt_opts.mo_oss_capa = 1;
4412 #ifdef CONFIG_FS_POSIX_ACL
4413 /* ACLs should be enabled by default (b=13829) */
4414 m->mdt_opts.mo_acl = 1;
4415 LCONSOLE_INFO("Enabling ACL\n");
4417 m->mdt_opts.mo_acl = 0;
4418 LCONSOLE_INFO("Disabling ACL\n");
4427 while (*p && *p != ',')
4431 if ((len == sizeof("user_xattr") - 1) &&
4432 (memcmp(options, "user_xattr", len) == 0)) {
4433 m->mdt_opts.mo_user_xattr = 1;
4434 LCONSOLE_INFO("Enabling user_xattr\n");
4435 } else if ((len == sizeof("nouser_xattr") - 1) &&
4436 (memcmp(options, "nouser_xattr", len) == 0)) {
4437 m->mdt_opts.mo_user_xattr = 0;
4438 LCONSOLE_INFO("Disabling user_xattr\n");
4439 } else if ((len == sizeof("noacl") - 1) &&
4440 (memcmp(options, "noacl", len) == 0)) {
4441 m->mdt_opts.mo_acl = 0;
4442 LCONSOLE_INFO("Disabling ACL\n");
4449 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4451 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4452 struct lu_device_type *ldt, struct lustre_cfg *cfg)
4454 struct lprocfs_static_vars lvars;
4455 struct mdt_thread_info *info;
4456 struct obd_device *obd;
4457 const char *dev = lustre_cfg_string(cfg, 0);
4458 const char *num = lustre_cfg_string(cfg, 2);
4459 struct lustre_mount_info *lmi = NULL;
4460 struct lustre_sb_info *lsi;
4461 struct lustre_disk_data *ldd;
4463 struct md_site *mite;
4464 const char *identity_upcall = "NONE";
4465 #ifdef HAVE_QUOTA_SUPPORT
4466 struct md_device *next;
4472 md_device_init(&m->mdt_md_dev, ldt);
4474 * Environment (env) might be missing mdt_thread_key values at that
4475 * point, if device is allocated when mdt_thread_key is in QUIESCENT
4478 * Usually device allocation path doesn't use module key values, but
4479 * mdt has to do a lot of work here, so allocate key value.
4481 rc = lu_env_refill((struct lu_env *)env);
4485 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4486 LASSERT(info != NULL);
4488 obd = class_name2obd(dev);
4489 LASSERT(obd != NULL);
4491 spin_lock_init(&m->mdt_transno_lock);
4493 m->mdt_max_mdsize = MAX_MD_SIZE;
4494 m->mdt_max_cookiesize = sizeof(struct llog_cookie);
4495 m->mdt_som_conf = 0;
4497 m->mdt_opts.mo_user_xattr = 0;
4498 m->mdt_opts.mo_acl = 0;
4499 m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4500 lmi = server_get_mount_2(dev);
4502 CERROR("Cannot get mount info for %s!\n", dev);
4505 lsi = s2lsi(lmi->lmi_sb);
4506 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
4507 /* CMD is supported only in IAM mode */
4510 node_id = simple_strtol(num, NULL, 10);
4511 if (!(ldd->ldd_flags & LDD_F_IAM_DIR) && node_id) {
4512 CERROR("CMD Operation not allowed in IOP mode\n");
4513 GOTO(err_lmi, rc = -EINVAL);
4517 rwlock_init(&m->mdt_sptlrpc_lock);
4518 sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4520 spin_lock_init(&m->mdt_ioepoch_lock);
4521 m->mdt_opts.mo_compat_resname = 0;
4522 m->mdt_capa_timeout = CAPA_TIMEOUT;
4523 m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4524 m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4525 m->mdt_squash_uid = 0;
4526 m->mdt_squash_gid = 0;
4527 CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4528 m->mdt_nosquash_str = NULL;
4529 m->mdt_nosquash_strlen = 0;
4530 init_rwsem(&m->mdt_squash_sem);
4532 spin_lock_init(&m->mdt_client_bitmap_lock);
4534 OBD_ALLOC_PTR(mite);
4536 GOTO(err_lmi, rc = -ENOMEM);
4540 m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4541 m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4542 /* set this lu_device to obd, because error handling need it */
4543 obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4545 rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
4547 CERROR("Can't init lu_site, rc %d\n", rc);
4548 GOTO(err_free_site, rc);
4551 lprocfs_mdt_init_vars(&lvars);
4552 rc = lprocfs_obd_setup(obd, lvars.obd_vars);
4554 CERROR("Can't init lprocfs, rc %d\n", rc);
4555 GOTO(err_fini_site, rc);
4557 ptlrpc_lprocfs_register_obd(obd);
4559 rc = mdt_procfs_init(m, dev);
4561 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4562 GOTO(err_fini_proc, rc);
4565 obd->obd_proc_exports_entry = proc_mkdir("exports",
4566 obd->obd_proc_entry);
4567 if (obd->obd_proc_exports_entry)
4568 lprocfs_add_simple(obd->obd_proc_exports_entry,
4569 "clear", lprocfs_nid_stats_clear_read,
4570 lprocfs_nid_stats_clear_write, obd, NULL);
4572 /* set server index */
4573 lu_site2md(s)->ms_node_id = node_id;
4575 /* failover is the default
4576 * FIXME: we do not failout mds0/mgs, which may cause some problems.
4577 * assumed whose ms_node_id == 0 XXX
4579 obd->obd_replayable = 1;
4580 /* No connection accepted until configurations will finish */
4581 obd->obd_no_conn = 1;
4583 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4584 char *str = lustre_cfg_string(cfg, 4);
4585 if (strchr(str, 'n')) {
4586 CWARN("%s: recovery disabled\n", obd->obd_name);
4587 obd->obd_replayable = 0;
4591 /* init the stack */
4592 rc = mdt_stack_init((struct lu_env *)env, m, cfg, lmi);
4594 CERROR("Can't init device stack, rc %d\n", rc);
4595 GOTO(err_fini_proc, rc);
4598 rc = lut_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4600 GOTO(err_fini_stack, rc);
4602 rc = mdt_fld_init(env, obd->obd_name, m);
4606 rc = mdt_seq_init(env, obd->obd_name, m);
4608 GOTO(err_fini_fld, rc);
4610 snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4611 LUSTRE_MDT_NAME"-%p", m);
4612 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4613 LDLM_NAMESPACE_SERVER,
4614 LDLM_NAMESPACE_GREEDY);
4615 if (m->mdt_namespace == NULL)
4616 GOTO(err_fini_seq, rc = -ENOMEM);
4618 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4619 /* set obd_namespace for compatibility with old code */
4620 obd->obd_namespace = m->mdt_namespace;
4622 /* XXX: to support suppgid for ACL, we enable identity_upcall
4623 * by default, otherwise, maybe got unexpected -EACCESS. */
4624 if (m->mdt_opts.mo_acl)
4625 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4627 m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
4628 &mdt_identity_upcall_cache_ops);
4629 if (IS_ERR(m->mdt_identity_cache)) {
4630 rc = PTR_ERR(m->mdt_identity_cache);
4631 m->mdt_identity_cache = NULL;
4632 GOTO(err_free_ns, rc);
4635 cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4637 rc = mdt_ck_thread_start(m);
4639 GOTO(err_free_ns, rc);
4641 rc = mdt_fs_setup(env, m, obd, lsi);
4645 rc = mdt_obd_llog_setup(obd, lsi);
4647 GOTO(err_fs_cleanup, rc);
4649 rc = mdt_llog_ctxt_clone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4651 GOTO(err_llog_cleanup, rc);
4653 mdt_adapt_sptlrpc_conf(obd, 1);
4655 #ifdef HAVE_QUOTA_SUPPORT
4656 next = m->mdt_child;
4657 rc = next->md_ops->mdo_quota.mqo_setup(env, next, lmi->lmi_mnt);
4659 GOTO(err_llog_cleanup, rc);
4662 server_put_mount_2(dev, lmi->lmi_mnt);
4665 target_recovery_init(&m->mdt_lut, mdt_recovery_handle);
4667 rc = mdt_start_ptlrpc_service(m);
4669 GOTO(err_recovery, rc);
4671 ping_evictor_start();
4673 rc = lu_site_init_finish(s);
4675 GOTO(err_stop_service, rc);
4677 if (obd->obd_recovering == 0)
4678 mdt_postrecov(env, m);
4680 mdt_init_capa_ctxt(env, m);
4682 /* Reduce the initial timeout on an MDS because it doesn't need such
4683 * a long timeout as an OST does. Adaptive timeouts will adjust this
4684 * value appropriately. */
4685 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4686 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4691 ping_evictor_stop();
4692 mdt_stop_ptlrpc_service(m);
4694 target_recovery_fini(obd);
4695 #ifdef HAVE_QUOTA_SUPPORT
4696 next->md_ops->mdo_quota.mqo_cleanup(env, next);
4699 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4700 mdt_obd_llog_cleanup(obd);
4702 mdt_fs_cleanup(env, m);
4704 cfs_timer_disarm(&m->mdt_ck_timer);
4705 mdt_ck_thread_stop(m);
4707 upcall_cache_cleanup(m->mdt_identity_cache);
4708 m->mdt_identity_cache = NULL;
4709 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4710 obd->obd_namespace = m->mdt_namespace = NULL;
4712 mdt_seq_fini(env, m);
4714 mdt_fld_fini(env, m);
4716 lut_fini(env, &m->mdt_lut);
4718 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4721 if (obd->obd_proc_exports_entry) {
4722 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
4723 obd->obd_proc_exports_entry = NULL;
4725 ptlrpc_lprocfs_unregister_obd(obd);
4726 lprocfs_obd_cleanup(obd);
4733 server_put_mount_2(dev, lmi->lmi_mnt);
4737 /* used by MGS to process specific configurations */
4738 static int mdt_process_config(const struct lu_env *env,
4739 struct lu_device *d, struct lustre_cfg *cfg)
4741 struct mdt_device *m = mdt_dev(d);
4742 struct md_device *md_next = m->mdt_child;
4743 struct lu_device *next = md2lu_dev(md_next);
4747 switch (cfg->lcfg_command) {
4749 struct lprocfs_static_vars lvars;
4750 struct obd_device *obd = d->ld_obd;
4753 * For interoperability between 1.8 and 2.0,
4754 * skip old "mdt.group_upcall" param.
4757 char *param = lustre_cfg_string(cfg, 1);
4758 if (param && !strncmp("mdt.group_upcall", param, 16)) {
4759 CWARN("For 1.8 interoperability, skip this"
4760 " mdt.group_upcall. It is obsolete\n");
4765 lprocfs_mdt_init_vars(&lvars);
4766 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4768 if (rc > 0 || rc == -ENOSYS)
4769 /* we don't understand; pass it on */
4770 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4775 * Add mdc hook to get first MDT uuid and connect it to
4776 * ls->controller to use for seq manager.
4778 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4780 CERROR("Can't add mdc, rc %d\n", rc);
4782 rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4785 /* others are passed further */
4786 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4792 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4793 const struct lu_object_header *hdr,
4794 struct lu_device *d)
4796 struct mdt_object *mo;
4802 struct lu_object *o;
4803 struct lu_object_header *h;
4805 o = &mo->mot_obj.mo_lu;
4806 h = &mo->mot_header;
4807 lu_object_header_init(h);
4808 lu_object_init(o, h, d);
4809 lu_object_add_top(h, o);
4810 o->lo_ops = &mdt_obj_ops;
4816 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4817 const struct lu_object_conf *unused)
4819 struct mdt_device *d = mdt_dev(o->lo_dev);
4820 struct lu_device *under;
4821 struct lu_object *below;
4825 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4826 PFID(lu_object_fid(o)));
4828 under = &d->mdt_child->md_lu_dev;
4829 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4830 if (below != NULL) {
4831 lu_object_add(o, below);
4838 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4840 struct mdt_object *mo = mdt_obj(o);
4841 struct lu_object_header *h;
4845 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4846 PFID(lu_object_fid(o)));
4849 lu_object_header_fini(h);
4854 static int mdt_object_print(const struct lu_env *env, void *cookie,
4855 lu_printer_t p, const struct lu_object *o)
4857 struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4858 return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch=%llu "
4859 "flags=%llx, epochcount=%d, writecount=%d)",
4860 mdto, mdto->mot_ioepoch, mdto->mot_flags,
4861 mdto->mot_epochcount, mdto->mot_writecount);
4864 static const struct lu_device_operations mdt_lu_ops = {
4865 .ldo_object_alloc = mdt_object_alloc,
4866 .ldo_process_config = mdt_process_config,
4869 static const struct lu_object_operations mdt_obj_ops = {
4870 .loo_object_init = mdt_object_init,
4871 .loo_object_free = mdt_object_free,
4872 .loo_object_print = mdt_object_print
4875 static int mdt_obd_set_info_async(struct obd_export *exp,
4876 __u32 keylen, void *key,
4877 __u32 vallen, void *val,
4878 struct ptlrpc_request_set *set)
4880 struct obd_device *obd = exp->exp_obd;
4886 if (KEY_IS(KEY_SPTLRPC_CONF)) {
4887 rc = mdt_adapt_sptlrpc_conf(obd, 0);
4894 /* mds_connect_internal */
4895 static int mdt_connect_internal(struct obd_export *exp,
4896 struct mdt_device *mdt,
4897 struct obd_connect_data *data)
4900 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4901 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4903 /* If no known bits (which should not happen, probably,
4904 as everybody should support LOOKUP and UPDATE bits at least)
4905 revert to compat mode with plain locks. */
4906 if (!data->ocd_ibits_known &&
4907 data->ocd_connect_flags & OBD_CONNECT_IBITS)
4908 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4910 if (!mdt->mdt_opts.mo_acl)
4911 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4913 if (!mdt->mdt_opts.mo_user_xattr)
4914 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4916 if (!mdt->mdt_som_conf)
4917 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4919 spin_lock(&exp->exp_lock);
4920 exp->exp_connect_flags = data->ocd_connect_flags;
4921 spin_unlock(&exp->exp_lock);
4922 data->ocd_version = LUSTRE_VERSION_CODE;
4923 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4927 if (mdt->mdt_opts.mo_acl &&
4928 ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4929 CWARN("%s: MDS requires ACL support but client does not\n",
4930 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4935 if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4936 CWARN("%s: MDS requires FID support, but client not\n",
4937 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4941 if (mdt->mdt_som_conf &&
4942 !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS) &&
4943 !(exp->exp_connect_flags & OBD_CONNECT_SOM)) {
4944 CWARN("%s: MDS has SOM enabled, but client does not support "
4945 "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4952 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
4953 struct obd_export *exp,
4954 struct ptlrpc_request *req)
4956 struct sptlrpc_flavor flvr;
4959 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4960 read_lock(&mdt->mdt_sptlrpc_lock);
4961 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
4965 read_unlock(&mdt->mdt_sptlrpc_lock);
4967 spin_lock(&exp->exp_lock);
4969 exp->exp_sp_peer = req->rq_sp_from;
4970 exp->exp_flvr = flvr;
4972 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
4973 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4974 CERROR("unauthorized rpc flavor %x from %s, "
4975 "expect %x\n", req->rq_flvr.sf_rpc,
4976 libcfs_nid2str(req->rq_peer.nid),
4977 exp->exp_flvr.sf_rpc);
4981 spin_unlock(&exp->exp_lock);
4983 if (exp->exp_sp_peer != req->rq_sp_from) {
4984 CERROR("RPC source %s doesn't match %s\n",
4985 sptlrpc_part2name(req->rq_sp_from),
4986 sptlrpc_part2name(exp->exp_sp_peer));
4989 rc = sptlrpc_target_export_check(exp, req);
4996 /* mds_connect copy */
4997 static int mdt_obd_connect(const struct lu_env *env,
4998 struct obd_export **exp, struct obd_device *obd,
4999 struct obd_uuid *cluuid,
5000 struct obd_connect_data *data,
5003 struct mdt_thread_info *info;
5004 struct lsd_client_data *lcd;
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 rc = mdt_connect_internal(lexp, mdt, data);
5035 struct mdt_thread_info *mti;
5036 mti = lu_context_key_get(&env->le_ctx,
5038 LASSERT(mti != NULL);
5039 mti->mti_exp = lexp;
5040 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5041 lexp->exp_mdt_data.med_lcd = lcd;
5042 rc = mdt_client_new(env, mdt);
5045 lexp->exp_mdt_data.med_lcd = NULL;
5047 mdt_export_stats_init(obd, lexp, localdata);
5055 class_disconnect(lexp);
5062 static int mdt_obd_reconnect(const struct lu_env *env,
5063 struct obd_export *exp, struct obd_device *obd,
5064 struct obd_uuid *cluuid,
5065 struct obd_connect_data *data,
5068 struct mdt_thread_info *info;
5069 struct mdt_device *mdt;
5070 struct ptlrpc_request *req;
5074 if (exp == NULL || obd == NULL || cluuid == NULL)
5077 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5078 req = info->mti_pill->rc_req;
5079 mdt = mdt_dev(obd->obd_lu_dev);
5081 rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5085 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5087 mdt_export_stats_init(obd, exp, localdata);
5091 static int mdt_mfd_cleanup(struct obd_export *exp)
5093 struct mdt_export_data *med = &exp->exp_mdt_data;
5094 struct obd_device *obd = exp->exp_obd;
5095 struct mdt_device *mdt;
5096 struct mdt_thread_info *info;
5098 CFS_LIST_HEAD(closing_list);
5099 struct mdt_file_data *mfd, *n;
5103 spin_lock(&med->med_open_lock);
5104 while (!list_empty(&med->med_open_head)) {
5105 struct list_head *tmp = med->med_open_head.next;
5106 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5108 /* Remove mfd handle so it can't be found again.
5109 * We are consuming the mfd_list reference here. */
5110 class_handle_unhash(&mfd->mfd_handle);
5111 list_move_tail(&mfd->mfd_list, &closing_list);
5113 spin_unlock(&med->med_open_lock);
5114 mdt = mdt_dev(obd->obd_lu_dev);
5115 LASSERT(mdt != NULL);
5117 rc = lu_env_init(&env, LCT_MD_THREAD);
5121 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5122 LASSERT(info != NULL);
5123 memset(info, 0, sizeof *info);
5124 info->mti_env = &env;
5125 info->mti_mdt = mdt;
5126 info->mti_exp = exp;
5128 if (!list_empty(&closing_list)) {
5129 struct md_attr *ma = &info->mti_attr;
5133 lmm_size = mdt->mdt_max_mdsize;
5134 OBD_ALLOC(ma->ma_lmm, lmm_size);
5135 if (ma->ma_lmm == NULL)
5136 GOTO(out_lmm, rc = -ENOMEM);
5138 cookie_size = mdt->mdt_max_cookiesize;
5139 OBD_ALLOC(ma->ma_cookie, cookie_size);
5140 if (ma->ma_cookie == NULL)
5141 GOTO(out_cookie, rc = -ENOMEM);
5143 /* Close any open files (which may also cause orphan unlinking). */
5144 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5145 list_del_init(&mfd->mfd_list);
5146 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
5147 ma->ma_lmm_size = lmm_size;
5148 ma->ma_cookie_size = cookie_size;
5150 /* It is not for setattr, just tell MDD to send
5151 * DESTROY RPC to OSS if needed */
5152 ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
5153 ma->ma_valid = MA_FLAGS;
5154 mdt_mfd_close(info, mfd);
5156 info->mti_mdt = NULL;
5157 OBD_FREE(ma->ma_cookie, cookie_size);
5158 ma->ma_cookie = NULL;
5160 OBD_FREE(ma->ma_lmm, lmm_size);
5169 static int mdt_obd_disconnect(struct obd_export *exp)
5171 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5176 class_export_get(exp);
5178 /* Disconnect early so that clients can't keep using export */
5179 rc = class_disconnect(exp);
5180 if (mdt->mdt_namespace != NULL || exp->exp_obd->obd_namespace != NULL)
5181 ldlm_cancel_locks_for_export(exp);
5183 /* release nid stat refererence */
5184 lprocfs_exp_cleanup(exp);
5186 /* complete all outstanding replies */
5187 spin_lock(&exp->exp_lock);
5188 while (!list_empty(&exp->exp_outstanding_replies)) {
5189 struct ptlrpc_reply_state *rs =
5190 list_entry(exp->exp_outstanding_replies.next,
5191 struct ptlrpc_reply_state, rs_exp_list);
5192 struct ptlrpc_service *svc = rs->rs_service;
5194 spin_lock(&svc->srv_lock);
5195 list_del_init(&rs->rs_exp_list);
5196 spin_lock(&rs->rs_lock);
5197 ptlrpc_schedule_difficult_reply(rs);
5198 spin_unlock(&rs->rs_lock);
5199 spin_unlock(&svc->srv_lock);
5201 spin_unlock(&exp->exp_lock);
5202 rc = mdt_mfd_cleanup(exp);
5203 class_export_put(exp);
5207 /* FIXME: Can we avoid using these two interfaces? */
5208 static int mdt_init_export(struct obd_export *exp)
5210 struct mdt_export_data *med = &exp->exp_mdt_data;
5214 CFS_INIT_LIST_HEAD(&med->med_open_head);
5215 spin_lock_init(&med->med_open_lock);
5216 sema_init(&med->med_idmap_sem, 1);
5217 med->med_idmap = NULL;
5218 spin_lock(&exp->exp_lock);
5219 exp->exp_connecting = 1;
5220 spin_unlock(&exp->exp_lock);
5221 rc = ldlm_init_export(exp);
5223 CERROR("Error %d while initializing export\n", rc);
5227 static int mdt_destroy_export(struct obd_export *export)
5229 struct mdt_export_data *med;
5230 struct obd_device *obd = export->exp_obd;
5231 struct mdt_device *mdt;
5232 struct mdt_thread_info *info;
5237 med = &export->exp_mdt_data;
5238 if (exp_connect_rmtclient(export))
5239 mdt_cleanup_idmap(med);
5241 target_destroy_export(export);
5242 ldlm_destroy_export(export);
5244 LASSERT(list_empty(&export->exp_outstanding_replies));
5245 LASSERT(list_empty(&med->med_open_head));
5246 if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
5249 mdt = mdt_dev(obd->obd_lu_dev);
5250 LASSERT(mdt != NULL);
5252 rc = lu_env_init(&env, LCT_MD_THREAD);
5256 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5257 LASSERT(info != NULL);
5258 memset(info, 0, sizeof *info);
5259 info->mti_env = &env;
5260 info->mti_exp = export;
5261 info->mti_mdt = NULL;
5262 mdt_client_del(&env, mdt);
5268 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
5270 if (flag & CONFIG_LOG)
5271 m->mdt_fl_cfglog = 1;
5273 /* also notify active event */
5274 if (flag & CONFIG_SYNC)
5275 m->mdt_fl_synced = 1;
5277 if (m->mdt_fl_cfglog && m->mdt_fl_synced)
5278 /* Open for clients */
5279 m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0;
5282 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
5283 enum md_upcall_event ev, void *data)
5285 struct mdt_device *m = mdt_dev(&md->md_lu_dev);
5286 struct md_device *next = m->mdt_child;
5287 struct mdt_thread_info *mti;
5293 rc = next->md_ops->mdo_maxsize_get(env, next,
5295 &m->mdt_max_cookiesize);
5296 CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
5297 m->mdt_max_mdsize, m->mdt_max_cookiesize);
5298 mdt_allow_cli(m, CONFIG_SYNC);
5300 (*(__u64 *)data) = m->mdt_mount_count;
5303 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5304 mti->mti_no_need_trans = 1;
5305 CDEBUG(D_INFO, "disable mdt trans for this thread\n");
5308 /* Check that MDT is not yet configured */
5309 LASSERT(!m->mdt_fl_cfglog);
5311 #ifdef HAVE_QUOTA_SUPPORT
5313 if (md->md_lu_dev.ld_obd->obd_recovering == 0 &&
5314 likely(md->md_lu_dev.ld_obd->obd_stopping == 0))
5315 next->md_ops->mdo_quota.mqo_recovery(env, next);
5319 CERROR("invalid event\n");
5326 static int mdt_obd_notify(struct obd_device *host,
5327 struct obd_device *watched,
5328 enum obd_notify_event ev, void *data)
5330 struct mdt_device *mdt = mdt_dev(host->obd_lu_dev);
5331 #ifdef HAVE_QUOTA_SUPPORT
5332 struct md_device *next = mdt->mdt_child;
5337 case OBD_NOTIFY_CONFIG:
5338 mdt_allow_cli(mdt, (unsigned long)data);
5340 #ifdef HAVE_QUOTA_SUPPORT
5341 /* quota_type has been processed, we can now handle
5342 * incoming quota requests */
5343 next->md_ops->mdo_quota.mqo_notify(NULL, next);
5347 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
5352 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5353 void *val, int vallen)
5355 struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev);
5356 struct getinfo_fid2path *fpout, *fpin;
5359 fpin = key + size_round(sizeof(KEY_FID2PATH));
5362 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5363 lustre_swab_fid2path(fpin);
5365 memcpy(fpout, fpin, sizeof(*fpin));
5366 if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5369 rc = mdt_fid2path(info->mti_env, mdt, fpout);
5373 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
5374 struct getinfo_fid2path *fp)
5376 struct mdt_object *obj;
5380 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5381 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5383 if (!fid_is_sane(&fp->gf_fid))
5386 obj = mdt_object_find(env, mdt, &fp->gf_fid);
5387 if (obj == NULL || IS_ERR(obj)) {
5388 CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid),
5393 rc = lu_object_exists(&obj->mot_obj.mo_lu);
5399 mdt_object_put(env, obj);
5400 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5401 PFID(&fp->gf_fid), rc);
5405 rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path,
5406 fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno);
5407 mdt_object_put(env, obj);
5412 static int mdt_get_info(struct mdt_thread_info *info)
5414 struct ptlrpc_request *req = mdt_info_req(info);
5422 key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5424 CDEBUG(D_IOCTL, "No GETINFO key");
5427 keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5430 vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5431 if (vallen == NULL) {
5432 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5436 req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5438 rc = req_capsule_server_pack(info->mti_pill);
5439 valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5440 if (valout == NULL) {
5441 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5445 if (KEY_IS(KEY_FID2PATH))
5446 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5450 lustre_msg_set_status(req->rq_repmsg, rc);
5455 /* Pass the ioc down */
5456 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5457 unsigned int cmd, int len, void *data)
5459 struct lu_context ioctl_session;
5460 struct md_device *next = mdt->mdt_child;
5464 rc = lu_context_init(&ioctl_session, LCT_SESSION);
5467 ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5468 lu_context_enter(&ioctl_session);
5469 env->le_ses = &ioctl_session;
5471 LASSERT(next->md_ops->mdo_iocontrol);
5472 rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5474 lu_context_exit(&ioctl_session);
5475 lu_context_fini(&ioctl_session);
5479 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5481 struct obd_ioctl_data *data = karg;
5482 struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1;
5484 struct mdt_object *obj;
5485 struct mdt_lock_handle *lh;
5488 CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5489 if (!fid_is_sane(fid))
5492 lh = &mti->mti_lh[MDT_LH_PARENT];
5493 mdt_lock_reg_init(lh, LCK_CR);
5495 obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5497 RETURN(PTR_ERR(obj));
5499 rc = mdt_object_exists(obj);
5503 * before calling version get the correct MDS should be
5504 * fid, this is error to find remote object here
5506 CERROR("nonlocal object "DFID"\n", PFID(fid));
5507 } else if (rc == 0) {
5509 CDEBUG(D_IOCTL, "no such object: "DFID"\n", PFID(fid));
5511 version = mo_version_get(mti->mti_env, mdt_object_child(obj));
5515 *(__u64 *)data->ioc_inlbuf2 = version;
5519 mdt_object_unlock_put(mti, obj, lh, 1);
5523 /* ioctls on obd dev */
5524 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5525 void *karg, void *uarg)
5528 struct obd_device *obd = exp->exp_obd;
5529 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5530 struct dt_device *dt = mdt->mdt_bottom;
5534 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5535 rc = lu_env_init(&env, LCT_MD_THREAD);
5541 rc = mdt_device_sync(&env, mdt);
5543 case OBD_IOC_SET_READONLY:
5544 dt->dd_ops->dt_ro(&env, dt);
5546 case OBD_IOC_ABORT_RECOVERY:
5547 CERROR("Aborting recovery for device %s\n", obd->obd_name);
5548 target_stop_recovery_thread(obd);
5551 case OBD_IOC_CHANGELOG_REG:
5552 case OBD_IOC_CHANGELOG_DEREG:
5553 case OBD_IOC_CHANGELOG_CLEAR:
5554 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5556 case OBD_IOC_GET_OBJ_VERSION: {
5557 struct mdt_thread_info *mti;
5558 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5559 memset(mti, 0, sizeof *mti);
5560 mti->mti_env = &env;
5564 rc = mdt_ioc_version_get(mti, karg);
5568 CERROR("Not supported cmd = %d for device %s\n",
5569 cmd, obd->obd_name);
5577 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5579 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5580 struct obd_device *obd = mdt2obd_dev(mdt);
5581 #ifdef HAVE_QUOTA_SUPPORT
5582 struct md_device *next = mdt->mdt_child;
5586 /* if some clients didn't participate in recovery then we can possibly
5587 * lost sequence. Now we should increase sequence for safe value */
5588 lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
5589 mdt_seq_adjust(env, mdt, lost);
5591 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5592 #ifdef HAVE_QUOTA_SUPPORT
5593 if (likely(obd->obd_stopping == 0))
5594 next->md_ops->mdo_quota.mqo_recovery(env, next);
5599 int mdt_obd_postrecov(struct obd_device *obd)
5604 rc = lu_env_init(&env, LCT_MD_THREAD);
5607 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5613 * Send a copytool req to a client
5614 * Note this sends a request RPC from a server (MDT) to a client (MDC),
5615 * backwards of normal comms.
5617 int mdt_hsm_copytool_send(struct obd_export *exp)
5620 struct hsm_action_list *hal;
5621 struct hsm_action_item *hai;
5625 CWARN("%s: writing to mdc at %s\n", exp->exp_obd->obd_name,
5626 libcfs_nid2str(exp->exp_connection->c_peer.nid));
5628 len = sizeof(*lh) + sizeof(*hal) + MTI_NAME_MAXLEN +
5629 /* for mockup below */ 2 * size_round(sizeof(*hai));
5634 lh->lnl_magic = LNL_MAGIC;
5635 lh->lnl_transport = LNL_TRANSPORT_HSM;
5636 lh->lnl_msgtype = HMT_ACTION_LIST;
5637 lh->lnl_msglen = len;
5639 hal = (struct hsm_action_list *)(lh + 1);
5640 hal->hal_version = HAL_VERSION;
5641 hal->hal_archive_num = 1;
5642 obd_uuid2fsname(hal->hal_fsname, exp->exp_obd->obd_name,
5645 /* mock up an action list */
5647 hai = hai_zero(hal);
5648 hai->hai_action = HSMA_ARCHIVE;
5649 hai->hai_fid.f_oid = 0xA00A;
5650 hai->hai_len = sizeof(*hai);
5651 hai = hai_next(hai);
5652 hai->hai_action = HSMA_RESTORE;
5653 hai->hai_fid.f_oid = 0xB00B;
5654 hai->hai_len = sizeof(*hai);
5656 /* Uses the ldlm reverse import; this rpc will be seen by
5657 the ldlm_callback_handler */
5658 rc = target_set_info_rpc(exp->exp_imp_reverse, LDLM_SET_INFO,
5659 sizeof(KEY_HSM_COPYTOOL_SEND),
5660 KEY_HSM_COPYTOOL_SEND,
5668 static struct obd_ops mdt_obd_device_ops = {
5669 .o_owner = THIS_MODULE,
5670 .o_set_info_async = mdt_obd_set_info_async,
5671 .o_connect = mdt_obd_connect,
5672 .o_reconnect = mdt_obd_reconnect,
5673 .o_disconnect = mdt_obd_disconnect,
5674 .o_init_export = mdt_init_export,
5675 .o_destroy_export = mdt_destroy_export,
5676 .o_iocontrol = mdt_iocontrol,
5677 .o_postrecov = mdt_obd_postrecov,
5678 .o_notify = mdt_obd_notify
5681 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5682 struct lu_device *d)
5684 struct mdt_device *m = mdt_dev(d);
5691 static struct lu_device *mdt_device_free(const struct lu_env *env,
5692 struct lu_device *d)
5694 struct mdt_device *m = mdt_dev(d);
5697 md_device_fini(&m->mdt_md_dev);
5702 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5703 struct lu_device_type *t,
5704 struct lustre_cfg *cfg)
5706 struct lu_device *l;
5707 struct mdt_device *m;
5713 l = &m->mdt_md_dev.md_lu_dev;
5714 rc = mdt_init0(env, m, t, cfg);
5716 mdt_device_free(env, l);
5720 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
5722 l = ERR_PTR(-ENOMEM);
5726 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5727 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
5729 /* context key: mdt_thread_key */
5730 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5732 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
5733 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
5735 struct lu_context_key mdt_txn_key = {
5736 .lct_tags = LCT_TX_HANDLE,
5737 .lct_init = mdt_txn_key_init,
5738 .lct_fini = mdt_txn_key_fini
5741 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
5743 return md_ucred(info->mti_env);
5747 * Enable/disable COS (Commit On Sharing).
5749 * Set/Clear the COS flag in mdt options.
5751 * \param mdt mdt device
5752 * \param val 0 disables COS, other values enable COS
5754 void mdt_enable_cos(struct mdt_device *mdt, int val)
5759 mdt->mdt_opts.mo_cos = !!val;
5760 rc = lu_env_init(&env, LCT_MD_THREAD);
5761 if (unlikely(rc != 0)) {
5762 CWARN("lu_env initialization failed with rc = %d,"
5763 "cannot sync\n", rc);
5766 mdt_device_sync(&env, mdt);
5771 * Check COS (Commit On Sharing) status.
5773 * Return COS flag status.
5775 * \param mdt mdt device
5777 int mdt_cos_is_enabled(struct mdt_device *mdt)
5779 return mdt->mdt_opts.mo_cos != 0;
5782 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
5783 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
5785 static struct lu_device_type_operations mdt_device_type_ops = {
5786 .ldto_init = mdt_type_init,
5787 .ldto_fini = mdt_type_fini,
5789 .ldto_start = mdt_type_start,
5790 .ldto_stop = mdt_type_stop,
5792 .ldto_device_alloc = mdt_device_alloc,
5793 .ldto_device_free = mdt_device_free,
5794 .ldto_device_fini = mdt_device_fini
5797 static struct lu_device_type mdt_device_type = {
5798 .ldt_tags = LU_DEVICE_MD,
5799 .ldt_name = LUSTRE_MDT_NAME,
5800 .ldt_ops = &mdt_device_type_ops,
5801 .ldt_ctx_tags = LCT_MD_THREAD
5804 static struct lu_local_obj_desc mdt_last_recv = {
5805 .llod_name = LAST_RCVD,
5806 .llod_oid = MDT_LAST_RECV_OID,
5810 static int __init mdt_mod_init(void)
5812 struct lprocfs_static_vars lvars;
5815 llo_local_obj_register(&mdt_last_recv);
5817 mdt_num_threads = MDT_NUM_THREADS;
5818 lprocfs_mdt_init_vars(&lvars);
5819 rc = class_register_type(&mdt_obd_device_ops, NULL,
5820 lvars.module_vars, LUSTRE_MDT_NAME,
5826 static void __exit mdt_mod_exit(void)
5828 llo_local_obj_unregister(&mdt_last_recv);
5829 class_unregister_type(LUSTRE_MDT_NAME);
5833 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt) \
5834 [prefix ## _ ## opc - prefix ## _ ## base] = { \
5836 .mh_fail_id = OBD_FAIL_ ## prefix ## _ ## opc ## suffix, \
5837 .mh_opc = prefix ## _ ## opc, \
5838 .mh_flags = flags, \
5843 #define DEF_MDT_HNDL(flags, name, fn, fmt) \
5844 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
5846 #define DEF_SEQ_HNDL(flags, name, fn, fmt) \
5847 DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
5849 #define DEF_FLD_HNDL(flags, name, fn, fmt) \
5850 DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
5852 * Request with a format known in advance
5854 #define DEF_MDT_HNDL_F(flags, name, fn) \
5855 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
5857 #define DEF_SEQ_HNDL_F(flags, name, fn) \
5858 DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
5860 #define DEF_FLD_HNDL_F(flags, name, fn) \
5861 DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
5863 * Request with a format we do not yet know
5865 #define DEF_MDT_HNDL_0(flags, name, fn) \
5866 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
5868 static struct mdt_handler mdt_mds_ops[] = {
5869 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5870 DEF_MDT_HNDL_F(0, DISCONNECT, mdt_disconnect),
5871 DEF_MDT_HNDL (0, SET_INFO, mdt_set_info,
5873 DEF_MDT_HNDL_F(0, GET_INFO, mdt_get_info),
5874 DEF_MDT_HNDL_F(0 |HABEO_REFERO, GETSTATUS, mdt_getstatus),
5875 DEF_MDT_HNDL_F(HABEO_CORPUS, GETATTR, mdt_getattr),
5876 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
5877 DEF_MDT_HNDL_F(HABEO_CORPUS, GETXATTR, mdt_getxattr),
5878 DEF_MDT_HNDL_F(0 |HABEO_REFERO, STATFS, mdt_statfs),
5879 DEF_MDT_HNDL_F(0 |MUTABOR, REINT, mdt_reint),
5880 DEF_MDT_HNDL_F(HABEO_CORPUS, CLOSE, mdt_close),
5881 DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing),
5882 DEF_MDT_HNDL_F(0 |HABEO_REFERO, PIN, mdt_pin),
5883 DEF_MDT_HNDL_0(0, SYNC, mdt_sync),
5884 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir),
5885 #ifdef HAVE_QUOTA_SUPPORT
5886 DEF_MDT_HNDL_F(0, QUOTACHECK, mdt_quotacheck_handle),
5887 DEF_MDT_HNDL_F(0, QUOTACTL, mdt_quotactl_handle)
5891 #define DEF_OBD_HNDL(flags, name, fn) \
5892 DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
5895 static struct mdt_handler mdt_obd_ops[] = {
5896 DEF_OBD_HNDL(0, PING, mdt_obd_ping),
5897 DEF_OBD_HNDL(0, LOG_CANCEL, mdt_obd_log_cancel),
5898 DEF_OBD_HNDL(0, QC_CALLBACK, mdt_obd_qc_callback)
5901 #define DEF_DLM_HNDL_0(flags, name, fn) \
5902 DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
5903 #define DEF_DLM_HNDL_F(flags, name, fn) \
5904 DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
5906 static struct mdt_handler mdt_dlm_ops[] = {
5907 DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE, mdt_enqueue),
5908 DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT, mdt_convert),
5909 DEF_DLM_HNDL_0(0, BL_CALLBACK, mdt_bl_callback),
5910 DEF_DLM_HNDL_0(0, CP_CALLBACK, mdt_cp_callback)
5913 #define DEF_LLOG_HNDL(flags, name, fn) \
5914 DEF_HNDL(LLOG, ORIGIN_HANDLE_CREATE, _NET, flags, name, fn, NULL)
5916 static struct mdt_handler mdt_llog_ops[] = {
5917 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CREATE, mdt_llog_create),
5918 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_NEXT_BLOCK, mdt_llog_next_block),
5919 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_READ_HEADER, mdt_llog_read_header),
5920 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_WRITE_REC, NULL),
5921 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CLOSE, NULL),
5922 DEF_LLOG_HNDL(0, ORIGIN_CONNECT, NULL),
5923 DEF_LLOG_HNDL(0, CATINFO, NULL),
5924 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_PREV_BLOCK, mdt_llog_prev_block),
5925 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_DESTROY, mdt_llog_destroy),
5928 #define DEF_SEC_CTX_HNDL(name, fn) \
5929 DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
5931 static struct mdt_handler mdt_sec_ctx_ops[] = {
5932 DEF_SEC_CTX_HNDL(INIT, mdt_sec_ctx_handle),
5933 DEF_SEC_CTX_HNDL(INIT_CONT, mdt_sec_ctx_handle),
5934 DEF_SEC_CTX_HNDL(FINI, mdt_sec_ctx_handle)
5937 static struct mdt_opc_slice mdt_regular_handlers[] = {
5939 .mos_opc_start = MDS_GETATTR,
5940 .mos_opc_end = MDS_LAST_OPC,
5941 .mos_hs = mdt_mds_ops
5944 .mos_opc_start = OBD_PING,
5945 .mos_opc_end = OBD_LAST_OPC,
5946 .mos_hs = mdt_obd_ops
5949 .mos_opc_start = LDLM_ENQUEUE,
5950 .mos_opc_end = LDLM_LAST_OPC,
5951 .mos_hs = mdt_dlm_ops
5954 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
5955 .mos_opc_end = LLOG_LAST_OPC,
5956 .mos_hs = mdt_llog_ops
5959 .mos_opc_start = SEC_CTX_INIT,
5960 .mos_opc_end = SEC_LAST_OPC,
5961 .mos_hs = mdt_sec_ctx_ops
5968 static struct mdt_handler mdt_readpage_ops[] = {
5969 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5970 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
5971 #ifdef HAVE_SPLIT_SUPPORT
5972 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
5976 * XXX: this is ugly and should be fixed one day, see mdc_close() for
5977 * detailed comments. --umka
5979 DEF_MDT_HNDL_F(HABEO_CORPUS, CLOSE, mdt_close),
5980 DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing),
5983 static struct mdt_opc_slice mdt_readpage_handlers[] = {
5985 .mos_opc_start = MDS_GETATTR,
5986 .mos_opc_end = MDS_LAST_OPC,
5987 .mos_hs = mdt_readpage_ops
5994 static struct mdt_handler mdt_xmds_ops[] = {
5995 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5996 DEF_MDT_HNDL_F(HABEO_CORPUS , GETATTR, mdt_getattr),
5997 DEF_MDT_HNDL_F(0 | MUTABOR , REINT, mdt_reint),
5998 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir),
6001 static struct mdt_opc_slice mdt_xmds_handlers[] = {
6003 .mos_opc_start = MDS_GETATTR,
6004 .mos_opc_end = MDS_LAST_OPC,
6005 .mos_hs = mdt_xmds_ops
6008 .mos_opc_start = OBD_PING,
6009 .mos_opc_end = OBD_LAST_OPC,
6010 .mos_hs = mdt_obd_ops
6013 .mos_opc_start = SEC_CTX_INIT,
6014 .mos_opc_end = SEC_LAST_OPC,
6015 .mos_hs = mdt_sec_ctx_ops
6022 static struct mdt_handler mdt_seq_ops[] = {
6023 DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
6026 static struct mdt_opc_slice mdt_seq_handlers[] = {
6028 .mos_opc_start = SEQ_QUERY,
6029 .mos_opc_end = SEQ_LAST_OPC,
6030 .mos_hs = mdt_seq_ops
6037 static struct mdt_handler mdt_fld_ops[] = {
6038 DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
6041 static struct mdt_opc_slice mdt_fld_handlers[] = {
6043 .mos_opc_start = FLD_QUERY,
6044 .mos_opc_end = FLD_LAST_OPC,
6045 .mos_hs = mdt_fld_ops
6052 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6053 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
6054 MODULE_LICENSE("GPL");
6056 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
6057 "number of mdt service threads to start");
6059 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);