1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * 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 static unsigned long mdt_num_threads;
103 static unsigned long mdt_min_threads;
104 static unsigned long mdt_max_threads;
106 /* ptlrpc request handler for MDT. All handlers are
107 * grouped into several slices - struct mdt_opc_slice,
108 * and stored in an array - mdt_handlers[].
111 /* The name of this handler. */
113 /* Fail id for this handler, checked at the beginning of this handler*/
115 /* Operation code for this handler */
117 /* flags are listed in enum mdt_handler_flags below. */
119 /* The actual handler function to execute. */
120 int (*mh_act)(struct mdt_thread_info *info);
121 /* Request format for this request. */
122 const struct req_format *mh_fmt;
125 enum mdt_handler_flags {
127 * struct mdt_body is passed in the incoming message, and object
128 * identified by this fid exists on disk.
130 * "habeo corpus" == "I have a body"
132 HABEO_CORPUS = (1 << 0),
134 * struct ldlm_request is passed in the incoming message.
136 * "habeo clavis" == "I have a key"
138 HABEO_CLAVIS = (1 << 1),
140 * this request has fixed reply format, so that reply message can be
141 * packed by generic code.
143 * "habeo refero" == "I have a reply"
145 HABEO_REFERO = (1 << 2),
147 * this request will modify something, so check whether the filesystem
148 * is readonly or not, then return -EROFS to client asap if necessary.
150 * "mutabor" == "I shall modify"
155 struct mdt_opc_slice {
158 struct mdt_handler *mos_hs;
161 static struct mdt_opc_slice mdt_regular_handlers[];
162 static struct mdt_opc_slice mdt_readpage_handlers[];
163 static struct mdt_opc_slice mdt_xmds_handlers[];
164 static struct mdt_opc_slice mdt_seq_handlers[];
165 static struct mdt_opc_slice mdt_fld_handlers[];
167 static struct mdt_device *mdt_dev(struct lu_device *d);
168 static int mdt_regular_handle(struct ptlrpc_request *req);
169 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
170 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
171 struct getinfo_fid2path *fp);
173 static const struct lu_object_operations mdt_obj_ops;
175 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
179 return (rep->lock_policy_res1 & flag);
182 void mdt_clear_disposition(struct mdt_thread_info *info,
183 struct ldlm_reply *rep, int flag)
186 info->mti_opdata &= ~flag;
188 rep->lock_policy_res1 &= ~flag;
191 void mdt_set_disposition(struct mdt_thread_info *info,
192 struct ldlm_reply *rep, int flag)
195 info->mti_opdata |= flag;
197 rep->lock_policy_res1 |= flag;
200 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
202 lh->mlh_pdo_hash = 0;
203 lh->mlh_reg_mode = lm;
204 lh->mlh_type = MDT_REG_LOCK;
207 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
208 const char *name, int namelen)
210 lh->mlh_reg_mode = lm;
211 lh->mlh_type = MDT_PDO_LOCK;
214 LASSERT(namelen > 0);
215 lh->mlh_pdo_hash = full_name_hash(name, namelen);
217 LASSERT(namelen == 0);
218 lh->mlh_pdo_hash = 0ull;
222 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
223 struct mdt_lock_handle *lh)
229 * Any dir access needs couple of locks:
231 * 1) on part of dir we gonna take lookup/modify;
233 * 2) on whole dir to protect it from concurrent splitting and/or to
234 * flush client's cache for readdir().
236 * so, for a given mode and object this routine decides what lock mode
237 * to use for lock #2:
239 * 1) if caller's gonna lookup in dir then we need to protect dir from
240 * being splitted only - LCK_CR
242 * 2) if caller's gonna modify dir then we need to protect dir from
243 * being splitted and to flush cache - LCK_CW
245 * 3) if caller's gonna modify dir and that dir seems ready for
246 * splitting then we need to protect it from any type of access
247 * (lookup/modify/split) - LCK_EX --bzzz
250 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
251 LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
254 * Ask underlaying level its opinion about preferable PDO lock mode
255 * having access type passed as regular lock mode:
257 * - MDL_MINMODE means that lower layer does not want to specify lock
260 * - MDL_NL means that no PDO lock should be taken. This is used in some
261 * cases. Say, for non-splittable directories no need to use PDO locks
264 mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
265 mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
267 if (mode != MDL_MINMODE) {
268 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
271 * Lower layer does not want to specify locking mode. We do it
272 * our selves. No special protection is needed, just flush
273 * client's cache on modification and allow concurrent
276 switch (lh->mlh_reg_mode) {
278 lh->mlh_pdo_mode = LCK_EX;
281 lh->mlh_pdo_mode = LCK_CR;
284 lh->mlh_pdo_mode = LCK_CW;
287 CERROR("Not expected lock type (0x%x)\n",
288 (int)lh->mlh_reg_mode);
293 LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
297 static int mdt_getstatus(struct mdt_thread_info *info)
299 struct mdt_device *mdt = info->mti_mdt;
300 struct md_device *next = mdt->mdt_child;
301 struct mdt_body *repbody;
306 rc = mdt_check_ucred(info);
308 RETURN(err_serious(rc));
310 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
311 RETURN(err_serious(-ENOMEM));
313 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
314 rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
318 repbody->valid |= OBD_MD_FLID;
320 if (mdt->mdt_opts.mo_mds_capa &&
321 info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
322 struct mdt_object *root;
323 struct lustre_capa *capa;
325 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
327 RETURN(PTR_ERR(root));
329 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
331 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
332 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
334 mdt_object_put(info->mti_env, root);
336 repbody->valid |= OBD_MD_FLMDSCAPA;
342 static int mdt_statfs(struct mdt_thread_info *info)
344 struct md_device *next = info->mti_mdt->mdt_child;
345 struct ptlrpc_service *svc;
346 struct obd_statfs *osfs;
351 svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
353 /* This will trigger a watchdog timeout */
354 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
355 (MDT_SERVICE_WATCHDOG_FACTOR *
356 at_get(&svc->srv_at_estimate)) + 1);
358 rc = mdt_check_ucred(info);
360 RETURN(err_serious(rc));
362 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
363 rc = err_serious(-ENOMEM);
365 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
366 rc = next->md_ops->mdo_statfs(info->mti_env, next,
368 statfs_pack(osfs, &info->mti_u.ksfs);
374 * Pack SOM attributes into the reply.
375 * Call under a DLM UPDATE lock.
377 static void mdt_pack_size2body(struct mdt_thread_info *info,
378 struct mdt_object *mo)
381 struct md_attr *ma = &info->mti_attr;
383 LASSERT(ma->ma_attr.la_valid & LA_MODE);
384 b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
386 /* Check if Size-on-MDS is supported, if this is a regular file,
387 * if SOM is enabled on the object and if SOM cache exists and valid.
388 * Otherwise do not pack Size-on-MDS attributes to the reply. */
389 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
390 !S_ISREG(ma->ma_attr.la_mode) ||
391 !mdt_object_is_som_enabled(mo) ||
392 !(ma->ma_valid & MA_SOM))
395 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
396 b->size = ma->ma_som->msd_size;
397 b->blocks = ma->ma_som->msd_blocks;
400 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
401 const struct lu_attr *attr, const struct lu_fid *fid)
403 struct md_attr *ma = &info->mti_attr;
405 LASSERT(ma->ma_valid & MA_INODE);
407 b->atime = attr->la_atime;
408 b->mtime = attr->la_mtime;
409 b->ctime = attr->la_ctime;
410 b->mode = attr->la_mode;
411 b->size = attr->la_size;
412 b->blocks = attr->la_blocks;
413 b->uid = attr->la_uid;
414 b->gid = attr->la_gid;
415 b->flags = attr->la_flags;
416 b->nlink = attr->la_nlink;
417 b->rdev = attr->la_rdev;
419 /*XXX should pack the reply body according to lu_valid*/
420 b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID |
421 OBD_MD_FLGID | OBD_MD_FLTYPE |
422 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
423 OBD_MD_FLATIME | OBD_MD_FLMTIME ;
425 if (!S_ISREG(attr->la_mode)) {
426 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
427 } else if (ma->ma_need & MA_LOV && ma->ma_lmm_size == 0) {
428 /* means no objects are allocated on osts. */
429 LASSERT(!(ma->ma_valid & MA_LOV));
430 /* just ignore blocks occupied by extend attributes on MDS */
432 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
433 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
438 b->valid |= OBD_MD_FLID;
440 /* FIXME: these should be fixed when new igif ready.*/
441 b->ino = fid_oid(fid); /* 1.6 compatibility */
442 b->generation = fid_ver(fid); /* 1.6 compatibility */
443 b->valid |= OBD_MD_FLGENER; /* 1.6 compatibility */
445 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
446 PFID(fid), b->nlink, b->mode, b->size);
450 mdt_body_reverse_idmap(info, b);
452 if (b->valid & OBD_MD_FLSIZE)
453 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
454 PFID(fid), (unsigned long long)b->size);
457 static inline int mdt_body_has_lov(const struct lu_attr *la,
458 const struct mdt_body *body)
460 return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
461 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
464 static int mdt_getattr_internal(struct mdt_thread_info *info,
465 struct mdt_object *o, int ma_need)
467 struct md_object *next = mdt_object_child(o);
468 const struct mdt_body *reqbody = info->mti_body;
469 struct ptlrpc_request *req = mdt_info_req(info);
470 struct md_attr *ma = &info->mti_attr;
471 struct lu_attr *la = &ma->ma_attr;
472 struct req_capsule *pill = info->mti_pill;
473 const struct lu_env *env = info->mti_env;
474 struct mdt_body *repbody;
475 struct lu_buf *buffer = &info->mti_buf;
479 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
480 RETURN(err_serious(-ENOMEM));
482 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
486 rc = mdt_object_exists(o);
488 /* This object is located on remote node.*/
489 repbody->fid1 = *mdt_object_fid(o);
490 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
494 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
495 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
497 /* If it is dir object and client require MEA, then we got MEA */
498 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
499 reqbody->valid & OBD_MD_MEA) {
500 /* Assumption: MDT_MD size is enough for lmv size. */
501 ma->ma_lmv = buffer->lb_buf;
502 ma->ma_lmv_size = buffer->lb_len;
503 ma->ma_need = MA_LMV | MA_INODE;
505 ma->ma_lmm = buffer->lb_buf;
506 ma->ma_lmm_size = buffer->lb_len;
507 ma->ma_need = MA_LOV | MA_INODE;
510 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
511 reqbody->valid & OBD_MD_FLDIREA &&
512 lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
513 /* get default stripe info for this dir. */
514 ma->ma_need |= MA_LOV_DEF;
516 ma->ma_need |= ma_need;
517 if (ma->ma_need & MA_SOM)
518 ma->ma_som = &info->mti_u.som.data;
520 rc = mo_attr_get(env, next, ma);
522 CERROR("getattr error for "DFID": %d\n",
523 PFID(mdt_object_fid(o)), rc);
527 if (likely(ma->ma_valid & MA_INODE))
528 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
532 if (mdt_body_has_lov(la, reqbody)) {
533 if (ma->ma_valid & MA_LOV) {
534 LASSERT(ma->ma_lmm_size);
535 mdt_dump_lmm(D_INFO, ma->ma_lmm);
536 repbody->eadatasize = ma->ma_lmm_size;
537 if (S_ISDIR(la->la_mode))
538 repbody->valid |= OBD_MD_FLDIREA;
540 repbody->valid |= OBD_MD_FLEASIZE;
542 if (ma->ma_valid & MA_LMV) {
543 LASSERT(S_ISDIR(la->la_mode));
544 repbody->eadatasize = ma->ma_lmv_size;
545 repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
547 } else if (S_ISLNK(la->la_mode) &&
548 reqbody->valid & OBD_MD_LINKNAME) {
549 buffer->lb_buf = ma->ma_lmm;
550 buffer->lb_len = reqbody->eadatasize;
551 rc = mo_readlink(env, next, buffer);
552 if (unlikely(rc <= 0)) {
553 CERROR("readlink failed: %d\n", rc);
556 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
558 repbody->valid |= OBD_MD_LINKNAME;
559 repbody->eadatasize = rc;
561 ((char*)ma->ma_lmm)[rc - 1] = 0;
562 CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
563 (char*)ma->ma_lmm, rc);
568 if (reqbody->valid & OBD_MD_FLMODEASIZE) {
569 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
570 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
571 repbody->valid |= OBD_MD_FLMODEASIZE;
572 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
573 "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
574 repbody->max_cookiesize);
577 if (exp_connect_rmtclient(info->mti_exp) &&
578 reqbody->valid & OBD_MD_FLRMTPERM) {
579 void *buf = req_capsule_server_get(pill, &RMF_ACL);
581 /* mdt_getattr_lock only */
582 rc = mdt_pack_remote_perm(info, o, buf);
584 repbody->valid &= ~OBD_MD_FLRMTPERM;
585 repbody->aclsize = 0;
588 repbody->valid |= OBD_MD_FLRMTPERM;
589 repbody->aclsize = sizeof(struct mdt_remote_perm);
592 #ifdef CONFIG_FS_POSIX_ACL
593 else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
594 (reqbody->valid & OBD_MD_FLACL)) {
595 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
596 buffer->lb_len = req_capsule_get_size(pill,
597 &RMF_ACL, RCL_SERVER);
598 if (buffer->lb_len > 0) {
599 rc = mo_xattr_get(env, next, buffer,
600 XATTR_NAME_ACL_ACCESS);
602 if (rc == -ENODATA) {
603 repbody->aclsize = 0;
604 repbody->valid |= OBD_MD_FLACL;
606 } else if (rc == -EOPNOTSUPP) {
609 CERROR("got acl size: %d\n", rc);
612 repbody->aclsize = rc;
613 repbody->valid |= OBD_MD_FLACL;
620 if (reqbody->valid & OBD_MD_FLMDSCAPA &&
621 info->mti_mdt->mdt_opts.mo_mds_capa &&
622 info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
623 struct lustre_capa *capa;
625 capa = req_capsule_server_get(pill, &RMF_CAPA1);
627 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
628 rc = mo_capa_get(env, next, capa, 0);
631 repbody->valid |= OBD_MD_FLMDSCAPA;
636 static int mdt_renew_capa(struct mdt_thread_info *info)
638 struct mdt_object *obj = info->mti_object;
639 struct mdt_body *body;
640 struct lustre_capa *capa, *c;
644 /* if object doesn't exist, or server has disabled capability,
645 * return directly, client will find body->valid OBD_MD_FLOSSCAPA
648 if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
649 !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
652 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
653 LASSERT(body != NULL);
655 c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
658 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
662 rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
664 body->valid |= OBD_MD_FLOSSCAPA;
668 static int mdt_getattr(struct mdt_thread_info *info)
670 struct mdt_object *obj = info->mti_object;
671 struct req_capsule *pill = info->mti_pill;
672 struct mdt_body *reqbody;
673 struct mdt_body *repbody;
679 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
682 if (reqbody->valid & OBD_MD_FLOSSCAPA) {
683 rc = req_capsule_server_pack(pill);
685 RETURN(err_serious(rc));
686 rc = mdt_renew_capa(info);
687 GOTO(out_shrink, rc);
690 LASSERT(obj != NULL);
691 LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
693 mode = lu_object_attr(&obj->mot_obj.mo_lu);
694 if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
695 (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
696 md_size = reqbody->eadatasize;
698 md_size = info->mti_mdt->mdt_max_mdsize;
700 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
702 rc = req_capsule_server_pack(pill);
703 if (unlikely(rc != 0))
704 RETURN(err_serious(rc));
706 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
707 LASSERT(repbody != NULL);
708 repbody->eadatasize = 0;
709 repbody->aclsize = 0;
711 if (reqbody->valid & OBD_MD_FLRMTPERM)
712 rc = mdt_init_ucred(info, reqbody);
714 rc = mdt_check_ucred(info);
716 GOTO(out_shrink, rc);
718 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
719 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
722 * Don't check capability at all, because rename might getattr for
723 * remote obj, and at that time no capability is available.
725 mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
726 rc = mdt_getattr_internal(info, obj, 0);
727 if (reqbody->valid & OBD_MD_FLRMTPERM)
728 mdt_exit_ucred(info);
731 mdt_shrink_reply(info);
735 static int mdt_is_subdir(struct mdt_thread_info *info)
737 struct mdt_object *o = info->mti_object;
738 struct req_capsule *pill = info->mti_pill;
739 const struct mdt_body *body = info->mti_body;
740 struct mdt_body *repbody;
746 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
749 * We save last checked parent fid to @repbody->fid1 for remote
752 LASSERT(fid_is_sane(&body->fid2));
753 LASSERT(mdt_object_exists(o) > 0);
754 rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
755 &body->fid2, &repbody->fid1);
756 if (rc == 0 || rc == -EREMOTE)
757 repbody->valid |= OBD_MD_FLID;
762 static int mdt_raw_lookup(struct mdt_thread_info *info,
763 struct mdt_object *parent,
764 const struct lu_name *lname,
765 struct ldlm_reply *ldlm_rep)
767 struct md_object *next = mdt_object_child(info->mti_object);
768 const struct mdt_body *reqbody = info->mti_body;
769 struct lu_fid *child_fid = &info->mti_tmp_fid1;
770 struct mdt_body *repbody;
774 if (reqbody->valid != OBD_MD_FLID)
777 LASSERT(!info->mti_cross_ref);
779 /* Only got the fid of this obj by name */
780 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
783 /* XXX is raw_lookup possible as intent operation? */
786 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
789 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
791 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
794 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
795 repbody->fid1 = *child_fid;
796 repbody->valid = OBD_MD_FLID;
802 * UPDATE lock should be taken against parent, and be release before exit;
803 * child_bits lock should be taken against child, and be returned back:
804 * (1)normal request should release the child lock;
805 * (2)intent request will grant the lock to client.
807 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
808 struct mdt_lock_handle *lhc,
810 struct ldlm_reply *ldlm_rep)
812 struct ptlrpc_request *req = mdt_info_req(info);
813 struct mdt_body *reqbody = NULL;
814 struct mdt_object *parent = info->mti_object;
815 struct mdt_object *child;
816 struct md_object *next = mdt_object_child(parent);
817 struct lu_fid *child_fid = &info->mti_tmp_fid1;
818 struct lu_name *lname = NULL;
819 const char *name = NULL;
821 struct mdt_lock_handle *lhp;
822 struct ldlm_lock *lock;
823 struct ldlm_res_id *res_id;
830 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
831 LASSERT(ergo(is_resent,
832 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
834 LASSERT(parent != NULL);
835 name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
837 RETURN(err_serious(-EFAULT));
839 namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
841 if (!info->mti_cross_ref) {
843 * XXX: Check for "namelen == 0" is for getattr by fid
844 * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
845 * that is the name must contain at least one character and
846 * the terminating '\0'
849 reqbody = req_capsule_client_get(info->mti_pill,
851 LASSERT(fid_is_sane(&reqbody->fid2));
854 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
856 PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
859 lname = mdt_name(info->mti_env, (char *)name, namelen);
860 CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
861 "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
865 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
867 rc = mdt_object_exists(parent);
868 if (unlikely(rc == 0)) {
869 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
870 &parent->mot_obj.mo_lu,
871 "Parent doesn't exist!\n");
873 } else if (!info->mti_cross_ref) {
874 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
875 PFID(mdt_object_fid(parent)));
878 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
886 if (info->mti_cross_ref) {
887 /* Only getattr on the child. Parent is on another node. */
888 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
890 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
891 "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
894 /* Do not take lock for resent case. */
895 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
896 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
897 lhc->mlh_reg_lh.cookie);
898 LASSERT(fid_res_name_eq(mdt_object_fid(child),
899 &lock->l_resource->lr_name));
903 mdt_lock_handle_init(lhc);
904 mdt_lock_reg_init(lhc, LCK_PR);
907 * Object's name is on another MDS, no lookup lock is
908 * needed here but update is.
910 child_bits &= ~MDS_INODELOCK_LOOKUP;
911 child_bits |= MDS_INODELOCK_UPDATE;
913 rc = mdt_object_lock(info, child, lhc, child_bits,
917 /* Finally, we can get attr for child. */
918 mdt_set_capainfo(info, 0, mdt_object_fid(child),
920 rc = mdt_getattr_internal(info, child, 0);
921 if (unlikely(rc != 0))
922 mdt_object_unlock(info, child, lhc, 1);
927 /* step 1: lock parent */
928 lhp = &info->mti_lh[MDT_LH_PARENT];
929 mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
930 rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
933 if (unlikely(rc != 0))
937 /* step 2: lookup child's fid by name */
938 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
943 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
944 GOTO(out_parent, rc);
946 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
948 *child_fid = reqbody->fid2;
949 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
953 *step 3: find the child object by fid & lock it.
954 * regardless if it is local or remote.
956 child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
958 if (unlikely(IS_ERR(child)))
959 GOTO(out_parent, rc = PTR_ERR(child));
961 /* Do not take lock for resent case. */
962 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
963 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
964 lhc->mlh_reg_lh.cookie);
966 res_id = &lock->l_resource->lr_name;
967 if (!fid_res_name_eq(mdt_object_fid(child),
968 &lock->l_resource->lr_name)) {
969 LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
970 &lock->l_resource->lr_name),
971 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
972 (unsigned long)res_id->name[0],
973 (unsigned long)res_id->name[1],
974 (unsigned long)res_id->name[2],
975 PFID(mdt_object_fid(parent)));
976 CWARN("Although resent, but still not get child lock"
977 "parent:"DFID" child:"DFID"\n",
978 PFID(mdt_object_fid(parent)),
979 PFID(mdt_object_fid(child)));
980 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
988 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
989 mdt_lock_handle_init(lhc);
990 mdt_lock_reg_init(lhc, LCK_PR);
992 if (mdt_object_exists(child) == 0) {
993 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
994 &child->mot_obj.mo_lu,
995 "Object doesn't exist!\n");
996 GOTO(out_child, rc = -ESTALE);
999 if (!(child_bits & MDS_INODELOCK_UPDATE)) {
1000 struct md_attr *ma = &info->mti_attr;
1003 ma->ma_need = MA_INODE;
1004 rc = mo_attr_get(info->mti_env,
1005 mdt_object_child(child), ma);
1006 if (unlikely(rc != 0))
1007 GOTO(out_child, rc);
1009 /* If the file has not been changed for some time, we
1010 * return not only a LOOKUP lock, but also an UPDATE
1011 * lock and this might save us RPC on later STAT. For
1012 * directories, it also let negative dentry starts
1013 * working for this dir. */
1014 if (ma->ma_valid & MA_INODE &&
1015 ma->ma_attr.la_valid & LA_CTIME &&
1016 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1017 ma->ma_attr.la_ctime < cfs_time_current_sec())
1018 child_bits |= MDS_INODELOCK_UPDATE;
1021 rc = mdt_object_lock(info, child, lhc, child_bits,
1024 if (unlikely(rc != 0))
1025 GOTO(out_child, rc);
1028 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1029 /* Get MA_SOM attributes if update lock is given. */
1031 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1032 S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1035 /* finally, we can get attr for child. */
1036 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1037 rc = mdt_getattr_internal(info, child, ma_need);
1038 if (unlikely(rc != 0)) {
1039 mdt_object_unlock(info, child, lhc, 1);
1041 /* Debugging code. */
1042 res_id = &lock->l_resource->lr_name;
1043 LDLM_DEBUG(lock, "Returning lock to client");
1044 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1045 &lock->l_resource->lr_name),
1046 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1047 (unsigned long)res_id->name[0],
1048 (unsigned long)res_id->name[1],
1049 (unsigned long)res_id->name[2],
1050 PFID(mdt_object_fid(child)));
1051 mdt_pack_size2body(info, child);
1054 LDLM_LOCK_PUT(lock);
1058 mdt_object_put(info->mti_env, child);
1060 mdt_object_unlock(info, parent, lhp, 1);
1064 /* normal handler: should release the child lock */
1065 static int mdt_getattr_name(struct mdt_thread_info *info)
1067 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1068 struct mdt_body *reqbody;
1069 struct mdt_body *repbody;
1073 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1074 LASSERT(reqbody != NULL);
1075 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1076 LASSERT(repbody != NULL);
1078 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1079 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1080 repbody->eadatasize = 0;
1081 repbody->aclsize = 0;
1083 rc = mdt_init_ucred(info, reqbody);
1085 GOTO(out_shrink, rc);
1087 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1088 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1089 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1090 lhc->mlh_reg_lh.cookie = 0;
1092 mdt_exit_ucred(info);
1095 mdt_shrink_reply(info);
1099 static const struct lu_device_operations mdt_lu_ops;
1101 static int lu_device_is_mdt(struct lu_device *d)
1103 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1106 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1107 void *karg, void *uarg);
1109 static int mdt_set_info(struct mdt_thread_info *info)
1111 struct ptlrpc_request *req = mdt_info_req(info);
1114 int keylen, vallen, rc = 0;
1117 rc = req_capsule_server_pack(info->mti_pill);
1121 key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1123 DEBUG_REQ(D_HA, req, "no set_info key");
1127 keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1130 val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1132 DEBUG_REQ(D_HA, req, "no set_info val");
1136 vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1139 /* Swab any part of val you need to here */
1140 if (KEY_IS(KEY_READ_ONLY)) {
1142 lustre_msg_set_status(req->rq_repmsg, 0);
1144 cfs_spin_lock(&req->rq_export->exp_lock);
1146 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1148 req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
1149 cfs_spin_unlock(&req->rq_export->exp_lock);
1151 } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1152 struct changelog_setinfo *cs =
1153 (struct changelog_setinfo *)val;
1154 if (vallen != sizeof(*cs)) {
1155 CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1158 if (ptlrpc_req_need_swab(req)) {
1159 __swab64s(&cs->cs_recno);
1160 __swab32s(&cs->cs_id);
1163 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1165 lustre_msg_set_status(req->rq_repmsg, rc);
1173 static int mdt_connect(struct mdt_thread_info *info)
1176 struct ptlrpc_request *req;
1178 req = mdt_info_req(info);
1179 rc = target_handle_connect(req);
1181 LASSERT(req->rq_export != NULL);
1182 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1183 rc = mdt_init_sec_level(info);
1185 rc = mdt_init_idmap(info);
1187 obd_disconnect(class_export_get(req->rq_export));
1189 rc = err_serious(rc);
1194 static int mdt_disconnect(struct mdt_thread_info *info)
1199 rc = target_handle_disconnect(mdt_info_req(info));
1201 rc = err_serious(rc);
1205 static int mdt_sendpage(struct mdt_thread_info *info,
1206 struct lu_rdpg *rdpg)
1208 struct ptlrpc_request *req = mdt_info_req(info);
1209 struct obd_export *exp = req->rq_export;
1210 struct ptlrpc_bulk_desc *desc;
1211 struct l_wait_info *lwi = &info->mti_u.rdpg.mti_wait_info;
1219 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1224 for (i = 0, tmpcount = rdpg->rp_count;
1225 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1226 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1227 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1230 LASSERT(desc->bd_nob == rdpg->rp_count);
1231 rc = sptlrpc_svc_wrap_bulk(req, desc);
1233 GOTO(free_desc, rc);
1235 rc = ptlrpc_start_bulk_transfer(desc);
1237 GOTO(free_desc, rc);
1239 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1240 GOTO(abort_bulk, rc = 0);
1243 timeout = (int) req->rq_deadline - cfs_time_current_sec();
1245 CERROR("Req deadline already passed %lu (now: %lu)\n",
1246 req->rq_deadline, cfs_time_current_sec());
1247 *lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(max(timeout, 1)),
1248 cfs_time_seconds(1), NULL, NULL);
1249 rc = l_wait_event(desc->bd_waitq,
1250 !ptlrpc_server_bulk_active(desc) ||
1252 exp->exp_abort_active_req, lwi);
1253 LASSERT (rc == 0 || rc == -ETIMEDOUT);
1254 } while ((rc == -ETIMEDOUT) &&
1255 (req->rq_deadline > cfs_time_current_sec()));
1258 if (desc->bd_success &&
1259 desc->bd_nob_transferred == rdpg->rp_count)
1260 GOTO(free_desc, rc);
1263 if (exp->exp_abort_active_req || exp->exp_failed)
1264 GOTO(abort_bulk, rc);
1267 DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1268 (rc == -ETIMEDOUT) ? "timeout" : "network error",
1269 desc->bd_nob_transferred, rdpg->rp_count,
1270 exp->exp_client_uuid.uuid,
1271 exp->exp_connection->c_remote_uuid.uuid);
1273 class_fail_export(exp);
1277 ptlrpc_abort_bulk(desc);
1279 ptlrpc_free_bulk(desc);
1283 #ifdef HAVE_SPLIT_SUPPORT
1285 * Retrieve dir entry from the page and insert it to the slave object, actually,
1286 * this should be in osd layer, but since it will not in the final product, so
1287 * just do it here and do not define more moo api anymore for this.
1289 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1292 struct mdt_object *object = info->mti_object;
1293 struct lu_fid *lf = &info->mti_tmp_fid2;
1294 struct md_attr *ma = &info->mti_attr;
1295 struct lu_dirpage *dp;
1296 struct lu_dirent *ent;
1297 int rc = 0, offset = 0;
1300 /* Make sure we have at least one entry. */
1305 * Disable trans for this name insert, since it will include many trans
1308 info->mti_no_need_trans = 1;
1310 * When write_dir_page, no need update parent's ctime,
1311 * and no permission check for name_insert.
1313 ma->ma_attr.la_ctime = 0;
1314 ma->ma_attr.la_valid = LA_MODE;
1315 ma->ma_valid = MA_INODE;
1318 dp = page_address(page);
1319 offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1321 for (ent = lu_dirent_start(dp); ent != NULL;
1322 ent = lu_dirent_next(ent)) {
1323 struct lu_name *lname;
1326 if (le16_to_cpu(ent->lde_namelen) == 0)
1329 fid_le_to_cpu(lf, &ent->lde_fid);
1330 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1331 ma->ma_attr.la_mode = S_IFDIR;
1333 ma->ma_attr.la_mode = 0;
1334 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1336 GOTO(out, rc = -ENOMEM);
1338 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1339 lname = mdt_name(info->mti_env, name,
1340 le16_to_cpu(ent->lde_namelen));
1341 ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE);
1342 rc = mdo_name_insert(info->mti_env,
1343 md_object_next(&object->mot_obj),
1345 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1347 CERROR("Can't insert %*.*s, rc %d\n",
1348 le16_to_cpu(ent->lde_namelen),
1349 le16_to_cpu(ent->lde_namelen),
1354 offset += lu_dirent_size(ent);
1364 static int mdt_bulk_timeout(void *data)
1368 CERROR("mdt bulk transfer timeout \n");
1373 static int mdt_writepage(struct mdt_thread_info *info)
1375 struct ptlrpc_request *req = mdt_info_req(info);
1376 struct mdt_body *reqbody;
1377 struct l_wait_info *lwi;
1378 struct ptlrpc_bulk_desc *desc;
1384 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1385 if (reqbody == NULL)
1386 RETURN(err_serious(-EFAULT));
1388 desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1390 RETURN(err_serious(-ENOMEM));
1392 /* allocate the page for the desc */
1393 page = cfs_alloc_page(CFS_ALLOC_STD);
1395 GOTO(desc_cleanup, rc = -ENOMEM);
1397 CDEBUG(D_INFO, "Received page offset %d size %d \n",
1398 (int)reqbody->size, (int)reqbody->nlink);
1400 ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1401 (int)reqbody->nlink);
1403 rc = sptlrpc_svc_prep_bulk(req, desc);
1405 GOTO(cleanup_page, rc);
1407 * Check if client was evicted while we were doing i/o before touching
1412 GOTO(cleanup_page, rc = -ENOMEM);
1414 if (desc->bd_export->exp_failed)
1417 rc = ptlrpc_start_bulk_transfer (desc);
1419 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * CFS_HZ / 4, CFS_HZ,
1420 mdt_bulk_timeout, desc);
1421 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1422 desc->bd_export->exp_failed, lwi);
1423 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1424 if (rc == -ETIMEDOUT) {
1425 DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1426 ptlrpc_abort_bulk(desc);
1427 } else if (desc->bd_export->exp_failed) {
1428 DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1430 ptlrpc_abort_bulk(desc);
1431 } else if (!desc->bd_success ||
1432 desc->bd_nob_transferred != desc->bd_nob) {
1433 DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1435 "truncated" : "network error on",
1436 desc->bd_nob_transferred, desc->bd_nob);
1437 /* XXX should this be a different errno? */
1441 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1444 GOTO(cleanup_lwi, rc);
1445 rc = mdt_write_dir_page(info, page, reqbody->nlink);
1450 cfs_free_page(page);
1452 ptlrpc_free_bulk(desc);
1457 static int mdt_readpage(struct mdt_thread_info *info)
1459 struct mdt_object *object = info->mti_object;
1460 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1461 struct mdt_body *reqbody;
1462 struct mdt_body *repbody;
1467 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1468 RETURN(err_serious(-ENOMEM));
1470 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1471 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1472 if (reqbody == NULL || repbody == NULL)
1473 RETURN(err_serious(-EFAULT));
1476 * prepare @rdpg before calling lower layers and transfer itself. Here
1477 * reqbody->size contains offset of where to start to read and
1478 * reqbody->nlink contains number bytes to read.
1480 rdpg->rp_hash = reqbody->size;
1481 if (rdpg->rp_hash != reqbody->size) {
1482 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1483 rdpg->rp_hash, reqbody->size);
1487 rdpg->rp_attrs = reqbody->mode;
1488 rdpg->rp_count = reqbody->nlink;
1489 rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1490 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1491 if (rdpg->rp_pages == NULL)
1494 for (i = 0; i < rdpg->rp_npages; ++i) {
1495 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1496 if (rdpg->rp_pages[i] == NULL)
1497 GOTO(free_rdpg, rc = -ENOMEM);
1500 /* call lower layers to fill allocated pages with directory data */
1501 rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1503 GOTO(free_rdpg, rc);
1505 /* send pages to client */
1506 rc = mdt_sendpage(info, rdpg);
1511 for (i = 0; i < rdpg->rp_npages; i++)
1512 if (rdpg->rp_pages[i] != NULL)
1513 cfs_free_page(rdpg->rp_pages[i]);
1514 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1516 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1522 static int mdt_reint_internal(struct mdt_thread_info *info,
1523 struct mdt_lock_handle *lhc,
1526 struct req_capsule *pill = info->mti_pill;
1527 struct mdt_device *mdt = info->mti_mdt;
1528 struct md_quota *mq = md_quota(info->mti_env);
1529 struct mdt_body *repbody;
1534 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1535 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1536 mdt->mdt_max_mdsize);
1537 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1538 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1539 mdt->mdt_max_cookiesize);
1541 rc = req_capsule_server_pack(pill);
1543 CERROR("Can't pack response, rc %d\n", rc);
1544 RETURN(err_serious(rc));
1547 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1548 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1550 repbody->eadatasize = 0;
1551 repbody->aclsize = 0;
1554 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1555 GOTO(out_shrink, rc = err_serious(-EFAULT));
1557 rc = mdt_reint_unpack(info, op);
1559 CERROR("Can't unpack reint, rc %d\n", rc);
1560 GOTO(out_shrink, rc = err_serious(rc));
1563 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1565 /* for replay no cookkie / lmm need, because client have this already */
1566 if (info->mti_spec.no_create == 1) {
1567 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1568 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1570 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1571 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1575 rc = mdt_init_ucred_reint(info);
1577 GOTO(out_shrink, rc);
1579 rc = mdt_fix_attr_ucred(info, op);
1581 GOTO(out_ucred, rc = err_serious(rc));
1583 if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1584 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1585 GOTO(out_ucred, rc);
1587 mq->mq_exp = info->mti_exp;
1588 rc = mdt_reint_rec(info, lhc);
1591 mdt_exit_ucred(info);
1593 mdt_shrink_reply(info);
1597 static long mdt_reint_opcode(struct mdt_thread_info *info,
1598 const struct req_format **fmt)
1600 struct mdt_rec_reint *rec;
1603 opc = err_serious(-EFAULT);
1604 rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1606 opc = rec->rr_opcode;
1607 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1608 if (opc < REINT_MAX && fmt[opc] != NULL)
1609 req_capsule_extend(info->mti_pill, fmt[opc]);
1611 CERROR("Unsupported opc: %ld\n", opc);
1612 opc = err_serious(opc);
1618 static int mdt_reint(struct mdt_thread_info *info)
1623 static const struct req_format *reint_fmts[REINT_MAX] = {
1624 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1625 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1626 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1627 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1628 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1629 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1630 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1635 opc = mdt_reint_opcode(info, reint_fmts);
1638 * No lock possible here from client to pass it to reint code
1641 rc = mdt_reint_internal(info, NULL, opc);
1646 info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1650 /* this should sync the whole device */
1651 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1653 struct dt_device *dt = mdt->mdt_bottom;
1657 rc = dt->dd_ops->dt_sync(env, dt);
1661 /* this should sync this object */
1662 static int mdt_object_sync(struct mdt_thread_info *info)
1664 struct md_object *next;
1668 if (!mdt_object_exists(info->mti_object)) {
1669 CWARN("Non existing object "DFID"!\n",
1670 PFID(mdt_object_fid(info->mti_object)));
1673 next = mdt_object_child(info->mti_object);
1674 rc = mo_object_sync(info->mti_env, next);
1679 static int mdt_sync(struct mdt_thread_info *info)
1681 struct req_capsule *pill = info->mti_pill;
1682 struct mdt_body *body;
1686 /* The fid may be zero, so we req_capsule_set manually */
1687 req_capsule_set(pill, &RQF_MDS_SYNC);
1689 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1691 RETURN(err_serious(-EINVAL));
1693 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1694 RETURN(err_serious(-ENOMEM));
1696 if (fid_seq(&body->fid1) == 0) {
1697 /* sync the whole device */
1698 rc = req_capsule_server_pack(pill);
1700 rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1702 rc = err_serious(rc);
1704 /* sync an object */
1705 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1707 rc = mdt_object_sync(info);
1709 struct md_object *next;
1710 const struct lu_fid *fid;
1711 struct lu_attr *la = &info->mti_attr.ma_attr;
1713 next = mdt_object_child(info->mti_object);
1714 info->mti_attr.ma_need = MA_INODE;
1715 info->mti_attr.ma_valid = 0;
1716 rc = mo_attr_get(info->mti_env, next,
1719 body = req_capsule_server_get(pill,
1721 fid = mdt_object_fid(info->mti_object);
1722 mdt_pack_attr2body(info, body, la, fid);
1726 rc = err_serious(rc);
1731 #ifdef HAVE_QUOTA_SUPPORT
1732 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1734 struct obd_quotactl *oqctl;
1735 struct req_capsule *pill = info->mti_pill;
1736 struct obd_export *exp = info->mti_exp;
1737 struct md_quota *mq = md_quota(info->mti_env);
1738 struct md_device *next = info->mti_mdt->mdt_child;
1742 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1746 /* remote client has no permission for quotacheck */
1747 if (unlikely(exp_connect_rmtclient(exp)))
1750 rc = req_capsule_server_pack(pill);
1755 rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next,
1760 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1762 struct obd_quotactl *oqctl, *repoqc;
1763 struct req_capsule *pill = info->mti_pill;
1764 struct obd_export *exp = info->mti_exp;
1765 struct md_quota *mq = md_quota(info->mti_env);
1766 struct md_device *next = info->mti_mdt->mdt_child;
1767 const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
1771 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1776 if (exp_connect_rmtclient(exp)) {
1777 struct ptlrpc_request *req = mdt_info_req(info);
1778 struct mdt_export_data *med = mdt_req2med(req);
1779 struct lustre_idmap_table *idmap = med->med_idmap;
1781 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1782 oqctl->qc_cmd != Q_GETINFO))
1786 if (oqctl->qc_type == USRQUOTA)
1787 id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1789 else if (oqctl->qc_type == GRPQUOTA)
1790 id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1795 if (id == CFS_IDMAP_NOTFOUND) {
1796 CDEBUG(D_QUOTA, "no mapping for id %u\n",
1802 rc = req_capsule_server_pack(pill);
1806 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1807 LASSERT(repoqc != NULL);
1810 switch (oqctl->qc_cmd) {
1812 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
1815 rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
1818 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
1822 rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id,
1826 rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id,
1830 rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id,
1834 rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id,
1838 rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id,
1841 case LUSTRE_Q_INVALIDATE:
1842 rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type);
1844 case LUSTRE_Q_FINVALIDATE:
1845 rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type);
1848 CERROR("unsupported mdt_quotactl command: %d\n",
1860 * OBD PING and other handlers.
1862 static int mdt_obd_ping(struct mdt_thread_info *info)
1867 req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1869 rc = target_handle_ping(mdt_info_req(info));
1871 rc = err_serious(rc);
1875 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1877 return err_serious(-EOPNOTSUPP);
1880 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1882 return err_serious(-EOPNOTSUPP);
1890 /** clone llog ctxt from child (mdd)
1891 * This allows remote llog (replicator) access.
1892 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
1893 * context was originally set up, or we can handle them directly.
1894 * I choose the latter, but that means I need any llog
1895 * contexts set up by child to be accessable by the mdt. So we clone the
1896 * context into our context list here.
1898 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
1901 struct md_device *next = mdt->mdt_child;
1902 struct llog_ctxt *ctxt;
1905 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
1908 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
1909 if (rc || ctxt == NULL) {
1910 CERROR("Can't get mdd ctxt %d\n", rc);
1914 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
1916 CERROR("Can't set mdt ctxt %d\n", rc);
1921 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
1922 struct mdt_device *mdt, int idx)
1924 struct llog_ctxt *ctxt;
1926 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
1929 /* Put once for the get we just did, and once for the clone */
1930 llog_ctxt_put(ctxt);
1931 llog_ctxt_put(ctxt);
1935 static int mdt_llog_create(struct mdt_thread_info *info)
1939 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
1940 rc = llog_origin_handle_create(mdt_info_req(info));
1941 return (rc < 0 ? err_serious(rc) : rc);
1944 static int mdt_llog_destroy(struct mdt_thread_info *info)
1948 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
1949 rc = llog_origin_handle_destroy(mdt_info_req(info));
1950 return (rc < 0 ? err_serious(rc) : rc);
1953 static int mdt_llog_read_header(struct mdt_thread_info *info)
1957 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
1958 rc = llog_origin_handle_read_header(mdt_info_req(info));
1959 return (rc < 0 ? err_serious(rc) : rc);
1962 static int mdt_llog_next_block(struct mdt_thread_info *info)
1966 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
1967 rc = llog_origin_handle_next_block(mdt_info_req(info));
1968 return (rc < 0 ? err_serious(rc) : rc);
1971 static int mdt_llog_prev_block(struct mdt_thread_info *info)
1975 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
1976 rc = llog_origin_handle_prev_block(mdt_info_req(info));
1977 return (rc < 0 ? err_serious(rc) : rc);
1984 static struct ldlm_callback_suite cbs = {
1985 .lcs_completion = ldlm_server_completion_ast,
1986 .lcs_blocking = ldlm_server_blocking_ast,
1990 static int mdt_enqueue(struct mdt_thread_info *info)
1992 struct ptlrpc_request *req;
1996 * info->mti_dlm_req already contains swapped and (if necessary)
1997 * converted dlm request.
1999 LASSERT(info->mti_dlm_req != NULL);
2001 req = mdt_info_req(info);
2002 rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
2003 req, info->mti_dlm_req, &cbs);
2004 info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
2005 return rc ? err_serious(rc) : req->rq_status;
2008 static int mdt_convert(struct mdt_thread_info *info)
2011 struct ptlrpc_request *req;
2013 LASSERT(info->mti_dlm_req);
2014 req = mdt_info_req(info);
2015 rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2016 return rc ? err_serious(rc) : req->rq_status;
2019 static int mdt_bl_callback(struct mdt_thread_info *info)
2021 CERROR("bl callbacks should not happen on MDS\n");
2023 return err_serious(-EOPNOTSUPP);
2026 static int mdt_cp_callback(struct mdt_thread_info *info)
2028 CERROR("cp callbacks should not happen on MDS\n");
2030 return err_serious(-EOPNOTSUPP);
2034 * sec context handlers
2036 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2040 rc = mdt_handle_idmap(info);
2043 struct ptlrpc_request *req = mdt_info_req(info);
2046 opc = lustre_msg_get_opc(req->rq_reqmsg);
2047 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2048 sptlrpc_svc_ctx_invalidate(req);
2051 OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
2056 static struct mdt_object *mdt_obj(struct lu_object *o)
2058 LASSERT(lu_device_is_mdt(o->lo_dev));
2059 return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2062 struct mdt_object *mdt_object_find(const struct lu_env *env,
2063 struct mdt_device *d,
2064 const struct lu_fid *f)
2066 struct lu_object *o;
2067 struct mdt_object *m;
2070 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2071 o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2072 if (unlikely(IS_ERR(o)))
2073 m = (struct mdt_object *)o;
2080 * Asyncronous commit for mdt device.
2082 * Pass asynchonous commit call down the MDS stack.
2084 * \param env environment
2085 * \param mdt the mdt device
2087 static void mdt_device_commit_async(const struct lu_env *env,
2088 struct mdt_device *mdt)
2090 struct dt_device *dt = mdt->mdt_bottom;
2093 rc = dt->dd_ops->dt_commit_async(env, dt);
2094 if (unlikely(rc != 0))
2095 CWARN("async commit start failed with rc = %d", rc);
2099 * Mark the lock as "synchonous".
2101 * Mark the lock to deffer transaction commit to the unlock time.
2103 * \param lock the lock to mark as "synchonous"
2105 * \see mdt_is_lock_sync
2106 * \see mdt_save_lock
2108 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2110 lock->l_ast_data = (void*)1;
2114 * Check whehter the lock "synchonous" or not.
2116 * \param lock the lock to check
2117 * \retval 1 the lock is "synchonous"
2118 * \retval 0 the lock isn't "synchronous"
2120 * \see mdt_set_lock_sync
2121 * \see mdt_save_lock
2123 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2125 return lock->l_ast_data != NULL;
2129 * Blocking AST for mdt locks.
2131 * Starts transaction commit if in case of COS lock conflict or
2132 * deffers such a commit to the mdt_save_lock.
2134 * \param lock the lock which blocks a request or cancelling lock
2135 * \param desc unused
2136 * \param data unused
2137 * \param flag indicates whether this cancelling or blocking callback
2139 * \see ldlm_blocking_ast_nocheck
2141 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2142 void *data, int flag)
2144 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2145 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2149 if (flag == LDLM_CB_CANCELING)
2151 lock_res_and_lock(lock);
2152 if (lock->l_blocking_ast != mdt_blocking_ast) {
2153 unlock_res_and_lock(lock);
2156 if (mdt_cos_is_enabled(mdt) &&
2157 lock->l_req_mode & (LCK_PW | LCK_EX) &&
2158 lock->l_blocking_lock != NULL &&
2159 lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2160 mdt_set_lock_sync(lock);
2162 rc = ldlm_blocking_ast_nocheck(lock);
2164 /* There is no lock conflict if l_blocking_lock == NULL,
2165 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2166 * when the last reference to a local lock was released */
2167 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2170 rc = lu_env_init(&env, LCT_MD_THREAD);
2171 if (unlikely(rc != 0))
2172 CWARN("lu_env initialization failed with rc = %d,"
2173 "cannot start asynchronous commit\n", rc);
2175 mdt_device_commit_async(&env, mdt);
2181 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2182 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2184 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2185 ldlm_policy_data_t *policy = &info->mti_policy;
2186 struct ldlm_res_id *res_id = &info->mti_res_id;
2190 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2191 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2192 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2193 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2195 if (mdt_object_exists(o) < 0) {
2196 if (locality == MDT_CROSS_LOCK) {
2197 /* cross-ref object fix */
2198 ibits &= ~MDS_INODELOCK_UPDATE;
2199 ibits |= MDS_INODELOCK_LOOKUP;
2201 LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2202 LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2204 /* No PDO lock on remote object */
2205 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2208 if (lh->mlh_type == MDT_PDO_LOCK) {
2209 /* check for exists after object is locked */
2210 if (mdt_object_exists(o) == 0) {
2211 /* Non-existent object shouldn't have PDO lock */
2214 /* Non-dir object shouldn't have PDO lock */
2215 if (!S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)))
2220 memset(policy, 0, sizeof(*policy));
2221 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2224 * Take PDO lock on whole directory and build correct @res_id for lock
2225 * on part of directory.
2227 if (lh->mlh_pdo_hash != 0) {
2228 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2229 mdt_lock_pdo_mode(info, o, lh);
2230 if (lh->mlh_pdo_mode != LCK_NL) {
2232 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2233 * is never going to be sent to client and we do not
2234 * want it slowed down due to possible cancels.
2236 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2237 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2238 policy, res_id, LDLM_FL_ATOMIC_CB,
2239 &info->mti_exp->exp_handle.h_cookie);
2245 * Finish res_id initializing by name hash marking part of
2246 * directory which is taking modification.
2248 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2251 policy->l_inodebits.bits = ibits;
2254 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2255 * going to be sent to client. If it is - mdt_intent_policy() path will
2256 * fix it up and turn FL_LOCAL flag off.
2258 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2259 res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
2260 &info->mti_exp->exp_handle.h_cookie);
2262 mdt_object_unlock(info, o, lh, 1);
2263 else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2264 lh->mlh_pdo_hash != 0 &&
2265 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2266 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 10);
2273 * Save a lock within request object.
2275 * Keep the lock referenced until whether client ACK or transaction
2276 * commit happens or release the lock immediately depending on input
2277 * parameters. If COS is ON, a write lock is converted to COS lock
2280 * \param info thead info object
2281 * \param h lock handle
2282 * \param mode lock mode
2283 * \param decref force immediate lock releasing
2286 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2287 ldlm_mode_t mode, int decref)
2291 if (lustre_handle_is_used(h)) {
2292 if (decref || !info->mti_has_trans ||
2293 !(mode & (LCK_PW | LCK_EX))){
2294 mdt_fid_unlock(h, mode);
2296 struct mdt_device *mdt = info->mti_mdt;
2297 struct ldlm_lock *lock = ldlm_handle2lock(h);
2298 struct ptlrpc_request *req = mdt_info_req(info);
2301 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2303 CDEBUG(D_HA, "request = %p reply state = %p"
2304 " transno = "LPD64"\n",
2305 req, req->rq_reply_state, req->rq_transno);
2306 if (mdt_cos_is_enabled(mdt)) {
2308 ldlm_lock_downgrade(lock, LCK_COS);
2311 ptlrpc_save_lock(req, h, mode, no_ack);
2312 if (mdt_is_lock_sync(lock)) {
2313 CDEBUG(D_HA, "found sync-lock,"
2314 " async commit started\n");
2315 mdt_device_commit_async(info->mti_env,
2318 LDLM_LOCK_PUT(lock);
2327 * Unlock mdt object.
2329 * Immeditely release the regular lock and the PDO lock or save the
2330 * lock in reqeuest and keep them referenced until client ACK or
2331 * transaction commit.
2333 * \param info thread info object
2334 * \param o mdt object
2335 * \param lh mdt lock handle referencing regular and PDO locks
2336 * \param decref force immediate lock releasing
2338 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2339 struct mdt_lock_handle *lh, int decref)
2343 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2344 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2349 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2350 const struct lu_fid *f,
2351 struct mdt_lock_handle *lh,
2354 struct mdt_object *o;
2356 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2360 rc = mdt_object_lock(info, o, lh, ibits,
2363 mdt_object_put(info->mti_env, o);
2370 void mdt_object_unlock_put(struct mdt_thread_info * info,
2371 struct mdt_object * o,
2372 struct mdt_lock_handle *lh,
2375 mdt_object_unlock(info, o, lh, decref);
2376 mdt_object_put(info->mti_env, o);
2379 static struct mdt_handler *mdt_handler_find(__u32 opc,
2380 struct mdt_opc_slice *supported)
2382 struct mdt_opc_slice *s;
2383 struct mdt_handler *h;
2386 for (s = supported; s->mos_hs != NULL; s++) {
2387 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2388 h = s->mos_hs + (opc - s->mos_opc_start);
2389 if (likely(h->mh_opc != 0))
2390 LASSERTF(h->mh_opc == opc,
2391 "opcode mismatch %d != %d\n",
2394 h = NULL; /* unsupported opc */
2401 static int mdt_lock_resname_compat(struct mdt_device *m,
2402 struct ldlm_request *req)
2404 /* XXX something... later. */
2408 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2410 /* XXX something... later. */
2415 * Generic code handling requests that have struct mdt_body passed in:
2417 * - extract mdt_body from request and save it in @info, if present;
2419 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2422 * - if HABEO_CORPUS flag is set for this request type check whether object
2423 * actually exists on storage (lu_object_exists()).
2426 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2428 const struct mdt_body *body;
2429 struct mdt_object *obj;
2430 const struct lu_env *env;
2431 struct req_capsule *pill;
2435 env = info->mti_env;
2436 pill = info->mti_pill;
2438 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2442 if (!(body->valid & OBD_MD_FLID))
2445 if (!fid_is_sane(&body->fid1)) {
2446 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2451 * Do not get size or any capa fields before we check that request
2452 * contains capa actually. There are some requests which do not, for
2453 * instance MDS_IS_SUBDIR.
2455 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2456 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2457 mdt_set_capainfo(info, 0, &body->fid1,
2458 req_capsule_client_get(pill, &RMF_CAPA1));
2460 obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2462 if ((flags & HABEO_CORPUS) &&
2463 !mdt_object_exists(obj)) {
2464 mdt_object_put(env, obj);
2465 /* for capability renew ENOENT will be handled in
2467 if (body->valid & OBD_MD_FLOSSCAPA)
2472 info->mti_object = obj;
2481 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2483 struct req_capsule *pill = info->mti_pill;
2487 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2488 rc = mdt_body_unpack(info, flags);
2492 if (rc == 0 && (flags & HABEO_REFERO)) {
2493 struct mdt_device *mdt = info->mti_mdt;
2497 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2498 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2499 mdt->mdt_max_mdsize);
2500 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2501 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2502 mdt->mdt_max_cookiesize);
2504 rc = req_capsule_server_pack(pill);
2509 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2511 struct md_device *next = m->mdt_child;
2513 return next->md_ops->mdo_init_capa_ctxt(env, next,
2514 m->mdt_opts.mo_mds_capa,
2515 m->mdt_capa_timeout,
2521 * Invoke handler for this request opc. Also do necessary preprocessing
2522 * (according to handler ->mh_flags), and post-processing (setting of
2523 * ->last_{xid,committed}).
2525 static int mdt_req_handle(struct mdt_thread_info *info,
2526 struct mdt_handler *h, struct ptlrpc_request *req)
2528 int rc, serious = 0;
2533 LASSERT(h->mh_act != NULL);
2534 LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2535 LASSERT(current->journal_info == NULL);
2538 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2539 * to put same checks into handlers like mdt_close(), mdt_reint(),
2540 * etc., without talking to mdt authors first. Checking same thing
2541 * there again is useless and returning 0 error without packing reply
2542 * is buggy! Handlers either pack reply or return error.
2544 * We return 0 here and do not send any reply in order to emulate
2545 * network failure. Do not send any reply in case any of NET related
2546 * fail_id has occured.
2548 if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2552 flags = h->mh_flags;
2553 LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2555 if (h->mh_fmt != NULL) {
2556 req_capsule_set(info->mti_pill, h->mh_fmt);
2557 rc = mdt_unpack_req_pack_rep(info, flags);
2560 if (rc == 0 && flags & MUTABOR &&
2561 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2562 /* should it be rq_status? */
2565 if (rc == 0 && flags & HABEO_CLAVIS) {
2566 struct ldlm_request *dlm_req;
2568 LASSERT(h->mh_fmt != NULL);
2570 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2571 if (dlm_req != NULL) {
2572 if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2574 dlm_req->lock_desc.l_policy_data.\
2575 l_inodebits.bits == 0)) {
2577 * Lock without inodebits makes no sense and
2578 * will oops later in ldlm. If client miss to
2579 * set such bits, do not trigger ASSERTION.
2581 * For liblustre flock case, it maybe zero.
2585 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2586 rc = mdt_lock_resname_compat(
2589 info->mti_dlm_req = dlm_req;
2596 /* capability setting changed via /proc, needs reinitialize ctxt */
2597 if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2598 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2599 info->mti_mdt->mdt_capa_conf = 0;
2602 if (likely(rc == 0)) {
2604 * Process request, there can be two types of rc:
2605 * 1) errors with msg unpack/pack, other failures outside the
2606 * operation itself. This is counted as serious errors;
2607 * 2) errors during fs operation, should be placed in rq_status
2610 rc = h->mh_act(info);
2612 !req->rq_no_reply && req->rq_reply_state == NULL) {
2613 DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2614 "pack reply and returned 0 error\n",
2618 serious = is_serious(rc);
2619 rc = clear_serious(rc);
2623 req->rq_status = rc;
2626 * ELDLM_* codes which > 0 should be in rq_status only as well as
2627 * all non-serious errors.
2629 if (rc > 0 || !serious)
2632 LASSERT(current->journal_info == NULL);
2634 if (rc == 0 && (flags & HABEO_CLAVIS) &&
2635 info->mti_mdt->mdt_opts.mo_compat_resname) {
2636 struct ldlm_reply *dlmrep;
2638 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2640 rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2643 /* If we're DISCONNECTing, the mdt_export_data is already freed */
2644 if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2645 target_committed_to_req(req);
2647 if (unlikely(req_is_replay(req) &&
2648 lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2649 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2653 target_send_reply(req, rc, info->mti_fail_id);
2657 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2659 lh->mlh_type = MDT_NUL_LOCK;
2660 lh->mlh_reg_lh.cookie = 0ull;
2661 lh->mlh_reg_mode = LCK_MINMODE;
2662 lh->mlh_pdo_lh.cookie = 0ull;
2663 lh->mlh_pdo_mode = LCK_MINMODE;
2666 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2668 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2669 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2673 * Initialize fields of struct mdt_thread_info. Other fields are left in
2674 * uninitialized state, because it's too expensive to zero out whole
2675 * mdt_thread_info (> 1K) on each request arrival.
2677 static void mdt_thread_info_init(struct ptlrpc_request *req,
2678 struct mdt_thread_info *info)
2681 struct md_capainfo *ci;
2683 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2684 info->mti_pill = &req->rq_pill;
2687 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2688 mdt_lock_handle_init(&info->mti_lh[i]);
2690 /* mdt device: it can be NULL while CONNECT */
2691 if (req->rq_export) {
2692 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2693 info->mti_exp = req->rq_export;
2695 info->mti_mdt = NULL;
2696 info->mti_env = req->rq_svc_thread->t_env;
2697 ci = md_capainfo(info->mti_env);
2698 memset(ci, 0, sizeof *ci);
2699 if (req->rq_export) {
2700 if (exp_connect_rmtclient(req->rq_export))
2701 ci->mc_auth = LC_ID_CONVERT;
2702 else if (req->rq_export->exp_connect_flags &
2703 OBD_CONNECT_MDS_CAPA)
2704 ci->mc_auth = LC_ID_PLAIN;
2706 ci->mc_auth = LC_ID_NONE;
2709 info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2710 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2711 info->mti_mos = NULL;
2713 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2714 info->mti_body = NULL;
2715 info->mti_object = NULL;
2716 info->mti_dlm_req = NULL;
2717 info->mti_has_trans = 0;
2718 info->mti_no_need_trans = 0;
2719 info->mti_cross_ref = 0;
2720 info->mti_opdata = 0;
2722 /* To not check for split by default. */
2723 info->mti_spec.sp_ck_split = 0;
2724 info->mti_spec.no_create = 0;
2727 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2731 req_capsule_fini(info->mti_pill);
2732 if (info->mti_object != NULL) {
2734 * freeing an object may lead to OSD level transaction, do not
2735 * let it mess with MDT. bz19385.
2737 info->mti_no_need_trans = 1;
2738 mdt_object_put(info->mti_env, info->mti_object);
2739 info->mti_object = NULL;
2741 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2742 mdt_lock_handle_fini(&info->mti_lh[i]);
2743 info->mti_env = NULL;
2746 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2747 struct obd_device *obd, int *process)
2749 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2750 case MDS_CONNECT: /* This will never get here, but for completeness. */
2751 case OST_CONNECT: /* This will never get here, but for completeness. */
2752 case MDS_DISCONNECT:
2753 case OST_DISCONNECT:
2758 case MDS_DONE_WRITING:
2759 case MDS_SYNC: /* used in unmounting */
2765 *process = target_queue_recovery_request(req, obd);
2769 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2776 * Handle recovery. Return:
2777 * +1: continue request processing;
2778 * -ve: abort immediately with the given error code;
2779 * 0: send reply with error code in req->rq_status;
2781 static int mdt_recovery(struct mdt_thread_info *info)
2783 struct ptlrpc_request *req = mdt_info_req(info);
2784 struct obd_device *obd;
2788 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2791 case SEC_CTX_INIT_CONT:
2797 rc = mdt_handle_idmap(info);
2806 if (unlikely(!class_connected_export(req->rq_export))) {
2807 CERROR("operation %d on unconnected MDS from %s\n",
2808 lustre_msg_get_opc(req->rq_reqmsg),
2809 libcfs_id2str(req->rq_peer));
2810 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2811 * mds_A will get -ENOTCONN(especially for ping req),
2812 * which will cause that mds_A deactive timeout, then when
2813 * mds_A cleanup, the cleanup process will be suspended since
2814 * deactive timeout is not zero.
2816 req->rq_status = -ENOTCONN;
2817 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2821 /* sanity check: if the xid matches, the request must be marked as a
2822 * resent or replayed */
2823 if (req_xid_is_last(req)) {
2824 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2825 (MSG_RESENT | MSG_REPLAY))) {
2826 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2827 "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2828 lustre_msg_get_flags(req->rq_reqmsg));
2830 req->rq_status = -ENOTCONN;
2835 /* else: note the opposite is not always true; a RESENT req after a
2836 * failover will usually not match the last_xid, since it was likely
2837 * never committed. A REPLAYed request will almost never match the
2838 * last xid, however it could for a committed, but still retained,
2841 obd = req->rq_export->exp_obd;
2843 /* Check for aborted recovery... */
2844 if (unlikely(obd->obd_recovering)) {
2847 DEBUG_REQ(D_INFO, req, "Got new replay");
2848 rc = mdt_filter_recovery_request(req, obd, &should_process);
2849 if (rc != 0 || !should_process)
2851 else if (should_process < 0) {
2852 req->rq_status = should_process;
2853 rc = ptlrpc_error(req);
2860 static int mdt_msg_check_version(struct lustre_msg *msg)
2864 switch (lustre_msg_get_opc(msg)) {
2866 case MDS_DISCONNECT:
2869 case SEC_CTX_INIT_CONT:
2871 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2873 CERROR("bad opc %u version %08x, expecting %08x\n",
2874 lustre_msg_get_opc(msg),
2875 lustre_msg_get_version(msg),
2876 LUSTRE_OBD_VERSION);
2880 case MDS_GETATTR_NAME:
2887 case MDS_DONE_WRITING:
2894 case MDS_QUOTACHECK:
2900 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2902 CERROR("bad opc %u version %08x, expecting %08x\n",
2903 lustre_msg_get_opc(msg),
2904 lustre_msg_get_version(msg),
2905 LUSTRE_MDS_VERSION);
2909 case LDLM_BL_CALLBACK:
2910 case LDLM_CP_CALLBACK:
2911 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2913 CERROR("bad opc %u version %08x, expecting %08x\n",
2914 lustre_msg_get_opc(msg),
2915 lustre_msg_get_version(msg),
2916 LUSTRE_DLM_VERSION);
2918 case OBD_LOG_CANCEL:
2919 case LLOG_ORIGIN_HANDLE_CREATE:
2920 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2921 case LLOG_ORIGIN_HANDLE_READ_HEADER:
2922 case LLOG_ORIGIN_HANDLE_CLOSE:
2923 case LLOG_ORIGIN_HANDLE_DESTROY:
2924 case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2926 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2928 CERROR("bad opc %u version %08x, expecting %08x\n",
2929 lustre_msg_get_opc(msg),
2930 lustre_msg_get_version(msg),
2931 LUSTRE_LOG_VERSION);
2934 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2940 static int mdt_handle0(struct ptlrpc_request *req,
2941 struct mdt_thread_info *info,
2942 struct mdt_opc_slice *supported)
2944 struct mdt_handler *h;
2945 struct lustre_msg *msg;
2950 if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2953 LASSERT(current->journal_info == NULL);
2955 msg = req->rq_reqmsg;
2956 rc = mdt_msg_check_version(msg);
2957 if (likely(rc == 0)) {
2958 rc = mdt_recovery(info);
2959 if (likely(rc == +1)) {
2960 h = mdt_handler_find(lustre_msg_get_opc(msg),
2962 if (likely(h != NULL)) {
2963 rc = mdt_req_handle(info, h, req);
2965 CERROR("The unsupported opc: 0x%x\n",
2966 lustre_msg_get_opc(msg) );
2967 req->rq_status = -ENOTSUPP;
2968 rc = ptlrpc_error(req);
2973 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2978 * MDT handler function called by ptlrpc service thread when request comes.
2980 * XXX common "target" functionality should be factored into separate module
2981 * shared by mdt, ost and stand-alone services like fld.
2983 static int mdt_handle_common(struct ptlrpc_request *req,
2984 struct mdt_opc_slice *supported)
2987 struct mdt_thread_info *info;
2991 env = req->rq_svc_thread->t_env;
2992 LASSERT(env != NULL);
2993 LASSERT(env->le_ses != NULL);
2994 LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2995 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2996 LASSERT(info != NULL);
2998 mdt_thread_info_init(req, info);
3000 rc = mdt_handle0(req, info, supported);
3002 mdt_thread_info_fini(info);
3007 * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
3010 int mdt_recovery_handle(struct ptlrpc_request *req)
3015 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3017 rc = mdt_handle_common(req, mdt_fld_handlers);
3020 rc = mdt_handle_common(req, mdt_seq_handlers);
3023 rc = mdt_handle_common(req, mdt_regular_handlers);
3030 static int mdt_regular_handle(struct ptlrpc_request *req)
3032 return mdt_handle_common(req, mdt_regular_handlers);
3035 static int mdt_readpage_handle(struct ptlrpc_request *req)
3037 return mdt_handle_common(req, mdt_readpage_handlers);
3040 static int mdt_xmds_handle(struct ptlrpc_request *req)
3042 return mdt_handle_common(req, mdt_xmds_handlers);
3045 static int mdt_mdsc_handle(struct ptlrpc_request *req)
3047 return mdt_handle_common(req, mdt_seq_handlers);
3050 static int mdt_mdss_handle(struct ptlrpc_request *req)
3052 return mdt_handle_common(req, mdt_seq_handlers);
3055 static int mdt_dtss_handle(struct ptlrpc_request *req)
3057 return mdt_handle_common(req, mdt_seq_handlers);
3060 static int mdt_fld_handle(struct ptlrpc_request *req)
3062 return mdt_handle_common(req, mdt_fld_handlers);
3078 static int mdt_intent_getattr(enum mdt_it_code opcode,
3079 struct mdt_thread_info *info,
3080 struct ldlm_lock **,
3082 static int mdt_intent_reint(enum mdt_it_code opcode,
3083 struct mdt_thread_info *info,
3084 struct ldlm_lock **,
3087 static struct mdt_it_flavor {
3088 const struct req_format *it_fmt;
3090 int (*it_act)(enum mdt_it_code ,
3091 struct mdt_thread_info *,
3092 struct ldlm_lock **,
3095 } mdt_it_flavor[] = {
3097 .it_fmt = &RQF_LDLM_INTENT,
3098 /*.it_flags = HABEO_REFERO,*/
3100 .it_act = mdt_intent_reint,
3101 .it_reint = REINT_OPEN
3104 .it_fmt = &RQF_LDLM_INTENT,
3105 .it_flags = MUTABOR,
3106 .it_act = mdt_intent_reint,
3107 .it_reint = REINT_OPEN
3110 .it_fmt = &RQF_LDLM_INTENT,
3111 .it_flags = MUTABOR,
3112 .it_act = mdt_intent_reint,
3113 .it_reint = REINT_CREATE
3115 [MDT_IT_GETATTR] = {
3116 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3117 .it_flags = HABEO_REFERO,
3118 .it_act = mdt_intent_getattr
3120 [MDT_IT_READDIR] = {
3126 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3127 .it_flags = HABEO_REFERO,
3128 .it_act = mdt_intent_getattr
3131 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
3132 .it_flags = MUTABOR,
3134 .it_reint = REINT_UNLINK
3138 .it_flags = MUTABOR,
3141 [MDT_IT_GETXATTR] = {
3148 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3149 struct ldlm_lock **lockp,
3150 struct ldlm_lock *new_lock,
3151 struct mdt_lock_handle *lh,
3154 struct ptlrpc_request *req = mdt_info_req(info);
3155 struct ldlm_lock *lock = *lockp;
3158 * Get new lock only for cases when possible resent did not find any
3161 if (new_lock == NULL)
3162 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3164 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3165 lh->mlh_reg_lh.cookie = 0;
3169 LASSERTF(new_lock != NULL,
3170 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3173 * If we've already given this lock to a client once, then we should
3174 * have no readers or writers. Otherwise, we should have one reader
3175 * _or_ writer ref (which will be zeroed below) before returning the
3178 if (new_lock->l_export == req->rq_export) {
3179 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3181 LASSERT(new_lock->l_export == NULL);
3182 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3187 if (new_lock->l_export == req->rq_export) {
3189 * Already gave this to the client, which means that we
3190 * reconstructed a reply.
3192 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3194 lh->mlh_reg_lh.cookie = 0;
3195 RETURN(ELDLM_LOCK_REPLACED);
3199 * Fixup the lock to be given to the client.
3201 lock_res_and_lock(new_lock);
3202 /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3203 * possible blocking AST. */
3204 while (new_lock->l_readers > 0) {
3205 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3206 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3207 new_lock->l_readers--;
3209 while (new_lock->l_writers > 0) {
3210 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3211 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3212 new_lock->l_writers--;
3215 new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3216 new_lock->l_blocking_ast = lock->l_blocking_ast;
3217 new_lock->l_completion_ast = lock->l_completion_ast;
3218 new_lock->l_remote_handle = lock->l_remote_handle;
3219 new_lock->l_flags &= ~LDLM_FL_LOCAL;
3221 unlock_res_and_lock(new_lock);
3223 cfs_hash_add(new_lock->l_export->exp_lock_hash,
3224 &new_lock->l_remote_handle,
3225 &new_lock->l_exp_hash);
3227 LDLM_LOCK_RELEASE(new_lock);
3228 lh->mlh_reg_lh.cookie = 0;
3230 RETURN(ELDLM_LOCK_REPLACED);
3233 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3234 struct ldlm_lock *new_lock,
3235 struct ldlm_lock **old_lock,
3236 struct mdt_lock_handle *lh)
3238 struct ptlrpc_request *req = mdt_info_req(info);
3239 struct obd_export *exp = req->rq_export;
3240 struct lustre_handle remote_hdl;
3241 struct ldlm_request *dlmreq;
3242 struct ldlm_lock *lock;
3244 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3247 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3248 remote_hdl = dlmreq->lock_handle[0];
3250 lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3252 if (lock != new_lock) {
3253 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3254 lh->mlh_reg_mode = lock->l_granted_mode;
3256 LDLM_DEBUG(lock, "Restoring lock cookie");
3257 DEBUG_REQ(D_DLMTRACE, req,
3258 "restoring lock cookie "LPX64,
3259 lh->mlh_reg_lh.cookie);
3261 *old_lock = LDLM_LOCK_GET(lock);
3262 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3266 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3270 * If the xid matches, then we know this is a resent request, and allow
3271 * it. (It's probably an OPEN, for which we don't send a lock.
3273 if (req_xid_is_last(req))
3277 * This remote handle isn't enqueued, so we never received or processed
3278 * this request. Clear MSG_RESENT, because it can be handled like any
3279 * normal request now.
3281 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3283 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3287 static int mdt_intent_getattr(enum mdt_it_code opcode,
3288 struct mdt_thread_info *info,
3289 struct ldlm_lock **lockp,
3292 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3293 struct ldlm_lock *new_lock = NULL;
3295 struct ldlm_reply *ldlm_rep;
3296 struct ptlrpc_request *req;
3297 struct mdt_body *reqbody;
3298 struct mdt_body *repbody;
3302 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3305 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3308 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
3309 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3310 repbody->eadatasize = 0;
3311 repbody->aclsize = 0;
3315 child_bits = MDS_INODELOCK_LOOKUP;
3317 case MDT_IT_GETATTR:
3318 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3321 CERROR("Unhandled till now");
3322 GOTO(out_shrink, rc = -EINVAL);
3325 rc = mdt_init_ucred(info, reqbody);
3327 GOTO(out_shrink, rc);
3329 req = info->mti_pill->rc_req;
3330 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3331 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3333 /* Get lock from request for possible resent case. */
3334 mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3336 ldlm_rep->lock_policy_res2 =
3337 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3339 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3340 ldlm_rep->lock_policy_res2 = 0;
3341 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3342 ldlm_rep->lock_policy_res2) {
3343 lhc->mlh_reg_lh.cookie = 0ull;
3344 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3347 rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3350 mdt_exit_ucred(info);
3352 mdt_shrink_reply(info);
3356 static int mdt_intent_reint(enum mdt_it_code opcode,
3357 struct mdt_thread_info *info,
3358 struct ldlm_lock **lockp,
3361 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3362 struct ldlm_reply *rep = NULL;
3366 static const struct req_format *intent_fmts[REINT_MAX] = {
3367 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
3368 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
3373 opc = mdt_reint_opcode(info, intent_fmts);
3377 if (mdt_it_flavor[opcode].it_reint != opc) {
3378 CERROR("Reint code %ld doesn't match intent: %d\n",
3380 RETURN(err_serious(-EPROTO));
3383 /* Get lock from request for possible resent case. */
3384 mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3386 rc = mdt_reint_internal(info, lhc, opc);
3388 /* Check whether the reply has been packed successfully. */
3389 if (mdt_info_req(info)->rq_repmsg != NULL)
3390 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3392 RETURN(err_serious(-EFAULT));
3394 /* MDC expects this in any case */
3396 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3398 /* Cross-ref case, the lock should be returned to the client */
3399 if (rc == -EREMOTE) {
3400 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3401 rep->lock_policy_res2 = 0;
3402 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3405 rep->lock_policy_res2 = clear_serious(rc);
3407 if (rc == -ENOTCONN || rc == -ENODEV ||
3408 rc == -EOVERFLOW) { /**< if VBR failure then return error */
3410 * If it is the disconnect error (ENODEV & ENOCONN), the error
3411 * will be returned by rq_status, and client at ptlrpc layer
3412 * will detect this, then disconnect, reconnect the import
3413 * immediately, instead of impacting the following the rpc.
3415 lhc->mlh_reg_lh.cookie = 0ull;
3419 * For other cases, the error will be returned by intent.
3420 * and client will retrieve the result from intent.
3423 * FIXME: when open lock is finished, that should be
3426 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3427 LASSERTF(rc == 0, "Error occurred but lock handle "
3428 "is still in use\n");
3429 rep->lock_policy_res2 = 0;
3430 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3433 lhc->mlh_reg_lh.cookie = 0ull;
3434 RETURN(ELDLM_LOCK_ABORTED);
3439 static int mdt_intent_code(long itcode)
3447 case IT_OPEN|IT_CREAT:
3454 rc = MDT_IT_READDIR;
3457 rc = MDT_IT_GETATTR;
3469 rc = MDT_IT_GETXATTR;
3472 CERROR("Unknown intent opcode: %ld\n", itcode);
3479 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3480 struct ldlm_lock **lockp, int flags)
3482 struct req_capsule *pill;
3483 struct mdt_it_flavor *flv;
3488 opc = mdt_intent_code(itopc);
3492 pill = info->mti_pill;
3493 flv = &mdt_it_flavor[opc];
3495 if (flv->it_fmt != NULL)
3496 req_capsule_extend(pill, flv->it_fmt);
3498 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3500 struct ptlrpc_request *req = mdt_info_req(info);
3501 if (flv->it_flags & MUTABOR &&
3502 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3505 if (rc == 0 && flv->it_act != NULL) {
3506 /* execute policy */
3507 rc = flv->it_act(opc, info, lockp, flags);
3514 static int mdt_intent_policy(struct ldlm_namespace *ns,
3515 struct ldlm_lock **lockp, void *req_cookie,
3516 ldlm_mode_t mode, int flags, void *data)
3518 struct mdt_thread_info *info;
3519 struct ptlrpc_request *req = req_cookie;
3520 struct ldlm_intent *it;
3521 struct req_capsule *pill;
3526 LASSERT(req != NULL);
3528 info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3530 LASSERT(info != NULL);
3531 pill = info->mti_pill;
3532 LASSERT(pill->rc_req == req);
3534 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3535 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3536 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3538 rc = mdt_intent_opc(it->opc, info, lockp, flags);
3542 /* Lock without inodebits makes no sense and will oops
3543 * later in ldlm. Let's check it now to see if we have
3544 * ibits corrupted somewhere in mdt_intent_opc().
3545 * The case for client miss to set ibits has been
3546 * processed by others. */
3547 LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3548 lr_type == LDLM_IBITS,
3549 info->mti_dlm_req->lock_desc.\
3550 l_policy_data.l_inodebits.bits != 0));
3552 rc = err_serious(-EFAULT);
3554 /* No intent was provided */
3555 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3556 rc = req_capsule_server_pack(pill);
3558 rc = err_serious(rc);
3563 static int mdt_seq_fini(const struct lu_env *env,
3564 struct mdt_device *m)
3566 struct md_site *ms = mdt_md_site(m);
3570 if (ms->ms_server_seq) {
3571 seq_server_fini(ms->ms_server_seq, env);
3572 OBD_FREE_PTR(ms->ms_server_seq);
3573 ms->ms_server_seq = NULL;
3576 if (ms->ms_control_seq) {
3577 seq_server_fini(ms->ms_control_seq, env);
3578 OBD_FREE_PTR(ms->ms_control_seq);
3579 ms->ms_control_seq = NULL;
3582 if (ms->ms_client_seq) {
3583 seq_client_fini(ms->ms_client_seq);
3584 OBD_FREE_PTR(ms->ms_client_seq);
3585 ms->ms_client_seq = NULL;
3592 static int mdt_seq_init(const struct lu_env *env,
3594 struct mdt_device *m)
3601 ms = mdt_md_site(m);
3604 * This is sequence-controller node. Init seq-controller server on local
3607 if (ms->ms_node_id == 0) {
3608 LASSERT(ms->ms_control_seq == NULL);
3610 OBD_ALLOC_PTR(ms->ms_control_seq);
3611 if (ms->ms_control_seq == NULL)
3614 rc = seq_server_init(ms->ms_control_seq,
3615 m->mdt_bottom, uuid,
3616 LUSTRE_SEQ_CONTROLLER,
3621 GOTO(out_seq_fini, rc);
3623 OBD_ALLOC_PTR(ms->ms_client_seq);
3624 if (ms->ms_client_seq == NULL)
3625 GOTO(out_seq_fini, rc = -ENOMEM);
3627 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3628 if (prefix == NULL) {
3629 OBD_FREE_PTR(ms->ms_client_seq);
3630 GOTO(out_seq_fini, rc = -ENOMEM);
3633 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3637 * Init seq-controller client after seq-controller server is
3638 * ready. Pass ms->ms_control_seq to it for direct talking.
3640 rc = seq_client_init(ms->ms_client_seq, NULL,
3641 LUSTRE_SEQ_METADATA, prefix,
3642 ms->ms_control_seq);
3643 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3646 GOTO(out_seq_fini, rc);
3649 /* Init seq-server on local MDT */
3650 LASSERT(ms->ms_server_seq == NULL);
3652 OBD_ALLOC_PTR(ms->ms_server_seq);
3653 if (ms->ms_server_seq == NULL)
3654 GOTO(out_seq_fini, rc = -ENOMEM);
3656 rc = seq_server_init(ms->ms_server_seq,
3657 m->mdt_bottom, uuid,
3662 GOTO(out_seq_fini, rc = -ENOMEM);
3664 /* Assign seq-controller client to local seq-server. */
3665 if (ms->ms_node_id == 0) {
3666 LASSERT(ms->ms_client_seq != NULL);
3668 rc = seq_server_set_cli(ms->ms_server_seq,
3676 mdt_seq_fini(env, m);
3681 * Init client sequence manager which is used by local MDS to talk to sequence
3682 * controller on remote node.
3684 static int mdt_seq_init_cli(const struct lu_env *env,
3685 struct mdt_device *m,
3686 struct lustre_cfg *cfg)
3688 struct md_site *ms = mdt_md_site(m);
3689 struct obd_device *mdc;
3690 struct obd_uuid *uuidp, *mdcuuidp;
3691 char *uuid_str, *mdc_uuid_str;
3694 struct mdt_thread_info *info;
3695 char *p, *index_string = lustre_cfg_string(cfg, 2);
3698 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3699 uuidp = &info->mti_u.uuid[0];
3700 mdcuuidp = &info->mti_u.uuid[1];
3702 LASSERT(index_string);
3704 index = simple_strtol(index_string, &p, 10);
3706 CERROR("Invalid index in lustre_cgf, offset 2\n");
3710 /* check if this is adding the first MDC and controller is not yet
3712 if (index != 0 || ms->ms_client_seq)
3715 uuid_str = lustre_cfg_string(cfg, 1);
3716 mdc_uuid_str = lustre_cfg_string(cfg, 4);
3717 obd_str2uuid(uuidp, uuid_str);
3718 obd_str2uuid(mdcuuidp, mdc_uuid_str);
3720 mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3722 CERROR("can't find controller MDC by uuid %s\n",
3725 } else if (!mdc->obd_set_up) {
3726 CERROR("target %s not set up\n", mdc->obd_name);
3729 LASSERT(ms->ms_control_exp);
3730 OBD_ALLOC_PTR(ms->ms_client_seq);
3731 if (ms->ms_client_seq != NULL) {
3734 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3738 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3741 rc = seq_client_init(ms->ms_client_seq,
3743 LUSTRE_SEQ_METADATA,
3745 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3752 LASSERT(ms->ms_server_seq != NULL);
3753 rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
3760 static void mdt_seq_fini_cli(struct mdt_device *m)
3766 ms = mdt_md_site(m);
3769 if (ms->ms_server_seq)
3770 seq_server_set_cli(ms->ms_server_seq,
3773 if (ms->ms_control_exp) {
3774 class_export_put(ms->ms_control_exp);
3775 ms->ms_control_exp = NULL;
3784 static int mdt_fld_fini(const struct lu_env *env,
3785 struct mdt_device *m)
3787 struct md_site *ms = mdt_md_site(m);
3790 if (ms && ms->ms_server_fld) {
3791 fld_server_fini(ms->ms_server_fld, env);
3792 OBD_FREE_PTR(ms->ms_server_fld);
3793 ms->ms_server_fld = NULL;
3799 static int mdt_fld_init(const struct lu_env *env,
3801 struct mdt_device *m)
3807 ms = mdt_md_site(m);
3809 OBD_ALLOC_PTR(ms->ms_server_fld);
3810 if (ms->ms_server_fld == NULL)
3811 RETURN(rc = -ENOMEM);
3813 rc = fld_server_init(ms->ms_server_fld,
3814 m->mdt_bottom, uuid,
3815 env, ms->ms_node_id);
3817 OBD_FREE_PTR(ms->ms_server_fld);
3818 ms->ms_server_fld = NULL;
3825 /* device init/fini methods */
3826 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3829 if (m->mdt_regular_service != NULL) {
3830 ptlrpc_unregister_service(m->mdt_regular_service);
3831 m->mdt_regular_service = NULL;
3833 if (m->mdt_readpage_service != NULL) {
3834 ptlrpc_unregister_service(m->mdt_readpage_service);
3835 m->mdt_readpage_service = NULL;
3837 if (m->mdt_xmds_service != NULL) {
3838 ptlrpc_unregister_service(m->mdt_xmds_service);
3839 m->mdt_xmds_service = NULL;
3841 if (m->mdt_setattr_service != NULL) {
3842 ptlrpc_unregister_service(m->mdt_setattr_service);
3843 m->mdt_setattr_service = NULL;
3845 if (m->mdt_mdsc_service != NULL) {
3846 ptlrpc_unregister_service(m->mdt_mdsc_service);
3847 m->mdt_mdsc_service = NULL;
3849 if (m->mdt_mdss_service != NULL) {
3850 ptlrpc_unregister_service(m->mdt_mdss_service);
3851 m->mdt_mdss_service = NULL;
3853 if (m->mdt_dtss_service != NULL) {
3854 ptlrpc_unregister_service(m->mdt_dtss_service);
3855 m->mdt_dtss_service = NULL;
3857 if (m->mdt_fld_service != NULL) {
3858 ptlrpc_unregister_service(m->mdt_fld_service);
3859 m->mdt_fld_service = NULL;
3864 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3867 static struct ptlrpc_service_conf conf;
3868 cfs_proc_dir_entry_t *procfs_entry;
3871 procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3873 conf = (typeof(conf)) {
3874 .psc_nbufs = MDS_NBUFS,
3875 .psc_bufsize = MDS_BUFSIZE,
3876 .psc_max_req_size = MDS_MAXREQSIZE,
3877 .psc_max_reply_size = MDS_MAXREPSIZE,
3878 .psc_req_portal = MDS_REQUEST_PORTAL,
3879 .psc_rep_portal = MDC_REPLY_PORTAL,
3880 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3882 * We'd like to have a mechanism to set this on a per-device
3883 * basis, but alas...
3885 .psc_min_threads = mdt_min_threads,
3886 .psc_max_threads = mdt_max_threads,
3887 .psc_ctx_tags = LCT_MD_THREAD
3890 m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3891 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3892 "mdt_ldlm_client", m->mdt_ldlm_client);
3894 m->mdt_regular_service =
3895 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3896 procfs_entry, target_print_req,
3898 if (m->mdt_regular_service == NULL)
3901 rc = ptlrpc_start_threads(m->mdt_regular_service);
3903 GOTO(err_mdt_svc, rc);
3906 * readpage service configuration. Parameters have to be adjusted,
3909 conf = (typeof(conf)) {
3910 .psc_nbufs = MDS_NBUFS,
3911 .psc_bufsize = MDS_BUFSIZE,
3912 .psc_max_req_size = MDS_MAXREQSIZE,
3913 .psc_max_reply_size = MDS_MAXREPSIZE,
3914 .psc_req_portal = MDS_READPAGE_PORTAL,
3915 .psc_rep_portal = MDC_REPLY_PORTAL,
3916 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3917 .psc_min_threads = mdt_min_threads,
3918 .psc_max_threads = mdt_max_threads,
3919 .psc_ctx_tags = LCT_MD_THREAD
3921 m->mdt_readpage_service =
3922 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3923 LUSTRE_MDT_NAME "_readpage",
3924 procfs_entry, target_print_req,"mdt_rdpg");
3926 if (m->mdt_readpage_service == NULL) {
3927 CERROR("failed to start readpage service\n");
3928 GOTO(err_mdt_svc, rc = -ENOMEM);
3931 rc = ptlrpc_start_threads(m->mdt_readpage_service);
3934 * setattr service configuration.
3936 conf = (typeof(conf)) {
3937 .psc_nbufs = MDS_NBUFS,
3938 .psc_bufsize = MDS_BUFSIZE,
3939 .psc_max_req_size = MDS_MAXREQSIZE,
3940 .psc_max_reply_size = MDS_MAXREPSIZE,
3941 .psc_req_portal = MDS_SETATTR_PORTAL,
3942 .psc_rep_portal = MDC_REPLY_PORTAL,
3943 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3944 .psc_min_threads = mdt_min_threads,
3945 .psc_max_threads = mdt_max_threads,
3946 .psc_ctx_tags = LCT_MD_THREAD
3949 m->mdt_setattr_service =
3950 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3951 LUSTRE_MDT_NAME "_setattr",
3952 procfs_entry, target_print_req,"mdt_attr");
3954 if (!m->mdt_setattr_service) {
3955 CERROR("failed to start setattr service\n");
3956 GOTO(err_mdt_svc, rc = -ENOMEM);
3959 rc = ptlrpc_start_threads(m->mdt_setattr_service);
3961 GOTO(err_mdt_svc, rc);
3964 * sequence controller service configuration
3966 conf = (typeof(conf)) {
3967 .psc_nbufs = MDS_NBUFS,
3968 .psc_bufsize = MDS_BUFSIZE,
3969 .psc_max_req_size = SEQ_MAXREQSIZE,
3970 .psc_max_reply_size = SEQ_MAXREPSIZE,
3971 .psc_req_portal = SEQ_CONTROLLER_PORTAL,
3972 .psc_rep_portal = MDC_REPLY_PORTAL,
3973 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3974 .psc_min_threads = mdt_min_threads,
3975 .psc_max_threads = mdt_max_threads,
3976 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3979 m->mdt_mdsc_service =
3980 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3981 LUSTRE_MDT_NAME"_mdsc",
3982 procfs_entry, target_print_req,"mdt_mdsc");
3983 if (!m->mdt_mdsc_service) {
3984 CERROR("failed to start seq controller service\n");
3985 GOTO(err_mdt_svc, rc = -ENOMEM);
3988 rc = ptlrpc_start_threads(m->mdt_mdsc_service);
3990 GOTO(err_mdt_svc, rc);
3993 * metadata sequence server service configuration
3995 conf = (typeof(conf)) {
3996 .psc_nbufs = MDS_NBUFS,
3997 .psc_bufsize = MDS_BUFSIZE,
3998 .psc_max_req_size = SEQ_MAXREQSIZE,
3999 .psc_max_reply_size = SEQ_MAXREPSIZE,
4000 .psc_req_portal = SEQ_METADATA_PORTAL,
4001 .psc_rep_portal = MDC_REPLY_PORTAL,
4002 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4003 .psc_min_threads = mdt_min_threads,
4004 .psc_max_threads = mdt_max_threads,
4005 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
4008 m->mdt_mdss_service =
4009 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
4010 LUSTRE_MDT_NAME"_mdss",
4011 procfs_entry, target_print_req,"mdt_mdss");
4012 if (!m->mdt_mdss_service) {
4013 CERROR("failed to start metadata seq server service\n");
4014 GOTO(err_mdt_svc, rc = -ENOMEM);
4017 rc = ptlrpc_start_threads(m->mdt_mdss_service);
4019 GOTO(err_mdt_svc, rc);
4023 * Data sequence server service configuration. We want to have really
4024 * cluster-wide sequences space. This is why we start only one sequence
4025 * controller which manages space.
4027 conf = (typeof(conf)) {
4028 .psc_nbufs = MDS_NBUFS,
4029 .psc_bufsize = MDS_BUFSIZE,
4030 .psc_max_req_size = SEQ_MAXREQSIZE,
4031 .psc_max_reply_size = SEQ_MAXREPSIZE,
4032 .psc_req_portal = SEQ_DATA_PORTAL,
4033 .psc_rep_portal = OSC_REPLY_PORTAL,
4034 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4035 .psc_min_threads = mdt_min_threads,
4036 .psc_max_threads = mdt_max_threads,
4037 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
4040 m->mdt_dtss_service =
4041 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
4042 LUSTRE_MDT_NAME"_dtss",
4043 procfs_entry, target_print_req,"mdt_dtss");
4044 if (!m->mdt_dtss_service) {
4045 CERROR("failed to start data seq server service\n");
4046 GOTO(err_mdt_svc, rc = -ENOMEM);
4049 rc = ptlrpc_start_threads(m->mdt_dtss_service);
4051 GOTO(err_mdt_svc, rc);
4053 /* FLD service start */
4054 conf = (typeof(conf)) {
4055 .psc_nbufs = MDS_NBUFS,
4056 .psc_bufsize = MDS_BUFSIZE,
4057 .psc_max_req_size = FLD_MAXREQSIZE,
4058 .psc_max_reply_size = FLD_MAXREPSIZE,
4059 .psc_req_portal = FLD_REQUEST_PORTAL,
4060 .psc_rep_portal = MDC_REPLY_PORTAL,
4061 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4062 .psc_min_threads = mdt_min_threads,
4063 .psc_max_threads = mdt_max_threads,
4064 .psc_ctx_tags = LCT_DT_THREAD|LCT_MD_THREAD
4067 m->mdt_fld_service =
4068 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
4069 LUSTRE_MDT_NAME"_fld",
4070 procfs_entry, target_print_req, "mdt_fld");
4071 if (!m->mdt_fld_service) {
4072 CERROR("failed to start fld service\n");
4073 GOTO(err_mdt_svc, rc = -ENOMEM);
4076 rc = ptlrpc_start_threads(m->mdt_fld_service);
4078 GOTO(err_mdt_svc, rc);
4081 * mds-mds service configuration. Separate portal is used to allow
4082 * mds-mds requests be not blocked during recovery.
4084 conf = (typeof(conf)) {
4085 .psc_nbufs = MDS_NBUFS,
4086 .psc_bufsize = MDS_BUFSIZE,
4087 .psc_max_req_size = MDS_MAXREQSIZE,
4088 .psc_max_reply_size = MDS_MAXREPSIZE,
4089 .psc_req_portal = MDS_MDS_PORTAL,
4090 .psc_rep_portal = MDC_REPLY_PORTAL,
4091 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4092 .psc_min_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 xmds service\n");
4103 GOTO(err_mdt_svc, rc = -ENOMEM);
4106 rc = ptlrpc_start_threads(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_exports_barrier(obd);
4341 obd_zombie_barrier();
4342 #ifdef HAVE_QUOTA_SUPPORT
4343 next->md_ops->mdo_quota.mqo_cleanup(env, next);
4345 lut_fini(env, &m->mdt_lut);
4346 mdt_fs_cleanup(env, m);
4347 upcall_cache_cleanup(m->mdt_identity_cache);
4348 m->mdt_identity_cache = NULL;
4350 if (m->mdt_namespace != NULL) {
4351 ldlm_namespace_free(m->mdt_namespace, NULL,
4352 d->ld_obd->obd_force);
4353 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4356 cfs_free_nidlist(&m->mdt_nosquash_nids);
4357 if (m->mdt_nosquash_str) {
4358 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4359 m->mdt_nosquash_str = NULL;
4360 m->mdt_nosquash_strlen = 0;
4363 mdt_seq_fini(env, m);
4364 mdt_seq_fini_cli(m);
4365 mdt_fld_fini(env, m);
4366 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4368 next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4369 cfs_timer_disarm(&m->mdt_ck_timer);
4370 mdt_ck_thread_stop(m);
4375 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4377 lprocfs_free_per_client_stats(obd);
4378 lprocfs_free_obd_stats(obd);
4382 struct md_site *mite;
4385 mite = lu_site2md(ls);
4389 LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
4394 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4396 struct mdt_device *m = mdt_dev(obd->obd_lu_dev);
4397 struct sptlrpc_rule_set tmp_rset;
4400 sptlrpc_rule_set_init(&tmp_rset);
4401 rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4403 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4408 sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4410 cfs_write_lock(&m->mdt_sptlrpc_lock);
4411 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4412 m->mdt_sptlrpc_rset = tmp_rset;
4413 cfs_write_unlock(&m->mdt_sptlrpc_lock);
4418 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
4422 m->mdt_opts.mo_mds_capa = 1;
4423 m->mdt_opts.mo_oss_capa = 1;
4424 #ifdef CONFIG_FS_POSIX_ACL
4425 /* ACLs should be enabled by default (b=13829) */
4426 m->mdt_opts.mo_acl = 1;
4427 LCONSOLE_INFO("Enabling ACL\n");
4429 m->mdt_opts.mo_acl = 0;
4430 LCONSOLE_INFO("Disabling ACL\n");
4439 while (*p && *p != ',')
4443 if ((len == sizeof("user_xattr") - 1) &&
4444 (memcmp(options, "user_xattr", len) == 0)) {
4445 m->mdt_opts.mo_user_xattr = 1;
4446 LCONSOLE_INFO("Enabling user_xattr\n");
4447 } else if ((len == sizeof("nouser_xattr") - 1) &&
4448 (memcmp(options, "nouser_xattr", len) == 0)) {
4449 m->mdt_opts.mo_user_xattr = 0;
4450 LCONSOLE_INFO("Disabling user_xattr\n");
4451 } else if ((len == sizeof("noacl") - 1) &&
4452 (memcmp(options, "noacl", len) == 0)) {
4453 m->mdt_opts.mo_acl = 0;
4454 LCONSOLE_INFO("Disabling ACL\n");
4464 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4466 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4467 struct lu_device_type *ldt, struct lustre_cfg *cfg)
4469 struct mdt_thread_info *info;
4470 struct obd_device *obd;
4471 const char *dev = lustre_cfg_string(cfg, 0);
4472 const char *num = lustre_cfg_string(cfg, 2);
4473 struct lustre_mount_info *lmi = NULL;
4474 struct lustre_sb_info *lsi;
4475 struct lustre_disk_data *ldd;
4477 struct md_site *mite;
4478 const char *identity_upcall = "NONE";
4479 #ifdef HAVE_QUOTA_SUPPORT
4480 struct md_device *next;
4486 md_device_init(&m->mdt_md_dev, ldt);
4488 * Environment (env) might be missing mdt_thread_key values at that
4489 * point, if device is allocated when mdt_thread_key is in QUIESCENT
4492 * Usually device allocation path doesn't use module key values, but
4493 * mdt has to do a lot of work here, so allocate key value.
4495 rc = lu_env_refill((struct lu_env *)env);
4499 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4500 LASSERT(info != NULL);
4502 obd = class_name2obd(dev);
4503 LASSERT(obd != NULL);
4505 m->mdt_max_mdsize = MAX_MD_SIZE;
4506 m->mdt_max_cookiesize = sizeof(struct llog_cookie);
4507 m->mdt_som_conf = 0;
4509 m->mdt_opts.mo_user_xattr = 0;
4510 m->mdt_opts.mo_acl = 0;
4511 m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4512 lmi = server_get_mount_2(dev);
4514 CERROR("Cannot get mount info for %s!\n", dev);
4517 lsi = s2lsi(lmi->lmi_sb);
4518 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
4519 /* CMD is supported only in IAM mode */
4522 node_id = simple_strtol(num, NULL, 10);
4523 if (!(ldd->ldd_flags & LDD_F_IAM_DIR) && node_id) {
4524 CERROR("CMD Operation not allowed in IOP mode\n");
4525 GOTO(err_lmi, rc = -EINVAL);
4527 /* Read recovery timeouts */
4528 if (lsi->lsi_lmd && lsi->lsi_lmd->lmd_recovery_time_soft)
4529 obd->obd_recovery_timeout =
4530 lsi->lsi_lmd->lmd_recovery_time_soft;
4532 if (lsi->lsi_lmd && lsi->lsi_lmd->lmd_recovery_time_hard)
4533 obd->obd_recovery_time_hard =
4534 lsi->lsi_lmd->lmd_recovery_time_hard;
4537 cfs_rwlock_init(&m->mdt_sptlrpc_lock);
4538 sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4540 cfs_spin_lock_init(&m->mdt_ioepoch_lock);
4541 m->mdt_opts.mo_compat_resname = 0;
4542 m->mdt_capa_timeout = CAPA_TIMEOUT;
4543 m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4544 m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4545 m->mdt_squash_uid = 0;
4546 m->mdt_squash_gid = 0;
4547 CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4548 m->mdt_nosquash_str = NULL;
4549 m->mdt_nosquash_strlen = 0;
4550 cfs_init_rwsem(&m->mdt_squash_sem);
4552 OBD_ALLOC_PTR(mite);
4554 GOTO(err_lmi, rc = -ENOMEM);
4558 m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4559 m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4560 /* set this lu_device to obd, because error handling need it */
4561 obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4563 rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
4565 CERROR("Can't init lu_site, rc %d\n", rc);
4566 GOTO(err_free_site, rc);
4569 rc = mdt_procfs_init(m, dev);
4571 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4572 GOTO(err_fini_proc, rc);
4575 /* set server index */
4576 lu_site2md(s)->ms_node_id = node_id;
4578 /* failover is the default
4579 * FIXME: we do not failout mds0/mgs, which may cause some problems.
4580 * assumed whose ms_node_id == 0 XXX
4582 obd->obd_replayable = 1;
4583 /* No connection accepted until configurations will finish */
4584 obd->obd_no_conn = 1;
4586 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4587 char *str = lustre_cfg_string(cfg, 4);
4588 if (strchr(str, 'n')) {
4589 CWARN("%s: recovery disabled\n", obd->obd_name);
4590 obd->obd_replayable = 0;
4594 /* init the stack */
4595 rc = mdt_stack_init((struct lu_env *)env, m, cfg, lmi);
4597 CERROR("Can't init device stack, rc %d\n", rc);
4598 GOTO(err_fini_proc, rc);
4601 rc = lut_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4603 GOTO(err_fini_stack, rc);
4605 rc = mdt_fld_init(env, obd->obd_name, m);
4609 rc = mdt_seq_init(env, obd->obd_name, m);
4611 GOTO(err_fini_fld, rc);
4613 snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4614 LUSTRE_MDT_NAME"-%p", m);
4615 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4616 LDLM_NAMESPACE_SERVER,
4617 LDLM_NAMESPACE_GREEDY,
4619 if (m->mdt_namespace == NULL)
4620 GOTO(err_fini_seq, rc = -ENOMEM);
4622 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4623 /* set obd_namespace for compatibility with old code */
4624 obd->obd_namespace = m->mdt_namespace;
4626 /* XXX: to support suppgid for ACL, we enable identity_upcall
4627 * by default, otherwise, maybe got unexpected -EACCESS. */
4628 if (m->mdt_opts.mo_acl)
4629 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4631 m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
4632 &mdt_identity_upcall_cache_ops);
4633 if (IS_ERR(m->mdt_identity_cache)) {
4634 rc = PTR_ERR(m->mdt_identity_cache);
4635 m->mdt_identity_cache = NULL;
4636 GOTO(err_free_ns, rc);
4639 cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4641 rc = mdt_ck_thread_start(m);
4643 GOTO(err_free_ns, rc);
4645 rc = mdt_fs_setup(env, m, obd, lsi);
4649 rc = mdt_obd_llog_setup(obd, lsi);
4651 GOTO(err_fs_cleanup, rc);
4653 rc = mdt_llog_ctxt_clone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4655 GOTO(err_llog_cleanup, rc);
4657 mdt_adapt_sptlrpc_conf(obd, 1);
4659 #ifdef HAVE_QUOTA_SUPPORT
4660 next = m->mdt_child;
4661 rc = next->md_ops->mdo_quota.mqo_setup(env, next, lmi->lmi_mnt);
4663 GOTO(err_llog_cleanup, rc);
4666 server_put_mount_2(dev, lmi->lmi_mnt);
4669 target_recovery_init(&m->mdt_lut, mdt_recovery_handle);
4671 rc = mdt_start_ptlrpc_service(m);
4673 GOTO(err_recovery, rc);
4675 ping_evictor_start();
4677 rc = lu_site_init_finish(s);
4679 GOTO(err_stop_service, rc);
4681 if (obd->obd_recovering == 0)
4682 mdt_postrecov(env, m);
4684 mdt_init_capa_ctxt(env, m);
4686 /* Reduce the initial timeout on an MDS because it doesn't need such
4687 * a long timeout as an OST does. Adaptive timeouts will adjust this
4688 * value appropriately. */
4689 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4690 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4695 ping_evictor_stop();
4696 mdt_stop_ptlrpc_service(m);
4698 target_recovery_fini(obd);
4699 #ifdef HAVE_QUOTA_SUPPORT
4700 next->md_ops->mdo_quota.mqo_cleanup(env, next);
4703 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4704 mdt_obd_llog_cleanup(obd);
4706 mdt_fs_cleanup(env, m);
4708 cfs_timer_disarm(&m->mdt_ck_timer);
4709 mdt_ck_thread_stop(m);
4711 upcall_cache_cleanup(m->mdt_identity_cache);
4712 m->mdt_identity_cache = NULL;
4713 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4714 obd->obd_namespace = m->mdt_namespace = NULL;
4716 mdt_seq_fini(env, m);
4718 mdt_fld_fini(env, m);
4720 lut_fini(env, &m->mdt_lut);
4722 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4730 server_put_mount_2(dev, lmi->lmi_mnt);
4734 /* used by MGS to process specific configurations */
4735 static int mdt_process_config(const struct lu_env *env,
4736 struct lu_device *d, struct lustre_cfg *cfg)
4738 struct mdt_device *m = mdt_dev(d);
4739 struct md_device *md_next = m->mdt_child;
4740 struct lu_device *next = md2lu_dev(md_next);
4744 switch (cfg->lcfg_command) {
4746 struct lprocfs_static_vars lvars;
4747 struct obd_device *obd = d->ld_obd;
4750 * For interoperability between 1.8 and 2.0,
4751 * skip old "mdt.group_upcall" param.
4754 char *param = lustre_cfg_string(cfg, 1);
4755 if (param && !strncmp("mdt.group_upcall", param, 16)) {
4756 CWARN("For 1.8 interoperability, skip this"
4757 " mdt.group_upcall. It is obsolete\n");
4762 lprocfs_mdt_init_vars(&lvars);
4763 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4765 if (rc > 0 || rc == -ENOSYS)
4766 /* we don't understand; pass it on */
4767 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4772 * Add mdc hook to get first MDT uuid and connect it to
4773 * ls->controller to use for seq manager.
4775 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4777 CERROR("Can't add mdc, rc %d\n", rc);
4779 rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4782 /* others are passed further */
4783 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4789 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4790 const struct lu_object_header *hdr,
4791 struct lu_device *d)
4793 struct mdt_object *mo;
4799 struct lu_object *o;
4800 struct lu_object_header *h;
4802 o = &mo->mot_obj.mo_lu;
4803 h = &mo->mot_header;
4804 lu_object_header_init(h);
4805 lu_object_init(o, h, d);
4806 lu_object_add_top(h, o);
4807 o->lo_ops = &mdt_obj_ops;
4808 cfs_sema_init(&mo->mot_ioepoch_sem, 1);
4814 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4815 const struct lu_object_conf *unused)
4817 struct mdt_device *d = mdt_dev(o->lo_dev);
4818 struct lu_device *under;
4819 struct lu_object *below;
4823 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4824 PFID(lu_object_fid(o)));
4826 under = &d->mdt_child->md_lu_dev;
4827 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4828 if (below != NULL) {
4829 lu_object_add(o, below);
4836 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4838 struct mdt_object *mo = mdt_obj(o);
4839 struct lu_object_header *h;
4843 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4844 PFID(lu_object_fid(o)));
4847 lu_object_header_fini(h);
4852 static int mdt_object_print(const struct lu_env *env, void *cookie,
4853 lu_printer_t p, const struct lu_object *o)
4855 struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4856 return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4857 "flags="LPX64", epochcount=%d, writecount=%d)",
4858 mdto, mdto->mot_ioepoch, mdto->mot_flags,
4859 mdto->mot_ioepoch_count, mdto->mot_writecount);
4862 static const struct lu_device_operations mdt_lu_ops = {
4863 .ldo_object_alloc = mdt_object_alloc,
4864 .ldo_process_config = mdt_process_config,
4867 static const struct lu_object_operations mdt_obj_ops = {
4868 .loo_object_init = mdt_object_init,
4869 .loo_object_free = mdt_object_free,
4870 .loo_object_print = mdt_object_print
4873 static int mdt_obd_set_info_async(struct obd_export *exp,
4874 __u32 keylen, void *key,
4875 __u32 vallen, void *val,
4876 struct ptlrpc_request_set *set)
4878 struct obd_device *obd = exp->exp_obd;
4884 if (KEY_IS(KEY_SPTLRPC_CONF)) {
4885 rc = mdt_adapt_sptlrpc_conf(obd, 0);
4892 /* mds_connect_internal */
4893 static int mdt_connect_internal(struct obd_export *exp,
4894 struct mdt_device *mdt,
4895 struct obd_connect_data *data)
4898 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4899 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4901 /* If no known bits (which should not happen, probably,
4902 as everybody should support LOOKUP and UPDATE bits at least)
4903 revert to compat mode with plain locks. */
4904 if (!data->ocd_ibits_known &&
4905 data->ocd_connect_flags & OBD_CONNECT_IBITS)
4906 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4908 if (!mdt->mdt_opts.mo_acl)
4909 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4911 if (!mdt->mdt_opts.mo_user_xattr)
4912 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4914 if (!mdt->mdt_som_conf)
4915 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4917 cfs_spin_lock(&exp->exp_lock);
4918 exp->exp_connect_flags = data->ocd_connect_flags;
4919 cfs_spin_unlock(&exp->exp_lock);
4920 data->ocd_version = LUSTRE_VERSION_CODE;
4921 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4925 if (mdt->mdt_opts.mo_acl &&
4926 ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4927 CWARN("%s: MDS requires ACL support but client does not\n",
4928 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4933 if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4934 CWARN("%s: MDS requires FID support, but client not\n",
4935 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4939 if (mdt->mdt_som_conf && !exp_connect_som(exp) &&
4940 !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS)) {
4941 CWARN("%s: MDS has SOM enabled, but client does not support "
4942 "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4949 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
4950 struct obd_export *exp,
4951 struct ptlrpc_request *req)
4953 struct sptlrpc_flavor flvr;
4956 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4957 cfs_read_lock(&mdt->mdt_sptlrpc_lock);
4958 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
4962 cfs_read_unlock(&mdt->mdt_sptlrpc_lock);
4964 cfs_spin_lock(&exp->exp_lock);
4966 exp->exp_sp_peer = req->rq_sp_from;
4967 exp->exp_flvr = flvr;
4969 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
4970 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4971 CERROR("unauthorized rpc flavor %x from %s, "
4972 "expect %x\n", req->rq_flvr.sf_rpc,
4973 libcfs_nid2str(req->rq_peer.nid),
4974 exp->exp_flvr.sf_rpc);
4978 cfs_spin_unlock(&exp->exp_lock);
4980 if (exp->exp_sp_peer != req->rq_sp_from) {
4981 CERROR("RPC source %s doesn't match %s\n",
4982 sptlrpc_part2name(req->rq_sp_from),
4983 sptlrpc_part2name(exp->exp_sp_peer));
4986 rc = sptlrpc_target_export_check(exp, req);
4993 /* mds_connect copy */
4994 static int mdt_obd_connect(const struct lu_env *env,
4995 struct obd_export **exp, struct obd_device *obd,
4996 struct obd_uuid *cluuid,
4997 struct obd_connect_data *data,
5000 struct mdt_thread_info *info;
5001 struct obd_export *lexp;
5002 struct lustre_handle conn = { 0 };
5003 struct mdt_device *mdt;
5004 struct ptlrpc_request *req;
5008 LASSERT(env != NULL);
5009 if (!exp || !obd || !cluuid)
5012 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5013 req = info->mti_pill->rc_req;
5014 mdt = mdt_dev(obd->obd_lu_dev);
5016 rc = class_connect(&conn, obd, cluuid);
5020 lexp = class_conn2export(&conn);
5021 LASSERT(lexp != NULL);
5023 rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
5027 rc = mdt_connect_internal(lexp, mdt, data);
5029 struct mdt_thread_info *mti;
5030 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5032 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5033 LASSERT(mti != NULL);
5034 mti->mti_exp = lexp;
5035 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5036 rc = mdt_client_new(env, mdt);
5038 mdt_export_stats_init(obd, lexp, 0, localdata);
5043 class_disconnect(lexp);
5052 static int mdt_obd_reconnect(const struct lu_env *env,
5053 struct obd_export *exp, struct obd_device *obd,
5054 struct obd_uuid *cluuid,
5055 struct obd_connect_data *data,
5058 struct mdt_thread_info *info;
5059 struct mdt_device *mdt;
5060 struct ptlrpc_request *req;
5064 if (exp == NULL || obd == NULL || cluuid == NULL)
5067 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5068 req = info->mti_pill->rc_req;
5069 mdt = mdt_dev(obd->obd_lu_dev);
5071 rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5075 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5077 mdt_export_stats_init(obd, exp, 1, localdata);
5081 static int mdt_export_cleanup(struct obd_export *exp)
5083 struct mdt_export_data *med = &exp->exp_mdt_data;
5084 struct obd_device *obd = exp->exp_obd;
5085 struct mdt_device *mdt;
5086 struct mdt_thread_info *info;
5088 CFS_LIST_HEAD(closing_list);
5089 struct mdt_file_data *mfd, *n;
5093 cfs_spin_lock(&med->med_open_lock);
5094 while (!cfs_list_empty(&med->med_open_head)) {
5095 cfs_list_t *tmp = med->med_open_head.next;
5096 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5098 /* Remove mfd handle so it can't be found again.
5099 * We are consuming the mfd_list reference here. */
5100 class_handle_unhash(&mfd->mfd_handle);
5101 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5103 cfs_spin_unlock(&med->med_open_lock);
5104 mdt = mdt_dev(obd->obd_lu_dev);
5105 LASSERT(mdt != NULL);
5107 rc = lu_env_init(&env, LCT_MD_THREAD);
5111 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5112 LASSERT(info != NULL);
5113 memset(info, 0, sizeof *info);
5114 info->mti_env = &env;
5115 info->mti_mdt = mdt;
5116 info->mti_exp = exp;
5118 if (!cfs_list_empty(&closing_list)) {
5119 struct md_attr *ma = &info->mti_attr;
5123 lmm_size = mdt->mdt_max_mdsize;
5124 OBD_ALLOC(ma->ma_lmm, lmm_size);
5125 if (ma->ma_lmm == NULL)
5126 GOTO(out_lmm, rc = -ENOMEM);
5128 cookie_size = mdt->mdt_max_cookiesize;
5129 OBD_ALLOC(ma->ma_cookie, cookie_size);
5130 if (ma->ma_cookie == NULL)
5131 GOTO(out_cookie, rc = -ENOMEM);
5133 /* Close any open files (which may also cause orphan unlinking). */
5134 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5135 cfs_list_del_init(&mfd->mfd_list);
5136 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
5137 ma->ma_lmm_size = lmm_size;
5138 ma->ma_cookie_size = cookie_size;
5140 /* It is not for setattr, just tell MDD to send
5141 * DESTROY RPC to OSS if needed */
5142 ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
5143 ma->ma_valid = MA_FLAGS;
5144 mdt_mfd_close(info, mfd);
5146 OBD_FREE(ma->ma_cookie, cookie_size);
5147 ma->ma_cookie = NULL;
5149 OBD_FREE(ma->ma_lmm, lmm_size);
5153 info->mti_mdt = NULL;
5154 /* cleanup client slot early */
5155 /* Do not erase record for recoverable client. */
5156 if (!obd->obd_fail || exp->exp_failed)
5157 mdt_client_del(&env, mdt);
5163 static int mdt_obd_disconnect(struct obd_export *exp)
5169 class_export_get(exp);
5171 rc = server_disconnect_export(exp);
5173 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5175 rc = mdt_export_cleanup(exp);
5176 class_export_put(exp);
5180 /* FIXME: Can we avoid using these two interfaces? */
5181 static int mdt_init_export(struct obd_export *exp)
5183 struct mdt_export_data *med = &exp->exp_mdt_data;
5187 CFS_INIT_LIST_HEAD(&med->med_open_head);
5188 cfs_spin_lock_init(&med->med_open_lock);
5189 cfs_sema_init(&med->med_idmap_sem, 1);
5190 med->med_idmap = NULL;
5191 cfs_spin_lock(&exp->exp_lock);
5192 exp->exp_connecting = 1;
5193 cfs_spin_unlock(&exp->exp_lock);
5194 rc = lut_client_alloc(exp);
5196 rc = ldlm_init_export(exp);
5199 CERROR("Error %d while initializing export\n", rc);
5203 static int mdt_destroy_export(struct obd_export *exp)
5205 struct mdt_export_data *med;
5209 med = &exp->exp_mdt_data;
5210 if (exp_connect_rmtclient(exp))
5211 mdt_cleanup_idmap(&exp->exp_mdt_data);
5213 target_destroy_export(exp);
5214 ldlm_destroy_export(exp);
5215 lut_client_free(exp);
5217 LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5218 LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5219 if (obd_uuid_equals(&exp->exp_client_uuid, &exp->exp_obd->obd_uuid))
5225 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
5227 if (flag & CONFIG_LOG)
5228 cfs_set_bit(MDT_FL_CFGLOG, &m->mdt_state);
5230 /* also notify active event */
5231 if (flag & CONFIG_SYNC)
5232 cfs_set_bit(MDT_FL_SYNCED, &m->mdt_state);
5234 if (cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state) &&
5235 cfs_test_bit(MDT_FL_SYNCED, &m->mdt_state)) {
5236 struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
5238 /* Open for clients */
5239 if (obd->obd_no_conn) {
5240 cfs_spin_lock(&obd->obd_dev_lock);
5241 obd->obd_no_conn = 0;
5242 cfs_spin_unlock(&obd->obd_dev_lock);
5247 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
5248 enum md_upcall_event ev, void *data)
5250 struct mdt_device *m = mdt_dev(&md->md_lu_dev);
5251 struct md_device *next = m->mdt_child;
5252 struct mdt_thread_info *mti;
5258 rc = next->md_ops->mdo_maxsize_get(env, next,
5260 &m->mdt_max_cookiesize);
5261 CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
5262 m->mdt_max_mdsize, m->mdt_max_cookiesize);
5263 mdt_allow_cli(m, CONFIG_SYNC);
5266 m->mdt_lut.lut_obd->u.obt.obt_mount_count;
5269 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5270 mti->mti_no_need_trans = 1;
5271 CDEBUG(D_INFO, "disable mdt trans for this thread\n");
5274 /* Check that MDT is not yet configured */
5275 LASSERT(!cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state));
5277 #ifdef HAVE_QUOTA_SUPPORT
5279 if (md->md_lu_dev.ld_obd->obd_recovering == 0 &&
5280 likely(md->md_lu_dev.ld_obd->obd_stopping == 0))
5281 next->md_ops->mdo_quota.mqo_recovery(env, next);
5285 CERROR("invalid event\n");
5292 static int mdt_obd_notify(struct obd_device *host,
5293 struct obd_device *watched,
5294 enum obd_notify_event ev, void *data)
5296 struct mdt_device *mdt = mdt_dev(host->obd_lu_dev);
5297 #ifdef HAVE_QUOTA_SUPPORT
5298 struct md_device *next = mdt->mdt_child;
5303 case OBD_NOTIFY_CONFIG:
5304 mdt_allow_cli(mdt, (unsigned long)data);
5306 #ifdef HAVE_QUOTA_SUPPORT
5307 /* quota_type has been processed, we can now handle
5308 * incoming quota requests */
5309 next->md_ops->mdo_quota.mqo_notify(NULL, next);
5313 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
5318 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5319 void *val, int vallen)
5321 struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev);
5322 struct getinfo_fid2path *fpout, *fpin;
5325 fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5328 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5329 lustre_swab_fid2path(fpin);
5331 memcpy(fpout, fpin, sizeof(*fpin));
5332 if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5335 rc = mdt_fid2path(info->mti_env, mdt, fpout);
5339 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
5340 struct getinfo_fid2path *fp)
5342 struct mdt_object *obj;
5346 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5347 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5349 if (!fid_is_sane(&fp->gf_fid))
5352 obj = mdt_object_find(env, mdt, &fp->gf_fid);
5353 if (obj == NULL || IS_ERR(obj)) {
5354 CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid),
5359 rc = lu_object_exists(&obj->mot_obj.mo_lu);
5365 mdt_object_put(env, obj);
5366 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5367 PFID(&fp->gf_fid), rc);
5371 rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path,
5372 fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno);
5373 mdt_object_put(env, obj);
5378 static int mdt_get_info(struct mdt_thread_info *info)
5380 struct ptlrpc_request *req = mdt_info_req(info);
5388 key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5390 CDEBUG(D_IOCTL, "No GETINFO key");
5393 keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5396 vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5397 if (vallen == NULL) {
5398 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5402 req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5404 rc = req_capsule_server_pack(info->mti_pill);
5405 valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5406 if (valout == NULL) {
5407 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5411 if (KEY_IS(KEY_FID2PATH))
5412 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5416 lustre_msg_set_status(req->rq_repmsg, rc);
5421 /* Pass the ioc down */
5422 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5423 unsigned int cmd, int len, void *data)
5425 struct lu_context ioctl_session;
5426 struct md_device *next = mdt->mdt_child;
5430 rc = lu_context_init(&ioctl_session, LCT_SESSION);
5433 ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5434 lu_context_enter(&ioctl_session);
5435 env->le_ses = &ioctl_session;
5437 LASSERT(next->md_ops->mdo_iocontrol);
5438 rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5440 lu_context_exit(&ioctl_session);
5441 lu_context_fini(&ioctl_session);
5445 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5447 struct obd_ioctl_data *data = karg;
5448 struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1;
5450 struct mdt_object *obj;
5451 struct mdt_lock_handle *lh;
5454 CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5455 if (!fid_is_sane(fid))
5458 lh = &mti->mti_lh[MDT_LH_PARENT];
5459 mdt_lock_reg_init(lh, LCK_CR);
5461 obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5463 RETURN(PTR_ERR(obj));
5465 rc = mdt_object_exists(obj);
5469 * before calling version get the correct MDS should be
5470 * fid, this is error to find remote object here
5472 CERROR("nonlocal object "DFID"\n", PFID(fid));
5474 version = mo_version_get(mti->mti_env, mdt_object_child(obj));
5475 *(__u64 *)data->ioc_inlbuf2 = version;
5478 mdt_object_unlock_put(mti, obj, lh, 1);
5482 /* ioctls on obd dev */
5483 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5484 void *karg, void *uarg)
5487 struct obd_device *obd = exp->exp_obd;
5488 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5489 struct dt_device *dt = mdt->mdt_bottom;
5493 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5494 rc = lu_env_init(&env, LCT_MD_THREAD);
5500 rc = mdt_device_sync(&env, mdt);
5502 case OBD_IOC_SET_READONLY:
5503 dt->dd_ops->dt_ro(&env, dt);
5505 case OBD_IOC_ABORT_RECOVERY:
5506 CERROR("Aborting recovery for device %s\n", obd->obd_name);
5507 target_stop_recovery_thread(obd);
5510 case OBD_IOC_CHANGELOG_REG:
5511 case OBD_IOC_CHANGELOG_DEREG:
5512 case OBD_IOC_CHANGELOG_CLEAR:
5513 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5515 case OBD_IOC_GET_OBJ_VERSION: {
5516 struct mdt_thread_info *mti;
5517 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5518 memset(mti, 0, sizeof *mti);
5519 mti->mti_env = &env;
5523 rc = mdt_ioc_version_get(mti, karg);
5527 CERROR("Not supported cmd = %d for device %s\n",
5528 cmd, obd->obd_name);
5536 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5538 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5539 #ifdef HAVE_QUOTA_SUPPORT
5540 struct obd_device *obd = mdt2obd_dev(mdt);
5541 struct md_device *next = mdt->mdt_child;
5546 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5547 #ifdef HAVE_QUOTA_SUPPORT
5548 if (likely(obd->obd_stopping == 0))
5549 next->md_ops->mdo_quota.mqo_recovery(env, next);
5554 int mdt_obd_postrecov(struct obd_device *obd)
5559 rc = lu_env_init(&env, LCT_MD_THREAD);
5562 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5568 * Send a copytool req to a client
5569 * Note this sends a request RPC from a server (MDT) to a client (MDC),
5570 * backwards of normal comms.
5572 int mdt_hsm_copytool_send(struct obd_export *exp)
5575 struct hsm_action_list *hal;
5576 struct hsm_action_item *hai;
5580 CWARN("%s: writing to mdc at %s\n", exp->exp_obd->obd_name,
5581 libcfs_nid2str(exp->exp_connection->c_peer.nid));
5583 len = sizeof(*lh) + sizeof(*hal) + MTI_NAME_MAXLEN +
5584 /* for mockup below */ 2 * cfs_size_round(sizeof(*hai));
5589 lh->kuc_magic = KUC_MAGIC;
5590 lh->kuc_transport = KUC_TRANSPORT_HSM;
5591 lh->kuc_msgtype = HMT_ACTION_LIST;
5592 lh->kuc_msglen = len;
5594 hal = (struct hsm_action_list *)(lh + 1);
5595 hal->hal_version = HAL_VERSION;
5596 hal->hal_archive_num = 1;
5597 obd_uuid2fsname(hal->hal_fsname, exp->exp_obd->obd_name,
5600 /* mock up an action list */
5602 hai = hai_zero(hal);
5603 hai->hai_action = HSMA_ARCHIVE;
5604 hai->hai_fid.f_oid = 0xA00A;
5605 hai->hai_len = sizeof(*hai);
5606 hai = hai_next(hai);
5607 hai->hai_action = HSMA_RESTORE;
5608 hai->hai_fid.f_oid = 0xB00B;
5609 hai->hai_len = sizeof(*hai);
5611 /* Uses the ldlm reverse import; this rpc will be seen by
5612 the ldlm_callback_handler */
5613 rc = do_set_info_async(exp->exp_imp_reverse,
5614 LDLM_SET_INFO, LUSTRE_OBD_VERSION,
5615 sizeof(KEY_HSM_COPYTOOL_SEND),
5616 KEY_HSM_COPYTOOL_SEND,
5624 static struct obd_ops mdt_obd_device_ops = {
5625 .o_owner = THIS_MODULE,
5626 .o_set_info_async = mdt_obd_set_info_async,
5627 .o_connect = mdt_obd_connect,
5628 .o_reconnect = mdt_obd_reconnect,
5629 .o_disconnect = mdt_obd_disconnect,
5630 .o_init_export = mdt_init_export,
5631 .o_destroy_export = mdt_destroy_export,
5632 .o_iocontrol = mdt_iocontrol,
5633 .o_postrecov = mdt_obd_postrecov,
5634 .o_notify = mdt_obd_notify
5637 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5638 struct lu_device *d)
5640 struct mdt_device *m = mdt_dev(d);
5647 static struct lu_device *mdt_device_free(const struct lu_env *env,
5648 struct lu_device *d)
5650 struct mdt_device *m = mdt_dev(d);
5653 md_device_fini(&m->mdt_md_dev);
5658 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5659 struct lu_device_type *t,
5660 struct lustre_cfg *cfg)
5662 struct lu_device *l;
5663 struct mdt_device *m;
5669 l = &m->mdt_md_dev.md_lu_dev;
5670 rc = mdt_init0(env, m, t, cfg);
5672 mdt_device_free(env, l);
5676 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
5678 l = ERR_PTR(-ENOMEM);
5682 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5683 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
5685 /* context key: mdt_thread_key */
5686 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5688 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
5689 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
5691 struct lu_context_key mdt_txn_key = {
5692 .lct_tags = LCT_TX_HANDLE,
5693 .lct_init = mdt_txn_key_init,
5694 .lct_fini = mdt_txn_key_fini
5697 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
5699 return md_ucred(info->mti_env);
5703 * Enable/disable COS (Commit On Sharing).
5705 * Set/Clear the COS flag in mdt options.
5707 * \param mdt mdt device
5708 * \param val 0 disables COS, other values enable COS
5710 void mdt_enable_cos(struct mdt_device *mdt, int val)
5715 mdt->mdt_opts.mo_cos = !!val;
5716 rc = lu_env_init(&env, LCT_MD_THREAD);
5717 if (unlikely(rc != 0)) {
5718 CWARN("lu_env initialization failed with rc = %d,"
5719 "cannot sync\n", rc);
5722 mdt_device_sync(&env, mdt);
5727 * Check COS (Commit On Sharing) status.
5729 * Return COS flag status.
5731 * \param mdt mdt device
5733 int mdt_cos_is_enabled(struct mdt_device *mdt)
5735 return mdt->mdt_opts.mo_cos != 0;
5738 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
5739 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
5741 static struct lu_device_type_operations mdt_device_type_ops = {
5742 .ldto_init = mdt_type_init,
5743 .ldto_fini = mdt_type_fini,
5745 .ldto_start = mdt_type_start,
5746 .ldto_stop = mdt_type_stop,
5748 .ldto_device_alloc = mdt_device_alloc,
5749 .ldto_device_free = mdt_device_free,
5750 .ldto_device_fini = mdt_device_fini
5753 static struct lu_device_type mdt_device_type = {
5754 .ldt_tags = LU_DEVICE_MD,
5755 .ldt_name = LUSTRE_MDT_NAME,
5756 .ldt_ops = &mdt_device_type_ops,
5757 .ldt_ctx_tags = LCT_MD_THREAD
5760 static struct lu_local_obj_desc mdt_last_recv = {
5761 .llod_name = LAST_RCVD,
5762 .llod_oid = MDT_LAST_RECV_OID,
5766 static int __init mdt_mod_init(void)
5768 struct lprocfs_static_vars lvars;
5771 llo_local_obj_register(&mdt_last_recv);
5773 if (mdt_num_threads > 0) {
5774 if (mdt_num_threads > MDT_MAX_THREADS)
5775 mdt_num_threads = MDT_MAX_THREADS;
5776 if (mdt_num_threads < MDT_MIN_THREADS)
5777 mdt_num_threads = MDT_MIN_THREADS;
5778 mdt_max_threads = mdt_min_threads = mdt_num_threads;
5780 mdt_max_threads = MDT_MAX_THREADS;
5781 mdt_min_threads = MDT_MIN_THREADS;
5784 lprocfs_mdt_init_vars(&lvars);
5785 rc = class_register_type(&mdt_obd_device_ops, NULL,
5786 lvars.module_vars, LUSTRE_MDT_NAME,
5792 static void __exit mdt_mod_exit(void)
5794 llo_local_obj_unregister(&mdt_last_recv);
5795 class_unregister_type(LUSTRE_MDT_NAME);
5799 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt) \
5800 [prefix ## _ ## opc - prefix ## _ ## base] = { \
5802 .mh_fail_id = OBD_FAIL_ ## prefix ## _ ## opc ## suffix, \
5803 .mh_opc = prefix ## _ ## opc, \
5804 .mh_flags = flags, \
5809 #define DEF_MDT_HNDL(flags, name, fn, fmt) \
5810 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
5812 #define DEF_SEQ_HNDL(flags, name, fn, fmt) \
5813 DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
5815 #define DEF_FLD_HNDL(flags, name, fn, fmt) \
5816 DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
5818 * Request with a format known in advance
5820 #define DEF_MDT_HNDL_F(flags, name, fn) \
5821 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
5823 #define DEF_SEQ_HNDL_F(flags, name, fn) \
5824 DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
5826 #define DEF_FLD_HNDL_F(flags, name, fn) \
5827 DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
5829 * Request with a format we do not yet know
5831 #define DEF_MDT_HNDL_0(flags, name, fn) \
5832 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
5834 static struct mdt_handler mdt_mds_ops[] = {
5835 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5836 DEF_MDT_HNDL_F(0, DISCONNECT, mdt_disconnect),
5837 DEF_MDT_HNDL (0, SET_INFO, mdt_set_info,
5839 DEF_MDT_HNDL_F(0, GET_INFO, mdt_get_info),
5840 DEF_MDT_HNDL_F(0 |HABEO_REFERO, GETSTATUS, mdt_getstatus),
5841 DEF_MDT_HNDL_F(HABEO_CORPUS, GETATTR, mdt_getattr),
5842 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
5843 DEF_MDT_HNDL_F(HABEO_CORPUS, GETXATTR, mdt_getxattr),
5844 DEF_MDT_HNDL_F(0 |HABEO_REFERO, STATFS, mdt_statfs),
5845 DEF_MDT_HNDL_F(0 |MUTABOR, REINT, mdt_reint),
5846 DEF_MDT_HNDL_F(HABEO_CORPUS, CLOSE, mdt_close),
5847 DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing),
5848 DEF_MDT_HNDL_F(0 |HABEO_REFERO, PIN, mdt_pin),
5849 DEF_MDT_HNDL_0(0, SYNC, mdt_sync),
5850 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir),
5851 #ifdef HAVE_QUOTA_SUPPORT
5852 DEF_MDT_HNDL_F(0, QUOTACHECK, mdt_quotacheck_handle),
5853 DEF_MDT_HNDL_F(0, QUOTACTL, mdt_quotactl_handle)
5857 #define DEF_OBD_HNDL(flags, name, fn) \
5858 DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
5861 static struct mdt_handler mdt_obd_ops[] = {
5862 DEF_OBD_HNDL(0, PING, mdt_obd_ping),
5863 DEF_OBD_HNDL(0, LOG_CANCEL, mdt_obd_log_cancel),
5864 DEF_OBD_HNDL(0, QC_CALLBACK, mdt_obd_qc_callback)
5867 #define DEF_DLM_HNDL_0(flags, name, fn) \
5868 DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
5869 #define DEF_DLM_HNDL_F(flags, name, fn) \
5870 DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
5872 static struct mdt_handler mdt_dlm_ops[] = {
5873 DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE, mdt_enqueue),
5874 DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT, mdt_convert),
5875 DEF_DLM_HNDL_0(0, BL_CALLBACK, mdt_bl_callback),
5876 DEF_DLM_HNDL_0(0, CP_CALLBACK, mdt_cp_callback)
5879 #define DEF_LLOG_HNDL(flags, name, fn) \
5880 DEF_HNDL(LLOG, ORIGIN_HANDLE_CREATE, _NET, flags, name, fn, NULL)
5882 static struct mdt_handler mdt_llog_ops[] = {
5883 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CREATE, mdt_llog_create),
5884 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_NEXT_BLOCK, mdt_llog_next_block),
5885 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_READ_HEADER, mdt_llog_read_header),
5886 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_WRITE_REC, NULL),
5887 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CLOSE, NULL),
5888 DEF_LLOG_HNDL(0, ORIGIN_CONNECT, NULL),
5889 DEF_LLOG_HNDL(0, CATINFO, NULL),
5890 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_PREV_BLOCK, mdt_llog_prev_block),
5891 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_DESTROY, mdt_llog_destroy),
5894 #define DEF_SEC_CTX_HNDL(name, fn) \
5895 DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
5897 static struct mdt_handler mdt_sec_ctx_ops[] = {
5898 DEF_SEC_CTX_HNDL(INIT, mdt_sec_ctx_handle),
5899 DEF_SEC_CTX_HNDL(INIT_CONT, mdt_sec_ctx_handle),
5900 DEF_SEC_CTX_HNDL(FINI, mdt_sec_ctx_handle)
5903 static struct mdt_opc_slice mdt_regular_handlers[] = {
5905 .mos_opc_start = MDS_GETATTR,
5906 .mos_opc_end = MDS_LAST_OPC,
5907 .mos_hs = mdt_mds_ops
5910 .mos_opc_start = OBD_PING,
5911 .mos_opc_end = OBD_LAST_OPC,
5912 .mos_hs = mdt_obd_ops
5915 .mos_opc_start = LDLM_ENQUEUE,
5916 .mos_opc_end = LDLM_LAST_OPC,
5917 .mos_hs = mdt_dlm_ops
5920 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
5921 .mos_opc_end = LLOG_LAST_OPC,
5922 .mos_hs = mdt_llog_ops
5925 .mos_opc_start = SEC_CTX_INIT,
5926 .mos_opc_end = SEC_LAST_OPC,
5927 .mos_hs = mdt_sec_ctx_ops
5934 static struct mdt_handler mdt_readpage_ops[] = {
5935 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5936 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
5937 #ifdef HAVE_SPLIT_SUPPORT
5938 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
5942 * XXX: this is ugly and should be fixed one day, see mdc_close() for
5943 * detailed comments. --umka
5945 DEF_MDT_HNDL_F(HABEO_CORPUS, CLOSE, mdt_close),
5946 DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing),
5949 static struct mdt_opc_slice mdt_readpage_handlers[] = {
5951 .mos_opc_start = MDS_GETATTR,
5952 .mos_opc_end = MDS_LAST_OPC,
5953 .mos_hs = mdt_readpage_ops
5960 static struct mdt_handler mdt_xmds_ops[] = {
5961 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5962 DEF_MDT_HNDL_F(HABEO_CORPUS , GETATTR, mdt_getattr),
5963 DEF_MDT_HNDL_F(0 | MUTABOR , REINT, mdt_reint),
5964 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir),
5967 static struct mdt_opc_slice mdt_xmds_handlers[] = {
5969 .mos_opc_start = MDS_GETATTR,
5970 .mos_opc_end = MDS_LAST_OPC,
5971 .mos_hs = mdt_xmds_ops
5974 .mos_opc_start = OBD_PING,
5975 .mos_opc_end = OBD_LAST_OPC,
5976 .mos_hs = mdt_obd_ops
5979 .mos_opc_start = SEC_CTX_INIT,
5980 .mos_opc_end = SEC_LAST_OPC,
5981 .mos_hs = mdt_sec_ctx_ops
5988 static struct mdt_handler mdt_seq_ops[] = {
5989 DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
5992 static struct mdt_opc_slice mdt_seq_handlers[] = {
5994 .mos_opc_start = SEQ_QUERY,
5995 .mos_opc_end = SEQ_LAST_OPC,
5996 .mos_hs = mdt_seq_ops
6003 static struct mdt_handler mdt_fld_ops[] = {
6004 DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
6007 static struct mdt_opc_slice mdt_fld_handlers[] = {
6009 .mos_opc_start = FLD_QUERY,
6010 .mos_opc_end = FLD_LAST_OPC,
6011 .mos_hs = mdt_fld_ops
6018 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6019 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
6020 MODULE_LICENSE("GPL");
6022 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
6023 "number of mdt service threads to start");
6025 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);