1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 # define EXPORT_SYMTAB
40 #define DEBUG_SUBSYSTEM S_MDC
43 # include <linux/module.h>
44 # include <linux/pagemap.h>
45 # include <linux/miscdevice.h>
46 # include <linux/init.h>
48 # include <liblustre.h>
51 #include <lustre_acl.h>
52 #include <obd_class.h>
53 #include <lustre_dlm.h>
54 #include <lustre_fid.h>
55 #include <md_object.h>
56 #include <lprocfs_status.h>
57 #include <lustre_param.h>
58 #include "mdc_internal.h"
60 #define REQUEST_MINOR 244
62 static quota_interface_t *quota_interface;
63 extern quota_interface_t mdc_quota_interface;
65 static int mdc_cleanup(struct obd_device *obd);
67 int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
68 const struct req_msg_field *field, struct obd_capa **oc)
70 struct lustre_capa *capa;
74 /* swabbed already in mdc_enqueue */
75 capa = req_capsule_server_get(&req->rq_pill, field);
79 c = alloc_capa(CAPA_SITE_CLIENT);
81 CDEBUG(D_INFO, "alloc capa failed!\n");
90 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
91 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
92 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
93 struct obd_capa **pc, int level, int msg_flags)
95 struct ptlrpc_request *req;
96 struct mdt_body *body;
100 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
101 LUSTRE_MDS_VERSION, MDS_GETSTATUS);
105 mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
106 lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
107 req->rq_send_state = level;
109 ptlrpc_request_set_replen(req);
111 rc = ptlrpc_queue_wait(req);
115 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
117 GOTO(out, rc = -EPROTO);
119 if (body->valid & OBD_MD_FLMDSCAPA) {
120 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
125 *rootfid = body->fid1;
127 "root fid="DFID", last_committed="LPU64", last_xid="LPU64"\n",
129 lustre_msg_get_last_committed(req->rq_repmsg),
130 lustre_msg_get_last_xid(req->rq_repmsg));
133 ptlrpc_req_finished(req);
137 /* This should be mdc_get_info("rootfid") */
138 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
139 struct obd_capa **pc)
141 return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
146 * This function now is known to always saying that it will receive 4 buffers
147 * from server. Even for cases when acl_size and md_size is zero, RPC header
148 * will contain 4 fields and RPC itself will contain zero size fields. This is
149 * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
150 * and thus zero, it shrinks it, making zero size. The same story about
151 * md_size. And this is course of problem when client waits for smaller number
152 * of fields. This issue will be fixed later when client gets aware of RPC
155 static int mdc_getattr_common(struct obd_export *exp,
156 struct ptlrpc_request *req)
158 struct req_capsule *pill = &req->rq_pill;
159 struct mdt_body *body;
164 /* Request message already built. */
165 rc = ptlrpc_queue_wait(req);
169 /* sanity check for the reply */
170 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
174 CDEBUG(D_NET, "mode: %o\n", body->mode);
176 if (body->eadatasize != 0) {
177 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
183 if (body->valid & OBD_MD_FLMODEASIZE) {
184 struct client_obd *cli = &exp->exp_obd->u.cli;
186 if (cli->cl_max_mds_easize < body->max_mdsize)
187 cli->cl_max_mds_easize = body->max_mdsize;
188 if (cli->cl_max_mds_cookiesize < body->max_cookiesize)
189 cli->cl_max_mds_cookiesize = body->max_cookiesize;
192 if (body->valid & OBD_MD_FLRMTPERM) {
193 struct mdt_remote_perm *perm;
195 LASSERT(client_is_remote(exp));
196 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
197 lustre_swab_mdt_remote_perm);
202 if (body->valid & OBD_MD_FLMDSCAPA) {
203 struct lustre_capa *capa;
204 capa = req_capsule_server_get(pill, &RMF_CAPA1);
212 int mdc_getattr(struct obd_export *exp, const struct lu_fid *fid,
213 struct obd_capa *oc, obd_valid valid, int ea_size,
214 struct ptlrpc_request **request)
216 struct ptlrpc_request *req;
221 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
225 mdc_set_capa_size(req, &RMF_CAPA1, oc);
227 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
229 ptlrpc_request_free(req);
233 /* MDS_BFLAG_EXT_FLAGS: request "new" flags(bug 9486) */
234 mdc_pack_body(req, fid, oc, valid, ea_size, -1, MDS_BFLAG_EXT_FLAGS);
236 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, ea_size);
237 if (valid & OBD_MD_FLRMTPERM) {
238 LASSERT(client_is_remote(exp));
239 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
240 sizeof(struct mdt_remote_perm));
242 ptlrpc_request_set_replen(req);
244 rc = mdc_getattr_common(exp, req);
246 ptlrpc_req_finished(req);
252 int mdc_getattr_name(struct obd_export *exp, const struct lu_fid *fid,
253 struct obd_capa *oc, const char *filename, int namelen,
254 obd_valid valid, int ea_size, __u32 suppgid,
255 struct ptlrpc_request **request)
257 struct ptlrpc_request *req;
262 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
263 &RQF_MDS_GETATTR_NAME);
267 mdc_set_capa_size(req, &RMF_CAPA1, oc);
268 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, namelen);
270 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
272 ptlrpc_request_free(req);
276 /* MDS_BFLAG_EXT_FLAGS: request "new" flags(bug 9486) */
277 mdc_pack_body(req, fid, oc, valid, ea_size, suppgid,
278 MDS_BFLAG_EXT_FLAGS);
281 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
282 LASSERT(strnlen(filename, namelen) == namelen - 1);
283 memcpy(name, filename, namelen);
286 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, ea_size);
287 ptlrpc_request_set_replen(req);
289 rc = mdc_getattr_common(exp, req);
291 ptlrpc_req_finished(req);
297 static int mdc_is_subdir(struct obd_export *exp,
298 const struct lu_fid *pfid,
299 const struct lu_fid *cfid,
300 struct ptlrpc_request **request)
302 struct ptlrpc_request *req;
308 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
309 &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
314 mdc_is_subdir_pack(req, pfid, cfid, 0);
315 ptlrpc_request_set_replen(req);
317 rc = ptlrpc_queue_wait(req);
318 if (rc && rc != -EREMOTE)
319 ptlrpc_req_finished(req);
325 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
326 const struct lu_fid *fid,
327 struct obd_capa *oc, int opcode, obd_valid valid,
328 const char *xattr_name, const char *input,
329 int input_size, int output_size, int flags,
330 __u32 suppgid, struct ptlrpc_request **request)
332 struct ptlrpc_request *req;
333 int xattr_namelen = 0;
339 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
343 mdc_set_capa_size(req, &RMF_CAPA1, oc);
345 xattr_namelen = strlen(xattr_name) + 1;
346 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
351 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
355 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
357 ptlrpc_request_free(req);
361 if (opcode == MDS_REINT) {
362 struct mdt_rec_setxattr *rec;
364 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
365 sizeof(struct mdt_rec_reint));
366 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
367 rec->sx_opcode = REINT_SETXATTR;
369 * cfs_curproc_fs{u,g}id() should replace
370 * current->fs{u,g}id for portability.
372 rec->sx_fsuid = current->fsuid;
373 rec->sx_fsgid = current->fsgid;
374 rec->sx_cap = cfs_curproc_cap_pack();
375 rec->sx_suppgid1 = suppgid;
376 rec->sx_suppgid2 = -1;
378 rec->sx_valid = valid | OBD_MD_FLCTIME;
379 rec->sx_time = cfs_time_current_sec();
380 rec->sx_size = output_size;
381 rec->sx_flags = flags;
383 mdc_pack_capa(req, &RMF_CAPA1, oc);
385 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
389 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
390 memcpy(tmp, xattr_name, xattr_namelen);
393 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
394 memcpy(tmp, input, input_size);
397 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
398 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
399 RCL_SERVER, output_size);
400 ptlrpc_request_set_replen(req);
403 if (opcode == MDS_REINT)
404 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
406 rc = ptlrpc_queue_wait(req);
408 if (opcode == MDS_REINT)
409 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
412 ptlrpc_req_finished(req);
418 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
419 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
420 const char *input, int input_size, int output_size,
421 int flags, __u32 suppgid, struct ptlrpc_request **request)
423 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
424 fid, oc, MDS_REINT, valid, xattr_name,
425 input, input_size, output_size, flags,
429 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
430 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
431 const char *input, int input_size, int output_size,
432 int flags, struct ptlrpc_request **request)
434 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
435 fid, oc, MDS_GETXATTR, valid, xattr_name,
436 input, input_size, output_size, flags,
440 #ifdef CONFIG_FS_POSIX_ACL
441 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
443 struct req_capsule *pill = &req->rq_pill;
444 struct mdt_body *body = md->body;
445 struct posix_acl *acl;
453 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
458 acl = posix_acl_from_xattr(buf, body->aclsize);
461 CERROR("convert xattr to acl: %d\n", rc);
465 rc = posix_acl_valid(acl);
467 CERROR("validate acl: %d\n", rc);
468 posix_acl_release(acl);
476 #define mdc_unpack_acl(req, md) 0
479 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
480 struct obd_export *dt_exp, struct obd_export *md_exp,
481 struct lustre_md *md)
483 struct req_capsule *pill = &req->rq_pill;
488 memset(md, 0, sizeof(*md));
490 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
491 LASSERT(md->body != NULL);
493 if (md->body->valid & OBD_MD_FLEASIZE) {
495 struct lov_mds_md *lmm;
497 if (!S_ISREG(md->body->mode)) {
498 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
499 "regular file, but is not\n");
500 GOTO(out, rc = -EPROTO);
503 if (md->body->eadatasize == 0) {
504 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
505 "but eadatasize 0\n");
506 GOTO(out, rc = -EPROTO);
508 lmmsize = md->body->eadatasize;
509 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
511 GOTO(out, rc = -EPROTO);
513 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
517 if (rc < sizeof(*md->lsm)) {
518 CDEBUG(D_INFO, "lsm size too small: "
519 "rc < sizeof (*md->lsm) (%d < %d)\n",
520 rc, (int)sizeof(*md->lsm));
521 GOTO(out, rc = -EPROTO);
524 } else if (md->body->valid & OBD_MD_FLDIREA) {
526 struct lov_mds_md *lmv;
528 if(!S_ISDIR(md->body->mode)) {
529 CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
530 "directory, but is not\n");
531 GOTO(out, rc = -EPROTO);
534 if (md->body->eadatasize == 0) {
535 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
536 "but eadatasize 0\n");
539 if (md->body->valid & OBD_MD_MEA) {
540 lmvsize = md->body->eadatasize;
541 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
544 GOTO(out, rc = -EPROTO);
546 rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
551 if (rc < sizeof(*md->mea)) {
552 CDEBUG(D_INFO, "size too small: "
553 "rc < sizeof(*md->mea) (%d < %d)\n",
554 rc, (int)sizeof(*md->mea));
555 GOTO(out, rc = -EPROTO);
561 if (md->body->valid & OBD_MD_FLRMTPERM) {
562 /* remote permission */
563 LASSERT(client_is_remote(exp));
564 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
565 lustre_swab_mdt_remote_perm);
566 if (!md->remote_perm)
567 GOTO(out, rc = -EPROTO);
569 else if (md->body->valid & OBD_MD_FLACL) {
570 /* for ACL, it's possible that FLACL is set but aclsize is zero.
571 * only when aclsize != 0 there's an actual segment for ACL
574 if (md->body->aclsize) {
575 rc = mdc_unpack_acl(req, md);
578 #ifdef CONFIG_FS_POSIX_ACL
580 md->posix_acl = NULL;
584 if (md->body->valid & OBD_MD_FLMDSCAPA) {
585 struct obd_capa *oc = NULL;
587 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
593 if (md->body->valid & OBD_MD_FLOSSCAPA) {
594 struct obd_capa *oc = NULL;
596 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
606 capa_put(md->oss_capa);
610 capa_put(md->mds_capa);
613 #ifdef CONFIG_FS_POSIX_ACL
614 posix_acl_release(md->posix_acl);
617 obd_free_memmd(dt_exp, &md->lsm);
622 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
628 static void mdc_replay_open(struct ptlrpc_request *req)
630 struct md_open_data *mod = req->rq_cb_data;
631 struct ptlrpc_request *cur, *tmp;
632 struct obd_client_handle *och;
633 struct lustre_handle old;
634 struct mdt_body *body;
638 DEBUG_REQ(D_ERROR, req,
639 "Can't properly replay without open data.");
644 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
645 LASSERT(body != NULL);
649 struct lustre_handle *file_fh;
651 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
653 file_fh = &och->och_fh;
654 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
655 file_fh->cookie, body->handle.cookie);
657 *file_fh = body->handle;
659 list_for_each_entry_safe(cur, tmp, &mod->mod_replay_list, rq_mod_list) {
660 int opc = lustre_msg_get_opc(cur->rq_reqmsg);
661 struct mdt_epoch *epoch = NULL;
663 if (opc == MDS_CLOSE || opc == MDS_DONE_WRITING) {
664 epoch = req_capsule_client_get(&cur->rq_pill,
667 DEBUG_REQ(D_HA, cur, "updating %s body with new fh",
668 opc == MDS_CLOSE ? "CLOSE" : "DONE_WRITING");
669 } else if (opc == MDS_REINT) {
670 struct mdt_rec_setattr *rec;
672 /* Check this is REINT_SETATTR. */
673 rec = req_capsule_client_get(&cur->rq_pill,
675 LASSERT(rec && rec->sa_opcode == REINT_SETATTR);
677 epoch = req_capsule_client_get(&cur->rq_pill,
680 DEBUG_REQ(D_HA, cur, "updating REINT_SETATTR body "
685 LASSERT(!memcmp(&old, &epoch->handle,
687 epoch->handle = body->handle;
693 void mdc_commit_delayed(struct ptlrpc_request *req)
695 struct md_open_data *mod = req->rq_cb_data;
696 struct ptlrpc_request *cur, *tmp;
698 DEBUG_REQ(D_HA, req, "req committed");
703 req->rq_cb_data = NULL;
704 req->rq_commit_cb = NULL;
705 list_del_init(&req->rq_mod_list);
706 if (req->rq_sequence) {
707 list_for_each_entry_safe(cur, tmp, &mod->mod_replay_list,
709 LASSERT(cur != LP_POISON);
710 LASSERT(cur->rq_type != LI_POISON);
711 DEBUG_REQ(D_HA, cur, "req balanced");
712 LASSERT(cur->rq_transno != 0);
713 LASSERT(cur->rq_import == req->rq_import);
715 /* We no longer want to preserve this for transno-
716 * unconditional replay. */
717 spin_lock(&cur->rq_lock);
719 spin_unlock(&cur->rq_lock);
723 if (list_empty(&mod->mod_replay_list)) {
724 if (mod->mod_och != NULL)
725 mod->mod_och->och_mod = NULL;
731 int mdc_set_open_replay_data(struct obd_export *exp,
732 struct obd_client_handle *och,
733 struct ptlrpc_request *open_req)
735 struct md_open_data *mod;
736 struct mdt_rec_create *rec;
737 struct mdt_body *body;
738 struct obd_import *imp = open_req->rq_import;
741 if (!open_req->rq_replay)
744 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
745 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
746 LASSERT(rec != NULL);
747 /* Incoming message in my byte order (it's been swabbed). */
748 /* Outgoing messages always in my byte order. */
749 LASSERT(body != NULL);
751 /* Only if the import is replayable, we set replay_open data */
752 if (och && imp->imp_replayable) {
755 DEBUG_REQ(D_ERROR, open_req,
756 "Can't allocate md_open_data");
759 CFS_INIT_LIST_HEAD(&mod->mod_replay_list);
761 spin_lock(&open_req->rq_lock);
764 open_req->rq_cb_data = mod;
765 list_add_tail(&open_req->rq_mod_list, &mod->mod_replay_list);
766 open_req->rq_commit_cb = mdc_commit_delayed;
767 spin_unlock(&open_req->rq_lock);
770 rec->cr_fid2 = body->fid1;
771 rec->cr_ioepoch = body->ioepoch;
772 rec->cr_old_handle.cookie = body->handle.cookie;
773 open_req->rq_replay_cb = mdc_replay_open;
774 if (!fid_is_sane(&body->fid1)) {
775 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
780 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
784 int mdc_clear_open_replay_data(struct obd_export *exp,
785 struct obd_client_handle *och)
787 struct md_open_data *mod = och->och_mod;
791 * Don't free the structure now (it happens in mdc_commit_delayed(),
792 * after the last request is removed from its replay list),
793 * but make sure that replay doesn't poke at the och, which is about to
796 LASSERT(mod != LP_POISON);
804 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
805 struct md_open_data *mod, struct ptlrpc_request **request)
807 struct obd_device *obd = class_exp2obd(exp);
808 struct ptlrpc_request *req;
813 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_CLOSE);
817 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
819 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
821 ptlrpc_request_free(req);
825 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
826 * portal whose threads are not taking any DLM locks and are therefore
827 * always progressing */
828 req->rq_request_portal = MDS_READPAGE_PORTAL;
829 ptlrpc_at_set_req_timeout(req);
831 /* Ensure that this close's handle is fixed up during replay. */
832 if (likely(mod != NULL))
833 list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
835 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
837 mdc_close_pack(req, op_data);
839 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
840 obd->u.cli.cl_max_mds_easize);
841 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
842 obd->u.cli.cl_max_mds_cookiesize);
844 ptlrpc_request_set_replen(req);
846 req->rq_commit_cb = mdc_commit_delayed;
848 LASSERT(req->rq_cb_data == NULL);
849 req->rq_cb_data = mod;
851 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
852 rc = ptlrpc_queue_wait(req);
853 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
855 if (req->rq_repmsg == NULL) {
856 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
859 rc = req->rq_status ?: -EIO;
860 } else if (rc == 0 || rc == -EAGAIN) {
861 struct mdt_body *body;
863 rc = lustre_msg_get_status(req->rq_repmsg);
864 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
865 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
869 } else if (mod == NULL) {
870 if (req->rq_import->imp_replayable)
871 CERROR("Unexpected: can't find md_open_data,"
872 "but close succeeded with replayable imp"
874 "http://bugzilla.lustre.org/\n");
877 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
883 if (rc != 0 && rc != -EAGAIN && req && req->rq_commit_cb)
884 req->rq_commit_cb(req);
890 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
891 struct md_open_data *mod)
893 struct obd_device *obd = class_exp2obd(exp);
894 struct ptlrpc_request *req;
898 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
899 &RQF_MDS_DONE_WRITING);
903 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
904 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
906 ptlrpc_request_free(req);
910 /* XXX: add DONE_WRITING request to och -- when Size-on-MDS
911 * recovery will be ready. */
912 mdc_close_pack(req, op_data);
913 ptlrpc_request_set_replen(req);
915 req->rq_cb_data = mod;
916 req->rq_commit_cb = mdc_commit_delayed;
917 if (likely(mod != NULL))
918 list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
920 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
922 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
923 rc = ptlrpc_queue_wait(req);
924 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
926 /* Close the open replay sequence if an error occured or no SOM
927 * attribute update is needed. */
929 ptlrpc_close_replay_seq(req);
931 if (rc && rc != -EAGAIN && req->rq_commit_cb)
932 req->rq_commit_cb(req);
934 ptlrpc_req_finished(req);
938 #ifdef HAVE_SPLIT_SUPPORT
939 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
940 const struct page *page, int offset)
942 struct ptlrpc_request *req;
943 struct ptlrpc_bulk_desc *desc;
947 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
951 /* FIXME: capa doesn't support split yet */
952 mdc_set_capa_size(req, &RMF_CAPA1, NULL);
954 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
956 ptlrpc_request_free(req);
960 req->rq_request_portal = MDS_READPAGE_PORTAL;
961 ptlrpc_at_set_req_timeout(req);
963 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
965 GOTO(out, rc = -ENOMEM);
967 /* NB req now owns desc and will free it when it gets freed. */
968 ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
969 mdc_readdir_pack(req, 0, offset, fid, NULL);
971 ptlrpc_request_set_replen(req);
972 rc = ptlrpc_queue_wait(req);
975 ptlrpc_req_finished(req);
978 EXPORT_SYMBOL(mdc_sendpage);
981 int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
982 struct obd_capa *oc, __u64 offset, struct page *page,
983 struct ptlrpc_request **request)
985 struct ptlrpc_request *req;
986 struct ptlrpc_bulk_desc *desc;
991 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
995 mdc_set_capa_size(req, &RMF_CAPA1, oc);
997 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
999 ptlrpc_request_free(req);
1003 req->rq_request_portal = MDS_READPAGE_PORTAL;
1004 ptlrpc_at_set_req_timeout(req);
1006 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
1008 ptlrpc_request_free(req);
1012 /* NB req now owns desc and will free it when it gets freed */
1013 ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
1014 mdc_readdir_pack(req, offset, CFS_PAGE_SIZE, fid, oc);
1016 ptlrpc_request_set_replen(req);
1017 rc = ptlrpc_queue_wait(req);
1019 ptlrpc_req_finished(req);
1023 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
1024 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1025 req->rq_bulk->bd_nob_transferred, CFS_PAGE_SIZE);
1026 ptlrpc_req_finished(req);
1034 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1035 void *karg, void *uarg)
1037 struct obd_device *obd = exp->exp_obd;
1038 struct obd_ioctl_data *data = karg;
1039 struct obd_import *imp = obd->u.cli.cl_import;
1040 struct llog_ctxt *ctxt;
1044 if (!try_module_get(THIS_MODULE)) {
1045 CERROR("Can't get module. Is it alive?");
1049 case OBD_IOC_CLIENT_RECOVER:
1050 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
1054 case IOC_OSC_SET_ACTIVE:
1055 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1057 case OBD_IOC_PARSE: {
1058 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1059 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
1060 llog_ctxt_put(ctxt);
1064 case OBD_IOC_LLOG_INFO:
1065 case OBD_IOC_LLOG_PRINT: {
1066 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1067 rc = llog_ioctl(ctxt, cmd, data);
1068 llog_ctxt_put(ctxt);
1072 case OBD_IOC_POLL_QUOTACHECK:
1073 rc = lquota_poll_check(quota_interface, exp,
1074 (struct if_quotacheck *)karg);
1077 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1078 GOTO(out, rc = -ENOTTY);
1081 module_put(THIS_MODULE);
1086 static int do_set_info_async(struct obd_export *exp,
1087 obd_count keylen, void *key,
1088 obd_count vallen, void *val,
1089 struct ptlrpc_request_set *set)
1091 struct obd_import *imp = class_exp2cliimp(exp);
1092 struct ptlrpc_request *req;
1097 if (vallen != sizeof(int))
1100 spin_lock(&imp->imp_lock);
1101 if (*((int *)val)) {
1102 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
1103 imp->imp_connect_data.ocd_connect_flags |= OBD_CONNECT_RDONLY;
1105 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
1106 imp->imp_connect_data.ocd_connect_flags &= ~OBD_CONNECT_RDONLY;
1108 spin_unlock(&imp->imp_lock);
1110 req = ptlrpc_request_alloc(imp, &RQF_MDS_SET_INFO);
1114 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
1115 RCL_CLIENT, keylen);
1116 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
1117 RCL_CLIENT, vallen);
1118 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SET_INFO);
1120 ptlrpc_request_free(req);
1124 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1125 memcpy(tmp, key, keylen);
1126 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
1127 memcpy(tmp, val, vallen);
1129 ptlrpc_request_set_replen(req);
1132 ptlrpc_set_add_req(set, req);
1133 ptlrpc_check_set(NULL, set);
1135 rc = ptlrpc_queue_wait(req);
1136 ptlrpc_req_finished(req);
1142 int mdc_set_info_async(struct obd_export *exp,
1143 obd_count keylen, void *key,
1144 obd_count vallen, void *val,
1145 struct ptlrpc_request_set *set)
1147 struct obd_import *imp = class_exp2cliimp(exp);
1151 if (KEY_IS(KEY_INIT_RECOV)) {
1152 if (vallen != sizeof(int))
1154 spin_lock(&imp->imp_lock);
1155 imp->imp_initial_recov = *(int *)val;
1156 spin_unlock(&imp->imp_lock);
1157 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
1158 exp->exp_obd->obd_name, imp->imp_initial_recov);
1161 /* Turn off initial_recov after we try all backup servers once */
1162 if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
1163 if (vallen != sizeof(int))
1165 spin_lock(&imp->imp_lock);
1166 imp->imp_initial_recov_bk = *(int *)val;
1167 if (imp->imp_initial_recov_bk)
1168 imp->imp_initial_recov = 1;
1169 spin_unlock(&imp->imp_lock);
1170 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
1171 exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
1174 if (KEY_IS(KEY_READ_ONLY)) {
1175 rc = do_set_info_async(exp, keylen, key, vallen, val, set);
1178 if (KEY_IS(KEY_FLUSH_CTX)) {
1179 sptlrpc_import_flush_my_ctx(imp);
1182 if (KEY_IS(KEY_MDS_CONN)) {
1183 struct obd_import *imp = class_exp2cliimp(exp);
1185 /* mds-mds import */
1186 spin_lock(&imp->imp_lock);
1187 imp->imp_server_timeout = 1;
1188 spin_unlock(&imp->imp_lock);
1189 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1190 CDEBUG(D_OTHER|D_WARNING, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1197 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
1198 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1202 if (KEY_IS(KEY_MAX_EASIZE)) {
1203 int mdsize, *max_easize;
1205 if (*vallen != sizeof(int))
1207 mdsize = *(int*)val;
1208 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
1209 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
1211 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
1214 if (KEY_IS(KEY_CONN_DATA)) {
1215 struct obd_import *imp = class_exp2cliimp(exp);
1216 struct obd_connect_data *data = val;
1218 if (*vallen != sizeof(*data))
1221 *data = imp->imp_connect_data;
1228 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1229 __u64 max_age, __u32 flags)
1231 struct ptlrpc_request *req;
1232 struct obd_statfs *msfs;
1233 struct obd_import *imp = NULL;
1238 /*Since the request might also come from lprocfs, so we need
1239 *sync this with client_disconnect_export Bug15684*/
1240 down_read(&obd->u.cli.cl_sem);
1241 if (obd->u.cli.cl_import)
1242 imp = class_import_get(obd->u.cli.cl_import);
1243 up_read(&obd->u.cli.cl_sem);
1247 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1248 LUSTRE_MDS_VERSION, MDS_STATFS);
1250 GOTO(output, rc = -ENOMEM);
1252 ptlrpc_request_set_replen(req);
1254 if (flags & OBD_STATFS_NODELAY) {
1255 /* procfs requests not want stay in wait for avoid deadlock */
1256 req->rq_no_resend = 1;
1257 req->rq_no_delay = 1;
1260 rc = ptlrpc_queue_wait(req);
1262 /* check connection error first */
1263 if (imp->imp_connect_error)
1264 rc = imp->imp_connect_error;
1268 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1270 GOTO(out, rc = -EPROTO);
1275 ptlrpc_req_finished(req);
1277 class_import_put(imp);
1281 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
1282 struct obd_capa *oc, struct obd_client_handle *handle,
1285 struct ptlrpc_request *req;
1286 struct mdt_body *body;
1290 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
1294 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1296 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
1298 ptlrpc_request_free(req);
1302 mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
1304 ptlrpc_request_set_replen(req);
1306 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1307 rc = ptlrpc_queue_wait(req);
1308 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1310 CERROR("Pin failed: %d\n", rc);
1314 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1316 GOTO(err_out, rc = -EPROTO);
1318 handle->och_fh = body->handle;
1319 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1321 OBD_ALLOC_PTR(handle->och_mod);
1322 if (handle->och_mod == NULL) {
1323 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
1324 GOTO(err_out, rc = -ENOMEM);
1326 /* will be dropped by unpin */
1327 CFS_INIT_LIST_HEAD(&handle->och_mod->mod_replay_list);
1328 list_add_tail(&req->rq_mod_list, &handle->och_mod->mod_replay_list);
1333 ptlrpc_req_finished(req);
1337 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
1340 struct ptlrpc_request *req;
1341 struct mdt_body *body;
1345 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
1346 LUSTRE_MDS_VERSION, MDS_UNPIN);
1350 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1351 body->handle = handle->och_fh;
1354 ptlrpc_request_set_replen(req);
1356 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1357 rc = ptlrpc_queue_wait(req);
1358 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1361 CERROR("Unpin failed: %d\n", rc);
1363 ptlrpc_req_finished(req);
1365 LASSERT(!list_empty(&handle->och_mod->mod_replay_list));
1366 req = list_entry(handle->och_mod->mod_replay_list.next,
1367 typeof(*req), rq_mod_list);
1368 list_del_init(&req->rq_mod_list);
1369 ptlrpc_req_finished(req);
1370 LASSERT(list_empty(&handle->och_mod->mod_replay_list));
1372 OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1376 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
1377 struct obd_capa *oc, struct ptlrpc_request **request)
1379 struct ptlrpc_request *req;
1384 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
1388 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1390 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
1392 ptlrpc_request_free(req);
1396 mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
1398 ptlrpc_request_set_replen(req);
1400 rc = ptlrpc_queue_wait(req);
1402 ptlrpc_req_finished(req);
1408 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1409 enum obd_import_event event)
1413 LASSERT(imp->imp_obd == obd);
1416 case IMP_EVENT_DISCON: {
1418 /* XXX Pass event up to OBDs stack. used only for FLD now */
1419 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
1423 case IMP_EVENT_INACTIVE: {
1424 struct client_obd *cli = &obd->u.cli;
1426 * Flush current sequence to make client obtain new one
1427 * from server in case of disconnect/reconnect.
1428 * If range is already empty then no need to flush it.
1430 if (cli->cl_seq != NULL &&
1431 !range_is_exhausted(&cli->cl_seq->lcs_space)) {
1432 seq_client_flush(cli->cl_seq);
1435 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1438 case IMP_EVENT_INVALIDATE: {
1439 struct ldlm_namespace *ns = obd->obd_namespace;
1441 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1445 case IMP_EVENT_ACTIVE: {
1446 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1450 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
1454 CERROR("Unknown import event %x\n", event);
1460 static int mdc_fid_init(struct obd_export *exp)
1462 struct client_obd *cli = &exp->exp_obd->u.cli;
1467 OBD_ALLOC_PTR(cli->cl_seq);
1468 if (cli->cl_seq == NULL)
1471 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
1473 GOTO(out_free_seq, rc = -ENOMEM);
1475 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
1476 exp->exp_obd->obd_name);
1478 /* Init client side sequence-manager */
1479 rc = seq_client_init(cli->cl_seq, exp,
1480 LUSTRE_SEQ_METADATA,
1482 OBD_FREE(prefix, MAX_OBD_NAME + 5);
1484 GOTO(out_free_seq, rc);
1488 OBD_FREE_PTR(cli->cl_seq);
1493 static int mdc_fid_fini(struct obd_export *exp)
1495 struct client_obd *cli = &exp->exp_obd->u.cli;
1498 if (cli->cl_seq != NULL) {
1499 seq_client_fini(cli->cl_seq);
1500 OBD_FREE_PTR(cli->cl_seq);
1507 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1508 struct md_op_data *op_data)
1510 struct client_obd *cli = &exp->exp_obd->u.cli;
1511 struct lu_client_seq *seq = cli->cl_seq;
1513 RETURN(seq_client_alloc_fid(seq, fid));
1516 /* XXX This method is used only to clear current fid seq
1517 * once fld/mds insert failed */
1518 static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1520 struct client_obd *cli = &exp->exp_obd->u.cli;
1522 seq_client_flush(cli->cl_seq);
1526 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
1527 struct client_obd *cli = &exp->exp_obd->u.cli;
1528 return &cli->cl_target_uuid;
1531 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
1533 struct client_obd *cli = &obd->u.cli;
1534 struct lprocfs_static_vars lvars = { 0 };
1538 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1539 if (!cli->cl_rpc_lock)
1541 mdc_init_rpc_lock(cli->cl_rpc_lock);
1545 OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1546 if (!cli->cl_setattr_lock)
1547 GOTO(err_rpc_lock, rc = -ENOMEM);
1548 mdc_init_rpc_lock(cli->cl_setattr_lock);
1550 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1551 if (!cli->cl_close_lock)
1552 GOTO(err_setattr_lock, rc = -ENOMEM);
1553 mdc_init_rpc_lock(cli->cl_close_lock);
1555 rc = client_obd_setup(obd, cfg);
1557 GOTO(err_close_lock, rc);
1558 lprocfs_mdc_init_vars(&lvars);
1559 lprocfs_obd_setup(obd, lvars.obd_vars);
1560 sptlrpc_lprocfs_cliobd_attach(obd);
1561 ptlrpc_lprocfs_register_obd(obd);
1563 rc = obd_llog_init(obd, &obd->obd_olg, obd, 0, NULL, NULL);
1566 CERROR("failed to setup llogging subsystems\n");
1572 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1574 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1576 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1581 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
1582 * us to make MDS RPCs with large enough reply buffers to hold the
1583 * maximum-sized (= maximum striped) EA and cookie without having to
1584 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1585 static int mdc_init_ea_size(struct obd_export *exp, int easize,
1586 int def_easize, int cookiesize)
1588 struct obd_device *obd = exp->exp_obd;
1589 struct client_obd *cli = &obd->u.cli;
1592 if (cli->cl_max_mds_easize < easize)
1593 cli->cl_max_mds_easize = easize;
1595 if (cli->cl_default_mds_easize < def_easize)
1596 cli->cl_default_mds_easize = def_easize;
1598 if (cli->cl_max_mds_cookiesize < cookiesize)
1599 cli->cl_max_mds_cookiesize = cookiesize;
1604 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1610 case OBD_CLEANUP_EARLY:
1611 case OBD_CLEANUP_EXPORTS:
1612 /* If we set up but never connected, the
1613 client import will not have been cleaned. */
1614 if (obd->u.cli.cl_import) {
1615 struct obd_import *imp;
1616 down_write(&obd->u.cli.cl_sem);
1617 imp = obd->u.cli.cl_import;
1618 CERROR("client import never connected\n");
1619 ptlrpc_invalidate_import(imp);
1620 class_destroy_import(imp);
1621 up_write(&obd->u.cli.cl_sem);
1622 obd->u.cli.cl_import = NULL;
1624 rc = obd_llog_finish(obd, 0);
1626 CERROR("failed to cleanup llogging subsystems\n");
1632 static int mdc_cleanup(struct obd_device *obd)
1634 struct client_obd *cli = &obd->u.cli;
1636 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1637 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1638 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1640 ptlrpc_lprocfs_unregister_obd(obd);
1641 lprocfs_obd_cleanup(obd);
1644 return client_obd_cleanup(obd);
1648 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
1649 struct obd_device *tgt, int count,
1650 struct llog_catid *logid, struct obd_uuid *uuid)
1652 struct llog_ctxt *ctxt;
1656 LASSERT(olg == &obd->obd_olg);
1658 rc = llog_setup(obd, olg, LLOG_LOVEA_REPL_CTXT, tgt, 0,
1659 NULL, &llog_client_ops);
1661 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1662 llog_initiator_connect(ctxt);
1663 llog_ctxt_put(ctxt);
1669 static int mdc_llog_finish(struct obd_device *obd, int count)
1671 struct llog_ctxt *ctxt;
1675 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1677 rc = llog_cleanup(ctxt);
1682 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1684 struct lustre_cfg *lcfg = buf;
1685 struct lprocfs_static_vars lvars = { 0 };
1688 lprocfs_mdc_init_vars(&lvars);
1689 switch (lcfg->lcfg_command) {
1690 case LCFG_SPTLRPC_CONF:
1691 rc = sptlrpc_cliobd_process_config(obd, lcfg);
1694 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
1704 /* get remote permission for current user on fid */
1705 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
1706 struct obd_capa *oc, __u32 suppgid,
1707 struct ptlrpc_request **request)
1709 struct ptlrpc_request *req;
1713 LASSERT(client_is_remote(exp));
1716 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1720 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1722 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1724 ptlrpc_request_free(req);
1728 mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
1730 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
1731 sizeof(struct mdt_remote_perm));
1733 ptlrpc_request_set_replen(req);
1735 rc = ptlrpc_queue_wait(req);
1737 ptlrpc_req_finished(req);
1743 static int mdc_interpret_renew_capa(const struct lu_env *env,
1744 struct ptlrpc_request *req, void *unused,
1747 struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
1748 renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
1749 struct mdt_body *body = NULL;
1750 struct lustre_capa *capa;
1754 GOTO(out, capa = ERR_PTR(status));
1756 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1758 GOTO(out, capa = ERR_PTR(-EFAULT));
1760 if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
1761 GOTO(out, capa = ERR_PTR(-ENOENT));
1763 capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
1765 GOTO(out, capa = ERR_PTR(-EFAULT));
1772 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
1775 struct ptlrpc_request *req;
1778 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
1779 LUSTRE_MDS_VERSION, MDS_GETATTR);
1783 /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
1784 * capa to renew is oss capa.
1786 mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
1787 ptlrpc_request_set_replen(req);
1789 req->rq_async_args.pointer_arg[0] = oc;
1790 req->rq_async_args.pointer_arg[1] = cb;
1791 req->rq_interpret_reply = mdc_interpret_renew_capa;
1792 ptlrpcd_add_req(req, PSCOPE_OTHER);
1796 static int mdc_connect(const struct lu_env *env,
1797 struct lustre_handle *dlm_handle,
1798 struct obd_device *obd, struct obd_uuid *cluuid,
1799 struct obd_connect_data *data,
1802 struct obd_import *imp = obd->u.cli.cl_import;
1804 /* mds-mds import features */
1805 if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
1806 spin_lock(&imp->imp_lock);
1807 imp->imp_server_timeout = 1;
1808 spin_unlock(&imp->imp_lock);
1809 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1810 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
1814 return client_connect_import(env, dlm_handle, obd, cluuid, data, NULL);
1817 struct obd_ops mdc_obd_ops = {
1818 .o_owner = THIS_MODULE,
1819 .o_setup = mdc_setup,
1820 .o_precleanup = mdc_precleanup,
1821 .o_cleanup = mdc_cleanup,
1822 .o_add_conn = client_import_add_conn,
1823 .o_del_conn = client_import_del_conn,
1824 .o_connect = mdc_connect,
1825 .o_disconnect = client_disconnect_export,
1826 .o_iocontrol = mdc_iocontrol,
1827 .o_set_info_async = mdc_set_info_async,
1828 .o_statfs = mdc_statfs,
1830 .o_unpin = mdc_unpin,
1831 .o_fid_init = mdc_fid_init,
1832 .o_fid_fini = mdc_fid_fini,
1833 .o_fid_alloc = mdc_fid_alloc,
1834 .o_fid_delete = mdc_fid_delete,
1835 .o_import_event = mdc_import_event,
1836 .o_llog_init = mdc_llog_init,
1837 .o_llog_finish = mdc_llog_finish,
1838 .o_get_info = mdc_get_info,
1839 .o_process_config = mdc_process_config,
1840 .o_get_uuid = mdc_get_uuid,
1843 struct md_ops mdc_md_ops = {
1844 .m_getstatus = mdc_getstatus,
1845 .m_change_cbdata = mdc_change_cbdata,
1846 .m_close = mdc_close,
1847 .m_create = mdc_create,
1848 .m_done_writing = mdc_done_writing,
1849 .m_enqueue = mdc_enqueue,
1850 .m_getattr = mdc_getattr,
1851 .m_getattr_name = mdc_getattr_name,
1852 .m_intent_lock = mdc_intent_lock,
1854 .m_is_subdir = mdc_is_subdir,
1855 .m_rename = mdc_rename,
1856 .m_setattr = mdc_setattr,
1857 .m_setxattr = mdc_setxattr,
1858 .m_getxattr = mdc_getxattr,
1860 .m_readpage = mdc_readpage,
1861 .m_unlink = mdc_unlink,
1862 .m_cancel_unused = mdc_cancel_unused,
1863 .m_init_ea_size = mdc_init_ea_size,
1864 .m_set_lock_data = mdc_set_lock_data,
1865 .m_lock_match = mdc_lock_match,
1866 .m_get_lustre_md = mdc_get_lustre_md,
1867 .m_free_lustre_md = mdc_free_lustre_md,
1868 .m_set_open_replay_data = mdc_set_open_replay_data,
1869 .m_clear_open_replay_data = mdc_clear_open_replay_data,
1870 .m_renew_capa = mdc_renew_capa,
1871 .m_unpack_capa = mdc_unpack_capa,
1872 .m_get_remote_perm = mdc_get_remote_perm,
1873 .m_intent_getattr_async = mdc_intent_getattr_async,
1874 .m_revalidate_lock = mdc_revalidate_lock
1877 int __init mdc_init(void)
1880 struct lprocfs_static_vars lvars = { 0 };
1881 lprocfs_mdc_init_vars(&lvars);
1883 request_module("lquota");
1884 quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1885 init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1887 rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1888 LUSTRE_MDC_NAME, NULL);
1889 if (rc && quota_interface)
1890 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1896 static void /*__exit*/ mdc_exit(void)
1898 if (quota_interface)
1899 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1901 class_unregister_type(LUSTRE_MDC_NAME);
1904 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1905 MODULE_DESCRIPTION("Lustre Metadata Client");
1906 MODULE_LICENSE("GPL");
1908 module_init(mdc_init);
1909 module_exit(mdc_exit);