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 obd_export *exp = req->rq_export;
1180 struct ptlrpc_bulk_desc *desc;
1181 struct l_wait_info *lwi = &info->mti_u.rdpg.mti_wait_info;
1189 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1194 for (i = 0, tmpcount = rdpg->rp_count;
1195 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1196 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1197 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1200 LASSERT(desc->bd_nob == rdpg->rp_count);
1201 rc = sptlrpc_svc_wrap_bulk(req, desc);
1203 GOTO(free_desc, rc);
1205 rc = ptlrpc_start_bulk_transfer(desc);
1207 GOTO(free_desc, rc);
1209 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1210 GOTO(abort_bulk, rc = 0);
1212 timeout = (int) req->rq_deadline - cfs_time_current_sec();
1214 CERROR("Req deadline already passed %lu (now: %lu)\n",
1215 req->rq_deadline, cfs_time_current_sec());
1216 *lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(max(timeout, 1)),
1217 cfs_time_seconds(1), NULL, NULL);
1218 rc = l_wait_event(desc->bd_waitq, !ptlrpc_server_bulk_active(desc) ||
1219 exp->exp_failed || exp->exp_abort_active_req, lwi);
1220 LASSERT (rc == 0 || rc == -ETIMEDOUT);
1223 if (desc->bd_success &&
1224 desc->bd_nob_transferred == rdpg->rp_count)
1225 GOTO(free_desc, rc);
1228 if (exp->exp_abort_active_req || exp->exp_failed)
1229 GOTO(abort_bulk, rc);
1232 DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1233 (rc == -ETIMEDOUT) ? "timeout" : "network error",
1234 desc->bd_nob_transferred, rdpg->rp_count,
1235 exp->exp_client_uuid.uuid,
1236 exp->exp_connection->c_remote_uuid.uuid);
1238 class_fail_export(exp);
1242 ptlrpc_abort_bulk(desc);
1244 ptlrpc_free_bulk(desc);
1248 #ifdef HAVE_SPLIT_SUPPORT
1250 * Retrieve dir entry from the page and insert it to the slave object, actually,
1251 * this should be in osd layer, but since it will not in the final product, so
1252 * just do it here and do not define more moo api anymore for this.
1254 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1257 struct mdt_object *object = info->mti_object;
1258 struct lu_fid *lf = &info->mti_tmp_fid2;
1259 struct md_attr *ma = &info->mti_attr;
1260 struct lu_dirpage *dp;
1261 struct lu_dirent *ent;
1262 int rc = 0, offset = 0;
1265 /* Make sure we have at least one entry. */
1270 * Disable trans for this name insert, since it will include many trans
1273 info->mti_no_need_trans = 1;
1275 * When write_dir_page, no need update parent's ctime,
1276 * and no permission check for name_insert.
1278 ma->ma_attr.la_ctime = 0;
1279 ma->ma_attr.la_valid = LA_MODE;
1280 ma->ma_valid = MA_INODE;
1283 dp = page_address(page);
1284 offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1286 for (ent = lu_dirent_start(dp); ent != NULL;
1287 ent = lu_dirent_next(ent)) {
1288 struct lu_name *lname;
1291 if (le16_to_cpu(ent->lde_namelen) == 0)
1294 fid_le_to_cpu(lf, &ent->lde_fid);
1295 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1296 ma->ma_attr.la_mode = S_IFDIR;
1298 ma->ma_attr.la_mode = 0;
1299 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1301 GOTO(out, rc = -ENOMEM);
1303 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1304 lname = mdt_name(info->mti_env, name,
1305 le16_to_cpu(ent->lde_namelen));
1306 ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE);
1307 rc = mdo_name_insert(info->mti_env,
1308 md_object_next(&object->mot_obj),
1310 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1312 CERROR("Can't insert %*.*s, rc %d\n",
1313 le16_to_cpu(ent->lde_namelen),
1314 le16_to_cpu(ent->lde_namelen),
1319 offset += lu_dirent_size(ent);
1329 static int mdt_bulk_timeout(void *data)
1333 CERROR("mdt bulk transfer timeout \n");
1338 static int mdt_writepage(struct mdt_thread_info *info)
1340 struct ptlrpc_request *req = mdt_info_req(info);
1341 struct mdt_body *reqbody;
1342 struct l_wait_info *lwi;
1343 struct ptlrpc_bulk_desc *desc;
1349 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1350 if (reqbody == NULL)
1351 RETURN(err_serious(-EFAULT));
1353 desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1355 RETURN(err_serious(-ENOMEM));
1357 /* allocate the page for the desc */
1358 page = cfs_alloc_page(CFS_ALLOC_STD);
1360 GOTO(desc_cleanup, rc = -ENOMEM);
1362 CDEBUG(D_INFO, "Received page offset %d size %d \n",
1363 (int)reqbody->size, (int)reqbody->nlink);
1365 ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1366 (int)reqbody->nlink);
1368 rc = sptlrpc_svc_prep_bulk(req, desc);
1370 GOTO(cleanup_page, rc);
1372 * Check if client was evicted while we were doing i/o before touching
1377 GOTO(cleanup_page, rc = -ENOMEM);
1379 if (desc->bd_export->exp_failed)
1382 rc = ptlrpc_start_bulk_transfer (desc);
1384 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
1385 mdt_bulk_timeout, desc);
1386 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1387 desc->bd_export->exp_failed, lwi);
1388 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1389 if (rc == -ETIMEDOUT) {
1390 DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1391 ptlrpc_abort_bulk(desc);
1392 } else if (desc->bd_export->exp_failed) {
1393 DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1395 ptlrpc_abort_bulk(desc);
1396 } else if (!desc->bd_success ||
1397 desc->bd_nob_transferred != desc->bd_nob) {
1398 DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1400 "truncated" : "network error on",
1401 desc->bd_nob_transferred, desc->bd_nob);
1402 /* XXX should this be a different errno? */
1406 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1409 GOTO(cleanup_lwi, rc);
1410 rc = mdt_write_dir_page(info, page, reqbody->nlink);
1415 cfs_free_page(page);
1417 ptlrpc_free_bulk(desc);
1422 static int mdt_readpage(struct mdt_thread_info *info)
1424 struct mdt_object *object = info->mti_object;
1425 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1426 struct mdt_body *reqbody;
1427 struct mdt_body *repbody;
1432 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1433 RETURN(err_serious(-ENOMEM));
1435 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1436 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1437 if (reqbody == NULL || repbody == NULL)
1438 RETURN(err_serious(-EFAULT));
1441 * prepare @rdpg before calling lower layers and transfer itself. Here
1442 * reqbody->size contains offset of where to start to read and
1443 * reqbody->nlink contains number bytes to read.
1445 rdpg->rp_hash = reqbody->size;
1446 if (rdpg->rp_hash != reqbody->size) {
1447 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1448 rdpg->rp_hash, reqbody->size);
1452 rdpg->rp_attrs = reqbody->mode;
1453 rdpg->rp_count = reqbody->nlink;
1454 rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1455 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1456 if (rdpg->rp_pages == NULL)
1459 for (i = 0; i < rdpg->rp_npages; ++i) {
1460 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1461 if (rdpg->rp_pages[i] == NULL)
1462 GOTO(free_rdpg, rc = -ENOMEM);
1465 /* call lower layers to fill allocated pages with directory data */
1466 rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1468 GOTO(free_rdpg, rc);
1470 /* send pages to client */
1471 rc = mdt_sendpage(info, rdpg);
1476 for (i = 0; i < rdpg->rp_npages; i++)
1477 if (rdpg->rp_pages[i] != NULL)
1478 cfs_free_page(rdpg->rp_pages[i]);
1479 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1481 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1487 static int mdt_reint_internal(struct mdt_thread_info *info,
1488 struct mdt_lock_handle *lhc,
1491 struct req_capsule *pill = info->mti_pill;
1492 struct mdt_device *mdt = info->mti_mdt;
1493 struct md_quota *mq = md_quota(info->mti_env);
1494 struct mdt_body *repbody;
1499 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1500 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1501 mdt->mdt_max_mdsize);
1502 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1503 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1504 mdt->mdt_max_cookiesize);
1506 rc = req_capsule_server_pack(pill);
1508 CERROR("Can't pack response, rc %d\n", rc);
1509 RETURN(err_serious(rc));
1512 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1513 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1515 repbody->eadatasize = 0;
1516 repbody->aclsize = 0;
1519 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1520 GOTO(out_shrink, rc = err_serious(-EFAULT));
1522 rc = mdt_reint_unpack(info, op);
1524 CERROR("Can't unpack reint, rc %d\n", rc);
1525 GOTO(out_shrink, rc = err_serious(rc));
1528 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1530 /* for replay no cookkie / lmm need, because client have this already */
1531 if (info->mti_spec.no_create == 1) {
1532 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1533 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1535 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1536 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1540 rc = mdt_init_ucred_reint(info);
1542 GOTO(out_shrink, rc);
1544 rc = mdt_fix_attr_ucred(info, op);
1546 GOTO(out_ucred, rc = err_serious(rc));
1548 if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1549 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1550 GOTO(out_ucred, rc);
1552 mq->mq_exp = info->mti_exp;
1553 rc = mdt_reint_rec(info, lhc);
1556 mdt_exit_ucred(info);
1558 mdt_shrink_reply(info);
1562 static long mdt_reint_opcode(struct mdt_thread_info *info,
1563 const struct req_format **fmt)
1565 struct mdt_rec_reint *rec;
1568 opc = err_serious(-EFAULT);
1569 rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1571 opc = rec->rr_opcode;
1572 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1573 if (opc < REINT_MAX && fmt[opc] != NULL)
1574 req_capsule_extend(info->mti_pill, fmt[opc]);
1576 CERROR("Unsupported opc: %ld\n", opc);
1577 opc = err_serious(opc);
1583 static int mdt_reint(struct mdt_thread_info *info)
1588 static const struct req_format *reint_fmts[REINT_MAX] = {
1589 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1590 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1591 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1592 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1593 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1594 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1595 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1600 opc = mdt_reint_opcode(info, reint_fmts);
1603 * No lock possible here from client to pass it to reint code
1606 rc = mdt_reint_internal(info, NULL, opc);
1611 info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1615 /* this should sync the whole device */
1616 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1618 struct dt_device *dt = mdt->mdt_bottom;
1622 rc = dt->dd_ops->dt_sync(env, dt);
1626 /* this should sync this object */
1627 static int mdt_object_sync(struct mdt_thread_info *info)
1629 struct md_object *next;
1633 if (!mdt_object_exists(info->mti_object)) {
1634 CWARN("Non existing object "DFID"!\n",
1635 PFID(mdt_object_fid(info->mti_object)));
1638 next = mdt_object_child(info->mti_object);
1639 rc = mo_object_sync(info->mti_env, next);
1644 static int mdt_sync(struct mdt_thread_info *info)
1646 struct req_capsule *pill = info->mti_pill;
1647 struct mdt_body *body;
1651 /* The fid may be zero, so we req_capsule_set manually */
1652 req_capsule_set(pill, &RQF_MDS_SYNC);
1654 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1656 RETURN(err_serious(-EINVAL));
1658 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1659 RETURN(err_serious(-ENOMEM));
1661 if (fid_seq(&body->fid1) == 0) {
1662 /* sync the whole device */
1663 rc = req_capsule_server_pack(pill);
1665 rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1667 rc = err_serious(rc);
1669 /* sync an object */
1670 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1672 rc = mdt_object_sync(info);
1674 struct md_object *next;
1675 const struct lu_fid *fid;
1676 struct lu_attr *la = &info->mti_attr.ma_attr;
1678 next = mdt_object_child(info->mti_object);
1679 info->mti_attr.ma_need = MA_INODE;
1680 info->mti_attr.ma_valid = 0;
1681 rc = mo_attr_get(info->mti_env, next,
1684 body = req_capsule_server_get(pill,
1686 fid = mdt_object_fid(info->mti_object);
1687 mdt_pack_attr2body(info, body, la, fid);
1691 rc = err_serious(rc);
1696 #ifdef HAVE_QUOTA_SUPPORT
1697 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1699 struct obd_quotactl *oqctl;
1700 struct req_capsule *pill = info->mti_pill;
1701 struct obd_export *exp = info->mti_exp;
1702 struct md_quota *mq = md_quota(info->mti_env);
1703 struct md_device *next = info->mti_mdt->mdt_child;
1707 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1711 /* remote client has no permission for quotacheck */
1712 if (unlikely(exp_connect_rmtclient(exp)))
1715 rc = req_capsule_server_pack(pill);
1720 rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next,
1725 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1727 struct obd_quotactl *oqctl, *repoqc;
1728 struct req_capsule *pill = info->mti_pill;
1729 struct obd_export *exp = info->mti_exp;
1730 struct md_quota *mq = md_quota(info->mti_env);
1731 struct md_device *next = info->mti_mdt->mdt_child;
1732 const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
1736 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1741 if (exp_connect_rmtclient(exp)) {
1742 struct ptlrpc_request *req = mdt_info_req(info);
1743 struct mdt_export_data *med = mdt_req2med(req);
1744 struct lustre_idmap_table *idmap = med->med_idmap;
1746 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1747 oqctl->qc_cmd != Q_GETINFO))
1751 if (oqctl->qc_type == USRQUOTA)
1752 id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1754 else if (oqctl->qc_type == GRPQUOTA)
1755 id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1760 if (id == CFS_IDMAP_NOTFOUND) {
1761 CDEBUG(D_QUOTA, "no mapping for id %u\n",
1767 rc = req_capsule_server_pack(pill);
1771 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1772 LASSERT(repoqc != NULL);
1775 switch (oqctl->qc_cmd) {
1777 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
1780 rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
1783 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
1787 rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id,
1791 rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id,
1795 rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id,
1799 rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id,
1803 rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id,
1806 case LUSTRE_Q_INVALIDATE:
1807 rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type);
1809 case LUSTRE_Q_FINVALIDATE:
1810 rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type);
1813 CERROR("unsupported mdt_quotactl command: %d\n",
1825 * OBD PING and other handlers.
1827 static int mdt_obd_ping(struct mdt_thread_info *info)
1832 req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1834 rc = target_handle_ping(mdt_info_req(info));
1836 rc = err_serious(rc);
1840 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1842 return err_serious(-EOPNOTSUPP);
1845 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1847 return err_serious(-EOPNOTSUPP);
1855 /** clone llog ctxt from child (mdd)
1856 * This allows remote llog (replicator) access.
1857 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
1858 * context was originally set up, or we can handle them directly.
1859 * I choose the latter, but that means I need any llog
1860 * contexts set up by child to be accessable by the mdt. So we clone the
1861 * context into our context list here.
1863 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
1866 struct md_device *next = mdt->mdt_child;
1867 struct llog_ctxt *ctxt;
1870 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
1873 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
1874 if (rc || ctxt == NULL) {
1875 CERROR("Can't get mdd ctxt %d\n", rc);
1879 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
1881 CERROR("Can't set mdt ctxt %d\n", rc);
1886 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
1887 struct mdt_device *mdt, int idx)
1889 struct llog_ctxt *ctxt;
1891 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
1894 /* Put once for the get we just did, and once for the clone */
1895 llog_ctxt_put(ctxt);
1896 llog_ctxt_put(ctxt);
1900 static int mdt_llog_create(struct mdt_thread_info *info)
1904 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
1905 rc = llog_origin_handle_create(mdt_info_req(info));
1906 return (rc < 0 ? err_serious(rc) : rc);
1909 static int mdt_llog_destroy(struct mdt_thread_info *info)
1913 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
1914 rc = llog_origin_handle_destroy(mdt_info_req(info));
1915 return (rc < 0 ? err_serious(rc) : rc);
1918 static int mdt_llog_read_header(struct mdt_thread_info *info)
1922 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
1923 rc = llog_origin_handle_read_header(mdt_info_req(info));
1924 return (rc < 0 ? err_serious(rc) : rc);
1927 static int mdt_llog_next_block(struct mdt_thread_info *info)
1931 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
1932 rc = llog_origin_handle_next_block(mdt_info_req(info));
1933 return (rc < 0 ? err_serious(rc) : rc);
1936 static int mdt_llog_prev_block(struct mdt_thread_info *info)
1940 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
1941 rc = llog_origin_handle_prev_block(mdt_info_req(info));
1942 return (rc < 0 ? err_serious(rc) : rc);
1949 static struct ldlm_callback_suite cbs = {
1950 .lcs_completion = ldlm_server_completion_ast,
1951 .lcs_blocking = ldlm_server_blocking_ast,
1955 static int mdt_enqueue(struct mdt_thread_info *info)
1957 struct ptlrpc_request *req;
1962 * info->mti_dlm_req already contains swapped and (if necessary)
1963 * converted dlm request.
1965 LASSERT(info->mti_dlm_req != NULL);
1967 req = mdt_info_req(info);
1970 * Lock without inodebits makes no sense and will oops later in
1971 * ldlm. Let's check it now to see if we have wrong lock from client or
1972 * bits get corrupted somewhere in mdt_intent_policy().
1974 req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1975 /* This is disabled because we need to support liblustre flock.
1976 * LASSERT(req_bits != 0);
1979 rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1980 req, info->mti_dlm_req, &cbs);
1981 info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1982 return rc ? err_serious(rc) : req->rq_status;
1985 static int mdt_convert(struct mdt_thread_info *info)
1988 struct ptlrpc_request *req;
1990 LASSERT(info->mti_dlm_req);
1991 req = mdt_info_req(info);
1992 rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1993 return rc ? err_serious(rc) : req->rq_status;
1996 static int mdt_bl_callback(struct mdt_thread_info *info)
1998 CERROR("bl callbacks should not happen on MDS\n");
2000 return err_serious(-EOPNOTSUPP);
2003 static int mdt_cp_callback(struct mdt_thread_info *info)
2005 CERROR("cp callbacks should not happen on MDS\n");
2007 return err_serious(-EOPNOTSUPP);
2011 * sec context handlers
2013 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2017 rc = mdt_handle_idmap(info);
2020 struct ptlrpc_request *req = mdt_info_req(info);
2023 opc = lustre_msg_get_opc(req->rq_reqmsg);
2024 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2025 sptlrpc_svc_ctx_invalidate(req);
2028 OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
2033 static struct mdt_object *mdt_obj(struct lu_object *o)
2035 LASSERT(lu_device_is_mdt(o->lo_dev));
2036 return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2039 struct mdt_object *mdt_object_find(const struct lu_env *env,
2040 struct mdt_device *d,
2041 const struct lu_fid *f)
2043 struct lu_object *o;
2044 struct mdt_object *m;
2047 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2048 o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2049 if (unlikely(IS_ERR(o)))
2050 m = (struct mdt_object *)o;
2057 * Asyncronous commit for mdt device.
2059 * Pass asynchonous commit call down the MDS stack.
2061 * \param env environment
2062 * \param mdt the mdt device
2064 static void mdt_device_commit_async(const struct lu_env *env,
2065 struct mdt_device *mdt)
2067 struct dt_device *dt = mdt->mdt_bottom;
2070 rc = dt->dd_ops->dt_commit_async(env, dt);
2071 if (unlikely(rc != 0))
2072 CWARN("async commit start failed with rc = %d", rc);
2076 * Mark the lock as "synchonous".
2078 * Mark the lock to deffer transaction commit to the unlock time.
2080 * \param lock the lock to mark as "synchonous"
2082 * \see mdt_is_lock_sync
2083 * \see mdt_save_lock
2085 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2087 lock->l_ast_data = (void*)1;
2091 * Check whehter the lock "synchonous" or not.
2093 * \param lock the lock to check
2094 * \retval 1 the lock is "synchonous"
2095 * \retval 0 the lock isn't "synchronous"
2097 * \see mdt_set_lock_sync
2098 * \see mdt_save_lock
2100 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2102 return lock->l_ast_data != NULL;
2106 * Blocking AST for mdt locks.
2108 * Starts transaction commit if in case of COS lock conflict or
2109 * deffers such a commit to the mdt_save_lock.
2111 * \param lock the lock which blocks a request or cancelling lock
2112 * \param desc unused
2113 * \param data unused
2114 * \param flag indicates whether this cancelling or blocking callback
2116 * \see ldlm_blocking_ast_nocheck
2118 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2119 void *data, int flag)
2121 struct obd_device *obd = lock->l_resource->lr_namespace->ns_obd;
2122 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2126 if (flag == LDLM_CB_CANCELING)
2128 lock_res_and_lock(lock);
2129 if (lock->l_blocking_ast != mdt_blocking_ast) {
2130 unlock_res_and_lock(lock);
2133 if (mdt_cos_is_enabled(mdt) &&
2134 lock->l_req_mode & (LCK_PW | LCK_EX) &&
2135 lock->l_blocking_lock != NULL &&
2136 lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2137 mdt_set_lock_sync(lock);
2139 rc = ldlm_blocking_ast_nocheck(lock);
2141 /* There is no lock conflict if l_blocking_lock == NULL,
2142 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2143 * when the last reference to a local lock was released */
2144 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2147 rc = lu_env_init(&env, LCT_MD_THREAD);
2148 if (unlikely(rc != 0))
2149 CWARN("lu_env initialization failed with rc = %d,"
2150 "cannot start asynchronous commit\n", rc);
2152 mdt_device_commit_async(&env, mdt);
2158 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2159 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2161 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2162 ldlm_policy_data_t *policy = &info->mti_policy;
2163 struct ldlm_res_id *res_id = &info->mti_res_id;
2167 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2168 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2169 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2170 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2172 if (mdt_object_exists(o) < 0) {
2173 if (locality == MDT_CROSS_LOCK) {
2174 /* cross-ref object fix */
2175 ibits &= ~MDS_INODELOCK_UPDATE;
2176 ibits |= MDS_INODELOCK_LOOKUP;
2178 LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2179 LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2181 /* No PDO lock on remote object */
2182 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2185 if (lh->mlh_type == MDT_PDO_LOCK) {
2186 /* check for exists after object is locked */
2187 if (mdt_object_exists(o) == 0) {
2188 /* Non-existent object shouldn't have PDO lock */
2191 /* Non-dir object shouldn't have PDO lock */
2192 LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
2196 memset(policy, 0, sizeof(*policy));
2197 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2200 * Take PDO lock on whole directory and build correct @res_id for lock
2201 * on part of directory.
2203 if (lh->mlh_pdo_hash != 0) {
2204 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2205 mdt_lock_pdo_mode(info, o, lh);
2206 if (lh->mlh_pdo_mode != LCK_NL) {
2208 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2209 * is never going to be sent to client and we do not
2210 * want it slowed down due to possible cancels.
2212 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2213 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2214 policy, res_id, LDLM_FL_ATOMIC_CB,
2215 &info->mti_exp->exp_handle.h_cookie);
2221 * Finish res_id initializing by name hash marking part of
2222 * directory which is taking modification.
2224 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2227 policy->l_inodebits.bits = ibits;
2230 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2231 * going to be sent to client. If it is - mdt_intent_policy() path will
2232 * fix it up and turn FL_LOCAL flag off.
2234 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2235 res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
2236 &info->mti_exp->exp_handle.h_cookie);
2242 mdt_object_unlock(info, o, lh, 1);
2249 * Save a lock within request object.
2251 * Keep the lock referenced until whether client ACK or transaction
2252 * commit happens or release the lock immediately depending on input
2253 * parameters. If COS is ON, a write lock is converted to COS lock
2256 * \param info thead info object
2257 * \param h lock handle
2258 * \param mode lock mode
2259 * \param decref force immediate lock releasing
2262 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2263 ldlm_mode_t mode, int decref)
2267 if (lustre_handle_is_used(h)) {
2268 if (decref || !info->mti_has_trans ||
2269 !(mode & (LCK_PW | LCK_EX))){
2270 mdt_fid_unlock(h, mode);
2272 struct mdt_device *mdt = info->mti_mdt;
2273 struct ldlm_lock *lock = ldlm_handle2lock(h);
2274 struct ptlrpc_request *req = mdt_info_req(info);
2277 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2279 CDEBUG(D_HA, "request = %p reply state = %p"
2280 " transno = "LPD64"\n",
2281 req, req->rq_reply_state, req->rq_transno);
2282 if (mdt_cos_is_enabled(mdt)) {
2284 ldlm_lock_downgrade(lock, LCK_COS);
2287 ptlrpc_save_lock(req, h, mode, no_ack);
2288 if (mdt_is_lock_sync(lock)) {
2289 CDEBUG(D_HA, "found sync-lock,"
2290 " async commit started\n");
2291 mdt_device_commit_async(info->mti_env,
2294 LDLM_LOCK_PUT(lock);
2303 * Unlock mdt object.
2305 * Immeditely release the regular lock and the PDO lock or save the
2306 * lock in reqeuest and keep them referenced until client ACK or
2307 * transaction commit.
2309 * \param info thread info object
2310 * \param o mdt object
2311 * \param lh mdt lock handle referencing regular and PDO locks
2312 * \param decref force immediate lock releasing
2314 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2315 struct mdt_lock_handle *lh, int decref)
2319 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2320 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2325 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2326 const struct lu_fid *f,
2327 struct mdt_lock_handle *lh,
2330 struct mdt_object *o;
2332 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2336 rc = mdt_object_lock(info, o, lh, ibits,
2339 mdt_object_put(info->mti_env, o);
2346 void mdt_object_unlock_put(struct mdt_thread_info * info,
2347 struct mdt_object * o,
2348 struct mdt_lock_handle *lh,
2351 mdt_object_unlock(info, o, lh, decref);
2352 mdt_object_put(info->mti_env, o);
2355 static struct mdt_handler *mdt_handler_find(__u32 opc,
2356 struct mdt_opc_slice *supported)
2358 struct mdt_opc_slice *s;
2359 struct mdt_handler *h;
2362 for (s = supported; s->mos_hs != NULL; s++) {
2363 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2364 h = s->mos_hs + (opc - s->mos_opc_start);
2365 if (likely(h->mh_opc != 0))
2366 LASSERTF(h->mh_opc == opc,
2367 "opcode mismatch %d != %d\n",
2370 h = NULL; /* unsupported opc */
2377 static int mdt_lock_resname_compat(struct mdt_device *m,
2378 struct ldlm_request *req)
2380 /* XXX something... later. */
2384 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2386 /* XXX something... later. */
2391 * Generic code handling requests that have struct mdt_body passed in:
2393 * - extract mdt_body from request and save it in @info, if present;
2395 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2398 * - if HABEO_CORPUS flag is set for this request type check whether object
2399 * actually exists on storage (lu_object_exists()).
2402 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2404 const struct mdt_body *body;
2405 struct mdt_object *obj;
2406 const struct lu_env *env;
2407 struct req_capsule *pill;
2411 env = info->mti_env;
2412 pill = info->mti_pill;
2414 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2418 if (!(body->valid & OBD_MD_FLID))
2421 if (!fid_is_sane(&body->fid1)) {
2422 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2427 * Do not get size or any capa fields before we check that request
2428 * contains capa actually. There are some requests which do not, for
2429 * instance MDS_IS_SUBDIR.
2431 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2432 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2433 mdt_set_capainfo(info, 0, &body->fid1,
2434 req_capsule_client_get(pill, &RMF_CAPA1));
2436 obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2438 if ((flags & HABEO_CORPUS) &&
2439 !mdt_object_exists(obj)) {
2440 mdt_object_put(env, obj);
2441 /* for capability renew ENOENT will be handled in
2443 if (body->valid & OBD_MD_FLOSSCAPA)
2448 info->mti_object = obj;
2457 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2459 struct req_capsule *pill = info->mti_pill;
2463 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2464 rc = mdt_body_unpack(info, flags);
2468 if (rc == 0 && (flags & HABEO_REFERO)) {
2469 struct mdt_device *mdt = info->mti_mdt;
2473 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2474 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2475 mdt->mdt_max_mdsize);
2476 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2477 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2478 mdt->mdt_max_cookiesize);
2480 rc = req_capsule_server_pack(pill);
2485 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2487 struct md_device *next = m->mdt_child;
2489 return next->md_ops->mdo_init_capa_ctxt(env, next,
2490 m->mdt_opts.mo_mds_capa,
2491 m->mdt_capa_timeout,
2497 * Invoke handler for this request opc. Also do necessary preprocessing
2498 * (according to handler ->mh_flags), and post-processing (setting of
2499 * ->last_{xid,committed}).
2501 static int mdt_req_handle(struct mdt_thread_info *info,
2502 struct mdt_handler *h, struct ptlrpc_request *req)
2504 int rc, serious = 0;
2509 LASSERT(h->mh_act != NULL);
2510 LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2511 LASSERT(current->journal_info == NULL);
2514 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2515 * to put same checks into handlers like mdt_close(), mdt_reint(),
2516 * etc., without talking to mdt authors first. Checking same thing
2517 * there again is useless and returning 0 error without packing reply
2518 * is buggy! Handlers either pack reply or return error.
2520 * We return 0 here and do not send any reply in order to emulate
2521 * network failure. Do not send any reply in case any of NET related
2522 * fail_id has occured.
2524 if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2528 flags = h->mh_flags;
2529 LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2531 if (h->mh_fmt != NULL) {
2532 req_capsule_set(info->mti_pill, h->mh_fmt);
2533 rc = mdt_unpack_req_pack_rep(info, flags);
2536 if (rc == 0 && flags & MUTABOR &&
2537 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2538 /* should it be rq_status? */
2541 if (rc == 0 && flags & HABEO_CLAVIS) {
2542 struct ldlm_request *dlm_req;
2544 LASSERT(h->mh_fmt != NULL);
2546 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2547 if (dlm_req != NULL) {
2548 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2549 rc = mdt_lock_resname_compat(info->mti_mdt,
2551 info->mti_dlm_req = dlm_req;
2557 /* capability setting changed via /proc, needs reinitialize ctxt */
2558 if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2559 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2560 info->mti_mdt->mdt_capa_conf = 0;
2563 if (likely(rc == 0)) {
2565 * Process request, there can be two types of rc:
2566 * 1) errors with msg unpack/pack, other failures outside the
2567 * operation itself. This is counted as serious errors;
2568 * 2) errors during fs operation, should be placed in rq_status
2571 rc = h->mh_act(info);
2573 !req->rq_no_reply && req->rq_reply_state == NULL) {
2574 DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2575 "pack reply and returned 0 error\n",
2579 serious = is_serious(rc);
2580 rc = clear_serious(rc);
2584 req->rq_status = rc;
2587 * ELDLM_* codes which > 0 should be in rq_status only as well as
2588 * all non-serious errors.
2590 if (rc > 0 || !serious)
2593 LASSERT(current->journal_info == NULL);
2595 if (rc == 0 && (flags & HABEO_CLAVIS) &&
2596 info->mti_mdt->mdt_opts.mo_compat_resname) {
2597 struct ldlm_reply *dlmrep;
2599 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2601 rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2604 /* If we're DISCONNECTing, the mdt_export_data is already freed */
2605 if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2606 target_committed_to_req(req);
2608 if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2609 lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2610 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2614 target_send_reply(req, rc, info->mti_fail_id);
2618 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2620 lh->mlh_type = MDT_NUL_LOCK;
2621 lh->mlh_reg_lh.cookie = 0ull;
2622 lh->mlh_reg_mode = LCK_MINMODE;
2623 lh->mlh_pdo_lh.cookie = 0ull;
2624 lh->mlh_pdo_mode = LCK_MINMODE;
2627 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2629 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2630 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2634 * Initialize fields of struct mdt_thread_info. Other fields are left in
2635 * uninitialized state, because it's too expensive to zero out whole
2636 * mdt_thread_info (> 1K) on each request arrival.
2638 static void mdt_thread_info_init(struct ptlrpc_request *req,
2639 struct mdt_thread_info *info)
2642 struct md_capainfo *ci;
2644 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2645 info->mti_pill = &req->rq_pill;
2648 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2649 mdt_lock_handle_init(&info->mti_lh[i]);
2651 /* mdt device: it can be NULL while CONNECT */
2652 if (req->rq_export) {
2653 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2654 info->mti_exp = req->rq_export;
2656 info->mti_mdt = NULL;
2657 info->mti_env = req->rq_svc_thread->t_env;
2658 ci = md_capainfo(info->mti_env);
2659 memset(ci, 0, sizeof *ci);
2660 if (req->rq_export) {
2661 if (exp_connect_rmtclient(req->rq_export))
2662 ci->mc_auth = LC_ID_CONVERT;
2663 else if (req->rq_export->exp_connect_flags &
2664 OBD_CONNECT_MDS_CAPA)
2665 ci->mc_auth = LC_ID_PLAIN;
2667 ci->mc_auth = LC_ID_NONE;
2670 info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2671 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2672 info->mti_mos[0] = NULL;
2673 info->mti_mos[1] = NULL;
2674 info->mti_mos[2] = NULL;
2675 info->mti_mos[3] = NULL;
2677 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2678 info->mti_body = NULL;
2679 info->mti_object = NULL;
2680 info->mti_dlm_req = NULL;
2681 info->mti_has_trans = 0;
2682 info->mti_no_need_trans = 0;
2683 info->mti_cross_ref = 0;
2684 info->mti_opdata = 0;
2686 /* To not check for split by default. */
2687 info->mti_spec.sp_ck_split = 0;
2688 info->mti_spec.no_create = 0;
2691 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2695 req_capsule_fini(info->mti_pill);
2696 if (info->mti_object != NULL) {
2698 * freeing an object may lead to OSD level transaction, do not
2699 * let it mess with MDT. bz19385.
2701 info->mti_no_need_trans = 1;
2702 mdt_object_put(info->mti_env, info->mti_object);
2703 info->mti_object = NULL;
2705 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2706 mdt_lock_handle_fini(&info->mti_lh[i]);
2707 info->mti_env = NULL;
2710 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2711 struct obd_device *obd, int *process)
2713 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2714 case MDS_CONNECT: /* This will never get here, but for completeness. */
2715 case OST_CONNECT: /* This will never get here, but for completeness. */
2716 case MDS_DISCONNECT:
2717 case OST_DISCONNECT:
2722 case MDS_DONE_WRITING:
2723 case MDS_SYNC: /* used in unmounting */
2729 *process = target_queue_recovery_request(req, obd);
2733 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2740 * Handle recovery. Return:
2741 * +1: continue request processing;
2742 * -ve: abort immediately with the given error code;
2743 * 0: send reply with error code in req->rq_status;
2745 static int mdt_recovery(struct mdt_thread_info *info)
2747 struct ptlrpc_request *req = mdt_info_req(info);
2749 struct obd_device *obd;
2753 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2756 case SEC_CTX_INIT_CONT:
2762 rc = mdt_handle_idmap(info);
2771 if (unlikely(req->rq_export == NULL)) {
2772 CERROR("operation %d on unconnected MDS from %s\n",
2773 lustre_msg_get_opc(req->rq_reqmsg),
2774 libcfs_id2str(req->rq_peer));
2775 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2776 * mds_A will get -ENOTCONN(especially for ping req),
2777 * which will cause that mds_A deactive timeout, then when
2778 * mds_A cleanup, the cleanup process will be suspended since
2779 * deactive timeout is not zero.
2781 req->rq_status = -ENOTCONN;
2782 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2786 /* sanity check: if the xid matches, the request must be marked as a
2787 * resent or replayed */
2788 if (req_xid_is_last(req)) {
2789 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2790 (MSG_RESENT | MSG_REPLAY))) {
2791 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2792 "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2793 lustre_msg_get_flags(req->rq_reqmsg));
2795 req->rq_status = -ENOTCONN;
2800 /* else: note the opposite is not always true; a RESENT req after a
2801 * failover will usually not match the last_xid, since it was likely
2802 * never committed. A REPLAYed request will almost never match the
2803 * last xid, however it could for a committed, but still retained,
2806 obd = req->rq_export->exp_obd;
2808 /* Check for aborted recovery... */
2809 spin_lock_bh(&obd->obd_processing_task_lock);
2810 recovering = obd->obd_recovering;
2811 spin_unlock_bh(&obd->obd_processing_task_lock);
2812 if (unlikely(recovering)) {
2815 DEBUG_REQ(D_INFO, req, "Got new replay");
2816 rc = mdt_filter_recovery_request(req, obd, &should_process);
2817 if (rc != 0 || !should_process)
2819 else if (should_process < 0) {
2820 req->rq_status = should_process;
2821 rc = ptlrpc_error(req);
2828 static int mdt_msg_check_version(struct lustre_msg *msg)
2832 switch (lustre_msg_get_opc(msg)) {
2834 case MDS_DISCONNECT:
2838 case SEC_CTX_INIT_CONT:
2840 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2842 CERROR("bad opc %u version %08x, expecting %08x\n",
2843 lustre_msg_get_opc(msg),
2844 lustre_msg_get_version(msg),
2845 LUSTRE_OBD_VERSION);
2849 case MDS_GETATTR_NAME:
2856 case MDS_DONE_WRITING:
2862 case MDS_QUOTACHECK:
2868 rc = lustre_msg_check_version(msg, LUSTRE_MDS_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_MDS_VERSION);
2877 case LDLM_BL_CALLBACK:
2878 case LDLM_CP_CALLBACK:
2879 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2881 CERROR("bad opc %u version %08x, expecting %08x\n",
2882 lustre_msg_get_opc(msg),
2883 lustre_msg_get_version(msg),
2884 LUSTRE_DLM_VERSION);
2886 case OBD_LOG_CANCEL:
2887 case LLOG_ORIGIN_HANDLE_CREATE:
2888 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2889 case LLOG_ORIGIN_HANDLE_READ_HEADER:
2890 case LLOG_ORIGIN_HANDLE_CLOSE:
2891 case LLOG_ORIGIN_HANDLE_DESTROY:
2892 case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2894 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2896 CERROR("bad opc %u version %08x, expecting %08x\n",
2897 lustre_msg_get_opc(msg),
2898 lustre_msg_get_version(msg),
2899 LUSTRE_LOG_VERSION);
2902 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2908 static int mdt_handle0(struct ptlrpc_request *req,
2909 struct mdt_thread_info *info,
2910 struct mdt_opc_slice *supported)
2912 struct mdt_handler *h;
2913 struct lustre_msg *msg;
2918 if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2921 LASSERT(current->journal_info == NULL);
2923 msg = req->rq_reqmsg;
2924 rc = mdt_msg_check_version(msg);
2925 if (likely(rc == 0)) {
2926 rc = mdt_recovery(info);
2927 if (likely(rc == +1)) {
2928 h = mdt_handler_find(lustre_msg_get_opc(msg),
2930 if (likely(h != NULL)) {
2931 rc = mdt_req_handle(info, h, req);
2933 CERROR("The unsupported opc: 0x%x\n",
2934 lustre_msg_get_opc(msg) );
2935 req->rq_status = -ENOTSUPP;
2936 rc = ptlrpc_error(req);
2941 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2946 * MDT handler function called by ptlrpc service thread when request comes.
2948 * XXX common "target" functionality should be factored into separate module
2949 * shared by mdt, ost and stand-alone services like fld.
2951 static int mdt_handle_common(struct ptlrpc_request *req,
2952 struct mdt_opc_slice *supported)
2955 struct mdt_thread_info *info;
2959 env = req->rq_svc_thread->t_env;
2960 LASSERT(env != NULL);
2961 LASSERT(env->le_ses != NULL);
2962 LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2963 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2964 LASSERT(info != NULL);
2966 mdt_thread_info_init(req, info);
2968 rc = mdt_handle0(req, info, supported);
2970 mdt_thread_info_fini(info);
2975 * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2978 int mdt_recovery_handle(struct ptlrpc_request *req)
2983 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2985 rc = mdt_handle_common(req, mdt_fld_handlers);
2988 rc = mdt_handle_common(req, mdt_seq_handlers);
2991 rc = mdt_handle_common(req, mdt_regular_handlers);
2998 static int mdt_regular_handle(struct ptlrpc_request *req)
3000 return mdt_handle_common(req, mdt_regular_handlers);
3003 static int mdt_readpage_handle(struct ptlrpc_request *req)
3005 return mdt_handle_common(req, mdt_readpage_handlers);
3008 static int mdt_xmds_handle(struct ptlrpc_request *req)
3010 return mdt_handle_common(req, mdt_xmds_handlers);
3013 static int mdt_mdsc_handle(struct ptlrpc_request *req)
3015 return mdt_handle_common(req, mdt_seq_handlers);
3018 static int mdt_mdss_handle(struct ptlrpc_request *req)
3020 return mdt_handle_common(req, mdt_seq_handlers);
3023 static int mdt_dtss_handle(struct ptlrpc_request *req)
3025 return mdt_handle_common(req, mdt_seq_handlers);
3028 static int mdt_fld_handle(struct ptlrpc_request *req)
3030 return mdt_handle_common(req, mdt_fld_handlers);
3046 static int mdt_intent_getattr(enum mdt_it_code opcode,
3047 struct mdt_thread_info *info,
3048 struct ldlm_lock **,
3050 static int mdt_intent_reint(enum mdt_it_code opcode,
3051 struct mdt_thread_info *info,
3052 struct ldlm_lock **,
3055 static struct mdt_it_flavor {
3056 const struct req_format *it_fmt;
3058 int (*it_act)(enum mdt_it_code ,
3059 struct mdt_thread_info *,
3060 struct ldlm_lock **,
3063 } mdt_it_flavor[] = {
3065 .it_fmt = &RQF_LDLM_INTENT,
3066 /*.it_flags = HABEO_REFERO,*/
3068 .it_act = mdt_intent_reint,
3069 .it_reint = REINT_OPEN
3072 .it_fmt = &RQF_LDLM_INTENT,
3073 .it_flags = MUTABOR,
3074 .it_act = mdt_intent_reint,
3075 .it_reint = REINT_OPEN
3078 .it_fmt = &RQF_LDLM_INTENT,
3079 .it_flags = MUTABOR,
3080 .it_act = mdt_intent_reint,
3081 .it_reint = REINT_CREATE
3083 [MDT_IT_GETATTR] = {
3084 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3085 .it_flags = HABEO_REFERO,
3086 .it_act = mdt_intent_getattr
3088 [MDT_IT_READDIR] = {
3094 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3095 .it_flags = HABEO_REFERO,
3096 .it_act = mdt_intent_getattr
3099 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
3100 .it_flags = MUTABOR,
3102 .it_reint = REINT_UNLINK
3106 .it_flags = MUTABOR,
3109 [MDT_IT_GETXATTR] = {
3116 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3117 struct ldlm_lock **lockp,
3118 struct ldlm_lock *new_lock,
3119 struct mdt_lock_handle *lh,
3122 struct ptlrpc_request *req = mdt_info_req(info);
3123 struct ldlm_lock *lock = *lockp;
3126 * Get new lock only for cases when possible resent did not find any
3129 if (new_lock == NULL)
3130 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3132 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3133 lh->mlh_reg_lh.cookie = 0;
3137 LASSERTF(new_lock != NULL,
3138 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3141 * If we've already given this lock to a client once, then we should
3142 * have no readers or writers. Otherwise, we should have one reader
3143 * _or_ writer ref (which will be zeroed below) before returning the
3146 if (new_lock->l_export == req->rq_export) {
3147 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3149 LASSERT(new_lock->l_export == NULL);
3150 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3155 if (new_lock->l_export == req->rq_export) {
3157 * Already gave this to the client, which means that we
3158 * reconstructed a reply.
3160 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3162 lh->mlh_reg_lh.cookie = 0;
3163 RETURN(ELDLM_LOCK_REPLACED);
3167 * Fixup the lock to be given to the client.
3169 lock_res_and_lock(new_lock);
3170 /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3171 * possible blocking AST. */
3172 while (new_lock->l_readers > 0) {
3173 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3174 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3175 new_lock->l_readers--;
3177 while (new_lock->l_writers > 0) {
3178 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3179 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3180 new_lock->l_writers--;
3183 new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3184 new_lock->l_blocking_ast = lock->l_blocking_ast;
3185 new_lock->l_completion_ast = lock->l_completion_ast;
3186 new_lock->l_remote_handle = lock->l_remote_handle;
3187 new_lock->l_flags &= ~LDLM_FL_LOCAL;
3189 unlock_res_and_lock(new_lock);
3191 cfs_hash_add(new_lock->l_export->exp_lock_hash,
3192 &new_lock->l_remote_handle,
3193 &new_lock->l_exp_hash);
3195 LDLM_LOCK_RELEASE(new_lock);
3196 lh->mlh_reg_lh.cookie = 0;
3198 RETURN(ELDLM_LOCK_REPLACED);
3201 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3202 struct ldlm_lock *new_lock,
3203 struct ldlm_lock **old_lock,
3204 struct mdt_lock_handle *lh)
3206 struct ptlrpc_request *req = mdt_info_req(info);
3207 struct obd_export *exp = req->rq_export;
3208 struct lustre_handle remote_hdl;
3209 struct ldlm_request *dlmreq;
3210 struct ldlm_lock *lock;
3212 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3215 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3216 remote_hdl = dlmreq->lock_handle[0];
3218 lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3220 if (lock != new_lock) {
3221 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3222 lh->mlh_reg_mode = lock->l_granted_mode;
3224 LDLM_DEBUG(lock, "Restoring lock cookie");
3225 DEBUG_REQ(D_DLMTRACE, req,
3226 "restoring lock cookie "LPX64,
3227 lh->mlh_reg_lh.cookie);
3229 *old_lock = LDLM_LOCK_GET(lock);
3230 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3234 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3238 * If the xid matches, then we know this is a resent request, and allow
3239 * it. (It's probably an OPEN, for which we don't send a lock.
3241 if (req_xid_is_last(req))
3245 * This remote handle isn't enqueued, so we never received or processed
3246 * this request. Clear MSG_RESENT, because it can be handled like any
3247 * normal request now.
3249 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3251 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3255 static int mdt_intent_getattr(enum mdt_it_code opcode,
3256 struct mdt_thread_info *info,
3257 struct ldlm_lock **lockp,
3260 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3261 struct ldlm_lock *new_lock = NULL;
3263 struct ldlm_reply *ldlm_rep;
3264 struct ptlrpc_request *req;
3265 struct mdt_body *reqbody;
3266 struct mdt_body *repbody;
3270 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3273 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3276 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
3277 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3278 repbody->eadatasize = 0;
3279 repbody->aclsize = 0;
3283 child_bits = MDS_INODELOCK_LOOKUP;
3285 case MDT_IT_GETATTR:
3286 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3289 CERROR("Unhandled till now");
3290 GOTO(out_shrink, rc = -EINVAL);
3293 rc = mdt_init_ucred(info, reqbody);
3295 GOTO(out_shrink, rc);
3297 req = info->mti_pill->rc_req;
3298 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3299 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3301 /* Get lock from request for possible resent case. */
3302 mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3304 ldlm_rep->lock_policy_res2 =
3305 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3307 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3308 ldlm_rep->lock_policy_res2 = 0;
3309 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3310 ldlm_rep->lock_policy_res2) {
3311 lhc->mlh_reg_lh.cookie = 0ull;
3312 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3315 rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3318 mdt_exit_ucred(info);
3320 mdt_shrink_reply(info);
3324 static int mdt_intent_reint(enum mdt_it_code opcode,
3325 struct mdt_thread_info *info,
3326 struct ldlm_lock **lockp,
3329 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3330 struct ldlm_reply *rep = NULL;
3334 static const struct req_format *intent_fmts[REINT_MAX] = {
3335 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
3336 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
3341 opc = mdt_reint_opcode(info, intent_fmts);
3345 if (mdt_it_flavor[opcode].it_reint != opc) {
3346 CERROR("Reint code %ld doesn't match intent: %d\n",
3348 RETURN(err_serious(-EPROTO));
3351 /* Get lock from request for possible resent case. */
3352 mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3354 rc = mdt_reint_internal(info, lhc, opc);
3356 /* Check whether the reply has been packed successfully. */
3357 if (mdt_info_req(info)->rq_repmsg != NULL)
3358 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3360 RETURN(err_serious(-EFAULT));
3362 /* MDC expects this in any case */
3364 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3366 /* Cross-ref case, the lock should be returned to the client */
3367 if (rc == -EREMOTE) {
3368 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3369 rep->lock_policy_res2 = 0;
3370 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3373 rep->lock_policy_res2 = clear_serious(rc);
3375 if (rc == -ENOTCONN || rc == -ENODEV ||
3376 rc == -EOVERFLOW) { /**< if VBR failure then return error */
3378 * If it is the disconnect error (ENODEV & ENOCONN), the error
3379 * will be returned by rq_status, and client at ptlrpc layer
3380 * will detect this, then disconnect, reconnect the import
3381 * immediately, instead of impacting the following the rpc.
3383 lhc->mlh_reg_lh.cookie = 0ull;
3387 * For other cases, the error will be returned by intent.
3388 * and client will retrieve the result from intent.
3391 * FIXME: when open lock is finished, that should be
3394 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3395 rep->lock_policy_res2 = 0;
3396 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3399 lhc->mlh_reg_lh.cookie = 0ull;
3400 RETURN(ELDLM_LOCK_ABORTED);
3405 static int mdt_intent_code(long itcode)
3413 case IT_OPEN|IT_CREAT:
3420 rc = MDT_IT_READDIR;
3423 rc = MDT_IT_GETATTR;
3435 rc = MDT_IT_GETXATTR;
3438 CERROR("Unknown intent opcode: %ld\n", itcode);
3445 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3446 struct ldlm_lock **lockp, int flags)
3448 struct req_capsule *pill;
3449 struct mdt_it_flavor *flv;
3454 opc = mdt_intent_code(itopc);
3458 pill = info->mti_pill;
3459 flv = &mdt_it_flavor[opc];
3461 if (flv->it_fmt != NULL)
3462 req_capsule_extend(pill, flv->it_fmt);
3464 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3466 struct ptlrpc_request *req = mdt_info_req(info);
3467 if (flv->it_flags & MUTABOR &&
3468 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3471 if (rc == 0 && flv->it_act != NULL) {
3472 /* execute policy */
3473 rc = flv->it_act(opc, info, lockp, flags);
3480 static int mdt_intent_policy(struct ldlm_namespace *ns,
3481 struct ldlm_lock **lockp, void *req_cookie,
3482 ldlm_mode_t mode, int flags, void *data)
3484 struct mdt_thread_info *info;
3485 struct ptlrpc_request *req = req_cookie;
3486 struct ldlm_intent *it;
3487 struct req_capsule *pill;
3492 LASSERT(req != NULL);
3494 info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3496 LASSERT(info != NULL);
3497 pill = info->mti_pill;
3498 LASSERT(pill->rc_req == req);
3500 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3501 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3502 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3504 const struct ldlm_request *dlmreq;
3507 rc = mdt_intent_opc(it->opc, info, lockp, flags);
3512 * Lock without inodebits makes no sense and will oops
3513 * later in ldlm. Let's check it now to see if we have
3514 * wrong lock from client or bits get corrupted
3515 * somewhere in mdt_intent_opc().
3517 dlmreq = info->mti_dlm_req;
3518 req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
3519 LASSERT(req_bits != 0);
3522 rc = err_serious(-EFAULT);
3524 /* No intent was provided */
3525 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3526 rc = req_capsule_server_pack(pill);
3528 rc = err_serious(rc);
3536 static void mdt_seq_adjust(const struct lu_env *env,
3537 struct mdt_device *m, int lost)
3539 struct md_site *ms = mdt_md_site(m);
3540 struct lu_seq_range out;
3543 LASSERT(ms && ms->ms_server_seq);
3545 /* get extra seq from seq_server, moving it's range up */
3546 while (lost-- > 0) {
3547 seq_server_alloc_meta(ms->ms_server_seq, NULL, &out, env);
3552 static int mdt_seq_fini(const struct lu_env *env,
3553 struct mdt_device *m)
3555 struct md_site *ms = mdt_md_site(m);
3559 if (ms->ms_server_seq) {
3560 seq_server_fini(ms->ms_server_seq, env);
3561 OBD_FREE_PTR(ms->ms_server_seq);
3562 ms->ms_server_seq = NULL;
3565 if (ms->ms_control_seq) {
3566 seq_server_fini(ms->ms_control_seq, env);
3567 OBD_FREE_PTR(ms->ms_control_seq);
3568 ms->ms_control_seq = NULL;
3571 if (ms->ms_client_seq) {
3572 seq_client_fini(ms->ms_client_seq);
3573 OBD_FREE_PTR(ms->ms_client_seq);
3574 ms->ms_client_seq = NULL;
3581 static int mdt_seq_init(const struct lu_env *env,
3583 struct mdt_device *m)
3590 ms = mdt_md_site(m);
3593 * This is sequence-controller node. Init seq-controller server on local
3596 if (ms->ms_node_id == 0) {
3597 LASSERT(ms->ms_control_seq == NULL);
3599 OBD_ALLOC_PTR(ms->ms_control_seq);
3600 if (ms->ms_control_seq == NULL)
3603 rc = seq_server_init(ms->ms_control_seq,
3604 m->mdt_bottom, uuid,
3605 LUSTRE_SEQ_CONTROLLER,
3610 GOTO(out_seq_fini, rc);
3612 OBD_ALLOC_PTR(ms->ms_client_seq);
3613 if (ms->ms_client_seq == NULL)
3614 GOTO(out_seq_fini, rc = -ENOMEM);
3616 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3617 if (prefix == NULL) {
3618 OBD_FREE_PTR(ms->ms_client_seq);
3619 GOTO(out_seq_fini, rc = -ENOMEM);
3622 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3626 * Init seq-controller client after seq-controller server is
3627 * ready. Pass ms->ms_control_seq to it for direct talking.
3629 rc = seq_client_init(ms->ms_client_seq, NULL,
3630 LUSTRE_SEQ_METADATA, prefix,
3631 ms->ms_control_seq);
3632 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3635 GOTO(out_seq_fini, rc);
3638 /* Init seq-server on local MDT */
3639 LASSERT(ms->ms_server_seq == NULL);
3641 OBD_ALLOC_PTR(ms->ms_server_seq);
3642 if (ms->ms_server_seq == NULL)
3643 GOTO(out_seq_fini, rc = -ENOMEM);
3645 rc = seq_server_init(ms->ms_server_seq,
3646 m->mdt_bottom, uuid,
3651 GOTO(out_seq_fini, rc = -ENOMEM);
3653 /* Assign seq-controller client to local seq-server. */
3654 if (ms->ms_node_id == 0) {
3655 LASSERT(ms->ms_client_seq != NULL);
3657 rc = seq_server_set_cli(ms->ms_server_seq,
3665 mdt_seq_fini(env, m);
3670 * Init client sequence manager which is used by local MDS to talk to sequence
3671 * controller on remote node.
3673 static int mdt_seq_init_cli(const struct lu_env *env,
3674 struct mdt_device *m,
3675 struct lustre_cfg *cfg)
3677 struct md_site *ms = mdt_md_site(m);
3678 struct obd_device *mdc;
3679 struct obd_uuid *uuidp, *mdcuuidp;
3680 char *uuid_str, *mdc_uuid_str;
3683 struct mdt_thread_info *info;
3684 char *p, *index_string = lustre_cfg_string(cfg, 2);
3687 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3688 uuidp = &info->mti_u.uuid[0];
3689 mdcuuidp = &info->mti_u.uuid[1];
3691 LASSERT(index_string);
3693 index = simple_strtol(index_string, &p, 10);
3695 CERROR("Invalid index in lustre_cgf, offset 2\n");
3699 /* check if this is adding the first MDC and controller is not yet
3701 if (index != 0 || ms->ms_client_seq)
3704 uuid_str = lustre_cfg_string(cfg, 1);
3705 mdc_uuid_str = lustre_cfg_string(cfg, 4);
3706 obd_str2uuid(uuidp, uuid_str);
3707 obd_str2uuid(mdcuuidp, mdc_uuid_str);
3709 mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3711 CERROR("can't find controller MDC by uuid %s\n",
3714 } else if (!mdc->obd_set_up) {
3715 CERROR("target %s not set up\n", mdc->obd_name);
3718 LASSERT(ms->ms_control_exp);
3719 OBD_ALLOC_PTR(ms->ms_client_seq);
3720 if (ms->ms_client_seq != NULL) {
3723 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3727 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3730 rc = seq_client_init(ms->ms_client_seq,
3732 LUSTRE_SEQ_METADATA,
3734 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3741 LASSERT(ms->ms_server_seq != NULL);
3742 rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
3749 static void mdt_seq_fini_cli(struct mdt_device *m)
3755 ms = mdt_md_site(m);
3758 if (ms->ms_server_seq)
3759 seq_server_set_cli(ms->ms_server_seq,
3762 if (ms->ms_control_exp) {
3763 class_export_put(ms->ms_control_exp);
3764 ms->ms_control_exp = NULL;
3773 static int mdt_fld_fini(const struct lu_env *env,
3774 struct mdt_device *m)
3776 struct md_site *ms = mdt_md_site(m);
3779 if (ms && ms->ms_server_fld) {
3780 fld_server_fini(ms->ms_server_fld, env);
3781 OBD_FREE_PTR(ms->ms_server_fld);
3782 ms->ms_server_fld = NULL;
3788 static int mdt_fld_init(const struct lu_env *env,
3790 struct mdt_device *m)
3796 ms = mdt_md_site(m);
3798 OBD_ALLOC_PTR(ms->ms_server_fld);
3799 if (ms->ms_server_fld == NULL)
3800 RETURN(rc = -ENOMEM);
3802 rc = fld_server_init(ms->ms_server_fld,
3803 m->mdt_bottom, uuid,
3804 env, ms->ms_node_id);
3806 OBD_FREE_PTR(ms->ms_server_fld);
3807 ms->ms_server_fld = NULL;
3814 /* device init/fini methods */
3815 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3818 if (m->mdt_regular_service != NULL) {
3819 ptlrpc_unregister_service(m->mdt_regular_service);
3820 m->mdt_regular_service = NULL;
3822 if (m->mdt_readpage_service != NULL) {
3823 ptlrpc_unregister_service(m->mdt_readpage_service);
3824 m->mdt_readpage_service = NULL;
3826 if (m->mdt_xmds_service != NULL) {
3827 ptlrpc_unregister_service(m->mdt_xmds_service);
3828 m->mdt_xmds_service = NULL;
3830 if (m->mdt_setattr_service != NULL) {
3831 ptlrpc_unregister_service(m->mdt_setattr_service);
3832 m->mdt_setattr_service = NULL;
3834 if (m->mdt_mdsc_service != NULL) {
3835 ptlrpc_unregister_service(m->mdt_mdsc_service);
3836 m->mdt_mdsc_service = NULL;
3838 if (m->mdt_mdss_service != NULL) {
3839 ptlrpc_unregister_service(m->mdt_mdss_service);
3840 m->mdt_mdss_service = NULL;
3842 if (m->mdt_dtss_service != NULL) {
3843 ptlrpc_unregister_service(m->mdt_dtss_service);
3844 m->mdt_dtss_service = NULL;
3846 if (m->mdt_fld_service != NULL) {
3847 ptlrpc_unregister_service(m->mdt_fld_service);
3848 m->mdt_fld_service = NULL;
3853 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3856 static struct ptlrpc_service_conf conf;
3857 cfs_proc_dir_entry_t *procfs_entry;
3860 procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3862 conf = (typeof(conf)) {
3863 .psc_nbufs = MDS_NBUFS,
3864 .psc_bufsize = MDS_BUFSIZE,
3865 .psc_max_req_size = MDS_MAXREQSIZE,
3866 .psc_max_reply_size = MDS_MAXREPSIZE,
3867 .psc_req_portal = MDS_REQUEST_PORTAL,
3868 .psc_rep_portal = MDC_REPLY_PORTAL,
3869 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3871 * We'd like to have a mechanism to set this on a per-device
3872 * basis, but alas...
3874 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3876 .psc_max_threads = MDT_MAX_THREADS,
3877 .psc_ctx_tags = LCT_MD_THREAD
3880 m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3881 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3882 "mdt_ldlm_client", m->mdt_ldlm_client);
3884 m->mdt_regular_service =
3885 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3886 procfs_entry, target_print_req,
3888 if (m->mdt_regular_service == NULL)
3891 rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3893 GOTO(err_mdt_svc, rc);
3896 * readpage service configuration. Parameters have to be adjusted,
3899 conf = (typeof(conf)) {
3900 .psc_nbufs = MDS_NBUFS,
3901 .psc_bufsize = MDS_BUFSIZE,
3902 .psc_max_req_size = MDS_MAXREQSIZE,
3903 .psc_max_reply_size = MDS_MAXREPSIZE,
3904 .psc_req_portal = MDS_READPAGE_PORTAL,
3905 .psc_rep_portal = MDC_REPLY_PORTAL,
3906 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3907 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3909 .psc_max_threads = MDT_MAX_THREADS,
3910 .psc_ctx_tags = LCT_MD_THREAD
3912 m->mdt_readpage_service =
3913 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3914 LUSTRE_MDT_NAME "_readpage",
3915 procfs_entry, target_print_req,"mdt_rdpg");
3917 if (m->mdt_readpage_service == NULL) {
3918 CERROR("failed to start readpage service\n");
3919 GOTO(err_mdt_svc, rc = -ENOMEM);
3922 rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3925 * setattr service configuration.
3927 conf = (typeof(conf)) {
3928 .psc_nbufs = MDS_NBUFS,
3929 .psc_bufsize = MDS_BUFSIZE,
3930 .psc_max_req_size = MDS_MAXREQSIZE,
3931 .psc_max_reply_size = MDS_MAXREPSIZE,
3932 .psc_req_portal = MDS_SETATTR_PORTAL,
3933 .psc_rep_portal = MDC_REPLY_PORTAL,
3934 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3935 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3937 .psc_max_threads = MDT_MAX_THREADS,
3938 .psc_ctx_tags = LCT_MD_THREAD
3941 m->mdt_setattr_service =
3942 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3943 LUSTRE_MDT_NAME "_setattr",
3944 procfs_entry, target_print_req,"mdt_attr");
3946 if (!m->mdt_setattr_service) {
3947 CERROR("failed to start setattr service\n");
3948 GOTO(err_mdt_svc, rc = -ENOMEM);
3951 rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3953 GOTO(err_mdt_svc, rc);
3956 * sequence controller service configuration
3958 conf = (typeof(conf)) {
3959 .psc_nbufs = MDS_NBUFS,
3960 .psc_bufsize = MDS_BUFSIZE,
3961 .psc_max_req_size = SEQ_MAXREQSIZE,
3962 .psc_max_reply_size = SEQ_MAXREPSIZE,
3963 .psc_req_portal = SEQ_CONTROLLER_PORTAL,
3964 .psc_rep_portal = MDC_REPLY_PORTAL,
3965 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3966 .psc_min_threads = SEQ_NUM_THREADS,
3967 .psc_max_threads = SEQ_NUM_THREADS,
3968 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3971 m->mdt_mdsc_service =
3972 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3973 LUSTRE_MDT_NAME"_mdsc",
3974 procfs_entry, target_print_req,"mdt_mdsc");
3975 if (!m->mdt_mdsc_service) {
3976 CERROR("failed to start seq controller service\n");
3977 GOTO(err_mdt_svc, rc = -ENOMEM);
3980 rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3982 GOTO(err_mdt_svc, rc);
3985 * metadata sequence server service configuration
3987 conf = (typeof(conf)) {
3988 .psc_nbufs = MDS_NBUFS,
3989 .psc_bufsize = MDS_BUFSIZE,
3990 .psc_max_req_size = SEQ_MAXREQSIZE,
3991 .psc_max_reply_size = SEQ_MAXREPSIZE,
3992 .psc_req_portal = SEQ_METADATA_PORTAL,
3993 .psc_rep_portal = MDC_REPLY_PORTAL,
3994 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3995 .psc_min_threads = SEQ_NUM_THREADS,
3996 .psc_max_threads = SEQ_NUM_THREADS,
3997 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
4000 m->mdt_mdss_service =
4001 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
4002 LUSTRE_MDT_NAME"_mdss",
4003 procfs_entry, target_print_req,"mdt_mdss");
4004 if (!m->mdt_mdss_service) {
4005 CERROR("failed to start metadata seq server service\n");
4006 GOTO(err_mdt_svc, rc = -ENOMEM);
4009 rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
4011 GOTO(err_mdt_svc, rc);
4015 * Data sequence server service configuration. We want to have really
4016 * cluster-wide sequences space. This is why we start only one sequence
4017 * controller which manages space.
4019 conf = (typeof(conf)) {
4020 .psc_nbufs = MDS_NBUFS,
4021 .psc_bufsize = MDS_BUFSIZE,
4022 .psc_max_req_size = SEQ_MAXREQSIZE,
4023 .psc_max_reply_size = SEQ_MAXREPSIZE,
4024 .psc_req_portal = SEQ_DATA_PORTAL,
4025 .psc_rep_portal = OSC_REPLY_PORTAL,
4026 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4027 .psc_min_threads = SEQ_NUM_THREADS,
4028 .psc_max_threads = SEQ_NUM_THREADS,
4029 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
4032 m->mdt_dtss_service =
4033 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
4034 LUSTRE_MDT_NAME"_dtss",
4035 procfs_entry, target_print_req,"mdt_dtss");
4036 if (!m->mdt_dtss_service) {
4037 CERROR("failed to start data seq server service\n");
4038 GOTO(err_mdt_svc, rc = -ENOMEM);
4041 rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
4043 GOTO(err_mdt_svc, rc);
4045 /* FLD service start */
4046 conf = (typeof(conf)) {
4047 .psc_nbufs = MDS_NBUFS,
4048 .psc_bufsize = MDS_BUFSIZE,
4049 .psc_max_req_size = FLD_MAXREQSIZE,
4050 .psc_max_reply_size = FLD_MAXREPSIZE,
4051 .psc_req_portal = FLD_REQUEST_PORTAL,
4052 .psc_rep_portal = MDC_REPLY_PORTAL,
4053 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4054 .psc_min_threads = FLD_NUM_THREADS,
4055 .psc_max_threads = FLD_NUM_THREADS,
4056 .psc_ctx_tags = LCT_DT_THREAD|LCT_MD_THREAD
4059 m->mdt_fld_service =
4060 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
4061 LUSTRE_MDT_NAME"_fld",
4062 procfs_entry, target_print_req, "mdt_fld");
4063 if (!m->mdt_fld_service) {
4064 CERROR("failed to start fld service\n");
4065 GOTO(err_mdt_svc, rc = -ENOMEM);
4068 rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
4070 GOTO(err_mdt_svc, rc);
4073 * mds-mds service configuration. Separate portal is used to allow
4074 * mds-mds requests be not blocked during recovery.
4076 conf = (typeof(conf)) {
4077 .psc_nbufs = MDS_NBUFS,
4078 .psc_bufsize = MDS_BUFSIZE,
4079 .psc_max_req_size = MDS_MAXREQSIZE,
4080 .psc_max_reply_size = MDS_MAXREPSIZE,
4081 .psc_req_portal = MDS_MDS_PORTAL,
4082 .psc_rep_portal = MDC_REPLY_PORTAL,
4083 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4084 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
4086 .psc_max_threads = MDT_MAX_THREADS,
4087 .psc_ctx_tags = LCT_MD_THREAD
4089 m->mdt_xmds_service =
4090 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
4091 LUSTRE_MDT_NAME "_mds",
4092 procfs_entry, target_print_req,"mdt_xmds");
4094 if (m->mdt_xmds_service == NULL) {
4095 CERROR("failed to start xmds service\n");
4096 GOTO(err_mdt_svc, rc = -ENOMEM);
4099 rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
4101 GOTO(err_mdt_svc, rc);
4106 mdt_stop_ptlrpc_service(m);
4111 static void mdt_stack_fini(const struct lu_env *env,
4112 struct mdt_device *m, struct lu_device *top)
4114 struct obd_device *obd = mdt2obd_dev(m);
4115 struct lustre_cfg_bufs *bufs;
4116 struct lustre_cfg *lcfg;
4117 struct mdt_thread_info *info;
4121 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4122 LASSERT(info != NULL);
4124 bufs = &info->mti_u.bufs;
4125 /* process cleanup, pass mdt obd name to get obd umount flags */
4126 lustre_cfg_bufs_reset(bufs, obd->obd_name);
4131 lustre_cfg_bufs_set_string(bufs, 1, flags);
4132 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4134 CERROR("Cannot alloc lcfg!\n");
4139 top->ld_ops->ldo_process_config(env, top, lcfg);
4140 lustre_cfg_free(lcfg);
4142 lu_stack_fini(env, top);
4143 m->mdt_child = NULL;
4144 m->mdt_bottom = NULL;
4147 static struct lu_device *mdt_layer_setup(struct lu_env *env,
4148 const char *typename,
4149 struct lu_device *child,
4150 struct lustre_cfg *cfg)
4152 const char *dev = lustre_cfg_string(cfg, 0);
4153 struct obd_type *type;
4154 struct lu_device_type *ldt;
4155 struct lu_device *d;
4160 type = class_get_type(typename);
4162 CERROR("Unknown type: '%s'\n", typename);
4163 GOTO(out, rc = -ENODEV);
4166 rc = lu_env_refill((struct lu_env *)env);
4168 CERROR("Failure to refill session: '%d'\n", rc);
4174 CERROR("type: '%s'\n", typename);
4175 GOTO(out_type, rc = -EINVAL);
4178 ldt->ldt_obd_type = type;
4179 d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
4181 CERROR("Cannot allocate device: '%s'\n", typename);
4182 GOTO(out_type, rc = -ENODEV);
4185 LASSERT(child->ld_site);
4186 d->ld_site = child->ld_site;
4189 rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
4191 CERROR("can't init device '%s', rc %d\n", typename, rc);
4192 GOTO(out_alloc, rc);
4195 lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
4200 ldt->ldt_ops->ldto_device_free(env, d);
4203 class_put_type(type);
4208 static int mdt_stack_init(struct lu_env *env,
4209 struct mdt_device *m,
4210 struct lustre_cfg *cfg,
4211 struct lustre_mount_info *lmi)
4213 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
4214 struct lu_device *tmp;
4215 struct md_device *md;
4216 struct lu_device *child_lu_dev;
4220 /* init the stack */
4221 tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
4223 RETURN(PTR_ERR(tmp));
4225 m->mdt_bottom = lu2dt_dev(tmp);
4227 tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
4229 GOTO(out, rc = PTR_ERR(tmp));
4234 tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
4236 GOTO(out, rc = PTR_ERR(tmp));
4239 /*set mdd upcall device*/
4240 md_upcall_dev_set(md, lu2md_dev(d));
4243 /*set cmm upcall device*/
4244 md_upcall_dev_set(md, &m->mdt_md_dev);
4246 m->mdt_child = lu2md_dev(d);
4248 /* process setup config */
4249 tmp = &m->mdt_md_dev.md_lu_dev;
4250 rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
4254 /* initialize local objects */
4255 child_lu_dev = &m->mdt_child->md_lu_dev;
4257 rc = child_lu_dev->ld_ops->ldo_prepare(env,
4258 &m->mdt_md_dev.md_lu_dev,
4261 /* fini from last known good lu_device */
4263 mdt_stack_fini(env, m, d);
4269 * setup CONFIG_ORIG context, used to access local config log.
4270 * this may need to be rewrite as part of llog rewrite for lu-api.
4272 static int mdt_obd_llog_setup(struct obd_device *obd,
4273 struct lustre_sb_info *lsi)
4277 LASSERT(obd->obd_fsops == NULL);
4279 obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
4280 if (IS_ERR(obd->obd_fsops))
4281 return PTR_ERR(obd->obd_fsops);
4283 rc = fsfilt_setup(obd, lsi->lsi_srv_mnt->mnt_sb);
4285 fsfilt_put_ops(obd->obd_fsops);
4289 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
4290 obd->obd_lvfs_ctxt.pwdmnt = lsi->lsi_srv_mnt;
4291 obd->obd_lvfs_ctxt.pwd = lsi->lsi_srv_mnt->mnt_root;
4292 obd->obd_lvfs_ctxt.fs = get_ds();
4294 rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd,
4295 0, NULL, &llog_lvfs_ops);
4297 CERROR("llog_setup() failed: %d\n", rc);
4298 fsfilt_put_ops(obd->obd_fsops);
4304 static void mdt_obd_llog_cleanup(struct obd_device *obd)
4306 struct llog_ctxt *ctxt;
4308 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
4312 if (obd->obd_fsops) {
4313 fsfilt_put_ops(obd->obd_fsops);
4314 obd->obd_fsops = NULL;
4318 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4320 struct md_device *next = m->mdt_child;
4321 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
4322 struct lu_site *ls = d->ld_site;
4323 struct obd_device *obd = mdt2obd_dev(m);
4326 target_recovery_fini(obd);
4328 ping_evictor_stop();
4330 mdt_stop_ptlrpc_service(m);
4331 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4332 mdt_obd_llog_cleanup(obd);
4333 obd_exports_barrier(obd);
4334 obd_zombie_barrier();
4335 #ifdef HAVE_QUOTA_SUPPORT
4336 next->md_ops->mdo_quota.mqo_cleanup(env, next);
4338 lut_fini(env, &m->mdt_lut);
4339 mdt_fs_cleanup(env, m);
4340 upcall_cache_cleanup(m->mdt_identity_cache);
4341 m->mdt_identity_cache = NULL;
4343 if (m->mdt_namespace != NULL) {
4344 ldlm_namespace_free(m->mdt_namespace, NULL,
4345 d->ld_obd->obd_force);
4346 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4349 cfs_free_nidlist(&m->mdt_nosquash_nids);
4350 if (m->mdt_nosquash_str) {
4351 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4352 m->mdt_nosquash_str = NULL;
4353 m->mdt_nosquash_strlen = 0;
4356 mdt_seq_fini(env, m);
4357 mdt_seq_fini_cli(m);
4358 mdt_fld_fini(env, m);
4359 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4361 next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4362 cfs_timer_disarm(&m->mdt_ck_timer);
4363 mdt_ck_thread_stop(m);
4368 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4371 if (obd->obd_proc_exports_entry) {
4372 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
4373 obd->obd_proc_exports_entry = NULL;
4375 lprocfs_free_per_client_stats(obd);
4376 lprocfs_free_obd_stats(obd);
4377 ptlrpc_lprocfs_unregister_obd(obd);
4378 lprocfs_obd_cleanup(obd);
4381 struct md_site *mite;
4384 mite = lu_site2md(ls);
4388 LASSERT(atomic_read(&d->ld_ref) == 0);
4393 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4395 struct mdt_device *m = mdt_dev(obd->obd_lu_dev);
4396 struct sptlrpc_rule_set tmp_rset;
4399 sptlrpc_rule_set_init(&tmp_rset);
4400 rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4402 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4407 sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4409 write_lock(&m->mdt_sptlrpc_lock);
4410 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4411 m->mdt_sptlrpc_rset = tmp_rset;
4412 write_unlock(&m->mdt_sptlrpc_lock);
4417 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
4421 m->mdt_opts.mo_mds_capa = 1;
4422 m->mdt_opts.mo_oss_capa = 1;
4423 #ifdef CONFIG_FS_POSIX_ACL
4424 /* ACLs should be enabled by default (b=13829) */
4425 m->mdt_opts.mo_acl = 1;
4426 LCONSOLE_INFO("Enabling ACL\n");
4428 m->mdt_opts.mo_acl = 0;
4429 LCONSOLE_INFO("Disabling ACL\n");
4438 while (*p && *p != ',')
4442 if ((len == sizeof("user_xattr") - 1) &&
4443 (memcmp(options, "user_xattr", len) == 0)) {
4444 m->mdt_opts.mo_user_xattr = 1;
4445 LCONSOLE_INFO("Enabling user_xattr\n");
4446 } else if ((len == sizeof("nouser_xattr") - 1) &&
4447 (memcmp(options, "nouser_xattr", len) == 0)) {
4448 m->mdt_opts.mo_user_xattr = 0;
4449 LCONSOLE_INFO("Disabling user_xattr\n");
4450 } else if ((len == sizeof("noacl") - 1) &&
4451 (memcmp(options, "noacl", len) == 0)) {
4452 m->mdt_opts.mo_acl = 0;
4453 LCONSOLE_INFO("Disabling ACL\n");
4460 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4462 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4463 struct lu_device_type *ldt, struct lustre_cfg *cfg)
4465 struct lprocfs_static_vars lvars;
4466 struct mdt_thread_info *info;
4467 struct obd_device *obd;
4468 const char *dev = lustre_cfg_string(cfg, 0);
4469 const char *num = lustre_cfg_string(cfg, 2);
4470 struct lustre_mount_info *lmi = NULL;
4471 struct lustre_sb_info *lsi;
4472 struct lustre_disk_data *ldd;
4474 struct md_site *mite;
4475 const char *identity_upcall = "NONE";
4476 #ifdef HAVE_QUOTA_SUPPORT
4477 struct md_device *next;
4483 md_device_init(&m->mdt_md_dev, ldt);
4485 * Environment (env) might be missing mdt_thread_key values at that
4486 * point, if device is allocated when mdt_thread_key is in QUIESCENT
4489 * Usually device allocation path doesn't use module key values, but
4490 * mdt has to do a lot of work here, so allocate key value.
4492 rc = lu_env_refill((struct lu_env *)env);
4496 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4497 LASSERT(info != NULL);
4499 obd = class_name2obd(dev);
4500 LASSERT(obd != NULL);
4502 spin_lock_init(&m->mdt_transno_lock);
4504 m->mdt_max_mdsize = MAX_MD_SIZE;
4505 m->mdt_max_cookiesize = sizeof(struct llog_cookie);
4506 m->mdt_som_conf = 0;
4508 m->mdt_opts.mo_user_xattr = 0;
4509 m->mdt_opts.mo_acl = 0;
4510 m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4511 lmi = server_get_mount_2(dev);
4513 CERROR("Cannot get mount info for %s!\n", dev);
4516 lsi = s2lsi(lmi->lmi_sb);
4517 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
4518 /* CMD is supported only in IAM mode */
4521 node_id = simple_strtol(num, NULL, 10);
4522 if (!(ldd->ldd_flags & LDD_F_IAM_DIR) && node_id) {
4523 CERROR("CMD Operation not allowed in IOP mode\n");
4524 GOTO(err_lmi, rc = -EINVAL);
4528 rwlock_init(&m->mdt_sptlrpc_lock);
4529 sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4531 spin_lock_init(&m->mdt_ioepoch_lock);
4532 m->mdt_opts.mo_compat_resname = 0;
4533 m->mdt_capa_timeout = CAPA_TIMEOUT;
4534 m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4535 m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4536 m->mdt_squash_uid = 0;
4537 m->mdt_squash_gid = 0;
4538 CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4539 m->mdt_nosquash_str = NULL;
4540 m->mdt_nosquash_strlen = 0;
4541 init_rwsem(&m->mdt_squash_sem);
4543 spin_lock_init(&m->mdt_client_bitmap_lock);
4545 OBD_ALLOC_PTR(mite);
4547 GOTO(err_lmi, rc = -ENOMEM);
4551 m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4552 m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4553 /* set this lu_device to obd, because error handling need it */
4554 obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4556 rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
4558 CERROR("Can't init lu_site, rc %d\n", rc);
4559 GOTO(err_free_site, rc);
4562 lprocfs_mdt_init_vars(&lvars);
4563 rc = lprocfs_obd_setup(obd, lvars.obd_vars);
4565 CERROR("Can't init lprocfs, rc %d\n", rc);
4566 GOTO(err_fini_site, rc);
4568 ptlrpc_lprocfs_register_obd(obd);
4570 rc = mdt_procfs_init(m, dev);
4572 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4573 GOTO(err_fini_proc, rc);
4576 obd->obd_proc_exports_entry = proc_mkdir("exports",
4577 obd->obd_proc_entry);
4578 if (obd->obd_proc_exports_entry)
4579 lprocfs_add_simple(obd->obd_proc_exports_entry,
4580 "clear", lprocfs_nid_stats_clear_read,
4581 lprocfs_nid_stats_clear_write, obd, NULL);
4583 /* set server index */
4584 lu_site2md(s)->ms_node_id = node_id;
4586 /* failover is the default
4587 * FIXME: we do not failout mds0/mgs, which may cause some problems.
4588 * assumed whose ms_node_id == 0 XXX
4590 obd->obd_replayable = 1;
4591 /* No connection accepted until configurations will finish */
4592 obd->obd_no_conn = 1;
4594 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4595 char *str = lustre_cfg_string(cfg, 4);
4596 if (strchr(str, 'n')) {
4597 CWARN("%s: recovery disabled\n", obd->obd_name);
4598 obd->obd_replayable = 0;
4602 /* init the stack */
4603 rc = mdt_stack_init((struct lu_env *)env, m, cfg, lmi);
4605 CERROR("Can't init device stack, rc %d\n", rc);
4606 GOTO(err_fini_proc, rc);
4609 rc = lut_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4611 GOTO(err_fini_stack, rc);
4613 rc = mdt_fld_init(env, obd->obd_name, m);
4617 rc = mdt_seq_init(env, obd->obd_name, m);
4619 GOTO(err_fini_fld, rc);
4621 snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4622 LUSTRE_MDT_NAME"-%p", m);
4623 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4624 LDLM_NAMESPACE_SERVER,
4625 LDLM_NAMESPACE_GREEDY);
4626 if (m->mdt_namespace == NULL)
4627 GOTO(err_fini_seq, rc = -ENOMEM);
4629 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4630 /* set obd_namespace for compatibility with old code */
4631 obd->obd_namespace = m->mdt_namespace;
4633 /* XXX: to support suppgid for ACL, we enable identity_upcall
4634 * by default, otherwise, maybe got unexpected -EACCESS. */
4635 if (m->mdt_opts.mo_acl)
4636 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4638 m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
4639 &mdt_identity_upcall_cache_ops);
4640 if (IS_ERR(m->mdt_identity_cache)) {
4641 rc = PTR_ERR(m->mdt_identity_cache);
4642 m->mdt_identity_cache = NULL;
4643 GOTO(err_free_ns, rc);
4646 cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4648 rc = mdt_ck_thread_start(m);
4650 GOTO(err_free_ns, rc);
4652 rc = mdt_fs_setup(env, m, obd, lsi);
4656 rc = mdt_obd_llog_setup(obd, lsi);
4658 GOTO(err_fs_cleanup, rc);
4660 rc = mdt_llog_ctxt_clone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4662 GOTO(err_llog_cleanup, rc);
4664 mdt_adapt_sptlrpc_conf(obd, 1);
4666 #ifdef HAVE_QUOTA_SUPPORT
4667 next = m->mdt_child;
4668 rc = next->md_ops->mdo_quota.mqo_setup(env, next, lmi->lmi_mnt);
4670 GOTO(err_llog_cleanup, rc);
4673 server_put_mount_2(dev, lmi->lmi_mnt);
4676 target_recovery_init(&m->mdt_lut, mdt_recovery_handle);
4678 rc = mdt_start_ptlrpc_service(m);
4680 GOTO(err_recovery, rc);
4682 ping_evictor_start();
4684 rc = lu_site_init_finish(s);
4686 GOTO(err_stop_service, rc);
4688 if (obd->obd_recovering == 0)
4689 mdt_postrecov(env, m);
4691 mdt_init_capa_ctxt(env, m);
4693 /* Reduce the initial timeout on an MDS because it doesn't need such
4694 * a long timeout as an OST does. Adaptive timeouts will adjust this
4695 * value appropriately. */
4696 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4697 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4702 ping_evictor_stop();
4703 mdt_stop_ptlrpc_service(m);
4705 target_recovery_fini(obd);
4706 #ifdef HAVE_QUOTA_SUPPORT
4707 next->md_ops->mdo_quota.mqo_cleanup(env, next);
4710 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4711 mdt_obd_llog_cleanup(obd);
4713 mdt_fs_cleanup(env, m);
4715 cfs_timer_disarm(&m->mdt_ck_timer);
4716 mdt_ck_thread_stop(m);
4718 upcall_cache_cleanup(m->mdt_identity_cache);
4719 m->mdt_identity_cache = NULL;
4720 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4721 obd->obd_namespace = m->mdt_namespace = NULL;
4723 mdt_seq_fini(env, m);
4725 mdt_fld_fini(env, m);
4727 lut_fini(env, &m->mdt_lut);
4729 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4732 if (obd->obd_proc_exports_entry) {
4733 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
4734 obd->obd_proc_exports_entry = NULL;
4736 ptlrpc_lprocfs_unregister_obd(obd);
4737 lprocfs_obd_cleanup(obd);
4744 server_put_mount_2(dev, lmi->lmi_mnt);
4748 /* used by MGS to process specific configurations */
4749 static int mdt_process_config(const struct lu_env *env,
4750 struct lu_device *d, struct lustre_cfg *cfg)
4752 struct mdt_device *m = mdt_dev(d);
4753 struct md_device *md_next = m->mdt_child;
4754 struct lu_device *next = md2lu_dev(md_next);
4758 switch (cfg->lcfg_command) {
4760 struct lprocfs_static_vars lvars;
4761 struct obd_device *obd = d->ld_obd;
4764 * For interoperability between 1.8 and 2.0,
4765 * skip old "mdt.group_upcall" param.
4768 char *param = lustre_cfg_string(cfg, 1);
4769 if (param && !strncmp("mdt.group_upcall", param, 16)) {
4770 CWARN("For 1.8 interoperability, skip this"
4771 " mdt.group_upcall. It is obsolete\n");
4776 lprocfs_mdt_init_vars(&lvars);
4777 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4779 if (rc > 0 || rc == -ENOSYS)
4780 /* we don't understand; pass it on */
4781 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4786 * Add mdc hook to get first MDT uuid and connect it to
4787 * ls->controller to use for seq manager.
4789 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4791 CERROR("Can't add mdc, rc %d\n", rc);
4793 rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4796 /* others are passed further */
4797 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4803 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4804 const struct lu_object_header *hdr,
4805 struct lu_device *d)
4807 struct mdt_object *mo;
4813 struct lu_object *o;
4814 struct lu_object_header *h;
4816 o = &mo->mot_obj.mo_lu;
4817 h = &mo->mot_header;
4818 lu_object_header_init(h);
4819 lu_object_init(o, h, d);
4820 lu_object_add_top(h, o);
4821 o->lo_ops = &mdt_obj_ops;
4827 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4828 const struct lu_object_conf *unused)
4830 struct mdt_device *d = mdt_dev(o->lo_dev);
4831 struct lu_device *under;
4832 struct lu_object *below;
4836 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4837 PFID(lu_object_fid(o)));
4839 under = &d->mdt_child->md_lu_dev;
4840 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4841 if (below != NULL) {
4842 lu_object_add(o, below);
4849 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4851 struct mdt_object *mo = mdt_obj(o);
4852 struct lu_object_header *h;
4856 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4857 PFID(lu_object_fid(o)));
4860 lu_object_header_fini(h);
4865 static int mdt_object_print(const struct lu_env *env, void *cookie,
4866 lu_printer_t p, const struct lu_object *o)
4868 struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4869 return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch=%llu "
4870 "flags=%llx, epochcount=%d, writecount=%d)",
4871 mdto, mdto->mot_ioepoch, mdto->mot_flags,
4872 mdto->mot_epochcount, mdto->mot_writecount);
4875 static const struct lu_device_operations mdt_lu_ops = {
4876 .ldo_object_alloc = mdt_object_alloc,
4877 .ldo_process_config = mdt_process_config,
4880 static const struct lu_object_operations mdt_obj_ops = {
4881 .loo_object_init = mdt_object_init,
4882 .loo_object_free = mdt_object_free,
4883 .loo_object_print = mdt_object_print
4886 static int mdt_obd_set_info_async(struct obd_export *exp,
4887 __u32 keylen, void *key,
4888 __u32 vallen, void *val,
4889 struct ptlrpc_request_set *set)
4891 struct obd_device *obd = exp->exp_obd;
4897 if (KEY_IS(KEY_SPTLRPC_CONF)) {
4898 rc = mdt_adapt_sptlrpc_conf(obd, 0);
4905 /* mds_connect_internal */
4906 static int mdt_connect_internal(struct obd_export *exp,
4907 struct mdt_device *mdt,
4908 struct obd_connect_data *data)
4911 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4912 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4914 /* If no known bits (which should not happen, probably,
4915 as everybody should support LOOKUP and UPDATE bits at least)
4916 revert to compat mode with plain locks. */
4917 if (!data->ocd_ibits_known &&
4918 data->ocd_connect_flags & OBD_CONNECT_IBITS)
4919 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4921 if (!mdt->mdt_opts.mo_acl)
4922 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4924 if (!mdt->mdt_opts.mo_user_xattr)
4925 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4927 if (!mdt->mdt_som_conf)
4928 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4930 spin_lock(&exp->exp_lock);
4931 exp->exp_connect_flags = data->ocd_connect_flags;
4932 spin_unlock(&exp->exp_lock);
4933 data->ocd_version = LUSTRE_VERSION_CODE;
4934 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4938 if (mdt->mdt_opts.mo_acl &&
4939 ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4940 CWARN("%s: MDS requires ACL support but client does not\n",
4941 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4946 if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4947 CWARN("%s: MDS requires FID support, but client not\n",
4948 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4952 if (mdt->mdt_som_conf &&
4953 !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS) &&
4954 !(exp->exp_connect_flags & OBD_CONNECT_SOM)) {
4955 CWARN("%s: MDS has SOM enabled, but client does not support "
4956 "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4963 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
4964 struct obd_export *exp,
4965 struct ptlrpc_request *req)
4967 struct sptlrpc_flavor flvr;
4970 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4971 read_lock(&mdt->mdt_sptlrpc_lock);
4972 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
4976 read_unlock(&mdt->mdt_sptlrpc_lock);
4978 spin_lock(&exp->exp_lock);
4980 exp->exp_sp_peer = req->rq_sp_from;
4981 exp->exp_flvr = flvr;
4983 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
4984 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4985 CERROR("unauthorized rpc flavor %x from %s, "
4986 "expect %x\n", req->rq_flvr.sf_rpc,
4987 libcfs_nid2str(req->rq_peer.nid),
4988 exp->exp_flvr.sf_rpc);
4992 spin_unlock(&exp->exp_lock);
4994 if (exp->exp_sp_peer != req->rq_sp_from) {
4995 CERROR("RPC source %s doesn't match %s\n",
4996 sptlrpc_part2name(req->rq_sp_from),
4997 sptlrpc_part2name(exp->exp_sp_peer));
5000 rc = sptlrpc_target_export_check(exp, req);
5007 /* mds_connect copy */
5008 static int mdt_obd_connect(const struct lu_env *env,
5009 struct obd_export **exp, struct obd_device *obd,
5010 struct obd_uuid *cluuid,
5011 struct obd_connect_data *data,
5014 struct mdt_thread_info *info;
5015 struct lsd_client_data *lcd;
5016 struct obd_export *lexp;
5017 struct lustre_handle conn = { 0 };
5018 struct mdt_device *mdt;
5019 struct ptlrpc_request *req;
5023 LASSERT(env != NULL);
5024 if (!exp || !obd || !cluuid)
5027 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5028 req = info->mti_pill->rc_req;
5029 mdt = mdt_dev(obd->obd_lu_dev);
5031 rc = class_connect(&conn, obd, cluuid);
5035 lexp = class_conn2export(&conn);
5036 LASSERT(lexp != NULL);
5038 rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
5042 rc = mdt_connect_internal(lexp, mdt, data);
5046 struct mdt_thread_info *mti;
5047 mti = lu_context_key_get(&env->le_ctx,
5049 LASSERT(mti != NULL);
5050 mti->mti_exp = lexp;
5051 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5052 lexp->exp_mdt_data.med_lcd = lcd;
5053 rc = mdt_client_new(env, mdt);
5056 lexp->exp_mdt_data.med_lcd = NULL;
5058 mdt_export_stats_init(obd, lexp, localdata);
5066 class_disconnect(lexp);
5073 static int mdt_obd_reconnect(const struct lu_env *env,
5074 struct obd_export *exp, struct obd_device *obd,
5075 struct obd_uuid *cluuid,
5076 struct obd_connect_data *data,
5079 struct mdt_thread_info *info;
5080 struct mdt_device *mdt;
5081 struct ptlrpc_request *req;
5085 if (exp == NULL || obd == NULL || cluuid == NULL)
5088 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5089 req = info->mti_pill->rc_req;
5090 mdt = mdt_dev(obd->obd_lu_dev);
5092 rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5096 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5098 mdt_export_stats_init(obd, exp, localdata);
5102 static int mdt_export_cleanup(struct obd_export *exp)
5104 struct mdt_export_data *med = &exp->exp_mdt_data;
5105 struct obd_device *obd = exp->exp_obd;
5106 struct mdt_device *mdt;
5107 struct mdt_thread_info *info;
5109 CFS_LIST_HEAD(closing_list);
5110 struct mdt_file_data *mfd, *n;
5114 spin_lock(&med->med_open_lock);
5115 while (!list_empty(&med->med_open_head)) {
5116 struct list_head *tmp = med->med_open_head.next;
5117 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5119 /* Remove mfd handle so it can't be found again.
5120 * We are consuming the mfd_list reference here. */
5121 class_handle_unhash(&mfd->mfd_handle);
5122 list_move_tail(&mfd->mfd_list, &closing_list);
5124 spin_unlock(&med->med_open_lock);
5125 mdt = mdt_dev(obd->obd_lu_dev);
5126 LASSERT(mdt != NULL);
5128 rc = lu_env_init(&env, LCT_MD_THREAD);
5132 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5133 LASSERT(info != NULL);
5134 memset(info, 0, sizeof *info);
5135 info->mti_env = &env;
5136 info->mti_mdt = mdt;
5137 info->mti_exp = exp;
5139 if (!list_empty(&closing_list)) {
5140 struct md_attr *ma = &info->mti_attr;
5144 lmm_size = mdt->mdt_max_mdsize;
5145 OBD_ALLOC(ma->ma_lmm, lmm_size);
5146 if (ma->ma_lmm == NULL)
5147 GOTO(out_lmm, rc = -ENOMEM);
5149 cookie_size = mdt->mdt_max_cookiesize;
5150 OBD_ALLOC(ma->ma_cookie, cookie_size);
5151 if (ma->ma_cookie == NULL)
5152 GOTO(out_cookie, rc = -ENOMEM);
5154 /* Close any open files (which may also cause orphan unlinking). */
5155 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5156 list_del_init(&mfd->mfd_list);
5157 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
5158 ma->ma_lmm_size = lmm_size;
5159 ma->ma_cookie_size = cookie_size;
5161 /* It is not for setattr, just tell MDD to send
5162 * DESTROY RPC to OSS if needed */
5163 ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
5164 ma->ma_valid = MA_FLAGS;
5165 mdt_mfd_close(info, mfd);
5167 OBD_FREE(ma->ma_cookie, cookie_size);
5168 ma->ma_cookie = NULL;
5170 OBD_FREE(ma->ma_lmm, lmm_size);
5174 info->mti_mdt = NULL;
5175 /* cleanup client slot early */
5176 mdt_client_del(&env, mdt);
5182 static int mdt_obd_disconnect(struct obd_export *exp)
5188 class_export_get(exp);
5190 rc = server_disconnect_export(exp);
5192 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5194 rc = mdt_export_cleanup(exp);
5195 class_export_put(exp);
5199 /* FIXME: Can we avoid using these two interfaces? */
5200 static int mdt_init_export(struct obd_export *exp)
5202 struct mdt_export_data *med = &exp->exp_mdt_data;
5206 CFS_INIT_LIST_HEAD(&med->med_open_head);
5207 spin_lock_init(&med->med_open_lock);
5208 sema_init(&med->med_idmap_sem, 1);
5209 med->med_idmap = NULL;
5210 spin_lock(&exp->exp_lock);
5211 exp->exp_connecting = 1;
5212 spin_unlock(&exp->exp_lock);
5213 rc = ldlm_init_export(exp);
5215 CERROR("Error %d while initializing export\n", rc);
5219 static int mdt_destroy_export(struct obd_export *exp)
5221 struct mdt_export_data *med;
5225 med = &exp->exp_mdt_data;
5226 if (exp_connect_rmtclient(exp))
5227 mdt_cleanup_idmap(&exp->exp_mdt_data);
5229 target_destroy_export(exp);
5230 ldlm_destroy_export(exp);
5232 LASSERT(list_empty(&exp->exp_outstanding_replies));
5233 LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5234 if (obd_uuid_equals(&exp->exp_client_uuid, &exp->exp_obd->obd_uuid))
5240 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
5242 if (flag & CONFIG_LOG)
5243 m->mdt_fl_cfglog = 1;
5245 /* also notify active event */
5246 if (flag & CONFIG_SYNC)
5247 m->mdt_fl_synced = 1;
5249 if (m->mdt_fl_cfglog && m->mdt_fl_synced)
5250 /* Open for clients */
5251 m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0;
5254 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
5255 enum md_upcall_event ev, void *data)
5257 struct mdt_device *m = mdt_dev(&md->md_lu_dev);
5258 struct md_device *next = m->mdt_child;
5259 struct mdt_thread_info *mti;
5265 rc = next->md_ops->mdo_maxsize_get(env, next,
5267 &m->mdt_max_cookiesize);
5268 CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
5269 m->mdt_max_mdsize, m->mdt_max_cookiesize);
5270 mdt_allow_cli(m, CONFIG_SYNC);
5272 (*(__u64 *)data) = m->mdt_mount_count;
5275 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5276 mti->mti_no_need_trans = 1;
5277 CDEBUG(D_INFO, "disable mdt trans for this thread\n");
5280 /* Check that MDT is not yet configured */
5281 LASSERT(!m->mdt_fl_cfglog);
5283 #ifdef HAVE_QUOTA_SUPPORT
5285 if (md->md_lu_dev.ld_obd->obd_recovering == 0 &&
5286 likely(md->md_lu_dev.ld_obd->obd_stopping == 0))
5287 next->md_ops->mdo_quota.mqo_recovery(env, next);
5291 CERROR("invalid event\n");
5298 static int mdt_obd_notify(struct obd_device *host,
5299 struct obd_device *watched,
5300 enum obd_notify_event ev, void *data)
5302 struct mdt_device *mdt = mdt_dev(host->obd_lu_dev);
5303 #ifdef HAVE_QUOTA_SUPPORT
5304 struct md_device *next = mdt->mdt_child;
5309 case OBD_NOTIFY_CONFIG:
5310 mdt_allow_cli(mdt, (unsigned long)data);
5312 #ifdef HAVE_QUOTA_SUPPORT
5313 /* quota_type has been processed, we can now handle
5314 * incoming quota requests */
5315 next->md_ops->mdo_quota.mqo_notify(NULL, next);
5319 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
5324 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5325 void *val, int vallen)
5327 struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev);
5328 struct getinfo_fid2path *fpout, *fpin;
5331 fpin = key + size_round(sizeof(KEY_FID2PATH));
5334 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5335 lustre_swab_fid2path(fpin);
5337 memcpy(fpout, fpin, sizeof(*fpin));
5338 if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5341 rc = mdt_fid2path(info->mti_env, mdt, fpout);
5345 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
5346 struct getinfo_fid2path *fp)
5348 struct mdt_object *obj;
5352 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5353 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5355 if (!fid_is_sane(&fp->gf_fid))
5358 obj = mdt_object_find(env, mdt, &fp->gf_fid);
5359 if (obj == NULL || IS_ERR(obj)) {
5360 CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid),
5365 rc = lu_object_exists(&obj->mot_obj.mo_lu);
5371 mdt_object_put(env, obj);
5372 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5373 PFID(&fp->gf_fid), rc);
5377 rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path,
5378 fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno);
5379 mdt_object_put(env, obj);
5384 static int mdt_get_info(struct mdt_thread_info *info)
5386 struct ptlrpc_request *req = mdt_info_req(info);
5394 key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5396 CDEBUG(D_IOCTL, "No GETINFO key");
5399 keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5402 vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5403 if (vallen == NULL) {
5404 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5408 req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5410 rc = req_capsule_server_pack(info->mti_pill);
5411 valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5412 if (valout == NULL) {
5413 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5417 if (KEY_IS(KEY_FID2PATH))
5418 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5422 lustre_msg_set_status(req->rq_repmsg, rc);
5427 /* Pass the ioc down */
5428 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5429 unsigned int cmd, int len, void *data)
5431 struct lu_context ioctl_session;
5432 struct md_device *next = mdt->mdt_child;
5436 rc = lu_context_init(&ioctl_session, LCT_SESSION);
5439 ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5440 lu_context_enter(&ioctl_session);
5441 env->le_ses = &ioctl_session;
5443 LASSERT(next->md_ops->mdo_iocontrol);
5444 rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5446 lu_context_exit(&ioctl_session);
5447 lu_context_fini(&ioctl_session);
5451 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5453 struct obd_ioctl_data *data = karg;
5454 struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1;
5456 struct mdt_object *obj;
5457 struct mdt_lock_handle *lh;
5460 CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5461 if (!fid_is_sane(fid))
5464 lh = &mti->mti_lh[MDT_LH_PARENT];
5465 mdt_lock_reg_init(lh, LCK_CR);
5467 obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5469 RETURN(PTR_ERR(obj));
5471 rc = mdt_object_exists(obj);
5475 * before calling version get the correct MDS should be
5476 * fid, this is error to find remote object here
5478 CERROR("nonlocal object "DFID"\n", PFID(fid));
5479 } else if (rc == 0) {
5481 CDEBUG(D_IOCTL, "no such object: "DFID"\n", PFID(fid));
5483 version = mo_version_get(mti->mti_env, mdt_object_child(obj));
5487 *(__u64 *)data->ioc_inlbuf2 = version;
5491 mdt_object_unlock_put(mti, obj, lh, 1);
5495 /* ioctls on obd dev */
5496 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5497 void *karg, void *uarg)
5500 struct obd_device *obd = exp->exp_obd;
5501 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5502 struct dt_device *dt = mdt->mdt_bottom;
5506 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5507 rc = lu_env_init(&env, LCT_MD_THREAD);
5513 rc = mdt_device_sync(&env, mdt);
5515 case OBD_IOC_SET_READONLY:
5516 dt->dd_ops->dt_ro(&env, dt);
5518 case OBD_IOC_ABORT_RECOVERY:
5519 CERROR("Aborting recovery for device %s\n", obd->obd_name);
5520 target_stop_recovery_thread(obd);
5523 case OBD_IOC_CHANGELOG_REG:
5524 case OBD_IOC_CHANGELOG_DEREG:
5525 case OBD_IOC_CHANGELOG_CLEAR:
5526 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5528 case OBD_IOC_GET_OBJ_VERSION: {
5529 struct mdt_thread_info *mti;
5530 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5531 memset(mti, 0, sizeof *mti);
5532 mti->mti_env = &env;
5536 rc = mdt_ioc_version_get(mti, karg);
5540 CERROR("Not supported cmd = %d for device %s\n",
5541 cmd, obd->obd_name);
5549 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5551 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5552 struct obd_device *obd = mdt2obd_dev(mdt);
5553 #ifdef HAVE_QUOTA_SUPPORT
5554 struct md_device *next = mdt->mdt_child;
5558 /* if some clients didn't participate in recovery then we can possibly
5559 * lost sequence. Now we should increase sequence for safe value */
5560 lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
5561 mdt_seq_adjust(env, mdt, lost);
5563 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5564 #ifdef HAVE_QUOTA_SUPPORT
5565 if (likely(obd->obd_stopping == 0))
5566 next->md_ops->mdo_quota.mqo_recovery(env, next);
5571 int mdt_obd_postrecov(struct obd_device *obd)
5576 rc = lu_env_init(&env, LCT_MD_THREAD);
5579 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5585 * Send a copytool req to a client
5586 * Note this sends a request RPC from a server (MDT) to a client (MDC),
5587 * backwards of normal comms.
5589 int mdt_hsm_copytool_send(struct obd_export *exp)
5592 struct hsm_action_list *hal;
5593 struct hsm_action_item *hai;
5597 CWARN("%s: writing to mdc at %s\n", exp->exp_obd->obd_name,
5598 libcfs_nid2str(exp->exp_connection->c_peer.nid));
5600 len = sizeof(*lh) + sizeof(*hal) + MTI_NAME_MAXLEN +
5601 /* for mockup below */ 2 * size_round(sizeof(*hai));
5606 lh->lnl_magic = LNL_MAGIC;
5607 lh->lnl_transport = LNL_TRANSPORT_HSM;
5608 lh->lnl_msgtype = HMT_ACTION_LIST;
5609 lh->lnl_msglen = len;
5611 hal = (struct hsm_action_list *)(lh + 1);
5612 hal->hal_version = HAL_VERSION;
5613 hal->hal_archive_num = 1;
5614 obd_uuid2fsname(hal->hal_fsname, exp->exp_obd->obd_name,
5617 /* mock up an action list */
5619 hai = hai_zero(hal);
5620 hai->hai_action = HSMA_ARCHIVE;
5621 hai->hai_fid.f_oid = 0xA00A;
5622 hai->hai_len = sizeof(*hai);
5623 hai = hai_next(hai);
5624 hai->hai_action = HSMA_RESTORE;
5625 hai->hai_fid.f_oid = 0xB00B;
5626 hai->hai_len = sizeof(*hai);
5628 /* Uses the ldlm reverse import; this rpc will be seen by
5629 the ldlm_callback_handler */
5630 rc = target_set_info_rpc(exp->exp_imp_reverse, LDLM_SET_INFO,
5631 sizeof(KEY_HSM_COPYTOOL_SEND),
5632 KEY_HSM_COPYTOOL_SEND,
5640 static struct obd_ops mdt_obd_device_ops = {
5641 .o_owner = THIS_MODULE,
5642 .o_set_info_async = mdt_obd_set_info_async,
5643 .o_connect = mdt_obd_connect,
5644 .o_reconnect = mdt_obd_reconnect,
5645 .o_disconnect = mdt_obd_disconnect,
5646 .o_init_export = mdt_init_export,
5647 .o_destroy_export = mdt_destroy_export,
5648 .o_iocontrol = mdt_iocontrol,
5649 .o_postrecov = mdt_obd_postrecov,
5650 .o_notify = mdt_obd_notify
5653 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5654 struct lu_device *d)
5656 struct mdt_device *m = mdt_dev(d);
5663 static struct lu_device *mdt_device_free(const struct lu_env *env,
5664 struct lu_device *d)
5666 struct mdt_device *m = mdt_dev(d);
5669 md_device_fini(&m->mdt_md_dev);
5674 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5675 struct lu_device_type *t,
5676 struct lustre_cfg *cfg)
5678 struct lu_device *l;
5679 struct mdt_device *m;
5685 l = &m->mdt_md_dev.md_lu_dev;
5686 rc = mdt_init0(env, m, t, cfg);
5688 mdt_device_free(env, l);
5692 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
5694 l = ERR_PTR(-ENOMEM);
5698 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5699 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
5701 /* context key: mdt_thread_key */
5702 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5704 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
5705 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
5707 struct lu_context_key mdt_txn_key = {
5708 .lct_tags = LCT_TX_HANDLE,
5709 .lct_init = mdt_txn_key_init,
5710 .lct_fini = mdt_txn_key_fini
5713 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
5715 return md_ucred(info->mti_env);
5719 * Enable/disable COS (Commit On Sharing).
5721 * Set/Clear the COS flag in mdt options.
5723 * \param mdt mdt device
5724 * \param val 0 disables COS, other values enable COS
5726 void mdt_enable_cos(struct mdt_device *mdt, int val)
5731 mdt->mdt_opts.mo_cos = !!val;
5732 rc = lu_env_init(&env, LCT_MD_THREAD);
5733 if (unlikely(rc != 0)) {
5734 CWARN("lu_env initialization failed with rc = %d,"
5735 "cannot sync\n", rc);
5738 mdt_device_sync(&env, mdt);
5743 * Check COS (Commit On Sharing) status.
5745 * Return COS flag status.
5747 * \param mdt mdt device
5749 int mdt_cos_is_enabled(struct mdt_device *mdt)
5751 return mdt->mdt_opts.mo_cos != 0;
5754 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
5755 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
5757 static struct lu_device_type_operations mdt_device_type_ops = {
5758 .ldto_init = mdt_type_init,
5759 .ldto_fini = mdt_type_fini,
5761 .ldto_start = mdt_type_start,
5762 .ldto_stop = mdt_type_stop,
5764 .ldto_device_alloc = mdt_device_alloc,
5765 .ldto_device_free = mdt_device_free,
5766 .ldto_device_fini = mdt_device_fini
5769 static struct lu_device_type mdt_device_type = {
5770 .ldt_tags = LU_DEVICE_MD,
5771 .ldt_name = LUSTRE_MDT_NAME,
5772 .ldt_ops = &mdt_device_type_ops,
5773 .ldt_ctx_tags = LCT_MD_THREAD
5776 static struct lu_local_obj_desc mdt_last_recv = {
5777 .llod_name = LAST_RCVD,
5778 .llod_oid = MDT_LAST_RECV_OID,
5782 static int __init mdt_mod_init(void)
5784 struct lprocfs_static_vars lvars;
5787 llo_local_obj_register(&mdt_last_recv);
5789 mdt_num_threads = MDT_NUM_THREADS;
5790 lprocfs_mdt_init_vars(&lvars);
5791 rc = class_register_type(&mdt_obd_device_ops, NULL,
5792 lvars.module_vars, LUSTRE_MDT_NAME,
5798 static void __exit mdt_mod_exit(void)
5800 llo_local_obj_unregister(&mdt_last_recv);
5801 class_unregister_type(LUSTRE_MDT_NAME);
5805 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt) \
5806 [prefix ## _ ## opc - prefix ## _ ## base] = { \
5808 .mh_fail_id = OBD_FAIL_ ## prefix ## _ ## opc ## suffix, \
5809 .mh_opc = prefix ## _ ## opc, \
5810 .mh_flags = flags, \
5815 #define DEF_MDT_HNDL(flags, name, fn, fmt) \
5816 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
5818 #define DEF_SEQ_HNDL(flags, name, fn, fmt) \
5819 DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
5821 #define DEF_FLD_HNDL(flags, name, fn, fmt) \
5822 DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
5824 * Request with a format known in advance
5826 #define DEF_MDT_HNDL_F(flags, name, fn) \
5827 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
5829 #define DEF_SEQ_HNDL_F(flags, name, fn) \
5830 DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
5832 #define DEF_FLD_HNDL_F(flags, name, fn) \
5833 DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
5835 * Request with a format we do not yet know
5837 #define DEF_MDT_HNDL_0(flags, name, fn) \
5838 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
5840 static struct mdt_handler mdt_mds_ops[] = {
5841 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5842 DEF_MDT_HNDL_F(0, DISCONNECT, mdt_disconnect),
5843 DEF_MDT_HNDL (0, SET_INFO, mdt_set_info,
5845 DEF_MDT_HNDL_F(0, GET_INFO, mdt_get_info),
5846 DEF_MDT_HNDL_F(0 |HABEO_REFERO, GETSTATUS, mdt_getstatus),
5847 DEF_MDT_HNDL_F(HABEO_CORPUS, GETATTR, mdt_getattr),
5848 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
5849 DEF_MDT_HNDL_F(HABEO_CORPUS, GETXATTR, mdt_getxattr),
5850 DEF_MDT_HNDL_F(0 |HABEO_REFERO, STATFS, mdt_statfs),
5851 DEF_MDT_HNDL_F(0 |MUTABOR, REINT, mdt_reint),
5852 DEF_MDT_HNDL_F(HABEO_CORPUS, CLOSE, mdt_close),
5853 DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing),
5854 DEF_MDT_HNDL_F(0 |HABEO_REFERO, PIN, mdt_pin),
5855 DEF_MDT_HNDL_0(0, SYNC, mdt_sync),
5856 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir),
5857 #ifdef HAVE_QUOTA_SUPPORT
5858 DEF_MDT_HNDL_F(0, QUOTACHECK, mdt_quotacheck_handle),
5859 DEF_MDT_HNDL_F(0, QUOTACTL, mdt_quotactl_handle)
5863 #define DEF_OBD_HNDL(flags, name, fn) \
5864 DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
5867 static struct mdt_handler mdt_obd_ops[] = {
5868 DEF_OBD_HNDL(0, PING, mdt_obd_ping),
5869 DEF_OBD_HNDL(0, LOG_CANCEL, mdt_obd_log_cancel),
5870 DEF_OBD_HNDL(0, QC_CALLBACK, mdt_obd_qc_callback)
5873 #define DEF_DLM_HNDL_0(flags, name, fn) \
5874 DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
5875 #define DEF_DLM_HNDL_F(flags, name, fn) \
5876 DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
5878 static struct mdt_handler mdt_dlm_ops[] = {
5879 DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE, mdt_enqueue),
5880 DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT, mdt_convert),
5881 DEF_DLM_HNDL_0(0, BL_CALLBACK, mdt_bl_callback),
5882 DEF_DLM_HNDL_0(0, CP_CALLBACK, mdt_cp_callback)
5885 #define DEF_LLOG_HNDL(flags, name, fn) \
5886 DEF_HNDL(LLOG, ORIGIN_HANDLE_CREATE, _NET, flags, name, fn, NULL)
5888 static struct mdt_handler mdt_llog_ops[] = {
5889 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CREATE, mdt_llog_create),
5890 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_NEXT_BLOCK, mdt_llog_next_block),
5891 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_READ_HEADER, mdt_llog_read_header),
5892 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_WRITE_REC, NULL),
5893 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CLOSE, NULL),
5894 DEF_LLOG_HNDL(0, ORIGIN_CONNECT, NULL),
5895 DEF_LLOG_HNDL(0, CATINFO, NULL),
5896 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_PREV_BLOCK, mdt_llog_prev_block),
5897 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_DESTROY, mdt_llog_destroy),
5900 #define DEF_SEC_CTX_HNDL(name, fn) \
5901 DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
5903 static struct mdt_handler mdt_sec_ctx_ops[] = {
5904 DEF_SEC_CTX_HNDL(INIT, mdt_sec_ctx_handle),
5905 DEF_SEC_CTX_HNDL(INIT_CONT, mdt_sec_ctx_handle),
5906 DEF_SEC_CTX_HNDL(FINI, mdt_sec_ctx_handle)
5909 static struct mdt_opc_slice mdt_regular_handlers[] = {
5911 .mos_opc_start = MDS_GETATTR,
5912 .mos_opc_end = MDS_LAST_OPC,
5913 .mos_hs = mdt_mds_ops
5916 .mos_opc_start = OBD_PING,
5917 .mos_opc_end = OBD_LAST_OPC,
5918 .mos_hs = mdt_obd_ops
5921 .mos_opc_start = LDLM_ENQUEUE,
5922 .mos_opc_end = LDLM_LAST_OPC,
5923 .mos_hs = mdt_dlm_ops
5926 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
5927 .mos_opc_end = LLOG_LAST_OPC,
5928 .mos_hs = mdt_llog_ops
5931 .mos_opc_start = SEC_CTX_INIT,
5932 .mos_opc_end = SEC_LAST_OPC,
5933 .mos_hs = mdt_sec_ctx_ops
5940 static struct mdt_handler mdt_readpage_ops[] = {
5941 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5942 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
5943 #ifdef HAVE_SPLIT_SUPPORT
5944 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
5948 * XXX: this is ugly and should be fixed one day, see mdc_close() for
5949 * detailed comments. --umka
5951 DEF_MDT_HNDL_F(HABEO_CORPUS, CLOSE, mdt_close),
5952 DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing),
5955 static struct mdt_opc_slice mdt_readpage_handlers[] = {
5957 .mos_opc_start = MDS_GETATTR,
5958 .mos_opc_end = MDS_LAST_OPC,
5959 .mos_hs = mdt_readpage_ops
5966 static struct mdt_handler mdt_xmds_ops[] = {
5967 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5968 DEF_MDT_HNDL_F(HABEO_CORPUS , GETATTR, mdt_getattr),
5969 DEF_MDT_HNDL_F(0 | MUTABOR , REINT, mdt_reint),
5970 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir),
5973 static struct mdt_opc_slice mdt_xmds_handlers[] = {
5975 .mos_opc_start = MDS_GETATTR,
5976 .mos_opc_end = MDS_LAST_OPC,
5977 .mos_hs = mdt_xmds_ops
5980 .mos_opc_start = OBD_PING,
5981 .mos_opc_end = OBD_LAST_OPC,
5982 .mos_hs = mdt_obd_ops
5985 .mos_opc_start = SEC_CTX_INIT,
5986 .mos_opc_end = SEC_LAST_OPC,
5987 .mos_hs = mdt_sec_ctx_ops
5994 static struct mdt_handler mdt_seq_ops[] = {
5995 DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
5998 static struct mdt_opc_slice mdt_seq_handlers[] = {
6000 .mos_opc_start = SEQ_QUERY,
6001 .mos_opc_end = SEQ_LAST_OPC,
6002 .mos_hs = mdt_seq_ops
6009 static struct mdt_handler mdt_fld_ops[] = {
6010 DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
6013 static struct mdt_opc_slice mdt_fld_handlers[] = {
6015 .mos_opc_start = FLD_QUERY,
6016 .mos_opc_end = FLD_LAST_OPC,
6017 .mos_hs = mdt_fld_ops
6024 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6025 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
6026 MODULE_LICENSE("GPL");
6028 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
6029 "number of mdt service threads to start");
6031 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);