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"
59 #include <lustre/lustre_idl.h>
61 #define REQUEST_MINOR 244
63 static quota_interface_t *quota_interface;
64 extern quota_interface_t mdc_quota_interface;
66 static int mdc_cleanup(struct obd_device *obd);
68 int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
69 const struct req_msg_field *field, struct obd_capa **oc)
71 struct lustre_capa *capa;
75 /* swabbed already in mdc_enqueue */
76 capa = req_capsule_server_get(&req->rq_pill, field);
80 c = alloc_capa(CAPA_SITE_CLIENT);
82 CDEBUG(D_INFO, "alloc capa failed!\n");
91 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
92 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
93 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
94 struct obd_capa **pc, int level, int msg_flags)
96 struct ptlrpc_request *req;
97 struct mdt_body *body;
101 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
102 LUSTRE_MDS_VERSION, MDS_GETSTATUS);
106 mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
107 lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
108 req->rq_send_state = level;
110 ptlrpc_request_set_replen(req);
112 rc = ptlrpc_queue_wait(req);
116 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
118 GOTO(out, rc = -EPROTO);
120 if (body->valid & OBD_MD_FLMDSCAPA) {
121 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
126 *rootfid = body->fid1;
128 "root fid="DFID", last_committed="LPU64", last_xid="LPU64"\n",
130 lustre_msg_get_last_committed(req->rq_repmsg),
131 lustre_msg_get_last_xid(req->rq_repmsg));
134 ptlrpc_req_finished(req);
138 /* This should be mdc_get_info("rootfid") */
139 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
140 struct obd_capa **pc)
142 return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
147 * This function now is known to always saying that it will receive 4 buffers
148 * from server. Even for cases when acl_size and md_size is zero, RPC header
149 * will contain 4 fields and RPC itself will contain zero size fields. This is
150 * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
151 * and thus zero, it shrinks it, making zero size. The same story about
152 * md_size. And this is course of problem when client waits for smaller number
153 * of fields. This issue will be fixed later when client gets aware of RPC
156 static int mdc_getattr_common(struct obd_export *exp,
157 struct ptlrpc_request *req)
159 struct req_capsule *pill = &req->rq_pill;
160 struct mdt_body *body;
165 /* Request message already built. */
166 rc = ptlrpc_queue_wait(req);
170 /* sanity check for the reply */
171 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
175 CDEBUG(D_NET, "mode: %o\n", body->mode);
177 if (body->eadatasize != 0) {
178 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
184 if (body->valid & OBD_MD_FLRMTPERM) {
185 struct mdt_remote_perm *perm;
187 LASSERT(client_is_remote(exp));
188 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
189 lustre_swab_mdt_remote_perm);
194 if (body->valid & OBD_MD_FLMDSCAPA) {
195 struct lustre_capa *capa;
196 capa = req_capsule_server_get(pill, &RMF_CAPA1);
204 int mdc_getattr(struct obd_export *exp, const struct lu_fid *fid,
205 struct obd_capa *oc, obd_valid valid, int ea_size,
206 struct ptlrpc_request **request)
208 struct ptlrpc_request *req;
213 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
217 mdc_set_capa_size(req, &RMF_CAPA1, oc);
219 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
221 ptlrpc_request_free(req);
225 /* MDS_BFLAG_EXT_FLAGS: request "new" flags(bug 9486) */
226 mdc_pack_body(req, fid, oc, valid, ea_size, -1, MDS_BFLAG_EXT_FLAGS);
228 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, ea_size);
229 if (valid & OBD_MD_FLRMTPERM) {
230 LASSERT(client_is_remote(exp));
231 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
232 sizeof(struct mdt_remote_perm));
234 ptlrpc_request_set_replen(req);
236 rc = mdc_getattr_common(exp, req);
238 ptlrpc_req_finished(req);
244 int mdc_getattr_name(struct obd_export *exp, const struct lu_fid *fid,
245 struct obd_capa *oc, const char *filename, int namelen,
246 obd_valid valid, int ea_size, __u32 suppgid,
247 struct ptlrpc_request **request)
249 struct ptlrpc_request *req;
254 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
255 &RQF_MDS_GETATTR_NAME);
259 mdc_set_capa_size(req, &RMF_CAPA1, oc);
260 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, namelen);
262 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
264 ptlrpc_request_free(req);
268 /* MDS_BFLAG_EXT_FLAGS: request "new" flags(bug 9486) */
269 mdc_pack_body(req, fid, oc, valid, ea_size, suppgid,
270 MDS_BFLAG_EXT_FLAGS);
273 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
274 LASSERT(strnlen(filename, namelen) == namelen - 1);
275 memcpy(name, filename, namelen);
278 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, ea_size);
279 ptlrpc_request_set_replen(req);
281 rc = mdc_getattr_common(exp, req);
283 ptlrpc_req_finished(req);
289 static int mdc_is_subdir(struct obd_export *exp,
290 const struct lu_fid *pfid,
291 const struct lu_fid *cfid,
292 struct ptlrpc_request **request)
294 struct ptlrpc_request *req;
300 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
301 &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
306 mdc_is_subdir_pack(req, pfid, cfid, 0);
307 ptlrpc_request_set_replen(req);
309 rc = ptlrpc_queue_wait(req);
310 if (rc && rc != -EREMOTE)
311 ptlrpc_req_finished(req);
317 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
318 const struct lu_fid *fid,
319 struct obd_capa *oc, int opcode, obd_valid valid,
320 const char *xattr_name, const char *input,
321 int input_size, int output_size, int flags,
322 __u32 suppgid, struct ptlrpc_request **request)
324 struct ptlrpc_request *req;
325 int xattr_namelen = 0;
331 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
335 mdc_set_capa_size(req, &RMF_CAPA1, oc);
337 xattr_namelen = strlen(xattr_name) + 1;
338 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
343 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
347 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
349 ptlrpc_request_free(req);
353 if (opcode == MDS_REINT) {
354 struct mdt_rec_setxattr *rec;
356 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
357 sizeof(struct mdt_rec_reint));
358 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
359 rec->sx_opcode = REINT_SETXATTR;
361 * cfs_curproc_fs{u,g}id() should replace
362 * current->fs{u,g}id for portability.
364 rec->sx_fsuid = current->fsuid;
365 rec->sx_fsgid = current->fsgid;
366 rec->sx_cap = cfs_curproc_cap_pack();
367 rec->sx_suppgid1 = suppgid;
368 rec->sx_suppgid2 = -1;
370 rec->sx_valid = valid | OBD_MD_FLCTIME;
371 rec->sx_time = cfs_time_current_sec();
372 rec->sx_size = output_size;
373 rec->sx_flags = flags;
375 mdc_pack_capa(req, &RMF_CAPA1, oc);
377 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
381 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
382 memcpy(tmp, xattr_name, xattr_namelen);
385 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
386 memcpy(tmp, input, input_size);
389 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
390 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
391 RCL_SERVER, output_size);
392 ptlrpc_request_set_replen(req);
395 if (opcode == MDS_REINT)
396 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
398 rc = ptlrpc_queue_wait(req);
400 if (opcode == MDS_REINT)
401 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
404 ptlrpc_req_finished(req);
410 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
411 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
412 const char *input, int input_size, int output_size,
413 int flags, __u32 suppgid, struct ptlrpc_request **request)
415 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
416 fid, oc, MDS_REINT, valid, xattr_name,
417 input, input_size, output_size, flags,
421 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
422 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
423 const char *input, int input_size, int output_size,
424 int flags, struct ptlrpc_request **request)
426 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
427 fid, oc, MDS_GETXATTR, valid, xattr_name,
428 input, input_size, output_size, flags,
432 #ifdef CONFIG_FS_POSIX_ACL
433 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
435 struct req_capsule *pill = &req->rq_pill;
436 struct mdt_body *body = md->body;
437 struct posix_acl *acl;
445 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
450 acl = posix_acl_from_xattr(buf, body->aclsize);
453 CERROR("convert xattr to acl: %d\n", rc);
457 rc = posix_acl_valid(acl);
459 CERROR("validate acl: %d\n", rc);
460 posix_acl_release(acl);
468 #define mdc_unpack_acl(req, md) 0
471 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
472 struct obd_export *dt_exp, struct obd_export *md_exp,
473 struct lustre_md *md)
475 struct req_capsule *pill = &req->rq_pill;
480 memset(md, 0, sizeof(*md));
482 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
483 LASSERT(md->body != NULL);
485 if (md->body->valid & OBD_MD_FLEASIZE) {
487 struct lov_mds_md *lmm;
489 if (!S_ISREG(md->body->mode)) {
490 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
491 "regular file, but is not\n");
492 GOTO(out, rc = -EPROTO);
495 if (md->body->eadatasize == 0) {
496 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
497 "but eadatasize 0\n");
498 GOTO(out, rc = -EPROTO);
500 lmmsize = md->body->eadatasize;
501 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
503 GOTO(out, rc = -EPROTO);
505 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
509 if (rc < sizeof(*md->lsm)) {
510 CDEBUG(D_INFO, "lsm size too small: "
511 "rc < sizeof (*md->lsm) (%d < %d)\n",
512 rc, (int)sizeof(*md->lsm));
513 GOTO(out, rc = -EPROTO);
516 } else if (md->body->valid & OBD_MD_FLDIREA) {
518 struct lov_mds_md *lmv;
520 if(!S_ISDIR(md->body->mode)) {
521 CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
522 "directory, but is not\n");
523 GOTO(out, rc = -EPROTO);
526 if (md->body->eadatasize == 0) {
527 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
528 "but eadatasize 0\n");
531 if (md->body->valid & OBD_MD_MEA) {
532 lmvsize = md->body->eadatasize;
533 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
536 GOTO(out, rc = -EPROTO);
538 rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
543 if (rc < sizeof(*md->mea)) {
544 CDEBUG(D_INFO, "size too small: "
545 "rc < sizeof(*md->mea) (%d < %d)\n",
546 rc, (int)sizeof(*md->mea));
547 GOTO(out, rc = -EPROTO);
553 if (md->body->valid & OBD_MD_FLRMTPERM) {
554 /* remote permission */
555 LASSERT(client_is_remote(exp));
556 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
557 lustre_swab_mdt_remote_perm);
558 if (!md->remote_perm)
559 GOTO(out, rc = -EPROTO);
561 else if (md->body->valid & OBD_MD_FLACL) {
562 /* for ACL, it's possible that FLACL is set but aclsize is zero.
563 * only when aclsize != 0 there's an actual segment for ACL
566 if (md->body->aclsize) {
567 rc = mdc_unpack_acl(req, md);
570 #ifdef CONFIG_FS_POSIX_ACL
572 md->posix_acl = NULL;
576 if (md->body->valid & OBD_MD_FLMDSCAPA) {
577 struct obd_capa *oc = NULL;
579 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
585 if (md->body->valid & OBD_MD_FLOSSCAPA) {
586 struct obd_capa *oc = NULL;
588 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
598 capa_put(md->oss_capa);
602 capa_put(md->mds_capa);
605 #ifdef CONFIG_FS_POSIX_ACL
606 posix_acl_release(md->posix_acl);
609 obd_free_memmd(dt_exp, &md->lsm);
614 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
621 * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
624 void mdc_replay_open(struct ptlrpc_request *req)
626 struct md_open_data *mod = req->rq_cb_data;
627 struct ptlrpc_request *close_req;
628 struct obd_client_handle *och;
629 struct lustre_handle old;
630 struct mdt_body *body;
634 DEBUG_REQ(D_ERROR, req,
635 "Can't properly replay without open data.");
640 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
641 LASSERT(body != NULL);
645 struct lustre_handle *file_fh;
647 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
649 file_fh = &och->och_fh;
650 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
651 file_fh->cookie, body->handle.cookie);
653 *file_fh = body->handle;
655 close_req = mod->mod_close_req;
656 if (close_req != NULL) {
657 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
658 struct mdt_epoch *epoch;
660 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
661 epoch = req_capsule_client_get(&close_req->rq_pill,
666 LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
667 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
668 epoch->handle = body->handle;
673 void mdc_commit_open(struct ptlrpc_request *req)
675 struct md_open_data *mod = req->rq_cb_data;
679 if (mod->mod_och != NULL)
680 mod->mod_och->och_mod = NULL;
682 OBD_FREE(mod, sizeof(*mod));
683 req->rq_cb_data = NULL;
686 int mdc_set_open_replay_data(struct obd_export *exp,
687 struct obd_client_handle *och,
688 struct ptlrpc_request *open_req)
690 struct md_open_data *mod;
691 struct mdt_rec_create *rec;
692 struct mdt_body *body;
693 struct obd_import *imp = open_req->rq_import;
696 if (!open_req->rq_replay)
699 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
700 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
701 LASSERT(rec != NULL);
702 /* Incoming message in my byte order (it's been swabbed). */
703 /* Outgoing messages always in my byte order. */
704 LASSERT(body != NULL);
706 /* Only if the import is replayable, we set replay_open data */
707 if (och && imp->imp_replayable) {
710 DEBUG_REQ(D_ERROR, open_req,
711 "Can't allocate md_open_data");
715 spin_lock(&open_req->rq_lock);
718 mod->mod_open_req = open_req;
719 open_req->rq_cb_data = mod;
720 open_req->rq_commit_cb = mdc_commit_open;
721 spin_unlock(&open_req->rq_lock);
724 rec->cr_fid2 = body->fid1;
725 rec->cr_ioepoch = body->ioepoch;
726 rec->cr_old_handle.cookie = body->handle.cookie;
727 open_req->rq_replay_cb = mdc_replay_open;
728 if (!fid_is_sane(&body->fid1)) {
729 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
734 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
738 int mdc_clear_open_replay_data(struct obd_export *exp,
739 struct obd_client_handle *och)
741 struct md_open_data *mod = och->och_mod;
745 * Don't free the structure now (it happens in mdc_commit_open(), after
746 * we're sure we won't need to fix up the close request in the future),
747 * but make sure that replay doesn't poke at the och, which is about to
750 LASSERT(mod != LP_POISON);
758 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
759 struct md_open_data *mod, struct ptlrpc_request **request)
761 struct obd_device *obd = class_exp2obd(exp);
762 struct ptlrpc_request *req;
767 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_CLOSE);
771 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
773 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
775 ptlrpc_request_free(req);
779 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
780 * portal whose threads are not taking any DLM locks and are therefore
781 * always progressing */
782 req->rq_request_portal = MDS_READPAGE_PORTAL;
783 ptlrpc_at_set_req_timeout(req);
785 /* Ensure that this close's handle is fixed up during replay. */
786 if (likely(mod != NULL)) {
787 LASSERTF(mod->mod_open_req->rq_type != LI_POISON,
788 "POISONED open %p!\n", mod->mod_open_req);
790 mod->mod_close_req = req;
791 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
792 /* We no longer want to preserve this open for replay even
793 * though the open was committed. b=3632, b=3633 */
794 spin_lock(&mod->mod_open_req->rq_lock);
795 mod->mod_open_req->rq_replay = 0;
796 spin_unlock(&mod->mod_open_req->rq_lock);
798 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
801 mdc_close_pack(req, op_data);
803 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
804 obd->u.cli.cl_max_mds_easize);
805 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
806 obd->u.cli.cl_max_mds_cookiesize);
808 ptlrpc_request_set_replen(req);
810 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
811 rc = ptlrpc_queue_wait(req);
812 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
814 if (req->rq_repmsg == NULL) {
815 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
818 rc = req->rq_status ?: -EIO;
819 } else if (rc == 0 || rc == -EAGAIN) {
820 struct mdt_body *body;
822 rc = lustre_msg_get_status(req->rq_repmsg);
823 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
824 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
829 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
835 mod->mod_close_req = NULL;
841 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
842 struct md_open_data *mod)
844 struct obd_device *obd = class_exp2obd(exp);
845 struct ptlrpc_request *req;
849 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
850 &RQF_MDS_DONE_WRITING);
854 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
855 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
857 ptlrpc_request_free(req);
862 LASSERTF(mod->mod_open_req->rq_type != LI_POISON,
863 "POISONED setattr %p!\n", mod->mod_open_req);
865 mod->mod_close_req = req;
866 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
867 /* We no longer want to preserve this open for replay even
868 * though the open was committed. b=3632, b=3633 */
869 spin_lock(&mod->mod_open_req->rq_lock);
870 mod->mod_open_req->rq_replay = 0;
871 spin_unlock(&mod->mod_open_req->rq_lock);
874 mdc_close_pack(req, op_data);
875 ptlrpc_request_set_replen(req);
877 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
878 rc = ptlrpc_queue_wait(req);
879 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
881 ptlrpc_req_finished(req);
885 #ifdef HAVE_SPLIT_SUPPORT
886 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
887 const struct page *page, int offset)
889 struct ptlrpc_request *req;
890 struct ptlrpc_bulk_desc *desc;
894 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
898 /* FIXME: capa doesn't support split yet */
899 mdc_set_capa_size(req, &RMF_CAPA1, NULL);
901 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
903 ptlrpc_request_free(req);
907 req->rq_request_portal = MDS_READPAGE_PORTAL;
908 ptlrpc_at_set_req_timeout(req);
910 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
912 GOTO(out, rc = -ENOMEM);
914 /* NB req now owns desc and will free it when it gets freed. */
915 ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
916 mdc_readdir_pack(req, 0, offset, fid, NULL);
918 ptlrpc_request_set_replen(req);
919 rc = ptlrpc_queue_wait(req);
923 rc = sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk);
925 ptlrpc_req_finished(req);
928 EXPORT_SYMBOL(mdc_sendpage);
931 int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
932 struct obd_capa *oc, __u64 offset, struct page *page,
933 struct ptlrpc_request **request)
935 struct ptlrpc_request *req;
936 struct ptlrpc_bulk_desc *desc;
941 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
945 mdc_set_capa_size(req, &RMF_CAPA1, oc);
947 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
949 ptlrpc_request_free(req);
953 req->rq_request_portal = MDS_READPAGE_PORTAL;
954 ptlrpc_at_set_req_timeout(req);
956 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
958 ptlrpc_request_free(req);
962 /* NB req now owns desc and will free it when it gets freed */
963 ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
964 mdc_readdir_pack(req, offset, CFS_PAGE_SIZE, fid, oc);
966 ptlrpc_request_set_replen(req);
967 rc = ptlrpc_queue_wait(req);
969 ptlrpc_req_finished(req);
973 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
974 req->rq_bulk->bd_nob_transferred);
976 ptlrpc_req_finished(req);
980 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
981 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
982 req->rq_bulk->bd_nob_transferred, CFS_PAGE_SIZE);
983 ptlrpc_req_finished(req);
991 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
993 __u32 keylen, vallen;
997 if (gf->gf_pathlen > PATH_MAX)
998 RETURN(-ENAMETOOLONG);
999 if (gf->gf_pathlen < 2)
1002 /* Key is KEY_FID2PATH + getinfo_fid2path description */
1003 keylen = size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1004 OBD_ALLOC(key, keylen);
1007 memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1008 memcpy(key + size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1010 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1011 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1013 if (!fid_is_sane(&gf->gf_fid))
1014 GOTO(out, rc = -EINVAL);
1016 /* Val is struct getinfo_fid2path result plus path */
1017 vallen = sizeof(*gf) + gf->gf_pathlen;
1019 rc = obd_get_info(exp, keylen, key, &vallen, gf, NULL);
1023 if (vallen <= sizeof(*gf))
1024 GOTO(out, rc = -EPROTO);
1025 else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1026 GOTO(out, rc = -EOVERFLOW);
1028 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1029 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1032 OBD_FREE(key, keylen);
1036 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1037 void *karg, void *uarg)
1039 struct obd_device *obd = exp->exp_obd;
1040 struct obd_ioctl_data *data = karg;
1041 struct obd_import *imp = obd->u.cli.cl_import;
1042 struct llog_ctxt *ctxt;
1046 if (!try_module_get(THIS_MODULE)) {
1047 CERROR("Can't get module. Is it alive?");
1051 case OBD_IOC_CHANGELOG_CLEAR: {
1052 struct ioc_changelog_clear *icc = karg;
1053 struct changelog_setinfo cs =
1054 {icc->icc_recno, icc->icc_id};
1055 rc = obd_set_info_async(exp, strlen(KEY_CHANGELOG_CLEAR),
1056 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1060 case OBD_IOC_FID2PATH: {
1061 rc = mdc_ioc_fid2path(exp, karg);
1064 case OBD_IOC_CLIENT_RECOVER:
1065 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
1069 case IOC_OSC_SET_ACTIVE:
1070 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1072 case OBD_IOC_PARSE: {
1073 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1074 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
1075 llog_ctxt_put(ctxt);
1079 case OBD_IOC_LLOG_INFO:
1080 case OBD_IOC_LLOG_PRINT: {
1081 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1082 rc = llog_ioctl(ctxt, cmd, data);
1083 llog_ctxt_put(ctxt);
1087 case OBD_IOC_POLL_QUOTACHECK:
1088 rc = lquota_poll_check(quota_interface, exp,
1089 (struct if_quotacheck *)karg);
1091 case OBD_IOC_PING_TARGET:
1092 rc = ptlrpc_obd_ping(obd);
1095 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1096 GOTO(out, rc = -ENOTTY);
1099 module_put(THIS_MODULE);
1104 static int do_set_info_async(struct obd_export *exp,
1105 obd_count keylen, void *key,
1106 obd_count vallen, void *val,
1107 struct ptlrpc_request_set *set)
1109 struct obd_import *imp = class_exp2cliimp(exp);
1110 struct ptlrpc_request *req;
1115 req = ptlrpc_request_alloc(imp, &RQF_MDS_SET_INFO);
1119 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
1120 RCL_CLIENT, keylen);
1121 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
1122 RCL_CLIENT, vallen);
1123 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SET_INFO);
1125 ptlrpc_request_free(req);
1129 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1130 memcpy(tmp, key, keylen);
1131 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
1132 memcpy(tmp, val, vallen);
1134 ptlrpc_request_set_replen(req);
1137 ptlrpc_set_add_req(set, req);
1138 ptlrpc_check_set(NULL, set);
1140 rc = ptlrpc_queue_wait(req);
1141 ptlrpc_req_finished(req);
1147 int mdc_get_info_rpc(struct obd_export *exp,
1148 obd_count keylen, void *key,
1149 int vallen, void *val)
1151 struct obd_import *imp = class_exp2cliimp(exp);
1152 struct ptlrpc_request *req;
1157 req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1161 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1162 RCL_CLIENT, keylen);
1163 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1164 RCL_CLIENT, sizeof(__u32));
1166 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1168 ptlrpc_request_free(req);
1172 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
1173 memcpy(tmp, key, keylen);
1174 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
1175 memcpy(tmp, &vallen, sizeof(__u32));
1177 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1178 RCL_SERVER, vallen);
1179 ptlrpc_request_set_replen(req);
1181 rc = ptlrpc_queue_wait(req);
1183 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
1184 memcpy(val, tmp, vallen);
1185 if (lustre_msg_swabbed(req->rq_repmsg)) {
1186 if (KEY_IS(KEY_FID2PATH)) {
1187 lustre_swab_fid2path(val);
1191 ptlrpc_req_finished(req);
1196 int mdc_set_info_async(struct obd_export *exp,
1197 obd_count keylen, void *key,
1198 obd_count vallen, void *val,
1199 struct ptlrpc_request_set *set)
1201 struct obd_import *imp = class_exp2cliimp(exp);
1205 if (KEY_IS(KEY_INIT_RECOV)) {
1206 if (vallen != sizeof(int))
1208 spin_lock(&imp->imp_lock);
1209 imp->imp_initial_recov = *(int *)val;
1210 spin_unlock(&imp->imp_lock);
1211 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
1212 exp->exp_obd->obd_name, imp->imp_initial_recov);
1215 /* Turn off initial_recov after we try all backup servers once */
1216 if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
1217 if (vallen != sizeof(int))
1219 spin_lock(&imp->imp_lock);
1220 imp->imp_initial_recov_bk = *(int *)val;
1221 if (imp->imp_initial_recov_bk)
1222 imp->imp_initial_recov = 1;
1223 spin_unlock(&imp->imp_lock);
1224 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
1225 exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
1228 if (KEY_IS(KEY_READ_ONLY)) {
1229 if (vallen != sizeof(int))
1232 spin_lock(&imp->imp_lock);
1233 if (*((int *)val)) {
1234 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
1235 imp->imp_connect_data.ocd_connect_flags |= OBD_CONNECT_RDONLY;
1237 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
1238 imp->imp_connect_data.ocd_connect_flags &= ~OBD_CONNECT_RDONLY;
1240 spin_unlock(&imp->imp_lock);
1242 rc = do_set_info_async(exp, keylen, key, vallen, val, set);
1245 if (KEY_IS(KEY_SPTLRPC_CONF)) {
1246 sptlrpc_conf_client_adapt(exp->exp_obd);
1249 if (KEY_IS(KEY_FLUSH_CTX)) {
1250 sptlrpc_import_flush_my_ctx(imp);
1253 if (KEY_IS(KEY_MDS_CONN)) {
1254 struct obd_import *imp = class_exp2cliimp(exp);
1256 /* mds-mds import */
1257 spin_lock(&imp->imp_lock);
1258 imp->imp_server_timeout = 1;
1259 spin_unlock(&imp->imp_lock);
1260 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1261 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1264 if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1265 rc = do_set_info_async(exp, keylen, key, vallen, val, set);
1272 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
1273 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1277 if (KEY_IS(KEY_MAX_EASIZE)) {
1278 int mdsize, *max_easize;
1280 if (*vallen != sizeof(int))
1282 mdsize = *(int*)val;
1283 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
1284 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
1286 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
1289 if (KEY_IS(KEY_CONN_DATA)) {
1290 struct obd_import *imp = class_exp2cliimp(exp);
1291 struct obd_connect_data *data = val;
1293 if (*vallen != sizeof(*data))
1296 *data = imp->imp_connect_data;
1300 rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
1305 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1306 __u64 max_age, __u32 flags)
1308 struct ptlrpc_request *req;
1309 struct obd_statfs *msfs;
1310 struct obd_import *imp = NULL;
1315 /*Since the request might also come from lprocfs, so we need
1316 *sync this with client_disconnect_export Bug15684*/
1317 down_read(&obd->u.cli.cl_sem);
1318 if (obd->u.cli.cl_import)
1319 imp = class_import_get(obd->u.cli.cl_import);
1320 up_read(&obd->u.cli.cl_sem);
1324 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1325 LUSTRE_MDS_VERSION, MDS_STATFS);
1327 GOTO(output, rc = -ENOMEM);
1329 ptlrpc_request_set_replen(req);
1331 if (flags & OBD_STATFS_NODELAY) {
1332 /* procfs requests not want stay in wait for avoid deadlock */
1333 req->rq_no_resend = 1;
1334 req->rq_no_delay = 1;
1337 rc = ptlrpc_queue_wait(req);
1339 /* check connection error first */
1340 if (imp->imp_connect_error)
1341 rc = imp->imp_connect_error;
1345 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1347 GOTO(out, rc = -EPROTO);
1352 ptlrpc_req_finished(req);
1354 class_import_put(imp);
1358 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
1359 struct obd_capa *oc, struct obd_client_handle *handle,
1362 struct ptlrpc_request *req;
1363 struct mdt_body *body;
1367 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
1371 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1373 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
1375 ptlrpc_request_free(req);
1379 mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
1381 ptlrpc_request_set_replen(req);
1383 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1384 rc = ptlrpc_queue_wait(req);
1385 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1387 CERROR("Pin failed: %d\n", rc);
1391 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1393 GOTO(err_out, rc = -EPROTO);
1395 handle->och_fh = body->handle;
1396 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1398 OBD_ALLOC_PTR(handle->och_mod);
1399 if (handle->och_mod == NULL) {
1400 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
1401 GOTO(err_out, rc = -ENOMEM);
1403 handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
1408 ptlrpc_req_finished(req);
1412 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
1415 struct ptlrpc_request *req;
1416 struct mdt_body *body;
1420 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
1421 LUSTRE_MDS_VERSION, MDS_UNPIN);
1425 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1426 body->handle = handle->och_fh;
1429 ptlrpc_request_set_replen(req);
1431 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1432 rc = ptlrpc_queue_wait(req);
1433 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1436 CERROR("Unpin failed: %d\n", rc);
1438 ptlrpc_req_finished(req);
1439 ptlrpc_req_finished(handle->och_mod->mod_open_req);
1441 OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1445 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
1446 struct obd_capa *oc, struct ptlrpc_request **request)
1448 struct ptlrpc_request *req;
1453 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
1457 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1459 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
1461 ptlrpc_request_free(req);
1465 mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
1467 ptlrpc_request_set_replen(req);
1469 rc = ptlrpc_queue_wait(req);
1471 ptlrpc_req_finished(req);
1477 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1478 enum obd_import_event event)
1482 LASSERT(imp->imp_obd == obd);
1485 case IMP_EVENT_DISCON: {
1487 /* XXX Pass event up to OBDs stack. used only for FLD now */
1488 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
1492 case IMP_EVENT_INACTIVE: {
1493 struct client_obd *cli = &obd->u.cli;
1495 * Flush current sequence to make client obtain new one
1496 * from server in case of disconnect/reconnect.
1497 * If range is already empty then no need to flush it.
1499 if (cli->cl_seq != NULL &&
1500 !range_is_exhausted(&cli->cl_seq->lcs_space)) {
1501 seq_client_flush(cli->cl_seq);
1504 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1507 case IMP_EVENT_INVALIDATE: {
1508 struct ldlm_namespace *ns = obd->obd_namespace;
1510 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1514 case IMP_EVENT_ACTIVE: {
1515 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1519 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
1523 CERROR("Unknown import event %x\n", event);
1529 static int mdc_fid_init(struct obd_export *exp)
1531 struct client_obd *cli = &exp->exp_obd->u.cli;
1536 OBD_ALLOC_PTR(cli->cl_seq);
1537 if (cli->cl_seq == NULL)
1540 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
1542 GOTO(out_free_seq, rc = -ENOMEM);
1544 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
1545 exp->exp_obd->obd_name);
1547 /* Init client side sequence-manager */
1548 rc = seq_client_init(cli->cl_seq, exp,
1549 LUSTRE_SEQ_METADATA,
1551 OBD_FREE(prefix, MAX_OBD_NAME + 5);
1553 GOTO(out_free_seq, rc);
1557 OBD_FREE_PTR(cli->cl_seq);
1562 static int mdc_fid_fini(struct obd_export *exp)
1564 struct client_obd *cli = &exp->exp_obd->u.cli;
1567 if (cli->cl_seq != NULL) {
1568 seq_client_fini(cli->cl_seq);
1569 OBD_FREE_PTR(cli->cl_seq);
1576 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1577 struct md_op_data *op_data)
1579 struct client_obd *cli = &exp->exp_obd->u.cli;
1580 struct lu_client_seq *seq = cli->cl_seq;
1582 RETURN(seq_client_alloc_fid(seq, fid));
1585 /* XXX This method is used only to clear current fid seq
1586 * once fld/mds insert failed */
1587 static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1589 struct client_obd *cli = &exp->exp_obd->u.cli;
1591 seq_client_flush(cli->cl_seq);
1595 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
1596 struct client_obd *cli = &exp->exp_obd->u.cli;
1597 return &cli->cl_target_uuid;
1600 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
1602 struct client_obd *cli = &obd->u.cli;
1603 struct lprocfs_static_vars lvars = { 0 };
1607 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1608 if (!cli->cl_rpc_lock)
1610 mdc_init_rpc_lock(cli->cl_rpc_lock);
1614 OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1615 if (!cli->cl_setattr_lock)
1616 GOTO(err_rpc_lock, rc = -ENOMEM);
1617 mdc_init_rpc_lock(cli->cl_setattr_lock);
1619 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1620 if (!cli->cl_close_lock)
1621 GOTO(err_setattr_lock, rc = -ENOMEM);
1622 mdc_init_rpc_lock(cli->cl_close_lock);
1624 rc = client_obd_setup(obd, cfg);
1626 GOTO(err_close_lock, rc);
1627 lprocfs_mdc_init_vars(&lvars);
1628 lprocfs_obd_setup(obd, lvars.obd_vars);
1629 sptlrpc_lprocfs_cliobd_attach(obd);
1630 ptlrpc_lprocfs_register_obd(obd);
1632 rc = obd_llog_init(obd, &obd->obd_olg, obd, 0, NULL, NULL);
1635 CERROR("failed to setup llogging subsystems\n");
1641 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1643 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1645 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1650 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
1651 * us to make MDS RPCs with large enough reply buffers to hold the
1652 * maximum-sized (= maximum striped) EA and cookie without having to
1653 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1654 static int mdc_init_ea_size(struct obd_export *exp, int easize,
1655 int def_easize, int cookiesize)
1657 struct obd_device *obd = exp->exp_obd;
1658 struct client_obd *cli = &obd->u.cli;
1661 if (cli->cl_max_mds_easize < easize)
1662 cli->cl_max_mds_easize = easize;
1664 if (cli->cl_default_mds_easize < def_easize)
1665 cli->cl_default_mds_easize = def_easize;
1667 if (cli->cl_max_mds_cookiesize < cookiesize)
1668 cli->cl_max_mds_cookiesize = cookiesize;
1673 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1679 case OBD_CLEANUP_EARLY:
1680 case OBD_CLEANUP_EXPORTS:
1681 /* If we set up but never connected, the
1682 client import will not have been cleaned. */
1683 if (obd->u.cli.cl_import) {
1684 struct obd_import *imp;
1685 down_write(&obd->u.cli.cl_sem);
1686 imp = obd->u.cli.cl_import;
1687 CERROR("client import never connected\n");
1688 ptlrpc_invalidate_import(imp);
1689 class_destroy_import(imp);
1690 up_write(&obd->u.cli.cl_sem);
1691 obd->u.cli.cl_import = NULL;
1693 rc = obd_llog_finish(obd, 0);
1695 CERROR("failed to cleanup llogging subsystems\n");
1701 static int mdc_cleanup(struct obd_device *obd)
1703 struct client_obd *cli = &obd->u.cli;
1705 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1706 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1707 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1709 ptlrpc_lprocfs_unregister_obd(obd);
1710 lprocfs_obd_cleanup(obd);
1713 return client_obd_cleanup(obd);
1717 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
1718 struct obd_device *tgt, int count,
1719 struct llog_catid *logid, struct obd_uuid *uuid)
1721 struct llog_ctxt *ctxt;
1725 LASSERT(olg == &obd->obd_olg);
1727 rc = llog_setup(obd, olg, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1731 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1732 llog_initiator_connect(ctxt);
1733 llog_ctxt_put(ctxt);
1735 rc = llog_setup(obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt, 0, NULL,
1738 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
1739 llog_initiator_connect(ctxt);
1740 llog_ctxt_put(ctxt);
1746 static int mdc_llog_finish(struct obd_device *obd, int count)
1748 struct llog_ctxt *ctxt;
1752 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1754 rc = llog_cleanup(ctxt);
1756 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
1758 rc = llog_cleanup(ctxt);
1763 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1765 struct lustre_cfg *lcfg = buf;
1766 struct lprocfs_static_vars lvars = { 0 };
1769 lprocfs_mdc_init_vars(&lvars);
1770 switch (lcfg->lcfg_command) {
1772 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
1782 /* get remote permission for current user on fid */
1783 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
1784 struct obd_capa *oc, __u32 suppgid,
1785 struct ptlrpc_request **request)
1787 struct ptlrpc_request *req;
1791 LASSERT(client_is_remote(exp));
1794 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1798 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1800 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1802 ptlrpc_request_free(req);
1806 mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
1808 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
1809 sizeof(struct mdt_remote_perm));
1811 ptlrpc_request_set_replen(req);
1813 rc = ptlrpc_queue_wait(req);
1815 ptlrpc_req_finished(req);
1821 static int mdc_interpret_renew_capa(const struct lu_env *env,
1822 struct ptlrpc_request *req, void *unused,
1825 struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
1826 renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
1827 struct mdt_body *body = NULL;
1828 struct lustre_capa *capa;
1832 GOTO(out, capa = ERR_PTR(status));
1834 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1836 GOTO(out, capa = ERR_PTR(-EFAULT));
1838 if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
1839 GOTO(out, capa = ERR_PTR(-ENOENT));
1841 capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
1843 GOTO(out, capa = ERR_PTR(-EFAULT));
1850 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
1853 struct ptlrpc_request *req;
1856 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
1857 LUSTRE_MDS_VERSION, MDS_GETATTR);
1861 /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
1862 * capa to renew is oss capa.
1864 mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
1865 ptlrpc_request_set_replen(req);
1867 req->rq_async_args.pointer_arg[0] = oc;
1868 req->rq_async_args.pointer_arg[1] = cb;
1869 req->rq_interpret_reply = mdc_interpret_renew_capa;
1870 ptlrpcd_add_req(req, PSCOPE_OTHER);
1874 static int mdc_connect(const struct lu_env *env,
1875 struct obd_export **exp,
1876 struct obd_device *obd, struct obd_uuid *cluuid,
1877 struct obd_connect_data *data,
1880 struct obd_import *imp = obd->u.cli.cl_import;
1882 /* mds-mds import features */
1883 if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
1884 spin_lock(&imp->imp_lock);
1885 imp->imp_server_timeout = 1;
1886 spin_unlock(&imp->imp_lock);
1887 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1888 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
1892 return client_connect_import(env, exp, obd, cluuid, data, NULL);
1895 struct obd_ops mdc_obd_ops = {
1896 .o_owner = THIS_MODULE,
1897 .o_setup = mdc_setup,
1898 .o_precleanup = mdc_precleanup,
1899 .o_cleanup = mdc_cleanup,
1900 .o_add_conn = client_import_add_conn,
1901 .o_del_conn = client_import_del_conn,
1902 .o_connect = mdc_connect,
1903 .o_disconnect = client_disconnect_export,
1904 .o_iocontrol = mdc_iocontrol,
1905 .o_set_info_async = mdc_set_info_async,
1906 .o_statfs = mdc_statfs,
1908 .o_unpin = mdc_unpin,
1909 .o_fid_init = mdc_fid_init,
1910 .o_fid_fini = mdc_fid_fini,
1911 .o_fid_alloc = mdc_fid_alloc,
1912 .o_fid_delete = mdc_fid_delete,
1913 .o_import_event = mdc_import_event,
1914 .o_llog_init = mdc_llog_init,
1915 .o_llog_finish = mdc_llog_finish,
1916 .o_get_info = mdc_get_info,
1917 .o_process_config = mdc_process_config,
1918 .o_get_uuid = mdc_get_uuid,
1921 struct md_ops mdc_md_ops = {
1922 .m_getstatus = mdc_getstatus,
1923 .m_change_cbdata = mdc_change_cbdata,
1924 .m_close = mdc_close,
1925 .m_create = mdc_create,
1926 .m_done_writing = mdc_done_writing,
1927 .m_enqueue = mdc_enqueue,
1928 .m_getattr = mdc_getattr,
1929 .m_getattr_name = mdc_getattr_name,
1930 .m_intent_lock = mdc_intent_lock,
1932 .m_is_subdir = mdc_is_subdir,
1933 .m_rename = mdc_rename,
1934 .m_setattr = mdc_setattr,
1935 .m_setxattr = mdc_setxattr,
1936 .m_getxattr = mdc_getxattr,
1938 .m_readpage = mdc_readpage,
1939 .m_unlink = mdc_unlink,
1940 .m_cancel_unused = mdc_cancel_unused,
1941 .m_init_ea_size = mdc_init_ea_size,
1942 .m_set_lock_data = mdc_set_lock_data,
1943 .m_lock_match = mdc_lock_match,
1944 .m_get_lustre_md = mdc_get_lustre_md,
1945 .m_free_lustre_md = mdc_free_lustre_md,
1946 .m_set_open_replay_data = mdc_set_open_replay_data,
1947 .m_clear_open_replay_data = mdc_clear_open_replay_data,
1948 .m_renew_capa = mdc_renew_capa,
1949 .m_unpack_capa = mdc_unpack_capa,
1950 .m_get_remote_perm = mdc_get_remote_perm,
1951 .m_intent_getattr_async = mdc_intent_getattr_async,
1952 .m_revalidate_lock = mdc_revalidate_lock
1955 int __init mdc_init(void)
1958 struct lprocfs_static_vars lvars = { 0 };
1959 lprocfs_mdc_init_vars(&lvars);
1961 request_module("lquota");
1962 quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1963 init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1965 rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1966 LUSTRE_MDC_NAME, NULL);
1967 if (rc && quota_interface)
1968 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1974 static void /*__exit*/ mdc_exit(void)
1976 if (quota_interface)
1977 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1979 class_unregister_type(LUSTRE_MDC_NAME);
1982 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1983 MODULE_DESCRIPTION("Lustre Metadata Client");
1984 MODULE_LICENSE("GPL");
1986 module_init(mdc_init);
1987 module_exit(mdc_exit);