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 repbody->valid |= OBD_MD_LINKNAME;
527 repbody->eadatasize = rc;
529 ((char*)ma->ma_lmm)[rc - 1] = 0;
530 CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
531 (char*)ma->ma_lmm, rc);
536 if (reqbody->valid & OBD_MD_FLMODEASIZE) {
537 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
538 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
539 repbody->valid |= OBD_MD_FLMODEASIZE;
540 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
541 "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
542 repbody->max_cookiesize);
545 if (exp_connect_rmtclient(info->mti_exp) &&
546 reqbody->valid & OBD_MD_FLRMTPERM) {
547 void *buf = req_capsule_server_get(pill, &RMF_ACL);
549 /* mdt_getattr_lock only */
550 rc = mdt_pack_remote_perm(info, o, buf);
552 repbody->valid &= ~OBD_MD_FLRMTPERM;
553 repbody->aclsize = 0;
556 repbody->valid |= OBD_MD_FLRMTPERM;
557 repbody->aclsize = sizeof(struct mdt_remote_perm);
560 #ifdef CONFIG_FS_POSIX_ACL
561 else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
562 (reqbody->valid & OBD_MD_FLACL)) {
563 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
564 buffer->lb_len = req_capsule_get_size(pill,
565 &RMF_ACL, RCL_SERVER);
566 if (buffer->lb_len > 0) {
567 rc = mo_xattr_get(env, next, buffer,
568 XATTR_NAME_ACL_ACCESS);
570 if (rc == -ENODATA) {
571 repbody->aclsize = 0;
572 repbody->valid |= OBD_MD_FLACL;
574 } else if (rc == -EOPNOTSUPP) {
577 CERROR("got acl size: %d\n", rc);
580 repbody->aclsize = rc;
581 repbody->valid |= OBD_MD_FLACL;
588 if (reqbody->valid & OBD_MD_FLMDSCAPA &&
589 info->mti_mdt->mdt_opts.mo_mds_capa &&
590 info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
591 struct lustre_capa *capa;
593 capa = req_capsule_server_get(pill, &RMF_CAPA1);
595 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
596 rc = mo_capa_get(env, next, capa, 0);
599 repbody->valid |= OBD_MD_FLMDSCAPA;
604 static int mdt_renew_capa(struct mdt_thread_info *info)
606 struct mdt_object *obj = info->mti_object;
607 struct mdt_body *body;
608 struct lustre_capa *capa, *c;
612 /* if object doesn't exist, or server has disabled capability,
613 * return directly, client will find body->valid OBD_MD_FLOSSCAPA
616 if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
617 !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
620 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
621 LASSERT(body != NULL);
623 c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
626 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
630 rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
632 body->valid |= OBD_MD_FLOSSCAPA;
636 static int mdt_getattr(struct mdt_thread_info *info)
638 struct mdt_object *obj = info->mti_object;
639 struct req_capsule *pill = info->mti_pill;
640 struct mdt_body *reqbody;
641 struct mdt_body *repbody;
647 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
650 if (reqbody->valid & OBD_MD_FLOSSCAPA) {
651 rc = req_capsule_server_pack(pill);
653 RETURN(err_serious(rc));
654 rc = mdt_renew_capa(info);
655 GOTO(out_shrink, rc);
658 LASSERT(obj != NULL);
659 LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
661 mode = lu_object_attr(&obj->mot_obj.mo_lu);
662 if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
663 (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
664 md_size = reqbody->eadatasize;
666 md_size = info->mti_mdt->mdt_max_mdsize;
668 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
670 rc = req_capsule_server_pack(pill);
671 if (unlikely(rc != 0))
672 RETURN(err_serious(rc));
674 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
675 LASSERT(repbody != NULL);
676 repbody->eadatasize = 0;
677 repbody->aclsize = 0;
679 if (reqbody->valid & OBD_MD_FLRMTPERM)
680 rc = mdt_init_ucred(info, reqbody);
682 rc = mdt_check_ucred(info);
684 GOTO(out_shrink, rc);
686 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
687 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
690 * Don't check capability at all, because rename might getattr for
691 * remote obj, and at that time no capability is available.
693 mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
694 rc = mdt_getattr_internal(info, obj);
695 if (reqbody->valid & OBD_MD_FLRMTPERM)
696 mdt_exit_ucred(info);
699 mdt_shrink_reply(info);
703 static int mdt_is_subdir(struct mdt_thread_info *info)
705 struct mdt_object *o = info->mti_object;
706 struct req_capsule *pill = info->mti_pill;
707 const struct mdt_body *body = info->mti_body;
708 struct mdt_body *repbody;
714 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
717 * We save last checked parent fid to @repbody->fid1 for remote
720 LASSERT(fid_is_sane(&body->fid2));
721 LASSERT(mdt_object_exists(o) > 0);
722 rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
723 &body->fid2, &repbody->fid1);
724 if (rc == 0 || rc == -EREMOTE)
725 repbody->valid |= OBD_MD_FLID;
730 static int mdt_raw_lookup(struct mdt_thread_info *info,
731 struct mdt_object *parent,
732 const struct lu_name *lname,
733 struct ldlm_reply *ldlm_rep)
735 struct md_object *next = mdt_object_child(info->mti_object);
736 const struct mdt_body *reqbody = info->mti_body;
737 struct lu_fid *child_fid = &info->mti_tmp_fid1;
738 struct mdt_body *repbody;
742 if (reqbody->valid != OBD_MD_FLID)
745 LASSERT(!info->mti_cross_ref);
747 /* Only got the fid of this obj by name */
748 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
751 /* XXX is raw_lookup possible as intent operation? */
754 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
757 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
759 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
762 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
763 repbody->fid1 = *child_fid;
764 repbody->valid = OBD_MD_FLID;
770 * UPDATE lock should be taken against parent, and be release before exit;
771 * child_bits lock should be taken against child, and be returned back:
772 * (1)normal request should release the child lock;
773 * (2)intent request will grant the lock to client.
775 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
776 struct mdt_lock_handle *lhc,
778 struct ldlm_reply *ldlm_rep)
780 struct ptlrpc_request *req = mdt_info_req(info);
781 struct mdt_body *reqbody = NULL;
782 struct mdt_object *parent = info->mti_object;
783 struct mdt_object *child;
784 struct md_object *next = mdt_object_child(parent);
785 struct lu_fid *child_fid = &info->mti_tmp_fid1;
786 struct lu_name *lname = NULL;
787 const char *name = NULL;
789 struct mdt_lock_handle *lhp;
790 struct ldlm_lock *lock;
791 struct ldlm_res_id *res_id;
797 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
798 LASSERT(ergo(is_resent,
799 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
801 LASSERT(parent != NULL);
802 name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
804 RETURN(err_serious(-EFAULT));
806 namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
808 if (!info->mti_cross_ref) {
810 * XXX: Check for "namelen == 0" is for getattr by fid
811 * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
812 * that is the name must contain at least one character and
813 * the terminating '\0'
816 reqbody = req_capsule_client_get(info->mti_pill,
818 LASSERT(fid_is_sane(&reqbody->fid2));
821 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
823 PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
826 lname = mdt_name(info->mti_env, (char *)name, namelen);
827 CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
828 "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
832 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
834 rc = mdt_object_exists(parent);
835 if (unlikely(rc == 0)) {
836 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
837 &parent->mot_obj.mo_lu,
838 "Parent doesn't exist!\n");
840 } else if (!info->mti_cross_ref) {
841 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
842 PFID(mdt_object_fid(parent)));
845 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
853 if (info->mti_cross_ref) {
854 /* Only getattr on the child. Parent is on another node. */
855 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
857 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
858 "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
861 /* Do not take lock for resent case. */
862 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
863 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
864 lhc->mlh_reg_lh.cookie);
865 LASSERT(fid_res_name_eq(mdt_object_fid(child),
866 &lock->l_resource->lr_name));
870 mdt_lock_handle_init(lhc);
871 mdt_lock_reg_init(lhc, LCK_PR);
874 * Object's name is on another MDS, no lookup lock is
875 * needed here but update is.
877 child_bits &= ~MDS_INODELOCK_LOOKUP;
878 child_bits |= MDS_INODELOCK_UPDATE;
880 rc = mdt_object_lock(info, child, lhc, child_bits,
884 /* Finally, we can get attr for child. */
885 mdt_set_capainfo(info, 0, mdt_object_fid(child),
887 rc = mdt_getattr_internal(info, child);
888 if (unlikely(rc != 0))
889 mdt_object_unlock(info, child, lhc, 1);
894 /* step 1: lock parent */
895 lhp = &info->mti_lh[MDT_LH_PARENT];
896 mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
897 rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
900 if (unlikely(rc != 0))
904 /* step 2: lookup child's fid by name */
905 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
910 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
911 GOTO(out_parent, rc);
913 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
915 *child_fid = reqbody->fid2;
916 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
920 *step 3: find the child object by fid & lock it.
921 * regardless if it is local or remote.
923 child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
925 if (unlikely(IS_ERR(child)))
926 GOTO(out_parent, rc = PTR_ERR(child));
928 /* Do not take lock for resent case. */
929 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
930 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
931 lhc->mlh_reg_lh.cookie);
933 res_id = &lock->l_resource->lr_name;
934 if (!fid_res_name_eq(mdt_object_fid(child),
935 &lock->l_resource->lr_name)) {
936 LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
937 &lock->l_resource->lr_name),
938 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
939 (unsigned long)res_id->name[0],
940 (unsigned long)res_id->name[1],
941 (unsigned long)res_id->name[2],
942 PFID(mdt_object_fid(parent)));
943 CWARN("Although resent, but still not get child lock"
944 "parent:"DFID" child:"DFID"\n",
945 PFID(mdt_object_fid(parent)),
946 PFID(mdt_object_fid(child)));
947 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
956 ma = &info->mti_attr;
958 mdt_lock_handle_init(lhc);
959 mdt_lock_reg_init(lhc, LCK_PR);
961 if (mdt_object_exists(child) == 0) {
962 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
963 &child->mot_obj.mo_lu,
964 "Object doesn't exist!\n");
965 GOTO(out_child, rc = -ESTALE);
969 ma->ma_need = MA_INODE;
970 rc = mo_attr_get(info->mti_env, next, ma);
971 if (unlikely(rc != 0))
974 /* If the file has not been changed for some time, we return
975 * not only a LOOKUP lock, but also an UPDATE lock and this
976 * might save us RPC on later STAT. For directories, it also
977 * let negative dentry starts working for this dir. */
978 if (ma->ma_valid & MA_INODE &&
979 ma->ma_attr.la_valid & LA_CTIME &&
980 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
981 ma->ma_attr.la_ctime < cfs_time_current_sec())
982 child_bits |= MDS_INODELOCK_UPDATE;
984 rc = mdt_object_lock(info, child, lhc, child_bits,
987 if (unlikely(rc != 0))
991 /* finally, we can get attr for child. */
992 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
993 rc = mdt_getattr_internal(info, child);
994 if (unlikely(rc != 0)) {
995 mdt_object_unlock(info, child, lhc, 1);
997 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
999 struct mdt_body *repbody;
1001 /* Debugging code. */
1002 res_id = &lock->l_resource->lr_name;
1003 LDLM_DEBUG(lock, "Returning lock to client\n");
1004 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1005 &lock->l_resource->lr_name),
1006 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1007 (unsigned long)res_id->name[0],
1008 (unsigned long)res_id->name[1],
1009 (unsigned long)res_id->name[2],
1010 PFID(mdt_object_fid(child)));
1012 * Pack Size-on-MDS inode attributes to the body if
1013 * update lock is given.
1015 repbody = req_capsule_server_get(info->mti_pill,
1017 if (lock->l_policy_data.l_inodebits.bits &
1018 MDS_INODELOCK_UPDATE)
1019 mdt_pack_size2body(info, child);
1020 LDLM_LOCK_PUT(lock);
1025 mdt_object_put(info->mti_env, child);
1027 mdt_object_unlock(info, parent, lhp, 1);
1031 /* normal handler: should release the child lock */
1032 static int mdt_getattr_name(struct mdt_thread_info *info)
1034 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1035 struct mdt_body *reqbody;
1036 struct mdt_body *repbody;
1040 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1041 LASSERT(reqbody != NULL);
1042 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1043 LASSERT(repbody != NULL);
1045 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1046 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1047 repbody->eadatasize = 0;
1048 repbody->aclsize = 0;
1050 rc = mdt_init_ucred(info, reqbody);
1052 GOTO(out_shrink, rc);
1054 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1055 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1056 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1057 lhc->mlh_reg_lh.cookie = 0;
1059 mdt_exit_ucred(info);
1062 mdt_shrink_reply(info);
1066 static const struct lu_device_operations mdt_lu_ops;
1068 static int lu_device_is_mdt(struct lu_device *d)
1070 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1073 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1074 void *karg, void *uarg);
1076 static int mdt_set_info(struct mdt_thread_info *info)
1078 struct ptlrpc_request *req = mdt_info_req(info);
1081 int keylen, vallen, rc = 0;
1084 rc = req_capsule_server_pack(info->mti_pill);
1088 key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1090 DEBUG_REQ(D_HA, req, "no set_info key");
1094 keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1097 val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1099 DEBUG_REQ(D_HA, req, "no set_info val");
1103 vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1106 if (KEY_IS(KEY_READ_ONLY)) {
1108 lustre_msg_set_status(req->rq_repmsg, 0);
1110 spin_lock(&req->rq_export->exp_lock);
1112 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1114 req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
1115 spin_unlock(&req->rq_export->exp_lock);
1117 } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1118 struct changelog_setinfo *cs =
1119 (struct changelog_setinfo *)val;
1120 if (vallen != sizeof(*cs)) {
1121 CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1124 if (lustre_msg_swabbed(req->rq_reqmsg)) {
1125 __swab64s(&cs->cs_recno);
1126 __swab32s(&cs->cs_id);
1129 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1131 lustre_msg_set_status(req->rq_repmsg, rc);
1139 static int mdt_connect(struct mdt_thread_info *info)
1142 struct ptlrpc_request *req;
1144 req = mdt_info_req(info);
1145 rc = target_handle_connect(req);
1147 LASSERT(req->rq_export != NULL);
1148 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1149 rc = mdt_init_sec_level(info);
1151 rc = mdt_init_idmap(info);
1153 obd_disconnect(class_export_get(req->rq_export));
1155 rc = err_serious(rc);
1160 static int mdt_disconnect(struct mdt_thread_info *info)
1165 rc = target_handle_disconnect(mdt_info_req(info));
1167 rc = err_serious(rc);
1171 static int mdt_sendpage(struct mdt_thread_info *info,
1172 struct lu_rdpg *rdpg)
1174 struct ptlrpc_request *req = mdt_info_req(info);
1175 struct ptlrpc_bulk_desc *desc;
1176 struct l_wait_info *lwi = &info->mti_u.rdpg.mti_wait_info;
1184 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1189 for (i = 0, tmpcount = rdpg->rp_count;
1190 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1191 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1192 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1195 LASSERT(desc->bd_nob == rdpg->rp_count);
1196 rc = sptlrpc_svc_wrap_bulk(req, desc);
1198 GOTO(free_desc, rc);
1200 rc = ptlrpc_start_bulk_transfer(desc);
1202 GOTO(free_desc, rc);
1204 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1205 GOTO(abort_bulk, rc = 0);
1207 timeout = (int) req->rq_deadline - cfs_time_current_sec();
1209 CERROR("Req deadline already passed %lu (now: %lu)\n",
1210 req->rq_deadline, cfs_time_current_sec());
1211 *lwi = LWI_TIMEOUT(cfs_time_seconds(max(timeout, 1)), NULL, NULL);
1212 rc = l_wait_event(desc->bd_waitq, !ptlrpc_server_bulk_active(desc), lwi);
1213 LASSERT (rc == 0 || rc == -ETIMEDOUT);
1216 if (desc->bd_success &&
1217 desc->bd_nob_transferred == rdpg->rp_count)
1218 GOTO(free_desc, rc);
1220 rc = -ETIMEDOUT; /* XXX should this be a different errno? */
1223 DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1224 (rc == -ETIMEDOUT) ? "timeout" : "network error",
1225 desc->bd_nob_transferred, rdpg->rp_count,
1226 req->rq_export->exp_client_uuid.uuid,
1227 req->rq_export->exp_connection->c_remote_uuid.uuid);
1229 class_fail_export(req->rq_export);
1233 ptlrpc_abort_bulk(desc);
1235 ptlrpc_free_bulk(desc);
1239 #ifdef HAVE_SPLIT_SUPPORT
1241 * Retrieve dir entry from the page and insert it to the slave object, actually,
1242 * this should be in osd layer, but since it will not in the final product, so
1243 * just do it here and do not define more moo api anymore for this.
1245 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1248 struct mdt_object *object = info->mti_object;
1249 struct lu_fid *lf = &info->mti_tmp_fid2;
1250 struct md_attr *ma = &info->mti_attr;
1251 struct lu_dirpage *dp;
1252 struct lu_dirent *ent;
1253 int rc = 0, offset = 0;
1256 /* Make sure we have at least one entry. */
1261 * Disable trans for this name insert, since it will include many trans
1264 info->mti_no_need_trans = 1;
1266 * When write_dir_page, no need update parent's ctime,
1267 * and no permission check for name_insert.
1269 ma->ma_attr.la_ctime = 0;
1270 ma->ma_attr.la_valid = LA_MODE;
1271 ma->ma_valid = MA_INODE;
1274 dp = page_address(page);
1275 offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1277 for (ent = lu_dirent_start(dp); ent != NULL;
1278 ent = lu_dirent_next(ent)) {
1279 struct lu_name *lname;
1282 if (le16_to_cpu(ent->lde_namelen) == 0)
1285 fid_le_to_cpu(lf, &ent->lde_fid);
1286 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1287 ma->ma_attr.la_mode = S_IFDIR;
1289 ma->ma_attr.la_mode = 0;
1290 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1292 GOTO(out, rc = -ENOMEM);
1294 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1295 lname = mdt_name(info->mti_env, name,
1296 le16_to_cpu(ent->lde_namelen));
1297 ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE);
1298 rc = mdo_name_insert(info->mti_env,
1299 md_object_next(&object->mot_obj),
1301 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1303 CERROR("Can't insert %*.*s, rc %d\n",
1304 le16_to_cpu(ent->lde_namelen),
1305 le16_to_cpu(ent->lde_namelen),
1310 offset += lu_dirent_size(ent);
1320 static int mdt_bulk_timeout(void *data)
1324 CERROR("mdt bulk transfer timeout \n");
1329 static int mdt_writepage(struct mdt_thread_info *info)
1331 struct ptlrpc_request *req = mdt_info_req(info);
1332 struct mdt_body *reqbody;
1333 struct l_wait_info *lwi;
1334 struct ptlrpc_bulk_desc *desc;
1340 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1341 if (reqbody == NULL)
1342 RETURN(err_serious(-EFAULT));
1344 desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1346 RETURN(err_serious(-ENOMEM));
1348 /* allocate the page for the desc */
1349 page = cfs_alloc_page(CFS_ALLOC_STD);
1351 GOTO(desc_cleanup, rc = -ENOMEM);
1353 CDEBUG(D_INFO, "Received page offset %d size %d \n",
1354 (int)reqbody->size, (int)reqbody->nlink);
1356 ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1357 (int)reqbody->nlink);
1359 rc = sptlrpc_svc_prep_bulk(req, desc);
1361 GOTO(cleanup_page, rc);
1363 * Check if client was evicted while we were doing i/o before touching
1368 GOTO(cleanup_page, rc = -ENOMEM);
1370 if (desc->bd_export->exp_failed)
1373 rc = ptlrpc_start_bulk_transfer (desc);
1375 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
1376 mdt_bulk_timeout, desc);
1377 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1378 desc->bd_export->exp_failed, lwi);
1379 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1380 if (rc == -ETIMEDOUT) {
1381 DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1382 ptlrpc_abort_bulk(desc);
1383 } else if (desc->bd_export->exp_failed) {
1384 DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1386 ptlrpc_abort_bulk(desc);
1387 } else if (!desc->bd_success ||
1388 desc->bd_nob_transferred != desc->bd_nob) {
1389 DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1391 "truncated" : "network error on",
1392 desc->bd_nob_transferred, desc->bd_nob);
1393 /* XXX should this be a different errno? */
1397 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1400 GOTO(cleanup_lwi, rc);
1401 rc = mdt_write_dir_page(info, page, reqbody->nlink);
1406 cfs_free_page(page);
1408 ptlrpc_free_bulk(desc);
1413 static int mdt_readpage(struct mdt_thread_info *info)
1415 struct mdt_object *object = info->mti_object;
1416 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1417 struct mdt_body *reqbody;
1418 struct mdt_body *repbody;
1423 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1424 RETURN(err_serious(-ENOMEM));
1426 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1427 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1428 if (reqbody == NULL || repbody == NULL)
1429 RETURN(err_serious(-EFAULT));
1432 * prepare @rdpg before calling lower layers and transfer itself. Here
1433 * reqbody->size contains offset of where to start to read and
1434 * reqbody->nlink contains number bytes to read.
1436 rdpg->rp_hash = reqbody->size;
1437 if (rdpg->rp_hash != reqbody->size) {
1438 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1439 rdpg->rp_hash, reqbody->size);
1443 rdpg->rp_attrs = reqbody->mode;
1444 rdpg->rp_count = reqbody->nlink;
1445 rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1446 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1447 if (rdpg->rp_pages == NULL)
1450 for (i = 0; i < rdpg->rp_npages; ++i) {
1451 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1452 if (rdpg->rp_pages[i] == NULL)
1453 GOTO(free_rdpg, rc = -ENOMEM);
1456 /* call lower layers to fill allocated pages with directory data */
1457 rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1459 GOTO(free_rdpg, rc);
1461 /* send pages to client */
1462 rc = mdt_sendpage(info, rdpg);
1467 for (i = 0; i < rdpg->rp_npages; i++)
1468 if (rdpg->rp_pages[i] != NULL)
1469 cfs_free_page(rdpg->rp_pages[i]);
1470 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1472 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1478 static int mdt_reint_internal(struct mdt_thread_info *info,
1479 struct mdt_lock_handle *lhc,
1482 struct req_capsule *pill = info->mti_pill;
1483 struct mdt_device *mdt = info->mti_mdt;
1484 struct mdt_body *repbody;
1489 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1490 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1491 mdt->mdt_max_mdsize);
1492 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1493 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1494 mdt->mdt_max_cookiesize);
1496 rc = req_capsule_server_pack(pill);
1498 CERROR("Can't pack response, rc %d\n", rc);
1499 RETURN(err_serious(rc));
1502 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1503 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1505 repbody->eadatasize = 0;
1506 repbody->aclsize = 0;
1509 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1510 GOTO(out_shrink, rc = err_serious(-EFAULT));
1512 rc = mdt_reint_unpack(info, op);
1514 CERROR("Can't unpack reint, rc %d\n", rc);
1515 GOTO(out_shrink, rc = err_serious(rc));
1518 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1520 /* for replay no cookkie / lmm need, because client have this already */
1521 if (info->mti_spec.no_create == 1) {
1522 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1523 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1525 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1526 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1530 rc = mdt_init_ucred_reint(info);
1532 GOTO(out_shrink, rc);
1534 rc = mdt_fix_attr_ucred(info, op);
1536 GOTO(out_ucred, rc = err_serious(rc));
1538 if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1539 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1540 GOTO(out_ucred, rc);
1542 rc = mdt_reint_rec(info, lhc);
1545 mdt_exit_ucred(info);
1547 mdt_shrink_reply(info);
1551 static long mdt_reint_opcode(struct mdt_thread_info *info,
1552 const struct req_format **fmt)
1554 struct mdt_rec_reint *rec;
1557 opc = err_serious(-EFAULT);
1558 rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1560 opc = rec->rr_opcode;
1561 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1562 if (opc < REINT_MAX && fmt[opc] != NULL)
1563 req_capsule_extend(info->mti_pill, fmt[opc]);
1565 CERROR("Unsupported opc: %ld\n", opc);
1566 opc = err_serious(opc);
1572 static int mdt_reint(struct mdt_thread_info *info)
1577 static const struct req_format *reint_fmts[REINT_MAX] = {
1578 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1579 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1580 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1581 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1582 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1583 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1584 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1589 opc = mdt_reint_opcode(info, reint_fmts);
1592 * No lock possible here from client to pass it to reint code
1595 rc = mdt_reint_internal(info, NULL, opc);
1600 info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1604 /* this should sync the whole device */
1605 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1607 struct dt_device *dt = mdt->mdt_bottom;
1611 rc = dt->dd_ops->dt_sync(env, dt);
1615 /* this should sync this object */
1616 static int mdt_object_sync(struct mdt_thread_info *info)
1618 struct md_object *next;
1622 if (!mdt_object_exists(info->mti_object)) {
1623 CWARN("Non existing object "DFID"!\n",
1624 PFID(mdt_object_fid(info->mti_object)));
1627 next = mdt_object_child(info->mti_object);
1628 rc = mo_object_sync(info->mti_env, next);
1633 static int mdt_sync(struct mdt_thread_info *info)
1635 struct req_capsule *pill = info->mti_pill;
1636 struct mdt_body *body;
1640 /* The fid may be zero, so we req_capsule_set manually */
1641 req_capsule_set(pill, &RQF_MDS_SYNC);
1643 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1645 RETURN(err_serious(-EINVAL));
1647 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1648 RETURN(err_serious(-ENOMEM));
1650 if (fid_seq(&body->fid1) == 0) {
1651 /* sync the whole device */
1652 rc = req_capsule_server_pack(pill);
1654 rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1656 rc = err_serious(rc);
1658 /* sync an object */
1659 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1661 rc = mdt_object_sync(info);
1663 struct md_object *next;
1664 const struct lu_fid *fid;
1665 struct lu_attr *la = &info->mti_attr.ma_attr;
1667 next = mdt_object_child(info->mti_object);
1668 info->mti_attr.ma_need = MA_INODE;
1669 info->mti_attr.ma_valid = 0;
1670 rc = mo_attr_get(info->mti_env, next,
1673 body = req_capsule_server_get(pill,
1675 fid = mdt_object_fid(info->mti_object);
1676 mdt_pack_attr2body(info, body, la, fid);
1680 rc = err_serious(rc);
1685 #ifdef HAVE_QUOTA_SUPPORT
1686 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1688 struct obd_quotactl *oqctl;
1689 struct req_capsule *pill = info->mti_pill;
1690 struct obd_export *exp = info->mti_exp;
1691 struct md_device *next = info->mti_mdt->mdt_child;
1695 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_QUOTACHECK_NET))
1698 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1702 /* remote client has no permission for quotacheck */
1703 if (unlikely(exp_connect_rmtclient(exp)))
1706 rc = req_capsule_server_pack(pill);
1710 rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next, exp,
1715 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1717 struct obd_quotactl *oqctl, *repoqc;
1718 struct req_capsule *pill = info->mti_pill;
1719 struct obd_export *exp = info->mti_exp;
1720 struct md_device *next = info->mti_mdt->mdt_child;
1721 const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
1725 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_QUOTACTL_NET))
1728 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1733 if (exp_connect_rmtclient(exp)) {
1734 struct ptlrpc_request *req = mdt_info_req(info);
1735 struct mdt_export_data *med = mdt_req2med(req);
1736 struct lustre_idmap_table *idmap = med->med_idmap;
1738 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1739 oqctl->qc_cmd != Q_GETINFO))
1743 if (oqctl->qc_type == USRQUOTA)
1744 id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1746 else if (oqctl->qc_type == GRPQUOTA)
1747 id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1752 if (id == CFS_IDMAP_NOTFOUND) {
1753 CDEBUG(D_QUOTA, "no mapping for id %u\n",
1759 rc = req_capsule_server_pack(pill);
1763 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1764 LASSERT(repoqc != NULL);
1766 switch (oqctl->qc_cmd) {
1768 if (info->mti_mdt->mdt_som_conf) {
1769 /* Quota cannot be used together with SOM while
1770 * SOM stored blocks in i_blocks but not in SOM EA. */
1771 LCONSOLE_ERROR("Fail to turn Quota on: SOM is enabled "
1772 "and temporary conflicts with quota.\n");
1775 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
1778 rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
1781 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
1785 rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id,
1789 rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id,
1793 rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id,
1797 rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id,
1801 rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id,
1804 case LUSTRE_Q_INVALIDATE:
1805 rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type);
1807 case LUSTRE_Q_FINVALIDATE:
1808 rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type);
1811 CERROR("unsupported mdt_quotactl command: %d\n",
1823 * OBD PING and other handlers.
1825 static int mdt_obd_ping(struct mdt_thread_info *info)
1830 req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1832 rc = target_handle_ping(mdt_info_req(info));
1834 rc = err_serious(rc);
1838 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1840 return err_serious(-EOPNOTSUPP);
1843 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1845 return err_serious(-EOPNOTSUPP);
1853 /** clone llog ctxt from child (mdd)
1854 * This allows remote llog (replicator) access.
1855 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
1856 * context was originally set up, or we can handle them directly.
1857 * I choose the latter, but that means I need any llog
1858 * contexts set up by child to be accessable by the mdt. So we clone the
1859 * context into our context list here.
1861 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
1864 struct md_device *next = mdt->mdt_child;
1865 struct llog_ctxt *ctxt;
1868 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
1871 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
1872 if (rc || ctxt == NULL) {
1873 CERROR("Can't get mdd ctxt %d\n", rc);
1877 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
1879 CERROR("Can't set mdt ctxt %d\n", rc);
1884 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
1885 struct mdt_device *mdt, int idx)
1887 struct llog_ctxt *ctxt;
1889 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
1892 /* Put once for the get we just did, and once for the clone */
1893 llog_ctxt_put(ctxt);
1894 llog_ctxt_put(ctxt);
1898 static int mdt_llog_create(struct mdt_thread_info *info)
1902 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
1903 rc = llog_origin_handle_create(mdt_info_req(info));
1904 return (rc < 0 ? err_serious(rc) : rc);
1907 static int mdt_llog_destroy(struct mdt_thread_info *info)
1911 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
1912 rc = llog_origin_handle_destroy(mdt_info_req(info));
1913 return (rc < 0 ? err_serious(rc) : rc);
1916 static int mdt_llog_read_header(struct mdt_thread_info *info)
1920 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
1921 rc = llog_origin_handle_read_header(mdt_info_req(info));
1922 return (rc < 0 ? err_serious(rc) : rc);
1925 static int mdt_llog_next_block(struct mdt_thread_info *info)
1929 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
1930 rc = llog_origin_handle_next_block(mdt_info_req(info));
1931 return (rc < 0 ? err_serious(rc) : rc);
1934 static int mdt_llog_prev_block(struct mdt_thread_info *info)
1938 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
1939 rc = llog_origin_handle_prev_block(mdt_info_req(info));
1940 return (rc < 0 ? err_serious(rc) : rc);
1947 static struct ldlm_callback_suite cbs = {
1948 .lcs_completion = ldlm_server_completion_ast,
1949 .lcs_blocking = ldlm_server_blocking_ast,
1953 static int mdt_enqueue(struct mdt_thread_info *info)
1955 struct ptlrpc_request *req;
1960 * info->mti_dlm_req already contains swapped and (if necessary)
1961 * converted dlm request.
1963 LASSERT(info->mti_dlm_req != NULL);
1965 req = mdt_info_req(info);
1968 * Lock without inodebits makes no sense and will oops later in
1969 * ldlm. Let's check it now to see if we have wrong lock from client or
1970 * bits get corrupted somewhere in mdt_intent_policy().
1972 req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1973 /* This is disabled because we need to support liblustre flock.
1974 * LASSERT(req_bits != 0);
1977 rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1978 req, info->mti_dlm_req, &cbs);
1979 info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1980 return rc ? err_serious(rc) : req->rq_status;
1983 static int mdt_convert(struct mdt_thread_info *info)
1986 struct ptlrpc_request *req;
1988 LASSERT(info->mti_dlm_req);
1989 req = mdt_info_req(info);
1990 rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1991 return rc ? err_serious(rc) : req->rq_status;
1994 static int mdt_bl_callback(struct mdt_thread_info *info)
1996 CERROR("bl callbacks should not happen on MDS\n");
1998 return err_serious(-EOPNOTSUPP);
2001 static int mdt_cp_callback(struct mdt_thread_info *info)
2003 CERROR("cp callbacks should not happen on MDS\n");
2005 return err_serious(-EOPNOTSUPP);
2009 * sec context handlers
2011 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2015 rc = mdt_handle_idmap(info);
2018 struct ptlrpc_request *req = mdt_info_req(info);
2021 opc = lustre_msg_get_opc(req->rq_reqmsg);
2022 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2023 sptlrpc_svc_ctx_invalidate(req);
2026 OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
2031 static struct mdt_object *mdt_obj(struct lu_object *o)
2033 LASSERT(lu_device_is_mdt(o->lo_dev));
2034 return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2037 struct mdt_object *mdt_object_find(const struct lu_env *env,
2038 struct mdt_device *d,
2039 const struct lu_fid *f)
2041 struct lu_object *o;
2042 struct mdt_object *m;
2045 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2046 o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2047 if (unlikely(IS_ERR(o)))
2048 m = (struct mdt_object *)o;
2055 * Asyncronous commit for mdt device.
2057 * Pass asynchonous commit call down the MDS stack.
2059 * \param env environment
2060 * \param mdt the mdt device
2062 static void mdt_device_commit_async(const struct lu_env *env,
2063 struct mdt_device *mdt)
2065 struct dt_device *dt = mdt->mdt_bottom;
2068 rc = dt->dd_ops->dt_commit_async(env, dt);
2069 if (unlikely(rc != 0))
2070 CWARN("async commit start failed with rc = %d", rc);
2074 * Mark the lock as "synchonous".
2076 * Mark the lock to deffer transaction commit to the unlock time.
2078 * \param lock the lock to mark as "synchonous"
2080 * \see mdt_is_lock_sync
2081 * \see mdt_save_lock
2083 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2085 lock->l_ast_data = (void*)1;
2089 * Check whehter the lock "synchonous" or not.
2091 * \param lock the lock to check
2092 * \retval 1 the lock is "synchonous"
2093 * \retval 0 the lock isn't "synchronous"
2095 * \see mdt_set_lock_sync
2096 * \see mdt_save_lock
2098 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2100 return lock->l_ast_data != NULL;
2104 * Blocking AST for mdt locks.
2106 * Starts transaction commit if in case of COS lock conflict or
2107 * deffers such a commit to the mdt_save_lock.
2109 * \param lock the lock which blocks a request or cancelling lock
2110 * \param desc unused
2111 * \param data unused
2112 * \param flag indicates whether this cancelling or blocking callback
2114 * \see ldlm_blocking_ast_nocheck
2116 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2117 void *data, int flag)
2119 struct obd_device *obd = lock->l_resource->lr_namespace->ns_obd;
2120 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2124 if (flag == LDLM_CB_CANCELING)
2126 lock_res_and_lock(lock);
2127 if (lock->l_blocking_ast != mdt_blocking_ast) {
2128 unlock_res_and_lock(lock);
2131 if (mdt_cos_is_enabled(mdt) &&
2132 lock->l_req_mode & (LCK_PW | LCK_EX) &&
2133 lock->l_blocking_lock != NULL &&
2134 lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2135 mdt_set_lock_sync(lock);
2137 rc = ldlm_blocking_ast_nocheck(lock);
2139 /* There is no lock conflict if l_blocking_lock == NULL,
2140 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2141 * when the last reference to a local lock was released */
2142 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2145 rc = lu_env_init(&env, LCT_MD_THREAD);
2146 if (unlikely(rc != 0))
2147 CWARN("lu_env initialization failed with rc = %d,"
2148 "cannot start asynchronous commit\n", rc);
2150 mdt_device_commit_async(&env, mdt);
2156 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2157 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2159 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2160 ldlm_policy_data_t *policy = &info->mti_policy;
2161 struct ldlm_res_id *res_id = &info->mti_res_id;
2165 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2166 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2167 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2168 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2170 if (mdt_object_exists(o) < 0) {
2171 if (locality == MDT_CROSS_LOCK) {
2172 /* cross-ref object fix */
2173 ibits &= ~MDS_INODELOCK_UPDATE;
2174 ibits |= MDS_INODELOCK_LOOKUP;
2176 LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2177 LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2179 /* No PDO lock on remote object */
2180 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2183 if (lh->mlh_type == MDT_PDO_LOCK) {
2184 /* check for exists after object is locked */
2185 if (mdt_object_exists(o) == 0) {
2186 /* Non-existent object shouldn't have PDO lock */
2189 /* Non-dir object shouldn't have PDO lock */
2190 LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
2194 memset(policy, 0, sizeof(*policy));
2195 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2198 * Take PDO lock on whole directory and build correct @res_id for lock
2199 * on part of directory.
2201 if (lh->mlh_pdo_hash != 0) {
2202 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2203 mdt_lock_pdo_mode(info, o, lh);
2204 if (lh->mlh_pdo_mode != LCK_NL) {
2206 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2207 * is never going to be sent to client and we do not
2208 * want it slowed down due to possible cancels.
2210 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2211 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2212 policy, res_id, LDLM_FL_ATOMIC_CB,
2213 &info->mti_exp->exp_handle.h_cookie);
2219 * Finish res_id initializing by name hash marking part of
2220 * directory which is taking modification.
2222 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2225 policy->l_inodebits.bits = ibits;
2228 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2229 * going to be sent to client. If it is - mdt_intent_policy() path will
2230 * fix it up and turn FL_LOCAL flag off.
2232 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2233 res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
2234 &info->mti_exp->exp_handle.h_cookie);
2240 mdt_object_unlock(info, o, lh, 1);
2247 * Save a lock within request object.
2249 * Keep the lock referenced until whether client ACK or transaction
2250 * commit happens or release the lock immediately depending on input
2251 * parameters. If COS is ON, a write lock is converted to COS lock
2254 * \param info thead info object
2255 * \param h lock handle
2256 * \param mode lock mode
2257 * \param decref force immediate lock releasing
2260 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2261 ldlm_mode_t mode, int decref)
2265 if (lustre_handle_is_used(h)) {
2266 if (decref || !info->mti_has_trans ||
2267 !(mode & (LCK_PW | LCK_EX))){
2268 mdt_fid_unlock(h, mode);
2270 struct mdt_device *mdt = info->mti_mdt;
2271 struct ldlm_lock *lock = ldlm_handle2lock(h);
2272 struct ptlrpc_request *req = mdt_info_req(info);
2275 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2277 CDEBUG(D_HA, "request = %p reply state = %p"
2278 " transno = "LPD64"\n",
2279 req, req->rq_reply_state, req->rq_transno);
2280 if (mdt_cos_is_enabled(mdt)) {
2282 ldlm_lock_downgrade(lock, LCK_COS);
2285 ptlrpc_save_lock(req, h, mode, no_ack);
2286 if (mdt_is_lock_sync(lock)) {
2287 CDEBUG(D_HA, "found sync-lock,"
2288 " async commit started\n");
2289 mdt_device_commit_async(info->mti_env,
2292 LDLM_LOCK_PUT(lock);
2301 * Unlock mdt object.
2303 * Immeditely release the regular lock and the PDO lock or save the
2304 * lock in reqeuest and keep them referenced until client ACK or
2305 * transaction commit.
2307 * \param info thread info object
2308 * \param o mdt object
2309 * \param h mdt lock handle referencing regular and PDO locks
2310 * \param decref force immediate lock releasing
2312 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2313 struct mdt_lock_handle *lh, int decref)
2317 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2318 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2323 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2324 const struct lu_fid *f,
2325 struct mdt_lock_handle *lh,
2328 struct mdt_object *o;
2330 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2334 rc = mdt_object_lock(info, o, lh, ibits,
2337 mdt_object_put(info->mti_env, o);
2344 void mdt_object_unlock_put(struct mdt_thread_info * info,
2345 struct mdt_object * o,
2346 struct mdt_lock_handle *lh,
2349 mdt_object_unlock(info, o, lh, decref);
2350 mdt_object_put(info->mti_env, o);
2353 static struct mdt_handler *mdt_handler_find(__u32 opc,
2354 struct mdt_opc_slice *supported)
2356 struct mdt_opc_slice *s;
2357 struct mdt_handler *h;
2360 for (s = supported; s->mos_hs != NULL; s++) {
2361 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2362 h = s->mos_hs + (opc - s->mos_opc_start);
2363 if (likely(h->mh_opc != 0))
2364 LASSERTF(h->mh_opc == opc,
2365 "opcode mismatch %d != %d\n",
2368 h = NULL; /* unsupported opc */
2375 static int mdt_lock_resname_compat(struct mdt_device *m,
2376 struct ldlm_request *req)
2378 /* XXX something... later. */
2382 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2384 /* XXX something... later. */
2389 * Generic code handling requests that have struct mdt_body passed in:
2391 * - extract mdt_body from request and save it in @info, if present;
2393 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2396 * - if HABEO_CORPUS flag is set for this request type check whether object
2397 * actually exists on storage (lu_object_exists()).
2400 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2402 const struct mdt_body *body;
2403 struct mdt_object *obj;
2404 const struct lu_env *env;
2405 struct req_capsule *pill;
2409 env = info->mti_env;
2410 pill = info->mti_pill;
2412 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2416 if (!(body->valid & OBD_MD_FLID))
2419 if (!fid_is_sane(&body->fid1)) {
2420 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2425 * Do not get size or any capa fields before we check that request
2426 * contains capa actually. There are some requests which do not, for
2427 * instance MDS_IS_SUBDIR.
2429 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2430 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2431 mdt_set_capainfo(info, 0, &body->fid1,
2432 req_capsule_client_get(pill, &RMF_CAPA1));
2434 obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2436 if ((flags & HABEO_CORPUS) &&
2437 !mdt_object_exists(obj)) {
2438 mdt_object_put(env, obj);
2439 /* for capability renew ENOENT will be handled in
2441 if (body->valid & OBD_MD_FLOSSCAPA)
2446 info->mti_object = obj;
2455 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2457 struct req_capsule *pill = info->mti_pill;
2461 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2462 rc = mdt_body_unpack(info, flags);
2466 if (rc == 0 && (flags & HABEO_REFERO)) {
2467 struct mdt_device *mdt = info->mti_mdt;
2471 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2472 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2473 mdt->mdt_max_mdsize);
2474 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2475 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2476 mdt->mdt_max_cookiesize);
2478 rc = req_capsule_server_pack(pill);
2483 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2485 struct md_device *next = m->mdt_child;
2487 return next->md_ops->mdo_init_capa_ctxt(env, next,
2488 m->mdt_opts.mo_mds_capa,
2489 m->mdt_capa_timeout,
2495 * Invoke handler for this request opc. Also do necessary preprocessing
2496 * (according to handler ->mh_flags), and post-processing (setting of
2497 * ->last_{xid,committed}).
2499 static int mdt_req_handle(struct mdt_thread_info *info,
2500 struct mdt_handler *h, struct ptlrpc_request *req)
2502 int rc, serious = 0;
2507 LASSERT(h->mh_act != NULL);
2508 LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2509 LASSERT(current->journal_info == NULL);
2512 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2513 * to put same checks into handlers like mdt_close(), mdt_reint(),
2514 * etc., without talking to mdt authors first. Checking same thing
2515 * there again is useless and returning 0 error wihtout packing reply
2516 * is buggy! Handlers either pack reply or return error.
2518 * We return 0 here and do not send any reply in order to emulate
2519 * network failure. Do not send any reply in case any of NET related
2520 * fail_id has occured.
2522 if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2526 flags = h->mh_flags;
2527 LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2529 if (h->mh_fmt != NULL) {
2530 req_capsule_set(info->mti_pill, h->mh_fmt);
2531 rc = mdt_unpack_req_pack_rep(info, flags);
2534 if (rc == 0 && flags & MUTABOR &&
2535 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2536 /* should it be rq_status? */
2539 if (rc == 0 && flags & HABEO_CLAVIS) {
2540 struct ldlm_request *dlm_req;
2542 LASSERT(h->mh_fmt != NULL);
2544 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2545 if (dlm_req != NULL) {
2546 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2547 rc = mdt_lock_resname_compat(info->mti_mdt,
2549 info->mti_dlm_req = dlm_req;
2555 /* capability setting changed via /proc, needs reinitialize ctxt */
2556 if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2557 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2558 info->mti_mdt->mdt_capa_conf = 0;
2561 if (likely(rc == 0)) {
2563 * Process request, there can be two types of rc:
2564 * 1) errors with msg unpack/pack, other failures outside the
2565 * operation itself. This is counted as serious errors;
2566 * 2) errors during fs operation, should be placed in rq_status
2569 rc = h->mh_act(info);
2571 !req->rq_no_reply && req->rq_reply_state == NULL) {
2572 DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2573 "pack reply and returned 0 error\n",
2577 serious = is_serious(rc);
2578 rc = clear_serious(rc);
2582 req->rq_status = rc;
2585 * ELDLM_* codes which > 0 should be in rq_status only as well as
2586 * all non-serious errors.
2588 if (rc > 0 || !serious)
2591 LASSERT(current->journal_info == NULL);
2593 if (rc == 0 && (flags & HABEO_CLAVIS) &&
2594 info->mti_mdt->mdt_opts.mo_compat_resname) {
2595 struct ldlm_reply *dlmrep;
2597 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2599 rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2602 /* If we're DISCONNECTing, the mdt_export_data is already freed */
2603 if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2604 target_committed_to_req(req);
2606 if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2607 lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2608 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2612 target_send_reply(req, rc, info->mti_fail_id);
2616 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2618 lh->mlh_type = MDT_NUL_LOCK;
2619 lh->mlh_reg_lh.cookie = 0ull;
2620 lh->mlh_reg_mode = LCK_MINMODE;
2621 lh->mlh_pdo_lh.cookie = 0ull;
2622 lh->mlh_pdo_mode = LCK_MINMODE;
2625 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2627 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2628 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2632 * Initialize fields of struct mdt_thread_info. Other fields are left in
2633 * uninitialized state, because it's too expensive to zero out whole
2634 * mdt_thread_info (> 1K) on each request arrival.
2636 static void mdt_thread_info_init(struct ptlrpc_request *req,
2637 struct mdt_thread_info *info)
2640 struct md_capainfo *ci;
2642 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2643 info->mti_pill = &req->rq_pill;
2646 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2647 mdt_lock_handle_init(&info->mti_lh[i]);
2649 /* mdt device: it can be NULL while CONNECT */
2650 if (req->rq_export) {
2651 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2652 info->mti_exp = req->rq_export;
2654 info->mti_mdt = NULL;
2655 info->mti_env = req->rq_svc_thread->t_env;
2656 ci = md_capainfo(info->mti_env);
2657 memset(ci, 0, sizeof *ci);
2658 if (req->rq_export) {
2659 if (exp_connect_rmtclient(req->rq_export))
2660 ci->mc_auth = LC_ID_CONVERT;
2661 else if (req->rq_export->exp_connect_flags &
2662 OBD_CONNECT_MDS_CAPA)
2663 ci->mc_auth = LC_ID_PLAIN;
2665 ci->mc_auth = LC_ID_NONE;
2668 info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2669 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2670 info->mti_mos[0] = NULL;
2671 info->mti_mos[1] = NULL;
2672 info->mti_mos[2] = NULL;
2673 info->mti_mos[3] = NULL;
2675 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2676 info->mti_body = NULL;
2677 info->mti_object = NULL;
2678 info->mti_dlm_req = NULL;
2679 info->mti_has_trans = 0;
2680 info->mti_no_need_trans = 0;
2681 info->mti_cross_ref = 0;
2682 info->mti_opdata = 0;
2684 /* To not check for split by default. */
2685 info->mti_spec.sp_ck_split = 0;
2686 info->mti_spec.no_create = 0;
2689 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2693 req_capsule_fini(info->mti_pill);
2694 if (info->mti_object != NULL) {
2696 * freeing an object may lead to OSD level transaction, do not
2697 * let it mess with MDT. bz19385.
2699 info->mti_no_need_trans = 1;
2700 mdt_object_put(info->mti_env, info->mti_object);
2701 info->mti_object = NULL;
2703 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2704 mdt_lock_handle_fini(&info->mti_lh[i]);
2705 info->mti_env = NULL;
2708 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2709 struct obd_device *obd, int *process)
2711 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2712 case MDS_CONNECT: /* This will never get here, but for completeness. */
2713 case OST_CONNECT: /* This will never get here, but for completeness. */
2714 case MDS_DISCONNECT:
2715 case OST_DISCONNECT:
2720 case MDS_DONE_WRITING:
2721 case MDS_SYNC: /* used in unmounting */
2727 *process = target_queue_recovery_request(req, obd);
2731 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2738 * Handle recovery. Return:
2739 * +1: continue request processing;
2740 * -ve: abort immediately with the given error code;
2741 * 0: send reply with error code in req->rq_status;
2743 static int mdt_recovery(struct mdt_thread_info *info)
2745 struct ptlrpc_request *req = mdt_info_req(info);
2747 struct obd_device *obd;
2751 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2754 case SEC_CTX_INIT_CONT:
2760 rc = mdt_handle_idmap(info);
2769 if (unlikely(req->rq_export == NULL)) {
2770 CERROR("operation %d on unconnected MDS from %s\n",
2771 lustre_msg_get_opc(req->rq_reqmsg),
2772 libcfs_id2str(req->rq_peer));
2773 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2774 * mds_A will get -ENOTCONN(especially for ping req),
2775 * which will cause that mds_A deactive timeout, then when
2776 * mds_A cleanup, the cleanup process will be suspended since
2777 * deactive timeout is not zero.
2779 req->rq_status = -ENOTCONN;
2780 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2784 /* sanity check: if the xid matches, the request must be marked as a
2785 * resent or replayed */
2786 if (req_xid_is_last(req)) {
2787 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2788 (MSG_RESENT | MSG_REPLAY))) {
2789 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2790 "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2791 lustre_msg_get_flags(req->rq_reqmsg));
2793 req->rq_status = -ENOTCONN;
2798 /* else: note the opposite is not always true; a RESENT req after a
2799 * failover will usually not match the last_xid, since it was likely
2800 * never committed. A REPLAYed request will almost never match the
2801 * last xid, however it could for a committed, but still retained,
2804 obd = req->rq_export->exp_obd;
2806 /* Check for aborted recovery... */
2807 spin_lock_bh(&obd->obd_processing_task_lock);
2808 recovering = obd->obd_recovering;
2809 spin_unlock_bh(&obd->obd_processing_task_lock);
2810 if (unlikely(recovering)) {
2813 DEBUG_REQ(D_INFO, req, "Got new replay");
2814 rc = mdt_filter_recovery_request(req, obd, &should_process);
2815 if (rc != 0 || !should_process)
2817 else if (should_process < 0) {
2818 req->rq_status = should_process;
2819 rc = ptlrpc_error(req);
2826 static int mdt_msg_check_version(struct lustre_msg *msg)
2830 switch (lustre_msg_get_opc(msg)) {
2832 case MDS_DISCONNECT:
2835 case SEC_CTX_INIT_CONT:
2837 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2839 CERROR("bad opc %u version %08x, expecting %08x\n",
2840 lustre_msg_get_opc(msg),
2841 lustre_msg_get_version(msg),
2842 LUSTRE_OBD_VERSION);
2846 case MDS_GETATTR_NAME:
2853 case MDS_DONE_WRITING:
2860 case MDS_QUOTACHECK:
2866 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2868 CERROR("bad opc %u version %08x, expecting %08x\n",
2869 lustre_msg_get_opc(msg),
2870 lustre_msg_get_version(msg),
2871 LUSTRE_MDS_VERSION);
2875 case LDLM_BL_CALLBACK:
2876 case LDLM_CP_CALLBACK:
2877 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2879 CERROR("bad opc %u version %08x, expecting %08x\n",
2880 lustre_msg_get_opc(msg),
2881 lustre_msg_get_version(msg),
2882 LUSTRE_DLM_VERSION);
2884 case OBD_LOG_CANCEL:
2885 case LLOG_ORIGIN_HANDLE_CREATE:
2886 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2887 case LLOG_ORIGIN_HANDLE_READ_HEADER:
2888 case LLOG_ORIGIN_HANDLE_CLOSE:
2889 case LLOG_ORIGIN_HANDLE_DESTROY:
2890 case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2892 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2894 CERROR("bad opc %u version %08x, expecting %08x\n",
2895 lustre_msg_get_opc(msg),
2896 lustre_msg_get_version(msg),
2897 LUSTRE_LOG_VERSION);
2900 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2906 static int mdt_handle0(struct ptlrpc_request *req,
2907 struct mdt_thread_info *info,
2908 struct mdt_opc_slice *supported)
2910 struct mdt_handler *h;
2911 struct lustre_msg *msg;
2916 if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2919 LASSERT(current->journal_info == NULL);
2921 msg = req->rq_reqmsg;
2922 rc = mdt_msg_check_version(msg);
2923 if (likely(rc == 0)) {
2924 rc = mdt_recovery(info);
2925 if (likely(rc == +1)) {
2926 switch (lustre_msg_get_opc(msg)) {
2928 req->rq_bulk_read = 1;
2931 req->rq_bulk_write = 1;
2935 h = mdt_handler_find(lustre_msg_get_opc(msg),
2937 if (likely(h != NULL)) {
2938 rc = mdt_req_handle(info, h, req);
2940 CERROR("The unsupported opc: 0x%x\n",
2941 lustre_msg_get_opc(msg) );
2942 req->rq_status = -ENOTSUPP;
2943 rc = ptlrpc_error(req);
2948 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2953 * MDT handler function called by ptlrpc service thread when request comes.
2955 * XXX common "target" functionality should be factored into separate module
2956 * shared by mdt, ost and stand-alone services like fld.
2958 static int mdt_handle_common(struct ptlrpc_request *req,
2959 struct mdt_opc_slice *supported)
2962 struct mdt_thread_info *info;
2966 env = req->rq_svc_thread->t_env;
2967 LASSERT(env != NULL);
2968 LASSERT(env->le_ses != NULL);
2969 LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2970 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2971 LASSERT(info != NULL);
2973 mdt_thread_info_init(req, info);
2975 rc = mdt_handle0(req, info, supported);
2977 mdt_thread_info_fini(info);
2982 * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2985 int mdt_recovery_handle(struct ptlrpc_request *req)
2990 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2992 rc = mdt_handle_common(req, mdt_fld_handlers);
2995 rc = mdt_handle_common(req, mdt_seq_handlers);
2998 rc = mdt_handle_common(req, mdt_regular_handlers);
3005 static int mdt_regular_handle(struct ptlrpc_request *req)
3007 return mdt_handle_common(req, mdt_regular_handlers);
3010 static int mdt_readpage_handle(struct ptlrpc_request *req)
3012 return mdt_handle_common(req, mdt_readpage_handlers);
3015 static int mdt_xmds_handle(struct ptlrpc_request *req)
3017 return mdt_handle_common(req, mdt_xmds_handlers);
3020 static int mdt_mdsc_handle(struct ptlrpc_request *req)
3022 return mdt_handle_common(req, mdt_seq_handlers);
3025 static int mdt_mdss_handle(struct ptlrpc_request *req)
3027 return mdt_handle_common(req, mdt_seq_handlers);
3030 static int mdt_dtss_handle(struct ptlrpc_request *req)
3032 return mdt_handle_common(req, mdt_seq_handlers);
3035 static int mdt_fld_handle(struct ptlrpc_request *req)
3037 return mdt_handle_common(req, mdt_fld_handlers);
3053 static int mdt_intent_getattr(enum mdt_it_code opcode,
3054 struct mdt_thread_info *info,
3055 struct ldlm_lock **,
3057 static int mdt_intent_reint(enum mdt_it_code opcode,
3058 struct mdt_thread_info *info,
3059 struct ldlm_lock **,
3062 static struct mdt_it_flavor {
3063 const struct req_format *it_fmt;
3065 int (*it_act)(enum mdt_it_code ,
3066 struct mdt_thread_info *,
3067 struct ldlm_lock **,
3070 } mdt_it_flavor[] = {
3072 .it_fmt = &RQF_LDLM_INTENT,
3073 /*.it_flags = HABEO_REFERO,*/
3075 .it_act = mdt_intent_reint,
3076 .it_reint = REINT_OPEN
3079 .it_fmt = &RQF_LDLM_INTENT,
3080 .it_flags = MUTABOR,
3081 .it_act = mdt_intent_reint,
3082 .it_reint = REINT_OPEN
3085 .it_fmt = &RQF_LDLM_INTENT,
3086 .it_flags = MUTABOR,
3087 .it_act = mdt_intent_reint,
3088 .it_reint = REINT_CREATE
3090 [MDT_IT_GETATTR] = {
3091 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3092 .it_flags = HABEO_REFERO,
3093 .it_act = mdt_intent_getattr
3095 [MDT_IT_READDIR] = {
3101 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3102 .it_flags = HABEO_REFERO,
3103 .it_act = mdt_intent_getattr
3106 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
3107 .it_flags = MUTABOR,
3109 .it_reint = REINT_UNLINK
3113 .it_flags = MUTABOR,
3116 [MDT_IT_GETXATTR] = {
3123 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3124 struct ldlm_lock **lockp,
3125 struct ldlm_lock *new_lock,
3126 struct mdt_lock_handle *lh,
3129 struct ptlrpc_request *req = mdt_info_req(info);
3130 struct ldlm_lock *lock = *lockp;
3133 * Get new lock only for cases when possible resent did not find any
3136 if (new_lock == NULL)
3137 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3139 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3140 lh->mlh_reg_lh.cookie = 0;
3144 LASSERTF(new_lock != NULL,
3145 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3148 * If we've already given this lock to a client once, then we should
3149 * have no readers or writers. Otherwise, we should have one reader
3150 * _or_ writer ref (which will be zeroed below) before returning the
3153 if (new_lock->l_export == req->rq_export) {
3154 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3156 LASSERT(new_lock->l_export == NULL);
3157 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3162 if (new_lock->l_export == req->rq_export) {
3164 * Already gave this to the client, which means that we
3165 * reconstructed a reply.
3167 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3169 lh->mlh_reg_lh.cookie = 0;
3170 RETURN(ELDLM_LOCK_REPLACED);
3174 * Fixup the lock to be given to the client.
3176 lock_res_and_lock(new_lock);
3177 /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3178 * possible blocking AST. */
3179 while (new_lock->l_readers > 0) {
3180 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3181 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3182 new_lock->l_readers--;
3184 while (new_lock->l_writers > 0) {
3185 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3186 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3187 new_lock->l_writers--;
3190 new_lock->l_export = class_export_get(req->rq_export);
3191 new_lock->l_blocking_ast = lock->l_blocking_ast;
3192 new_lock->l_completion_ast = lock->l_completion_ast;
3193 new_lock->l_remote_handle = lock->l_remote_handle;
3194 new_lock->l_flags &= ~LDLM_FL_LOCAL;
3196 unlock_res_and_lock(new_lock);
3198 lustre_hash_add(new_lock->l_export->exp_lock_hash,
3199 &new_lock->l_remote_handle,
3200 &new_lock->l_exp_hash);
3202 LDLM_LOCK_RELEASE(new_lock);
3203 lh->mlh_reg_lh.cookie = 0;
3205 RETURN(ELDLM_LOCK_REPLACED);
3208 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3209 struct ldlm_lock *new_lock,
3210 struct ldlm_lock **old_lock,
3211 struct mdt_lock_handle *lh)
3213 struct ptlrpc_request *req = mdt_info_req(info);
3214 struct obd_export *exp = req->rq_export;
3215 struct lustre_handle remote_hdl;
3216 struct ldlm_request *dlmreq;
3217 struct ldlm_lock *lock;
3219 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3222 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3223 remote_hdl = dlmreq->lock_handle[0];
3225 lock = lustre_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3227 if (lock != new_lock) {
3228 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3229 lh->mlh_reg_mode = lock->l_granted_mode;
3231 LDLM_DEBUG(lock, "Restoring lock cookie");
3232 DEBUG_REQ(D_DLMTRACE, req,
3233 "restoring lock cookie "LPX64,
3234 lh->mlh_reg_lh.cookie);
3236 *old_lock = LDLM_LOCK_GET(lock);
3237 lh_put(exp->exp_lock_hash, &lock->l_exp_hash);
3241 lh_put(exp->exp_lock_hash, &lock->l_exp_hash);
3245 * If the xid matches, then we know this is a resent request, and allow
3246 * it. (It's probably an OPEN, for which we don't send a lock.
3248 if (req_xid_is_last(req))
3252 * This remote handle isn't enqueued, so we never received or processed
3253 * this request. Clear MSG_RESENT, because it can be handled like any
3254 * normal request now.
3256 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3258 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3262 static int mdt_intent_getattr(enum mdt_it_code opcode,
3263 struct mdt_thread_info *info,
3264 struct ldlm_lock **lockp,
3267 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3268 struct ldlm_lock *new_lock = NULL;
3270 struct ldlm_reply *ldlm_rep;
3271 struct ptlrpc_request *req;
3272 struct mdt_body *reqbody;
3273 struct mdt_body *repbody;
3277 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3280 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3283 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
3284 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3285 repbody->eadatasize = 0;
3286 repbody->aclsize = 0;
3290 child_bits = MDS_INODELOCK_LOOKUP;
3292 case MDT_IT_GETATTR:
3293 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3296 CERROR("Unhandled till now");
3297 GOTO(out_shrink, rc = -EINVAL);
3300 rc = mdt_init_ucred(info, reqbody);
3302 GOTO(out_shrink, rc);
3304 req = info->mti_pill->rc_req;
3305 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3306 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3308 /* Get lock from request for possible resent case. */
3309 mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3311 ldlm_rep->lock_policy_res2 =
3312 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3314 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3315 ldlm_rep->lock_policy_res2 = 0;
3316 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3317 ldlm_rep->lock_policy_res2) {
3318 lhc->mlh_reg_lh.cookie = 0ull;
3319 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3322 rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3325 mdt_exit_ucred(info);
3327 mdt_shrink_reply(info);
3331 static int mdt_intent_reint(enum mdt_it_code opcode,
3332 struct mdt_thread_info *info,
3333 struct ldlm_lock **lockp,
3336 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3337 struct ldlm_reply *rep = NULL;
3341 static const struct req_format *intent_fmts[REINT_MAX] = {
3342 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
3343 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
3348 opc = mdt_reint_opcode(info, intent_fmts);
3352 if (mdt_it_flavor[opcode].it_reint != opc) {
3353 CERROR("Reint code %ld doesn't match intent: %d\n",
3355 RETURN(err_serious(-EPROTO));
3358 /* Get lock from request for possible resent case. */
3359 mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3361 rc = mdt_reint_internal(info, lhc, opc);
3363 /* Check whether the reply has been packed successfully. */
3364 if (mdt_info_req(info)->rq_repmsg != NULL)
3365 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3367 RETURN(err_serious(-EFAULT));
3369 /* MDC expects this in any case */
3371 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3373 /* Cross-ref case, the lock should be returned to the client */
3374 if (rc == -EREMOTE) {
3375 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3376 rep->lock_policy_res2 = 0;
3377 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3380 rep->lock_policy_res2 = clear_serious(rc);
3382 if (rc == -ENOTCONN || rc == -ENODEV ||
3383 rc == -EOVERFLOW) { /**< if VBR failure then return error */
3385 * If it is the disconnect error (ENODEV & ENOCONN), the error
3386 * will be returned by rq_status, and client at ptlrpc layer
3387 * will detect this, then disconnect, reconnect the import
3388 * immediately, instead of impacting the following the rpc.
3390 lhc->mlh_reg_lh.cookie = 0ull;
3394 * For other cases, the error will be returned by intent.
3395 * and client will retrieve the result from intent.
3398 * FIXME: when open lock is finished, that should be
3401 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3402 rep->lock_policy_res2 = 0;
3403 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3406 lhc->mlh_reg_lh.cookie = 0ull;
3407 RETURN(ELDLM_LOCK_ABORTED);
3412 static int mdt_intent_code(long itcode)
3420 case IT_OPEN|IT_CREAT:
3427 rc = MDT_IT_READDIR;
3430 rc = MDT_IT_GETATTR;
3442 rc = MDT_IT_GETXATTR;
3445 CERROR("Unknown intent opcode: %ld\n", itcode);
3452 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3453 struct ldlm_lock **lockp, int flags)
3455 struct req_capsule *pill;
3456 struct mdt_it_flavor *flv;
3461 opc = mdt_intent_code(itopc);
3465 pill = info->mti_pill;
3466 flv = &mdt_it_flavor[opc];
3468 if (flv->it_fmt != NULL)
3469 req_capsule_extend(pill, flv->it_fmt);
3471 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3473 struct ptlrpc_request *req = mdt_info_req(info);
3474 if (flv->it_flags & MUTABOR &&
3475 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3478 if (rc == 0 && flv->it_act != NULL) {
3479 /* execute policy */
3480 rc = flv->it_act(opc, info, lockp, flags);
3487 static int mdt_intent_policy(struct ldlm_namespace *ns,
3488 struct ldlm_lock **lockp, void *req_cookie,
3489 ldlm_mode_t mode, int flags, void *data)
3491 struct mdt_thread_info *info;
3492 struct ptlrpc_request *req = req_cookie;
3493 struct ldlm_intent *it;
3494 struct req_capsule *pill;
3499 LASSERT(req != NULL);
3501 info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3503 LASSERT(info != NULL);
3504 pill = info->mti_pill;
3505 LASSERT(pill->rc_req == req);
3507 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3508 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3509 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3511 const struct ldlm_request *dlmreq;
3514 rc = mdt_intent_opc(it->opc, info, lockp, flags);
3519 * Lock without inodebits makes no sense and will oops
3520 * later in ldlm. Let's check it now to see if we have
3521 * wrong lock from client or bits get corrupted
3522 * somewhere in mdt_intent_opc().
3524 dlmreq = info->mti_dlm_req;
3525 req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
3526 LASSERT(req_bits != 0);
3529 rc = err_serious(-EFAULT);
3531 /* No intent was provided */
3532 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3533 rc = req_capsule_server_pack(pill);
3535 rc = err_serious(rc);
3543 static void mdt_seq_adjust(const struct lu_env *env,
3544 struct mdt_device *m, int lost)
3546 struct md_site *ms = mdt_md_site(m);
3547 struct lu_seq_range out;
3550 LASSERT(ms && ms->ms_server_seq);
3552 /* get extra seq from seq_server, moving it's range up */
3553 while (lost-- > 0) {
3554 seq_server_alloc_meta(ms->ms_server_seq, NULL, &out, env);
3559 static int mdt_seq_fini(const struct lu_env *env,
3560 struct mdt_device *m)
3562 struct md_site *ms = mdt_md_site(m);
3566 if (ms->ms_server_seq) {
3567 seq_server_fini(ms->ms_server_seq, env);
3568 OBD_FREE_PTR(ms->ms_server_seq);
3569 ms->ms_server_seq = NULL;
3572 if (ms->ms_control_seq) {
3573 seq_server_fini(ms->ms_control_seq, env);
3574 OBD_FREE_PTR(ms->ms_control_seq);
3575 ms->ms_control_seq = NULL;
3578 if (ms->ms_client_seq) {
3579 seq_client_fini(ms->ms_client_seq);
3580 OBD_FREE_PTR(ms->ms_client_seq);
3581 ms->ms_client_seq = NULL;
3588 static int mdt_seq_init(const struct lu_env *env,
3590 struct mdt_device *m)
3597 ms = mdt_md_site(m);
3600 * This is sequence-controller node. Init seq-controller server on local
3603 if (ms->ms_node_id == 0) {
3604 LASSERT(ms->ms_control_seq == NULL);
3606 OBD_ALLOC_PTR(ms->ms_control_seq);
3607 if (ms->ms_control_seq == NULL)
3610 rc = seq_server_init(ms->ms_control_seq,
3611 m->mdt_bottom, uuid,
3612 LUSTRE_SEQ_CONTROLLER,
3617 GOTO(out_seq_fini, rc);
3619 OBD_ALLOC_PTR(ms->ms_client_seq);
3620 if (ms->ms_client_seq == NULL)
3621 GOTO(out_seq_fini, rc = -ENOMEM);
3623 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3624 if (prefix == NULL) {
3625 OBD_FREE_PTR(ms->ms_client_seq);
3626 GOTO(out_seq_fini, rc = -ENOMEM);
3629 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3633 * Init seq-controller client after seq-controller server is
3634 * ready. Pass ms->ms_control_seq to it for direct talking.
3636 rc = seq_client_init(ms->ms_client_seq, NULL,
3637 LUSTRE_SEQ_METADATA, prefix,
3638 ms->ms_control_seq);
3639 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3642 GOTO(out_seq_fini, rc);
3645 /* Init seq-server on local MDT */
3646 LASSERT(ms->ms_server_seq == NULL);
3648 OBD_ALLOC_PTR(ms->ms_server_seq);
3649 if (ms->ms_server_seq == NULL)
3650 GOTO(out_seq_fini, rc = -ENOMEM);
3652 rc = seq_server_init(ms->ms_server_seq,
3653 m->mdt_bottom, uuid,
3658 GOTO(out_seq_fini, rc = -ENOMEM);
3660 /* Assign seq-controller client to local seq-server. */
3661 if (ms->ms_node_id == 0) {
3662 LASSERT(ms->ms_client_seq != NULL);
3664 rc = seq_server_set_cli(ms->ms_server_seq,
3672 mdt_seq_fini(env, m);
3677 * Init client sequence manager which is used by local MDS to talk to sequence
3678 * controller on remote node.
3680 static int mdt_seq_init_cli(const struct lu_env *env,
3681 struct mdt_device *m,
3682 struct lustre_cfg *cfg)
3684 struct md_site *ms = mdt_md_site(m);
3685 struct obd_device *mdc;
3686 struct obd_uuid *uuidp, *mdcuuidp;
3687 char *uuid_str, *mdc_uuid_str;
3690 struct mdt_thread_info *info;
3691 char *p, *index_string = lustre_cfg_string(cfg, 2);
3694 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3695 uuidp = &info->mti_u.uuid[0];
3696 mdcuuidp = &info->mti_u.uuid[1];
3698 LASSERT(index_string);
3700 index = simple_strtol(index_string, &p, 10);
3702 CERROR("Invalid index in lustre_cgf, offset 2\n");
3706 /* check if this is adding the first MDC and controller is not yet
3708 if (index != 0 || ms->ms_client_seq)
3711 uuid_str = lustre_cfg_string(cfg, 1);
3712 mdc_uuid_str = lustre_cfg_string(cfg, 4);
3713 obd_str2uuid(uuidp, uuid_str);
3714 obd_str2uuid(mdcuuidp, mdc_uuid_str);
3716 mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3718 CERROR("can't find controller MDC by uuid %s\n",
3721 } else if (!mdc->obd_set_up) {
3722 CERROR("target %s not set up\n", mdc->obd_name);
3725 LASSERT(ms->ms_control_exp);
3726 OBD_ALLOC_PTR(ms->ms_client_seq);
3727 if (ms->ms_client_seq != NULL) {
3730 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3734 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3737 rc = seq_client_init(ms->ms_client_seq,
3739 LUSTRE_SEQ_METADATA,
3741 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3748 LASSERT(ms->ms_server_seq != NULL);
3749 rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
3756 static void mdt_seq_fini_cli(struct mdt_device *m)
3762 ms = mdt_md_site(m);
3765 if (ms->ms_server_seq)
3766 seq_server_set_cli(ms->ms_server_seq,
3769 if (ms->ms_control_exp) {
3770 class_export_put(ms->ms_control_exp);
3771 ms->ms_control_exp = NULL;
3780 static int mdt_fld_fini(const struct lu_env *env,
3781 struct mdt_device *m)
3783 struct md_site *ms = mdt_md_site(m);
3786 if (ms && ms->ms_server_fld) {
3787 fld_server_fini(ms->ms_server_fld, env);
3788 OBD_FREE_PTR(ms->ms_server_fld);
3789 ms->ms_server_fld = NULL;
3795 static int mdt_fld_init(const struct lu_env *env,
3797 struct mdt_device *m)
3803 ms = mdt_md_site(m);
3805 OBD_ALLOC_PTR(ms->ms_server_fld);
3806 if (ms->ms_server_fld == NULL)
3807 RETURN(rc = -ENOMEM);
3809 rc = fld_server_init(ms->ms_server_fld,
3810 m->mdt_bottom, uuid,
3811 env, ms->ms_node_id);
3813 OBD_FREE_PTR(ms->ms_server_fld);
3814 ms->ms_server_fld = NULL;
3821 /* device init/fini methods */
3822 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3825 if (m->mdt_regular_service != NULL) {
3826 ptlrpc_unregister_service(m->mdt_regular_service);
3827 m->mdt_regular_service = NULL;
3829 if (m->mdt_readpage_service != NULL) {
3830 ptlrpc_unregister_service(m->mdt_readpage_service);
3831 m->mdt_readpage_service = NULL;
3833 if (m->mdt_xmds_service != NULL) {
3834 ptlrpc_unregister_service(m->mdt_xmds_service);
3835 m->mdt_xmds_service = NULL;
3837 if (m->mdt_setattr_service != NULL) {
3838 ptlrpc_unregister_service(m->mdt_setattr_service);
3839 m->mdt_setattr_service = NULL;
3841 if (m->mdt_mdsc_service != NULL) {
3842 ptlrpc_unregister_service(m->mdt_mdsc_service);
3843 m->mdt_mdsc_service = NULL;
3845 if (m->mdt_mdss_service != NULL) {
3846 ptlrpc_unregister_service(m->mdt_mdss_service);
3847 m->mdt_mdss_service = NULL;
3849 if (m->mdt_dtss_service != NULL) {
3850 ptlrpc_unregister_service(m->mdt_dtss_service);
3851 m->mdt_dtss_service = NULL;
3853 if (m->mdt_fld_service != NULL) {
3854 ptlrpc_unregister_service(m->mdt_fld_service);
3855 m->mdt_fld_service = NULL;
3860 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3863 static struct ptlrpc_service_conf conf;
3864 cfs_proc_dir_entry_t *procfs_entry;
3867 procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3869 conf = (typeof(conf)) {
3870 .psc_nbufs = MDS_NBUFS,
3871 .psc_bufsize = MDS_BUFSIZE,
3872 .psc_max_req_size = MDS_MAXREQSIZE,
3873 .psc_max_reply_size = MDS_MAXREPSIZE,
3874 .psc_req_portal = MDS_REQUEST_PORTAL,
3875 .psc_rep_portal = MDC_REPLY_PORTAL,
3876 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3878 * We'd like to have a mechanism to set this on a per-device
3879 * basis, but alas...
3881 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3883 .psc_max_threads = MDT_MAX_THREADS,
3884 .psc_ctx_tags = LCT_MD_THREAD
3887 m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3888 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3889 "mdt_ldlm_client", m->mdt_ldlm_client);
3891 m->mdt_regular_service =
3892 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3893 procfs_entry, target_print_req,
3895 if (m->mdt_regular_service == NULL)
3898 rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3900 GOTO(err_mdt_svc, rc);
3903 * readpage service configuration. Parameters have to be adjusted,
3906 conf = (typeof(conf)) {
3907 .psc_nbufs = MDS_NBUFS,
3908 .psc_bufsize = MDS_BUFSIZE,
3909 .psc_max_req_size = MDS_MAXREQSIZE,
3910 .psc_max_reply_size = MDS_MAXREPSIZE,
3911 .psc_req_portal = MDS_READPAGE_PORTAL,
3912 .psc_rep_portal = MDC_REPLY_PORTAL,
3913 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3914 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3916 .psc_max_threads = MDT_MAX_THREADS,
3917 .psc_ctx_tags = LCT_MD_THREAD
3919 m->mdt_readpage_service =
3920 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3921 LUSTRE_MDT_NAME "_readpage",
3922 procfs_entry, target_print_req,"mdt_rdpg");
3924 if (m->mdt_readpage_service == NULL) {
3925 CERROR("failed to start readpage service\n");
3926 GOTO(err_mdt_svc, rc = -ENOMEM);
3929 rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3932 * setattr service configuration.
3934 conf = (typeof(conf)) {
3935 .psc_nbufs = MDS_NBUFS,
3936 .psc_bufsize = MDS_BUFSIZE,
3937 .psc_max_req_size = MDS_MAXREQSIZE,
3938 .psc_max_reply_size = MDS_MAXREPSIZE,
3939 .psc_req_portal = MDS_SETATTR_PORTAL,
3940 .psc_rep_portal = MDC_REPLY_PORTAL,
3941 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3942 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3944 .psc_max_threads = MDT_MAX_THREADS,
3945 .psc_ctx_tags = LCT_MD_THREAD
3948 m->mdt_setattr_service =
3949 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3950 LUSTRE_MDT_NAME "_setattr",
3951 procfs_entry, target_print_req,"mdt_attr");
3953 if (!m->mdt_setattr_service) {
3954 CERROR("failed to start setattr service\n");
3955 GOTO(err_mdt_svc, rc = -ENOMEM);
3958 rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3960 GOTO(err_mdt_svc, rc);
3963 * sequence controller service configuration
3965 conf = (typeof(conf)) {
3966 .psc_nbufs = MDS_NBUFS,
3967 .psc_bufsize = MDS_BUFSIZE,
3968 .psc_max_req_size = SEQ_MAXREQSIZE,
3969 .psc_max_reply_size = SEQ_MAXREPSIZE,
3970 .psc_req_portal = SEQ_CONTROLLER_PORTAL,
3971 .psc_rep_portal = MDC_REPLY_PORTAL,
3972 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3973 .psc_min_threads = SEQ_NUM_THREADS,
3974 .psc_max_threads = SEQ_NUM_THREADS,
3975 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3978 m->mdt_mdsc_service =
3979 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3980 LUSTRE_MDT_NAME"_mdsc",
3981 procfs_entry, target_print_req,"mdt_mdsc");
3982 if (!m->mdt_mdsc_service) {
3983 CERROR("failed to start seq controller service\n");
3984 GOTO(err_mdt_svc, rc = -ENOMEM);
3987 rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3989 GOTO(err_mdt_svc, rc);
3992 * metadata sequence server service configuration
3994 conf = (typeof(conf)) {
3995 .psc_nbufs = MDS_NBUFS,
3996 .psc_bufsize = MDS_BUFSIZE,
3997 .psc_max_req_size = SEQ_MAXREQSIZE,
3998 .psc_max_reply_size = SEQ_MAXREPSIZE,
3999 .psc_req_portal = SEQ_METADATA_PORTAL,
4000 .psc_rep_portal = MDC_REPLY_PORTAL,
4001 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4002 .psc_min_threads = SEQ_NUM_THREADS,
4003 .psc_max_threads = SEQ_NUM_THREADS,
4004 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
4007 m->mdt_mdss_service =
4008 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
4009 LUSTRE_MDT_NAME"_mdss",
4010 procfs_entry, target_print_req,"mdt_mdss");
4011 if (!m->mdt_mdss_service) {
4012 CERROR("failed to start metadata seq server service\n");
4013 GOTO(err_mdt_svc, rc = -ENOMEM);
4016 rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
4018 GOTO(err_mdt_svc, rc);
4022 * Data sequence server service configuration. We want to have really
4023 * cluster-wide sequences space. This is why we start only one sequence
4024 * controller which manages space.
4026 conf = (typeof(conf)) {
4027 .psc_nbufs = MDS_NBUFS,
4028 .psc_bufsize = MDS_BUFSIZE,
4029 .psc_max_req_size = SEQ_MAXREQSIZE,
4030 .psc_max_reply_size = SEQ_MAXREPSIZE,
4031 .psc_req_portal = SEQ_DATA_PORTAL,
4032 .psc_rep_portal = OSC_REPLY_PORTAL,
4033 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4034 .psc_min_threads = SEQ_NUM_THREADS,
4035 .psc_max_threads = SEQ_NUM_THREADS,
4036 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
4039 m->mdt_dtss_service =
4040 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
4041 LUSTRE_MDT_NAME"_dtss",
4042 procfs_entry, target_print_req,"mdt_dtss");
4043 if (!m->mdt_dtss_service) {
4044 CERROR("failed to start data seq server service\n");
4045 GOTO(err_mdt_svc, rc = -ENOMEM);
4048 rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
4050 GOTO(err_mdt_svc, rc);
4052 /* FLD service start */
4053 conf = (typeof(conf)) {
4054 .psc_nbufs = MDS_NBUFS,
4055 .psc_bufsize = MDS_BUFSIZE,
4056 .psc_max_req_size = FLD_MAXREQSIZE,
4057 .psc_max_reply_size = FLD_MAXREPSIZE,
4058 .psc_req_portal = FLD_REQUEST_PORTAL,
4059 .psc_rep_portal = MDC_REPLY_PORTAL,
4060 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4061 .psc_min_threads = FLD_NUM_THREADS,
4062 .psc_max_threads = FLD_NUM_THREADS,
4063 .psc_ctx_tags = LCT_DT_THREAD|LCT_MD_THREAD
4066 m->mdt_fld_service =
4067 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
4068 LUSTRE_MDT_NAME"_fld",
4069 procfs_entry, target_print_req, "mdt_fld");
4070 if (!m->mdt_fld_service) {
4071 CERROR("failed to start fld service\n");
4072 GOTO(err_mdt_svc, rc = -ENOMEM);
4075 rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
4077 GOTO(err_mdt_svc, rc);
4080 * mds-mds service configuration. Separate portal is used to allow
4081 * mds-mds requests be not blocked during recovery.
4083 conf = (typeof(conf)) {
4084 .psc_nbufs = MDS_NBUFS,
4085 .psc_bufsize = MDS_BUFSIZE,
4086 .psc_max_req_size = MDS_MAXREQSIZE,
4087 .psc_max_reply_size = MDS_MAXREPSIZE,
4088 .psc_req_portal = MDS_MDS_PORTAL,
4089 .psc_rep_portal = MDC_REPLY_PORTAL,
4090 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4091 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
4093 .psc_max_threads = MDT_MAX_THREADS,
4094 .psc_ctx_tags = LCT_MD_THREAD
4096 m->mdt_xmds_service =
4097 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
4098 LUSTRE_MDT_NAME "_mds",
4099 procfs_entry, target_print_req,"mdt_xmds");
4101 if (m->mdt_xmds_service == NULL) {
4102 CERROR("failed to start readpage service\n");
4103 GOTO(err_mdt_svc, rc = -ENOMEM);
4106 rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
4108 GOTO(err_mdt_svc, rc);
4113 mdt_stop_ptlrpc_service(m);
4118 static void mdt_stack_fini(const struct lu_env *env,
4119 struct mdt_device *m, struct lu_device *top)
4121 struct obd_device *obd = mdt2obd_dev(m);
4122 struct lustre_cfg_bufs *bufs;
4123 struct lustre_cfg *lcfg;
4124 struct mdt_thread_info *info;
4128 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4129 LASSERT(info != NULL);
4131 bufs = &info->mti_u.bufs;
4132 /* process cleanup, pass mdt obd name to get obd umount flags */
4133 lustre_cfg_bufs_reset(bufs, obd->obd_name);
4138 lustre_cfg_bufs_set_string(bufs, 1, flags);
4139 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4141 CERROR("Cannot alloc lcfg!\n");
4146 top->ld_ops->ldo_process_config(env, top, lcfg);
4147 lustre_cfg_free(lcfg);
4149 lu_stack_fini(env, top);
4150 m->mdt_child = NULL;
4151 m->mdt_bottom = NULL;
4154 static struct lu_device *mdt_layer_setup(struct lu_env *env,
4155 const char *typename,
4156 struct lu_device *child,
4157 struct lustre_cfg *cfg)
4159 const char *dev = lustre_cfg_string(cfg, 0);
4160 struct obd_type *type;
4161 struct lu_device_type *ldt;
4162 struct lu_device *d;
4167 type = class_get_type(typename);
4169 CERROR("Unknown type: '%s'\n", typename);
4170 GOTO(out, rc = -ENODEV);
4173 rc = lu_env_refill((struct lu_env *)env);
4175 CERROR("Failure to refill session: '%d'\n", rc);
4181 CERROR("type: '%s'\n", typename);
4182 GOTO(out_type, rc = -EINVAL);
4185 ldt->ldt_obd_type = type;
4186 d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
4188 CERROR("Cannot allocate device: '%s'\n", typename);
4189 GOTO(out_type, rc = -ENODEV);
4192 LASSERT(child->ld_site);
4193 d->ld_site = child->ld_site;
4196 rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
4198 CERROR("can't init device '%s', rc %d\n", typename, rc);
4199 GOTO(out_alloc, rc);
4202 lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
4207 ldt->ldt_ops->ldto_device_free(env, d);
4210 class_put_type(type);
4215 static int mdt_stack_init(struct lu_env *env,
4216 struct mdt_device *m,
4217 struct lustre_cfg *cfg,
4218 struct lustre_mount_info *lmi)
4220 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
4221 struct lu_device *tmp;
4222 struct md_device *md;
4223 struct lu_device *child_lu_dev;
4227 /* init the stack */
4228 tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
4230 RETURN(PTR_ERR(tmp));
4232 m->mdt_bottom = lu2dt_dev(tmp);
4234 tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
4236 GOTO(out, rc = PTR_ERR(tmp));
4241 tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
4243 GOTO(out, rc = PTR_ERR(tmp));
4246 /*set mdd upcall device*/
4247 md_upcall_dev_set(md, lu2md_dev(d));
4250 /*set cmm upcall device*/
4251 md_upcall_dev_set(md, &m->mdt_md_dev);
4253 m->mdt_child = lu2md_dev(d);
4255 /* process setup config */
4256 tmp = &m->mdt_md_dev.md_lu_dev;
4257 rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
4261 /* initialize local objects */
4262 child_lu_dev = &m->mdt_child->md_lu_dev;
4264 rc = child_lu_dev->ld_ops->ldo_prepare(env,
4265 &m->mdt_md_dev.md_lu_dev,
4268 /* fini from last known good lu_device */
4270 mdt_stack_fini(env, m, d);
4276 * setup CONFIG_ORIG context, used to access local config log.
4277 * this may need to be rewrite as part of llog rewrite for lu-api.
4279 static int mdt_obd_llog_setup(struct obd_device *obd,
4280 struct lustre_sb_info *lsi)
4284 LASSERT(obd->obd_fsops == NULL);
4286 obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
4287 if (IS_ERR(obd->obd_fsops))
4288 return PTR_ERR(obd->obd_fsops);
4290 rc = fsfilt_setup(obd, lsi->lsi_srv_mnt->mnt_sb);
4292 fsfilt_put_ops(obd->obd_fsops);
4296 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
4297 obd->obd_lvfs_ctxt.pwdmnt = lsi->lsi_srv_mnt;
4298 obd->obd_lvfs_ctxt.pwd = lsi->lsi_srv_mnt->mnt_root;
4299 obd->obd_lvfs_ctxt.fs = get_ds();
4301 rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd,
4302 0, NULL, &llog_lvfs_ops);
4304 CERROR("llog_setup() failed: %d\n", rc);
4305 fsfilt_put_ops(obd->obd_fsops);
4311 static void mdt_obd_llog_cleanup(struct obd_device *obd)
4313 struct llog_ctxt *ctxt;
4315 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
4319 if (obd->obd_fsops) {
4320 fsfilt_put_ops(obd->obd_fsops);
4321 obd->obd_fsops = NULL;
4325 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4327 struct md_device *next = m->mdt_child;
4328 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
4329 struct lu_site *ls = d->ld_site;
4330 struct obd_device *obd = mdt2obd_dev(m);
4333 target_recovery_fini(obd);
4335 ping_evictor_stop();
4337 mdt_stop_ptlrpc_service(m);
4338 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4339 mdt_obd_llog_cleanup(obd);
4340 obd_zombie_barrier();
4341 #ifdef HAVE_QUOTA_SUPPORT
4342 next->md_ops->mdo_quota.mqo_cleanup(env, next);
4344 lut_fini(env, &m->mdt_lut);
4345 mdt_fs_cleanup(env, m);
4346 upcall_cache_cleanup(m->mdt_identity_cache);
4347 m->mdt_identity_cache = NULL;
4349 if (m->mdt_namespace != NULL) {
4350 ldlm_namespace_free(m->mdt_namespace, NULL, d->ld_obd->obd_force);
4351 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4354 cfs_free_nidlist(&m->mdt_nosquash_nids);
4355 if (m->mdt_nosquash_str) {
4356 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4357 m->mdt_nosquash_str = NULL;
4358 m->mdt_nosquash_strlen = 0;
4361 mdt_seq_fini(env, m);
4362 mdt_seq_fini_cli(m);
4363 mdt_fld_fini(env, m);
4364 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4366 next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4367 cfs_timer_disarm(&m->mdt_ck_timer);
4368 mdt_ck_thread_stop(m);
4373 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4376 if (obd->obd_proc_exports_entry) {
4377 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
4378 obd->obd_proc_exports_entry = NULL;
4380 lprocfs_free_per_client_stats(obd);
4381 lprocfs_free_obd_stats(obd);
4382 ptlrpc_lprocfs_unregister_obd(obd);
4383 lprocfs_obd_cleanup(obd);
4386 struct md_site *mite;
4389 mite = lu_site2md(ls);
4393 LASSERT(atomic_read(&d->ld_ref) == 0);
4398 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4400 struct mdt_device *m = mdt_dev(obd->obd_lu_dev);
4401 struct sptlrpc_rule_set tmp_rset;
4404 sptlrpc_rule_set_init(&tmp_rset);
4405 rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4407 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4412 sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4414 write_lock(&m->mdt_sptlrpc_lock);
4415 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4416 m->mdt_sptlrpc_rset = tmp_rset;
4417 write_unlock(&m->mdt_sptlrpc_lock);
4422 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
4426 m->mdt_opts.mo_mds_capa = 1;
4427 m->mdt_opts.mo_oss_capa = 1;
4428 #ifdef CONFIG_FS_POSIX_ACL
4429 /* ACLs should be enabled by default (b=13829) */
4430 m->mdt_opts.mo_acl = 1;
4431 LCONSOLE_INFO("Enabling ACL\n");
4433 m->mdt_opts.mo_acl = 0;
4434 LCONSOLE_INFO("Disabling ACL\n");
4443 while (*p && *p != ',')
4447 if ((len == sizeof("user_xattr") - 1) &&
4448 (memcmp(options, "user_xattr", len) == 0)) {
4449 m->mdt_opts.mo_user_xattr = 1;
4450 LCONSOLE_INFO("Enabling user_xattr\n");
4451 } else if ((len == sizeof("nouser_xattr") - 1) &&
4452 (memcmp(options, "nouser_xattr", len) == 0)) {
4453 m->mdt_opts.mo_user_xattr = 0;
4454 LCONSOLE_INFO("Disabling user_xattr\n");
4455 } else if ((len == sizeof("noacl") - 1) &&
4456 (memcmp(options, "noacl", len) == 0)) {
4457 m->mdt_opts.mo_acl = 0;
4458 LCONSOLE_INFO("Disabling ACL\n");
4465 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4467 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4468 struct lu_device_type *ldt, struct lustre_cfg *cfg)
4470 struct lprocfs_static_vars lvars;
4471 struct mdt_thread_info *info;
4472 struct obd_device *obd;
4473 const char *dev = lustre_cfg_string(cfg, 0);
4474 const char *num = lustre_cfg_string(cfg, 2);
4475 struct lustre_mount_info *lmi = NULL;
4476 struct lustre_sb_info *lsi;
4477 struct lustre_disk_data *ldd;
4479 struct md_site *mite;
4480 const char *identity_upcall = "NONE";
4481 #ifdef HAVE_QUOTA_SUPPORT
4482 struct md_device *next;
4488 md_device_init(&m->mdt_md_dev, ldt);
4490 * Environment (env) might be missing mdt_thread_key values at that
4491 * point, if device is allocated when mdt_thread_key is in QUIESCENT
4494 * Usually device allocation path doesn't use module key values, but
4495 * mdt has to do a lot of work here, so allocate key value.
4497 rc = lu_env_refill((struct lu_env *)env);
4501 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4502 LASSERT(info != NULL);
4504 obd = class_name2obd(dev);
4505 LASSERT(obd != NULL);
4507 spin_lock_init(&m->mdt_transno_lock);
4509 m->mdt_max_mdsize = MAX_MD_SIZE;
4510 m->mdt_max_cookiesize = sizeof(struct llog_cookie);
4511 m->mdt_som_conf = 0;
4513 m->mdt_opts.mo_user_xattr = 0;
4514 m->mdt_opts.mo_acl = 0;
4515 m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4516 lmi = server_get_mount_2(dev);
4518 CERROR("Cannot get mount info for %s!\n", dev);
4521 lsi = s2lsi(lmi->lmi_sb);
4522 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
4523 /* CMD is supported only in IAM mode */
4526 node_id = simple_strtol(num, NULL, 10);
4527 if (!(ldd->ldd_flags & LDD_F_IAM_DIR) && node_id) {
4528 CERROR("CMD Operation not allowed in IOP mode\n");
4529 GOTO(err_lmi, rc = -EINVAL);
4533 rwlock_init(&m->mdt_sptlrpc_lock);
4534 sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4536 spin_lock_init(&m->mdt_ioepoch_lock);
4537 m->mdt_opts.mo_compat_resname = 0;
4538 m->mdt_capa_timeout = CAPA_TIMEOUT;
4539 m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4540 m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4541 m->mdt_squash_uid = 0;
4542 m->mdt_squash_gid = 0;
4543 CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4544 m->mdt_nosquash_str = NULL;
4545 m->mdt_nosquash_strlen = 0;
4546 init_rwsem(&m->mdt_squash_sem);
4548 spin_lock_init(&m->mdt_client_bitmap_lock);
4550 OBD_ALLOC_PTR(mite);
4552 GOTO(err_lmi, rc = -ENOMEM);
4556 m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4557 m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4558 /* set this lu_device to obd, because error handling need it */
4559 obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4561 rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
4563 CERROR("Can't init lu_site, rc %d\n", rc);
4564 GOTO(err_free_site, rc);
4567 lprocfs_mdt_init_vars(&lvars);
4568 rc = lprocfs_obd_setup(obd, lvars.obd_vars);
4570 CERROR("Can't init lprocfs, rc %d\n", rc);
4571 GOTO(err_fini_site, rc);
4573 ptlrpc_lprocfs_register_obd(obd);
4575 rc = mdt_procfs_init(m, dev);
4577 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4578 GOTO(err_fini_proc, rc);
4581 obd->obd_proc_exports_entry = proc_mkdir("exports",
4582 obd->obd_proc_entry);
4583 if (obd->obd_proc_exports_entry)
4584 lprocfs_add_simple(obd->obd_proc_exports_entry,
4585 "clear", lprocfs_nid_stats_clear_read,
4586 lprocfs_nid_stats_clear_write, obd, NULL);
4588 /* set server index */
4589 lu_site2md(s)->ms_node_id = node_id;
4591 /* failover is the default
4592 * FIXME: we do not failout mds0/mgs, which may cause some problems.
4593 * assumed whose ms_node_id == 0 XXX
4595 obd->obd_replayable = 1;
4596 /* No connection accepted until configurations will finish */
4597 obd->obd_no_conn = 1;
4599 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4600 char *str = lustre_cfg_string(cfg, 4);
4601 if (strchr(str, 'n')) {
4602 CWARN("%s: recovery disabled\n", obd->obd_name);
4603 obd->obd_replayable = 0;
4607 /* init the stack */
4608 rc = mdt_stack_init((struct lu_env *)env, m, cfg, lmi);
4610 CERROR("Can't init device stack, rc %d\n", rc);
4611 GOTO(err_fini_proc, rc);
4614 rc = lut_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4616 GOTO(err_fini_stack, rc);
4618 rc = mdt_fld_init(env, obd->obd_name, m);
4622 rc = mdt_seq_init(env, obd->obd_name, m);
4624 GOTO(err_fini_fld, rc);
4626 snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4627 LUSTRE_MDT_NAME"-%p", m);
4628 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4629 LDLM_NAMESPACE_SERVER,
4630 LDLM_NAMESPACE_GREEDY);
4631 if (m->mdt_namespace == NULL)
4632 GOTO(err_fini_seq, rc = -ENOMEM);
4634 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4635 /* set obd_namespace for compatibility with old code */
4636 obd->obd_namespace = m->mdt_namespace;
4638 /* XXX: to support suppgid for ACL, we enable identity_upcall
4639 * by default, otherwise, maybe got unexpected -EACCESS. */
4640 if (m->mdt_opts.mo_acl)
4641 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4643 m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
4644 &mdt_identity_upcall_cache_ops);
4645 if (IS_ERR(m->mdt_identity_cache)) {
4646 rc = PTR_ERR(m->mdt_identity_cache);
4647 m->mdt_identity_cache = NULL;
4648 GOTO(err_free_ns, rc);
4651 cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4653 rc = mdt_ck_thread_start(m);
4655 GOTO(err_free_ns, rc);
4657 rc = mdt_fs_setup(env, m, obd, lsi);
4661 rc = mdt_obd_llog_setup(obd, lsi);
4663 GOTO(err_fs_cleanup, rc);
4665 rc = mdt_llog_ctxt_clone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4667 GOTO(err_llog_cleanup, rc);
4669 mdt_adapt_sptlrpc_conf(obd, 1);
4671 #ifdef HAVE_QUOTA_SUPPORT
4672 next = m->mdt_child;
4673 rc = next->md_ops->mdo_quota.mqo_setup(env, next, lmi->lmi_mnt);
4675 GOTO(err_llog_cleanup, rc);
4678 server_put_mount_2(dev, lmi->lmi_mnt);
4681 target_recovery_init(&m->mdt_lut, mdt_recovery_handle);
4683 rc = mdt_start_ptlrpc_service(m);
4685 GOTO(err_recovery, rc);
4687 ping_evictor_start();
4689 rc = lu_site_init_finish(s);
4691 GOTO(err_stop_service, rc);
4693 if (obd->obd_recovering == 0)
4694 mdt_postrecov(env, m);
4696 mdt_init_capa_ctxt(env, m);
4698 /* Reduce the initial timeout on an MDS because it doesn't need such
4699 * a long timeout as an OST does. Adaptive timeouts will adjust this
4700 * value appropriately. */
4701 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4702 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4707 ping_evictor_stop();
4708 mdt_stop_ptlrpc_service(m);
4710 target_recovery_fini(obd);
4711 #ifdef HAVE_QUOTA_SUPPORT
4712 next->md_ops->mdo_quota.mqo_cleanup(env, next);
4715 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4716 mdt_obd_llog_cleanup(obd);
4718 mdt_fs_cleanup(env, m);
4720 cfs_timer_disarm(&m->mdt_ck_timer);
4721 mdt_ck_thread_stop(m);
4723 upcall_cache_cleanup(m->mdt_identity_cache);
4724 m->mdt_identity_cache = NULL;
4725 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4726 obd->obd_namespace = m->mdt_namespace = NULL;
4728 mdt_seq_fini(env, m);
4730 mdt_fld_fini(env, m);
4732 lut_fini(env, &m->mdt_lut);
4734 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4737 if (obd->obd_proc_exports_entry) {
4738 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
4739 obd->obd_proc_exports_entry = NULL;
4741 ptlrpc_lprocfs_unregister_obd(obd);
4742 lprocfs_obd_cleanup(obd);
4749 server_put_mount_2(dev, lmi->lmi_mnt);
4753 /* used by MGS to process specific configurations */
4754 static int mdt_process_config(const struct lu_env *env,
4755 struct lu_device *d, struct lustre_cfg *cfg)
4757 struct mdt_device *m = mdt_dev(d);
4758 struct md_device *md_next = m->mdt_child;
4759 struct lu_device *next = md2lu_dev(md_next);
4763 switch (cfg->lcfg_command) {
4765 struct lprocfs_static_vars lvars;
4766 struct obd_device *obd = d->ld_obd;
4769 * For interoperability between 1.8 and 2.0,
4770 * skip old "mdt.group_upcall" param.
4773 char *param = lustre_cfg_string(cfg, 1);
4774 if (param && !strncmp("mdt.group_upcall", param, 16)) {
4775 CWARN("For 1.8 interoperability, skip this"
4776 " mdt.group_upcall. It is obsolete\n");
4781 lprocfs_mdt_init_vars(&lvars);
4782 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4784 if (rc > 0 || rc == -ENOSYS)
4785 /* we don't understand; pass it on */
4786 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4791 * Add mdc hook to get first MDT uuid and connect it to
4792 * ls->controller to use for seq manager.
4794 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4796 CERROR("Can't add mdc, rc %d\n", rc);
4798 rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4801 /* others are passed further */
4802 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4808 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4809 const struct lu_object_header *hdr,
4810 struct lu_device *d)
4812 struct mdt_object *mo;
4818 struct lu_object *o;
4819 struct lu_object_header *h;
4821 o = &mo->mot_obj.mo_lu;
4822 h = &mo->mot_header;
4823 lu_object_header_init(h);
4824 lu_object_init(o, h, d);
4825 lu_object_add_top(h, o);
4826 o->lo_ops = &mdt_obj_ops;
4832 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4833 const struct lu_object_conf *unused)
4835 struct mdt_device *d = mdt_dev(o->lo_dev);
4836 struct lu_device *under;
4837 struct lu_object *below;
4841 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4842 PFID(lu_object_fid(o)));
4844 under = &d->mdt_child->md_lu_dev;
4845 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4846 if (below != NULL) {
4847 lu_object_add(o, below);
4854 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4856 struct mdt_object *mo = mdt_obj(o);
4857 struct lu_object_header *h;
4861 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4862 PFID(lu_object_fid(o)));
4865 lu_object_header_fini(h);
4870 static int mdt_object_print(const struct lu_env *env, void *cookie,
4871 lu_printer_t p, const struct lu_object *o)
4873 struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4874 return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch=%llu "
4875 "flags=%llx, epochcount=%d, writecount=%d)",
4876 mdto, mdto->mot_ioepoch, mdto->mot_flags,
4877 mdto->mot_epochcount, mdto->mot_writecount);
4880 static const struct lu_device_operations mdt_lu_ops = {
4881 .ldo_object_alloc = mdt_object_alloc,
4882 .ldo_process_config = mdt_process_config,
4885 static const struct lu_object_operations mdt_obj_ops = {
4886 .loo_object_init = mdt_object_init,
4887 .loo_object_free = mdt_object_free,
4888 .loo_object_print = mdt_object_print
4891 static int mdt_obd_set_info_async(struct obd_export *exp,
4892 __u32 keylen, void *key,
4893 __u32 vallen, void *val,
4894 struct ptlrpc_request_set *set)
4896 struct obd_device *obd = exp->exp_obd;
4902 if (KEY_IS(KEY_SPTLRPC_CONF)) {
4903 rc = mdt_adapt_sptlrpc_conf(obd, 0);
4910 /* mds_connect_internal */
4911 static int mdt_connect_internal(struct obd_export *exp,
4912 struct mdt_device *mdt,
4913 struct obd_connect_data *data)
4916 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4917 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4919 /* If no known bits (which should not happen, probably,
4920 as everybody should support LOOKUP and UPDATE bits at least)
4921 revert to compat mode with plain locks. */
4922 if (!data->ocd_ibits_known &&
4923 data->ocd_connect_flags & OBD_CONNECT_IBITS)
4924 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4926 if (!mdt->mdt_opts.mo_acl)
4927 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4929 if (!mdt->mdt_opts.mo_user_xattr)
4930 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4932 if (!mdt->mdt_som_conf)
4933 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4935 spin_lock(&exp->exp_lock);
4936 exp->exp_connect_flags = data->ocd_connect_flags;
4937 spin_unlock(&exp->exp_lock);
4938 data->ocd_version = LUSTRE_VERSION_CODE;
4939 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4943 if (mdt->mdt_opts.mo_acl &&
4944 ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4945 CWARN("%s: MDS requires ACL support but client does not\n",
4946 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4951 if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4952 CWARN("%s: MDS requires FID support, but client not\n",
4953 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4957 if (mdt->mdt_som_conf &&
4958 !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS) &&
4959 !(exp->exp_connect_flags & OBD_CONNECT_SOM)) {
4960 CWARN("%s: MDS has SOM enabled, but client does not support "
4961 "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4968 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
4969 struct obd_export *exp,
4970 struct ptlrpc_request *req)
4972 struct sptlrpc_flavor flvr;
4975 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4976 read_lock(&mdt->mdt_sptlrpc_lock);
4977 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
4981 read_unlock(&mdt->mdt_sptlrpc_lock);
4983 spin_lock(&exp->exp_lock);
4985 exp->exp_sp_peer = req->rq_sp_from;
4986 exp->exp_flvr = flvr;
4988 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
4989 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4990 CERROR("unauthorized rpc flavor %x from %s, "
4991 "expect %x\n", req->rq_flvr.sf_rpc,
4992 libcfs_nid2str(req->rq_peer.nid),
4993 exp->exp_flvr.sf_rpc);
4997 spin_unlock(&exp->exp_lock);
4999 if (exp->exp_sp_peer != req->rq_sp_from) {
5000 CERROR("RPC source %s doesn't match %s\n",
5001 sptlrpc_part2name(req->rq_sp_from),
5002 sptlrpc_part2name(exp->exp_sp_peer));
5005 rc = sptlrpc_target_export_check(exp, req);
5012 /* mds_connect copy */
5013 static int mdt_obd_connect(const struct lu_env *env,
5014 struct obd_export **exp, struct obd_device *obd,
5015 struct obd_uuid *cluuid,
5016 struct obd_connect_data *data,
5019 struct mdt_thread_info *info;
5020 struct lsd_client_data *lcd;
5021 struct obd_export *lexp;
5022 struct lustre_handle conn = { 0 };
5023 struct mdt_device *mdt;
5024 struct ptlrpc_request *req;
5028 LASSERT(env != NULL);
5029 if (!exp || !obd || !cluuid)
5032 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5033 req = info->mti_pill->rc_req;
5034 mdt = mdt_dev(obd->obd_lu_dev);
5036 rc = class_connect(&conn, obd, cluuid);
5040 lexp = class_conn2export(&conn);
5041 LASSERT(lexp != NULL);
5043 rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
5047 rc = mdt_connect_internal(lexp, mdt, data);
5051 struct mdt_thread_info *mti;
5052 mti = lu_context_key_get(&env->le_ctx,
5054 LASSERT(mti != NULL);
5055 mti->mti_exp = lexp;
5056 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5057 lexp->exp_mdt_data.med_lcd = lcd;
5058 rc = mdt_client_new(env, mdt);
5061 lexp->exp_mdt_data.med_lcd = NULL;
5063 mdt_export_stats_init(obd, lexp, localdata);
5071 class_disconnect(lexp);
5078 static int mdt_obd_reconnect(const struct lu_env *env,
5079 struct obd_export *exp, struct obd_device *obd,
5080 struct obd_uuid *cluuid,
5081 struct obd_connect_data *data,
5084 struct mdt_thread_info *info;
5085 struct mdt_device *mdt;
5086 struct ptlrpc_request *req;
5090 if (exp == NULL || obd == NULL || cluuid == NULL)
5093 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5094 req = info->mti_pill->rc_req;
5095 mdt = mdt_dev(obd->obd_lu_dev);
5097 rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5101 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5103 mdt_export_stats_init(obd, exp, localdata);
5107 static int mdt_mfd_cleanup(struct obd_export *exp)
5109 struct mdt_export_data *med = &exp->exp_mdt_data;
5110 struct obd_device *obd = exp->exp_obd;
5111 struct mdt_device *mdt;
5112 struct mdt_thread_info *info;
5114 CFS_LIST_HEAD(closing_list);
5115 struct mdt_file_data *mfd, *n;
5119 spin_lock(&med->med_open_lock);
5120 while (!list_empty(&med->med_open_head)) {
5121 struct list_head *tmp = med->med_open_head.next;
5122 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5124 /* Remove mfd handle so it can't be found again.
5125 * We are consuming the mfd_list reference here. */
5126 class_handle_unhash(&mfd->mfd_handle);
5127 list_move_tail(&mfd->mfd_list, &closing_list);
5129 spin_unlock(&med->med_open_lock);
5130 mdt = mdt_dev(obd->obd_lu_dev);
5131 LASSERT(mdt != NULL);
5133 rc = lu_env_init(&env, LCT_MD_THREAD);
5137 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5138 LASSERT(info != NULL);
5139 memset(info, 0, sizeof *info);
5140 info->mti_env = &env;
5141 info->mti_mdt = mdt;
5142 info->mti_exp = exp;
5144 if (!list_empty(&closing_list)) {
5145 struct md_attr *ma = &info->mti_attr;
5149 lmm_size = mdt->mdt_max_mdsize;
5150 OBD_ALLOC(ma->ma_lmm, lmm_size);
5151 if (ma->ma_lmm == NULL)
5152 GOTO(out_lmm, rc = -ENOMEM);
5154 cookie_size = mdt->mdt_max_cookiesize;
5155 OBD_ALLOC(ma->ma_cookie, cookie_size);
5156 if (ma->ma_cookie == NULL)
5157 GOTO(out_cookie, rc = -ENOMEM);
5159 /* Close any open files (which may also cause orphan unlinking). */
5160 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5161 list_del_init(&mfd->mfd_list);
5162 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
5163 ma->ma_lmm_size = lmm_size;
5164 ma->ma_cookie_size = cookie_size;
5166 /* It is not for setattr, just tell MDD to send
5167 * DESTROY RPC to OSS if needed */
5168 ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
5169 ma->ma_valid = MA_FLAGS;
5170 mdt_mfd_close(info, mfd);
5172 info->mti_mdt = NULL;
5173 OBD_FREE(ma->ma_cookie, cookie_size);
5174 ma->ma_cookie = NULL;
5176 OBD_FREE(ma->ma_lmm, lmm_size);
5185 static int mdt_obd_disconnect(struct obd_export *exp)
5187 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5192 class_export_get(exp);
5194 /* Disconnect early so that clients can't keep using export */
5195 rc = class_disconnect(exp);
5196 if (mdt->mdt_namespace != NULL || exp->exp_obd->obd_namespace != NULL)
5197 ldlm_cancel_locks_for_export(exp);
5199 /* release nid stat refererence */
5200 lprocfs_exp_cleanup(exp);
5202 /* complete all outstanding replies */
5203 spin_lock(&exp->exp_lock);
5204 while (!list_empty(&exp->exp_outstanding_replies)) {
5205 struct ptlrpc_reply_state *rs =
5206 list_entry(exp->exp_outstanding_replies.next,
5207 struct ptlrpc_reply_state, rs_exp_list);
5208 struct ptlrpc_service *svc = rs->rs_service;
5210 spin_lock(&svc->srv_lock);
5211 list_del_init(&rs->rs_exp_list);
5212 spin_lock(&rs->rs_lock);
5213 ptlrpc_schedule_difficult_reply(rs);
5214 spin_unlock(&rs->rs_lock);
5215 spin_unlock(&svc->srv_lock);
5217 spin_unlock(&exp->exp_lock);
5218 rc = mdt_mfd_cleanup(exp);
5219 class_export_put(exp);
5223 /* FIXME: Can we avoid using these two interfaces? */
5224 static int mdt_init_export(struct obd_export *exp)
5226 struct mdt_export_data *med = &exp->exp_mdt_data;
5230 CFS_INIT_LIST_HEAD(&med->med_open_head);
5231 spin_lock_init(&med->med_open_lock);
5232 sema_init(&med->med_idmap_sem, 1);
5233 med->med_idmap = NULL;
5234 spin_lock(&exp->exp_lock);
5235 exp->exp_connecting = 1;
5236 spin_unlock(&exp->exp_lock);
5237 rc = ldlm_init_export(exp);
5239 CERROR("Error %d while initializing export\n", rc);
5243 static int mdt_destroy_export(struct obd_export *export)
5245 struct mdt_export_data *med;
5246 struct obd_device *obd = export->exp_obd;
5247 struct mdt_device *mdt;
5248 struct mdt_thread_info *info;
5253 med = &export->exp_mdt_data;
5254 if (exp_connect_rmtclient(export))
5255 mdt_cleanup_idmap(med);
5257 target_destroy_export(export);
5258 ldlm_destroy_export(export);
5260 LASSERT(list_empty(&export->exp_outstanding_replies));
5261 LASSERT(list_empty(&med->med_open_head));
5262 if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
5265 mdt = mdt_dev(obd->obd_lu_dev);
5266 LASSERT(mdt != NULL);
5268 rc = lu_env_init(&env, LCT_MD_THREAD);
5272 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5273 LASSERT(info != NULL);
5274 memset(info, 0, sizeof *info);
5275 info->mti_env = &env;
5276 info->mti_exp = export;
5277 info->mti_mdt = NULL;
5278 mdt_client_del(&env, mdt);
5284 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
5286 if (flag & CONFIG_LOG)
5287 m->mdt_fl_cfglog = 1;
5289 /* also notify active event */
5290 if (flag & CONFIG_SYNC)
5291 m->mdt_fl_synced = 1;
5293 if (m->mdt_fl_cfglog && m->mdt_fl_synced)
5294 /* Open for clients */
5295 m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0;
5298 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
5299 enum md_upcall_event ev, void *data)
5301 struct mdt_device *m = mdt_dev(&md->md_lu_dev);
5302 struct md_device *next = m->mdt_child;
5303 struct mdt_thread_info *mti;
5309 rc = next->md_ops->mdo_maxsize_get(env, next,
5311 &m->mdt_max_cookiesize);
5312 CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
5313 m->mdt_max_mdsize, m->mdt_max_cookiesize);
5314 mdt_allow_cli(m, CONFIG_SYNC);
5316 (*(__u64 *)data) = m->mdt_mount_count;
5319 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5320 mti->mti_no_need_trans = 1;
5321 CDEBUG(D_INFO, "disable mdt trans for this thread\n");
5324 /* Check that MDT is not yet configured */
5325 LASSERT(!m->mdt_fl_cfglog);
5327 #ifdef HAVE_QUOTA_SUPPORT
5329 if (md->md_lu_dev.ld_obd->obd_recovering == 0 &&
5330 likely(md->md_lu_dev.ld_obd->obd_stopping == 0))
5331 next->md_ops->mdo_quota.mqo_recovery(env, next);
5335 CERROR("invalid event\n");
5342 static int mdt_obd_notify(struct obd_device *host,
5343 struct obd_device *watched,
5344 enum obd_notify_event ev, void *data)
5346 struct mdt_device *mdt = mdt_dev(host->obd_lu_dev);
5347 #ifdef HAVE_QUOTA_SUPPORT
5348 struct md_device *next = mdt->mdt_child;
5353 case OBD_NOTIFY_CONFIG:
5354 mdt_allow_cli(mdt, (unsigned long)data);
5356 #ifdef HAVE_QUOTA_SUPPORT
5357 /* quota_type has been processed, we can now handle
5358 * incoming quota requests */
5359 next->md_ops->mdo_quota.mqo_notify(NULL, next);
5363 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
5368 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5369 void *val, int vallen)
5371 struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev);
5372 struct getinfo_fid2path *fpout, *fpin;
5375 fpin = key + size_round(sizeof(KEY_FID2PATH));
5378 if (lustre_msg_swabbed(mdt_info_req(info)->rq_reqmsg))
5379 lustre_swab_fid2path(fpin);
5381 memcpy(fpout, fpin, sizeof(*fpin));
5382 if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5385 rc = mdt_fid2path(info->mti_env, mdt, fpout);
5389 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
5390 struct getinfo_fid2path *fp)
5392 struct mdt_object *obj;
5396 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5397 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5399 if (!fid_is_sane(&fp->gf_fid))
5402 obj = mdt_object_find(env, mdt, &fp->gf_fid);
5403 if (obj == NULL || IS_ERR(obj)) {
5404 CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid),
5409 rc = lu_object_exists(&obj->mot_obj.mo_lu);
5415 mdt_object_put(env, obj);
5416 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5417 PFID(&fp->gf_fid), rc);
5421 rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path,
5422 fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno);
5423 mdt_object_put(env, obj);
5428 static int mdt_get_info(struct mdt_thread_info *info)
5430 struct ptlrpc_request *req = mdt_info_req(info);
5438 key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5440 CDEBUG(D_IOCTL, "No GETINFO key");
5443 keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5446 vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5447 if (vallen == NULL) {
5448 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5452 req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5454 rc = req_capsule_server_pack(info->mti_pill);
5455 valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5456 if (valout == NULL) {
5457 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5461 if (KEY_IS(KEY_FID2PATH))
5462 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5466 lustre_msg_set_status(req->rq_repmsg, rc);
5471 /* Pass the ioc down */
5472 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5473 unsigned int cmd, int len, void *data)
5475 struct lu_context ioctl_session;
5476 struct md_device *next = mdt->mdt_child;
5480 rc = lu_context_init(&ioctl_session, LCT_SESSION);
5483 ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5484 lu_context_enter(&ioctl_session);
5485 env->le_ses = &ioctl_session;
5487 LASSERT(next->md_ops->mdo_iocontrol);
5488 rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5490 lu_context_exit(&ioctl_session);
5491 lu_context_fini(&ioctl_session);
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);
5529 CERROR("Not supported cmd = %d for device %s\n",
5530 cmd, obd->obd_name);
5538 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5540 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5541 struct obd_device *obd = mdt2obd_dev(mdt);
5542 #ifdef HAVE_QUOTA_SUPPORT
5543 struct md_device *next = mdt->mdt_child;
5547 /* if some clients didn't participate in recovery then we can possibly
5548 * lost sequence. Now we should increase sequence for safe value */
5549 lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
5550 mdt_seq_adjust(env, mdt, lost);
5552 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5553 #ifdef HAVE_QUOTA_SUPPORT
5554 if (likely(obd->obd_stopping == 0))
5555 next->md_ops->mdo_quota.mqo_recovery(env, next);
5560 int mdt_obd_postrecov(struct obd_device *obd)
5565 rc = lu_env_init(&env, LCT_MD_THREAD);
5568 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5573 static struct obd_ops mdt_obd_device_ops = {
5574 .o_owner = THIS_MODULE,
5575 .o_set_info_async = mdt_obd_set_info_async,
5576 .o_connect = mdt_obd_connect,
5577 .o_reconnect = mdt_obd_reconnect,
5578 .o_disconnect = mdt_obd_disconnect,
5579 .o_init_export = mdt_init_export,
5580 .o_destroy_export = mdt_destroy_export,
5581 .o_iocontrol = mdt_iocontrol,
5582 .o_postrecov = mdt_obd_postrecov,
5583 .o_notify = mdt_obd_notify
5586 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5587 struct lu_device *d)
5589 struct mdt_device *m = mdt_dev(d);
5596 static struct lu_device *mdt_device_free(const struct lu_env *env,
5597 struct lu_device *d)
5599 struct mdt_device *m = mdt_dev(d);
5602 md_device_fini(&m->mdt_md_dev);
5607 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5608 struct lu_device_type *t,
5609 struct lustre_cfg *cfg)
5611 struct lu_device *l;
5612 struct mdt_device *m;
5618 l = &m->mdt_md_dev.md_lu_dev;
5619 rc = mdt_init0(env, m, t, cfg);
5621 mdt_device_free(env, l);
5625 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
5627 l = ERR_PTR(-ENOMEM);
5631 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5632 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
5634 /* context key: mdt_thread_key */
5635 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5637 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
5638 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
5640 struct lu_context_key mdt_txn_key = {
5641 .lct_tags = LCT_TX_HANDLE,
5642 .lct_init = mdt_txn_key_init,
5643 .lct_fini = mdt_txn_key_fini
5646 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
5648 return md_ucred(info->mti_env);
5652 * Enable/disable COS.
5654 * Set/Clear the COS flag in mdt options.
5656 * \param mdt mdt device
5657 * \param val 0 disables COS, other values enable COS
5659 void mdt_enable_cos(struct mdt_device *mdt, int val)
5664 mdt->mdt_opts.mo_cos = !!val;
5665 rc = lu_env_init(&env, LCT_MD_THREAD);
5666 if (unlikely(rc != 0)) {
5667 CWARN("lu_env initialization failed with rc = %d,"
5668 "cannot sync\n", rc);
5671 mdt_device_sync(&env, mdt);
5678 * Return COS flag status/
5680 * \param mdt mdt device
5682 int mdt_cos_is_enabled(struct mdt_device *mdt)
5684 return mdt->mdt_opts.mo_cos != 0;
5687 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
5688 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
5690 static struct lu_device_type_operations mdt_device_type_ops = {
5691 .ldto_init = mdt_type_init,
5692 .ldto_fini = mdt_type_fini,
5694 .ldto_start = mdt_type_start,
5695 .ldto_stop = mdt_type_stop,
5697 .ldto_device_alloc = mdt_device_alloc,
5698 .ldto_device_free = mdt_device_free,
5699 .ldto_device_fini = mdt_device_fini
5702 static struct lu_device_type mdt_device_type = {
5703 .ldt_tags = LU_DEVICE_MD,
5704 .ldt_name = LUSTRE_MDT_NAME,
5705 .ldt_ops = &mdt_device_type_ops,
5706 .ldt_ctx_tags = LCT_MD_THREAD
5709 static struct lu_local_obj_desc mdt_last_recv = {
5710 .llod_name = LAST_RCVD,
5711 .llod_oid = MDT_LAST_RECV_OID,
5715 static int __init mdt_mod_init(void)
5717 struct lprocfs_static_vars lvars;
5720 llo_local_obj_register(&mdt_last_recv);
5722 mdt_num_threads = MDT_NUM_THREADS;
5723 lprocfs_mdt_init_vars(&lvars);
5724 rc = class_register_type(&mdt_obd_device_ops, NULL,
5725 lvars.module_vars, LUSTRE_MDT_NAME,
5731 static void __exit mdt_mod_exit(void)
5733 llo_local_obj_unregister(&mdt_last_recv);
5734 class_unregister_type(LUSTRE_MDT_NAME);
5738 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt) \
5739 [prefix ## _ ## opc - prefix ## _ ## base] = { \
5741 .mh_fail_id = OBD_FAIL_ ## prefix ## _ ## opc ## suffix, \
5742 .mh_opc = prefix ## _ ## opc, \
5743 .mh_flags = flags, \
5748 #define DEF_MDT_HNDL(flags, name, fn, fmt) \
5749 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
5751 #define DEF_SEQ_HNDL(flags, name, fn, fmt) \
5752 DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
5754 #define DEF_FLD_HNDL(flags, name, fn, fmt) \
5755 DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
5757 * Request with a format known in advance
5759 #define DEF_MDT_HNDL_F(flags, name, fn) \
5760 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
5762 #define DEF_SEQ_HNDL_F(flags, name, fn) \
5763 DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
5765 #define DEF_FLD_HNDL_F(flags, name, fn) \
5766 DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
5768 * Request with a format we do not yet know
5770 #define DEF_MDT_HNDL_0(flags, name, fn) \
5771 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
5773 static struct mdt_handler mdt_mds_ops[] = {
5774 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5775 DEF_MDT_HNDL_F(0, DISCONNECT, mdt_disconnect),
5776 DEF_MDT_HNDL_F(0, SET_INFO, mdt_set_info),
5777 DEF_MDT_HNDL_F(0, GET_INFO, mdt_get_info),
5778 DEF_MDT_HNDL_F(0 |HABEO_REFERO, GETSTATUS, mdt_getstatus),
5779 DEF_MDT_HNDL_F(HABEO_CORPUS, GETATTR, mdt_getattr),
5780 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
5781 DEF_MDT_HNDL_F(HABEO_CORPUS, GETXATTR, mdt_getxattr),
5782 DEF_MDT_HNDL_F(0 |HABEO_REFERO, STATFS, mdt_statfs),
5783 DEF_MDT_HNDL_F(0 |MUTABOR, REINT, mdt_reint),
5784 DEF_MDT_HNDL_F(HABEO_CORPUS, CLOSE, mdt_close),
5785 DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing),
5786 DEF_MDT_HNDL_F(0 |HABEO_REFERO, PIN, mdt_pin),
5787 DEF_MDT_HNDL_0(0, SYNC, mdt_sync),
5788 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir),
5789 #ifdef HAVE_QUOTA_SUPPORT
5790 DEF_MDT_HNDL_F(0, QUOTACHECK, mdt_quotacheck_handle),
5791 DEF_MDT_HNDL_F(0, QUOTACTL, mdt_quotactl_handle)
5795 #define DEF_OBD_HNDL(flags, name, fn) \
5796 DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
5799 static struct mdt_handler mdt_obd_ops[] = {
5800 DEF_OBD_HNDL(0, PING, mdt_obd_ping),
5801 DEF_OBD_HNDL(0, LOG_CANCEL, mdt_obd_log_cancel),
5802 DEF_OBD_HNDL(0, QC_CALLBACK, mdt_obd_qc_callback)
5805 #define DEF_DLM_HNDL_0(flags, name, fn) \
5806 DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
5807 #define DEF_DLM_HNDL_F(flags, name, fn) \
5808 DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
5810 static struct mdt_handler mdt_dlm_ops[] = {
5811 DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE, mdt_enqueue),
5812 DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT, mdt_convert),
5813 DEF_DLM_HNDL_0(0, BL_CALLBACK, mdt_bl_callback),
5814 DEF_DLM_HNDL_0(0, CP_CALLBACK, mdt_cp_callback)
5817 #define DEF_LLOG_HNDL(flags, name, fn) \
5818 DEF_HNDL(LLOG, ORIGIN_HANDLE_CREATE, _NET, flags, name, fn, NULL)
5820 static struct mdt_handler mdt_llog_ops[] = {
5821 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CREATE, mdt_llog_create),
5822 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_NEXT_BLOCK, mdt_llog_next_block),
5823 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_READ_HEADER, mdt_llog_read_header),
5824 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_WRITE_REC, NULL),
5825 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CLOSE, NULL),
5826 DEF_LLOG_HNDL(0, ORIGIN_CONNECT, NULL),
5827 DEF_LLOG_HNDL(0, CATINFO, NULL),
5828 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_PREV_BLOCK, mdt_llog_prev_block),
5829 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_DESTROY, mdt_llog_destroy),
5832 #define DEF_SEC_CTX_HNDL(name, fn) \
5833 DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
5835 static struct mdt_handler mdt_sec_ctx_ops[] = {
5836 DEF_SEC_CTX_HNDL(INIT, mdt_sec_ctx_handle),
5837 DEF_SEC_CTX_HNDL(INIT_CONT, mdt_sec_ctx_handle),
5838 DEF_SEC_CTX_HNDL(FINI, mdt_sec_ctx_handle)
5841 static struct mdt_opc_slice mdt_regular_handlers[] = {
5843 .mos_opc_start = MDS_GETATTR,
5844 .mos_opc_end = MDS_LAST_OPC,
5845 .mos_hs = mdt_mds_ops
5848 .mos_opc_start = OBD_PING,
5849 .mos_opc_end = OBD_LAST_OPC,
5850 .mos_hs = mdt_obd_ops
5853 .mos_opc_start = LDLM_ENQUEUE,
5854 .mos_opc_end = LDLM_LAST_OPC,
5855 .mos_hs = mdt_dlm_ops
5858 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
5859 .mos_opc_end = LLOG_LAST_OPC,
5860 .mos_hs = mdt_llog_ops
5863 .mos_opc_start = SEC_CTX_INIT,
5864 .mos_opc_end = SEC_LAST_OPC,
5865 .mos_hs = mdt_sec_ctx_ops
5872 static struct mdt_handler mdt_readpage_ops[] = {
5873 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5874 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
5875 #ifdef HAVE_SPLIT_SUPPORT
5876 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
5880 * XXX: this is ugly and should be fixed one day, see mdc_close() for
5881 * detailed comments. --umka
5883 DEF_MDT_HNDL_F(HABEO_CORPUS, CLOSE, mdt_close),
5884 DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing),
5887 static struct mdt_opc_slice mdt_readpage_handlers[] = {
5889 .mos_opc_start = MDS_GETATTR,
5890 .mos_opc_end = MDS_LAST_OPC,
5891 .mos_hs = mdt_readpage_ops
5898 static struct mdt_handler mdt_xmds_ops[] = {
5899 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5900 DEF_MDT_HNDL_F(HABEO_CORPUS , GETATTR, mdt_getattr),
5901 DEF_MDT_HNDL_F(0 | MUTABOR , REINT, mdt_reint),
5902 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir),
5905 static struct mdt_opc_slice mdt_xmds_handlers[] = {
5907 .mos_opc_start = MDS_GETATTR,
5908 .mos_opc_end = MDS_LAST_OPC,
5909 .mos_hs = mdt_xmds_ops
5912 .mos_opc_start = OBD_PING,
5913 .mos_opc_end = OBD_LAST_OPC,
5914 .mos_hs = mdt_obd_ops
5917 .mos_opc_start = SEC_CTX_INIT,
5918 .mos_opc_end = SEC_LAST_OPC,
5919 .mos_hs = mdt_sec_ctx_ops
5926 static struct mdt_handler mdt_seq_ops[] = {
5927 DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
5930 static struct mdt_opc_slice mdt_seq_handlers[] = {
5932 .mos_opc_start = SEQ_QUERY,
5933 .mos_opc_end = SEQ_LAST_OPC,
5934 .mos_hs = mdt_seq_ops
5941 static struct mdt_handler mdt_fld_ops[] = {
5942 DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
5945 static struct mdt_opc_slice mdt_fld_handlers[] = {
5947 .mos_opc_start = FLD_QUERY,
5948 .mos_opc_end = FLD_LAST_OPC,
5949 .mos_hs = mdt_fld_ops
5956 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5957 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
5958 MODULE_LICENSE("GPL");
5960 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
5961 "number of mdt service threads to start");
5963 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);