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_FLRMTPERM) {
184 struct mdt_remote_perm *perm;
186 LASSERT(client_is_remote(exp));
187 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
188 lustre_swab_mdt_remote_perm);
193 if (body->valid & OBD_MD_FLMDSCAPA) {
194 struct lustre_capa *capa;
195 capa = req_capsule_server_get(pill, &RMF_CAPA1);
203 int mdc_getattr(struct obd_export *exp, const struct lu_fid *fid,
204 struct obd_capa *oc, obd_valid valid, int ea_size,
205 struct ptlrpc_request **request)
207 struct ptlrpc_request *req;
212 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
216 mdc_set_capa_size(req, &RMF_CAPA1, oc);
218 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
220 ptlrpc_request_free(req);
224 /* MDS_BFLAG_EXT_FLAGS: request "new" flags(bug 9486) */
225 mdc_pack_body(req, fid, oc, valid, ea_size, -1, MDS_BFLAG_EXT_FLAGS);
227 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, ea_size);
228 if (valid & OBD_MD_FLRMTPERM) {
229 LASSERT(client_is_remote(exp));
230 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
231 sizeof(struct mdt_remote_perm));
233 ptlrpc_request_set_replen(req);
235 rc = mdc_getattr_common(exp, req);
237 ptlrpc_req_finished(req);
243 int mdc_getattr_name(struct obd_export *exp, const struct lu_fid *fid,
244 struct obd_capa *oc, const char *filename, int namelen,
245 obd_valid valid, int ea_size, __u32 suppgid,
246 struct ptlrpc_request **request)
248 struct ptlrpc_request *req;
253 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
254 &RQF_MDS_GETATTR_NAME);
258 mdc_set_capa_size(req, &RMF_CAPA1, oc);
259 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, namelen);
261 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
263 ptlrpc_request_free(req);
267 /* MDS_BFLAG_EXT_FLAGS: request "new" flags(bug 9486) */
268 mdc_pack_body(req, fid, oc, valid, ea_size, suppgid,
269 MDS_BFLAG_EXT_FLAGS);
272 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
273 LASSERT(strnlen(filename, namelen) == namelen - 1);
274 memcpy(name, filename, namelen);
277 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, ea_size);
278 ptlrpc_request_set_replen(req);
280 rc = mdc_getattr_common(exp, req);
282 ptlrpc_req_finished(req);
288 static int mdc_is_subdir(struct obd_export *exp,
289 const struct lu_fid *pfid,
290 const struct lu_fid *cfid,
291 struct ptlrpc_request **request)
293 struct ptlrpc_request *req;
299 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
300 &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
305 mdc_is_subdir_pack(req, pfid, cfid, 0);
306 ptlrpc_request_set_replen(req);
308 rc = ptlrpc_queue_wait(req);
309 if (rc && rc != -EREMOTE)
310 ptlrpc_req_finished(req);
316 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
317 const struct lu_fid *fid,
318 struct obd_capa *oc, int opcode, obd_valid valid,
319 const char *xattr_name, const char *input,
320 int input_size, int output_size, int flags,
321 __u32 suppgid, struct ptlrpc_request **request)
323 struct ptlrpc_request *req;
324 int xattr_namelen = 0;
330 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
334 mdc_set_capa_size(req, &RMF_CAPA1, oc);
336 xattr_namelen = strlen(xattr_name) + 1;
337 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
342 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
346 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
348 ptlrpc_request_free(req);
352 if (opcode == MDS_REINT) {
353 struct mdt_rec_setxattr *rec;
355 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
356 sizeof(struct mdt_rec_reint));
357 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
358 rec->sx_opcode = REINT_SETXATTR;
360 * cfs_curproc_fs{u,g}id() should replace
361 * current->fs{u,g}id for portability.
363 rec->sx_fsuid = current->fsuid;
364 rec->sx_fsgid = current->fsgid;
365 rec->sx_cap = cfs_curproc_cap_pack();
366 rec->sx_suppgid1 = suppgid;
367 rec->sx_suppgid2 = -1;
369 rec->sx_valid = valid | OBD_MD_FLCTIME;
370 rec->sx_time = cfs_time_current_sec();
371 rec->sx_size = output_size;
372 rec->sx_flags = flags;
374 mdc_pack_capa(req, &RMF_CAPA1, oc);
376 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
380 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
381 memcpy(tmp, xattr_name, xattr_namelen);
384 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
385 memcpy(tmp, input, input_size);
388 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
389 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
390 RCL_SERVER, output_size);
391 ptlrpc_request_set_replen(req);
394 if (opcode == MDS_REINT)
395 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
397 rc = ptlrpc_queue_wait(req);
399 if (opcode == MDS_REINT)
400 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
403 ptlrpc_req_finished(req);
409 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
410 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
411 const char *input, int input_size, int output_size,
412 int flags, __u32 suppgid, struct ptlrpc_request **request)
414 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
415 fid, oc, MDS_REINT, valid, xattr_name,
416 input, input_size, output_size, flags,
420 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
421 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
422 const char *input, int input_size, int output_size,
423 int flags, struct ptlrpc_request **request)
425 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
426 fid, oc, MDS_GETXATTR, valid, xattr_name,
427 input, input_size, output_size, flags,
431 #ifdef CONFIG_FS_POSIX_ACL
432 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
434 struct req_capsule *pill = &req->rq_pill;
435 struct mdt_body *body = md->body;
436 struct posix_acl *acl;
444 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
449 acl = posix_acl_from_xattr(buf, body->aclsize);
452 CERROR("convert xattr to acl: %d\n", rc);
456 rc = posix_acl_valid(acl);
458 CERROR("validate acl: %d\n", rc);
459 posix_acl_release(acl);
467 #define mdc_unpack_acl(req, md) 0
470 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
471 struct obd_export *dt_exp, struct obd_export *md_exp,
472 struct lustre_md *md)
474 struct req_capsule *pill = &req->rq_pill;
479 memset(md, 0, sizeof(*md));
481 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
482 LASSERT(md->body != NULL);
484 if (md->body->valid & OBD_MD_FLEASIZE) {
486 struct lov_mds_md *lmm;
488 if (!S_ISREG(md->body->mode)) {
489 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
490 "regular file, but is not\n");
491 GOTO(out, rc = -EPROTO);
494 if (md->body->eadatasize == 0) {
495 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
496 "but eadatasize 0\n");
497 GOTO(out, rc = -EPROTO);
499 lmmsize = md->body->eadatasize;
500 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
502 GOTO(out, rc = -EPROTO);
504 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
508 if (rc < sizeof(*md->lsm)) {
509 CDEBUG(D_INFO, "lsm size too small: "
510 "rc < sizeof (*md->lsm) (%d < %d)\n",
511 rc, (int)sizeof(*md->lsm));
512 GOTO(out, rc = -EPROTO);
515 } else if (md->body->valid & OBD_MD_FLDIREA) {
517 struct lov_mds_md *lmv;
519 if(!S_ISDIR(md->body->mode)) {
520 CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
521 "directory, but is not\n");
522 GOTO(out, rc = -EPROTO);
525 if (md->body->eadatasize == 0) {
526 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
527 "but eadatasize 0\n");
530 if (md->body->valid & OBD_MD_MEA) {
531 lmvsize = md->body->eadatasize;
532 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
535 GOTO(out, rc = -EPROTO);
537 rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
542 if (rc < sizeof(*md->mea)) {
543 CDEBUG(D_INFO, "size too small: "
544 "rc < sizeof(*md->mea) (%d < %d)\n",
545 rc, (int)sizeof(*md->mea));
546 GOTO(out, rc = -EPROTO);
552 if (md->body->valid & OBD_MD_FLRMTPERM) {
553 /* remote permission */
554 LASSERT(client_is_remote(exp));
555 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
556 lustre_swab_mdt_remote_perm);
557 if (!md->remote_perm)
558 GOTO(out, rc = -EPROTO);
560 else if (md->body->valid & OBD_MD_FLACL) {
561 /* for ACL, it's possible that FLACL is set but aclsize is zero.
562 * only when aclsize != 0 there's an actual segment for ACL
565 if (md->body->aclsize) {
566 rc = mdc_unpack_acl(req, md);
569 #ifdef CONFIG_FS_POSIX_ACL
571 md->posix_acl = NULL;
575 if (md->body->valid & OBD_MD_FLMDSCAPA) {
576 struct obd_capa *oc = NULL;
578 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
584 if (md->body->valid & OBD_MD_FLOSSCAPA) {
585 struct obd_capa *oc = NULL;
587 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
597 capa_put(md->oss_capa);
601 capa_put(md->mds_capa);
604 #ifdef CONFIG_FS_POSIX_ACL
605 posix_acl_release(md->posix_acl);
608 obd_free_memmd(dt_exp, &md->lsm);
613 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
620 * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
623 void mdc_replay_open(struct ptlrpc_request *req)
625 struct md_open_data *mod = req->rq_cb_data;
626 struct ptlrpc_request *close_req;
627 struct obd_client_handle *och;
628 struct lustre_handle old;
629 struct mdt_body *body;
633 DEBUG_REQ(D_ERROR, req,
634 "Can't properly replay without open data.");
639 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
640 LASSERT(body != NULL);
644 struct lustre_handle *file_fh;
646 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
648 file_fh = &och->och_fh;
649 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
650 file_fh->cookie, body->handle.cookie);
652 *file_fh = body->handle;
654 close_req = mod->mod_close_req;
655 if (close_req != NULL) {
656 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
657 struct mdt_epoch *epoch;
659 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
660 epoch = req_capsule_client_get(&close_req->rq_pill,
665 LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
666 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
667 epoch->handle = body->handle;
672 void mdc_commit_open(struct ptlrpc_request *req)
674 struct md_open_data *mod = req->rq_cb_data;
678 if (mod->mod_close_req != NULL)
679 mod->mod_close_req->rq_cb_data = NULL;
681 if (mod->mod_och != NULL)
682 mod->mod_och->och_mod = NULL;
684 OBD_FREE(mod, sizeof(*mod));
685 req->rq_cb_data = NULL;
688 int mdc_set_open_replay_data(struct obd_export *exp,
689 struct obd_client_handle *och,
690 struct ptlrpc_request *open_req)
692 struct md_open_data *mod;
693 struct mdt_rec_create *rec;
694 struct mdt_body *body;
695 struct obd_import *imp = open_req->rq_import;
698 if (!open_req->rq_replay)
701 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
702 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
703 LASSERT(rec != NULL);
704 /* Incoming message in my byte order (it's been swabbed). */
705 /* Outgoing messages always in my byte order. */
706 LASSERT(body != NULL);
708 /* Only if the import is replayable, we set replay_open data */
709 if (och && imp->imp_replayable) {
712 DEBUG_REQ(D_ERROR, open_req,
713 "Can't allocate md_open_data");
717 spin_lock(&open_req->rq_lock);
720 mod->mod_open_req = open_req;
721 open_req->rq_cb_data = mod;
722 open_req->rq_commit_cb = mdc_commit_open;
723 spin_unlock(&open_req->rq_lock);
726 rec->cr_fid2 = body->fid1;
727 rec->cr_ioepoch = body->ioepoch;
728 rec->cr_old_handle.cookie = body->handle.cookie;
729 open_req->rq_replay_cb = mdc_replay_open;
730 if (!fid_is_sane(&body->fid1)) {
731 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
736 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
740 int mdc_clear_open_replay_data(struct obd_export *exp,
741 struct obd_client_handle *och)
743 struct md_open_data *mod = och->och_mod;
747 * Don't free the structure now (it happens in mdc_commit_open(), after
748 * we're sure we won't need to fix up the close request in the future),
749 * but make sure that replay doesn't poke at the och, which is about to
752 LASSERT(mod != LP_POISON);
760 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
761 struct md_open_data *mod, struct ptlrpc_request **request)
763 struct obd_device *obd = class_exp2obd(exp);
764 struct ptlrpc_request *req;
769 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_CLOSE);
773 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
775 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
777 ptlrpc_request_free(req);
781 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
782 * portal whose threads are not taking any DLM locks and are therefore
783 * always progressing */
784 req->rq_request_portal = MDS_READPAGE_PORTAL;
785 ptlrpc_at_set_req_timeout(req);
787 /* Ensure that this close's handle is fixed up during replay. */
788 if (likely(mod != NULL)) {
789 LASSERTF(mod->mod_open_req->rq_type != LI_POISON,
790 "POISONED open %p!\n", mod->mod_open_req);
792 mod->mod_close_req = req;
793 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
794 /* We no longer want to preserve this open for replay even
795 * though the open was committed. b=3632, b=3633 */
796 spin_lock(&mod->mod_open_req->rq_lock);
797 mod->mod_open_req->rq_replay = 0;
798 spin_unlock(&mod->mod_open_req->rq_lock);
800 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
803 mdc_close_pack(req, op_data);
805 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
806 obd->u.cli.cl_max_mds_easize);
807 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
808 obd->u.cli.cl_max_mds_cookiesize);
810 ptlrpc_request_set_replen(req);
812 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
813 rc = ptlrpc_queue_wait(req);
814 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
816 if (req->rq_repmsg == NULL) {
817 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
820 rc = req->rq_status ?: -EIO;
821 } else if (rc == 0 || rc == -EAGAIN) {
822 struct mdt_body *body;
824 rc = lustre_msg_get_status(req->rq_repmsg);
825 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
826 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
830 } else if (mod == NULL) {
831 if (req->rq_import->imp_replayable)
832 CERROR("Unexpected: can't find md_open_data,"
833 "but close succeeded with replayable imp"
835 "http://bugzilla.lustre.org/\n");
838 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
847 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
848 struct md_open_data *mod)
850 struct obd_device *obd = class_exp2obd(exp);
851 struct ptlrpc_request *req;
855 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
856 &RQF_MDS_DONE_WRITING);
860 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
861 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
863 ptlrpc_request_free(req);
868 LASSERTF(mod->mod_open_req->rq_type != LI_POISON,
869 "POISONED setattr %p!\n", mod->mod_open_req);
871 mod->mod_close_req = req;
872 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
873 /* We no longer want to preserve this open for replay even
874 * though the open was committed. b=3632, b=3633 */
875 spin_lock(&mod->mod_open_req->rq_lock);
876 mod->mod_open_req->rq_replay = 0;
877 spin_unlock(&mod->mod_open_req->rq_lock);
880 mdc_close_pack(req, op_data);
881 ptlrpc_request_set_replen(req);
883 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
884 rc = ptlrpc_queue_wait(req);
885 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
887 ptlrpc_req_finished(req);
891 #ifdef HAVE_SPLIT_SUPPORT
892 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
893 const struct page *page, int offset)
895 struct ptlrpc_request *req;
896 struct ptlrpc_bulk_desc *desc;
900 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
904 /* FIXME: capa doesn't support split yet */
905 mdc_set_capa_size(req, &RMF_CAPA1, NULL);
907 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
909 ptlrpc_request_free(req);
913 req->rq_request_portal = MDS_READPAGE_PORTAL;
914 ptlrpc_at_set_req_timeout(req);
916 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
918 GOTO(out, rc = -ENOMEM);
920 /* NB req now owns desc and will free it when it gets freed. */
921 ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
922 mdc_readdir_pack(req, 0, offset, fid, NULL);
924 ptlrpc_request_set_replen(req);
925 rc = ptlrpc_queue_wait(req);
929 rc = sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk);
931 ptlrpc_req_finished(req);
934 EXPORT_SYMBOL(mdc_sendpage);
937 int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
938 struct obd_capa *oc, __u64 offset, struct page *page,
939 struct ptlrpc_request **request)
941 struct ptlrpc_request *req;
942 struct ptlrpc_bulk_desc *desc;
947 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
951 mdc_set_capa_size(req, &RMF_CAPA1, oc);
953 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
955 ptlrpc_request_free(req);
959 req->rq_request_portal = MDS_READPAGE_PORTAL;
960 ptlrpc_at_set_req_timeout(req);
962 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
964 ptlrpc_request_free(req);
968 /* NB req now owns desc and will free it when it gets freed */
969 ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
970 mdc_readdir_pack(req, offset, CFS_PAGE_SIZE, fid, oc);
972 ptlrpc_request_set_replen(req);
973 rc = ptlrpc_queue_wait(req);
975 ptlrpc_req_finished(req);
979 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
980 req->rq_bulk->bd_nob_transferred);
982 ptlrpc_req_finished(req);
986 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
987 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
988 req->rq_bulk->bd_nob_transferred, CFS_PAGE_SIZE);
989 ptlrpc_req_finished(req);
997 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
998 void *karg, void *uarg)
1000 struct obd_device *obd = exp->exp_obd;
1001 struct obd_ioctl_data *data = karg;
1002 struct obd_import *imp = obd->u.cli.cl_import;
1003 struct llog_ctxt *ctxt;
1007 if (!try_module_get(THIS_MODULE)) {
1008 CERROR("Can't get module. Is it alive?");
1012 case OBD_IOC_CHANGELOG_CLEAR: {
1013 struct changelog_setinfo cs =
1014 {data->ioc_u64_1, data->ioc_u32_1};
1015 rc = obd_set_info_async(exp, strlen(KEY_CHANGELOG_CLEAR),
1016 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1020 case OBD_IOC_CLIENT_RECOVER:
1021 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
1025 case IOC_OSC_SET_ACTIVE:
1026 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1028 case OBD_IOC_PARSE: {
1029 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1030 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
1031 llog_ctxt_put(ctxt);
1035 case OBD_IOC_LLOG_INFO:
1036 case OBD_IOC_LLOG_PRINT: {
1037 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1038 rc = llog_ioctl(ctxt, cmd, data);
1039 llog_ctxt_put(ctxt);
1043 case OBD_IOC_POLL_QUOTACHECK:
1044 rc = lquota_poll_check(quota_interface, exp,
1045 (struct if_quotacheck *)karg);
1047 case OBD_IOC_PING_TARGET:
1048 rc = ptlrpc_obd_ping(obd);
1051 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1052 GOTO(out, rc = -ENOTTY);
1055 module_put(THIS_MODULE);
1060 static int do_set_info_async(struct obd_export *exp,
1061 obd_count keylen, void *key,
1062 obd_count vallen, void *val,
1063 struct ptlrpc_request_set *set)
1065 struct obd_import *imp = class_exp2cliimp(exp);
1066 struct ptlrpc_request *req;
1071 req = ptlrpc_request_alloc(imp, &RQF_MDS_SET_INFO);
1075 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
1076 RCL_CLIENT, keylen);
1077 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
1078 RCL_CLIENT, vallen);
1079 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SET_INFO);
1081 ptlrpc_request_free(req);
1085 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1086 memcpy(tmp, key, keylen);
1087 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
1088 memcpy(tmp, val, vallen);
1090 ptlrpc_request_set_replen(req);
1093 ptlrpc_set_add_req(set, req);
1094 ptlrpc_check_set(NULL, set);
1096 rc = ptlrpc_queue_wait(req);
1097 ptlrpc_req_finished(req);
1103 int mdc_set_info_async(struct obd_export *exp,
1104 obd_count keylen, void *key,
1105 obd_count vallen, void *val,
1106 struct ptlrpc_request_set *set)
1108 struct obd_import *imp = class_exp2cliimp(exp);
1112 if (KEY_IS(KEY_INIT_RECOV)) {
1113 if (vallen != sizeof(int))
1115 spin_lock(&imp->imp_lock);
1116 imp->imp_initial_recov = *(int *)val;
1117 spin_unlock(&imp->imp_lock);
1118 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
1119 exp->exp_obd->obd_name, imp->imp_initial_recov);
1122 /* Turn off initial_recov after we try all backup servers once */
1123 if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
1124 if (vallen != sizeof(int))
1126 spin_lock(&imp->imp_lock);
1127 imp->imp_initial_recov_bk = *(int *)val;
1128 if (imp->imp_initial_recov_bk)
1129 imp->imp_initial_recov = 1;
1130 spin_unlock(&imp->imp_lock);
1131 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
1132 exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
1135 if (KEY_IS(KEY_READ_ONLY)) {
1136 if (vallen != sizeof(int))
1139 spin_lock(&imp->imp_lock);
1140 if (*((int *)val)) {
1141 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
1142 imp->imp_connect_data.ocd_connect_flags |= OBD_CONNECT_RDONLY;
1144 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
1145 imp->imp_connect_data.ocd_connect_flags &= ~OBD_CONNECT_RDONLY;
1147 spin_unlock(&imp->imp_lock);
1149 rc = do_set_info_async(exp, keylen, key, vallen, val, set);
1152 if (KEY_IS(KEY_SPTLRPC_CONF)) {
1153 sptlrpc_conf_client_adapt(exp->exp_obd);
1156 if (KEY_IS(KEY_FLUSH_CTX)) {
1157 sptlrpc_import_flush_my_ctx(imp);
1160 if (KEY_IS(KEY_MDS_CONN)) {
1161 struct obd_import *imp = class_exp2cliimp(exp);
1163 /* mds-mds import */
1164 spin_lock(&imp->imp_lock);
1165 imp->imp_server_timeout = 1;
1166 spin_unlock(&imp->imp_lock);
1167 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1168 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1171 if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1172 rc = do_set_info_async(exp, keylen, key, vallen, val, set);
1179 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
1180 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1184 if (KEY_IS(KEY_MAX_EASIZE)) {
1185 int mdsize, *max_easize;
1187 if (*vallen != sizeof(int))
1189 mdsize = *(int*)val;
1190 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
1191 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
1193 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
1196 if (KEY_IS(KEY_CONN_DATA)) {
1197 struct obd_import *imp = class_exp2cliimp(exp);
1198 struct obd_connect_data *data = val;
1200 if (*vallen != sizeof(*data))
1203 *data = imp->imp_connect_data;
1210 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1211 __u64 max_age, __u32 flags)
1213 struct ptlrpc_request *req;
1214 struct obd_statfs *msfs;
1215 struct obd_import *imp = NULL;
1220 /*Since the request might also come from lprocfs, so we need
1221 *sync this with client_disconnect_export Bug15684*/
1222 down_read(&obd->u.cli.cl_sem);
1223 if (obd->u.cli.cl_import)
1224 imp = class_import_get(obd->u.cli.cl_import);
1225 up_read(&obd->u.cli.cl_sem);
1229 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1230 LUSTRE_MDS_VERSION, MDS_STATFS);
1232 GOTO(output, rc = -ENOMEM);
1234 ptlrpc_request_set_replen(req);
1236 if (flags & OBD_STATFS_NODELAY) {
1237 /* procfs requests not want stay in wait for avoid deadlock */
1238 req->rq_no_resend = 1;
1239 req->rq_no_delay = 1;
1242 rc = ptlrpc_queue_wait(req);
1244 /* check connection error first */
1245 if (imp->imp_connect_error)
1246 rc = imp->imp_connect_error;
1250 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1252 GOTO(out, rc = -EPROTO);
1257 ptlrpc_req_finished(req);
1259 class_import_put(imp);
1263 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
1264 struct obd_capa *oc, struct obd_client_handle *handle,
1267 struct ptlrpc_request *req;
1268 struct mdt_body *body;
1272 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
1276 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1278 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
1280 ptlrpc_request_free(req);
1284 mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
1286 ptlrpc_request_set_replen(req);
1288 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1289 rc = ptlrpc_queue_wait(req);
1290 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1292 CERROR("Pin failed: %d\n", rc);
1296 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1298 GOTO(err_out, rc = -EPROTO);
1300 handle->och_fh = body->handle;
1301 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1303 OBD_ALLOC_PTR(handle->och_mod);
1304 if (handle->och_mod == NULL) {
1305 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
1306 GOTO(err_out, rc = -ENOMEM);
1308 handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
1313 ptlrpc_req_finished(req);
1317 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
1320 struct ptlrpc_request *req;
1321 struct mdt_body *body;
1325 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
1326 LUSTRE_MDS_VERSION, MDS_UNPIN);
1330 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1331 body->handle = handle->och_fh;
1334 ptlrpc_request_set_replen(req);
1336 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1337 rc = ptlrpc_queue_wait(req);
1338 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1341 CERROR("Unpin failed: %d\n", rc);
1343 ptlrpc_req_finished(req);
1344 ptlrpc_req_finished(handle->och_mod->mod_open_req);
1346 OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1350 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
1351 struct obd_capa *oc, struct ptlrpc_request **request)
1353 struct ptlrpc_request *req;
1358 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
1362 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1364 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
1366 ptlrpc_request_free(req);
1370 mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
1372 ptlrpc_request_set_replen(req);
1374 rc = ptlrpc_queue_wait(req);
1376 ptlrpc_req_finished(req);
1382 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1383 enum obd_import_event event)
1387 LASSERT(imp->imp_obd == obd);
1390 case IMP_EVENT_DISCON: {
1392 /* XXX Pass event up to OBDs stack. used only for FLD now */
1393 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
1397 case IMP_EVENT_INACTIVE: {
1398 struct client_obd *cli = &obd->u.cli;
1400 * Flush current sequence to make client obtain new one
1401 * from server in case of disconnect/reconnect.
1402 * If range is already empty then no need to flush it.
1404 if (cli->cl_seq != NULL &&
1405 !range_is_exhausted(&cli->cl_seq->lcs_space)) {
1406 seq_client_flush(cli->cl_seq);
1409 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1412 case IMP_EVENT_INVALIDATE: {
1413 struct ldlm_namespace *ns = obd->obd_namespace;
1415 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1419 case IMP_EVENT_ACTIVE: {
1420 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1424 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
1428 CERROR("Unknown import event %x\n", event);
1434 static int mdc_fid_init(struct obd_export *exp)
1436 struct client_obd *cli = &exp->exp_obd->u.cli;
1441 OBD_ALLOC_PTR(cli->cl_seq);
1442 if (cli->cl_seq == NULL)
1445 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
1447 GOTO(out_free_seq, rc = -ENOMEM);
1449 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
1450 exp->exp_obd->obd_name);
1452 /* Init client side sequence-manager */
1453 rc = seq_client_init(cli->cl_seq, exp,
1454 LUSTRE_SEQ_METADATA,
1456 OBD_FREE(prefix, MAX_OBD_NAME + 5);
1458 GOTO(out_free_seq, rc);
1462 OBD_FREE_PTR(cli->cl_seq);
1467 static int mdc_fid_fini(struct obd_export *exp)
1469 struct client_obd *cli = &exp->exp_obd->u.cli;
1472 if (cli->cl_seq != NULL) {
1473 seq_client_fini(cli->cl_seq);
1474 OBD_FREE_PTR(cli->cl_seq);
1481 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1482 struct md_op_data *op_data)
1484 struct client_obd *cli = &exp->exp_obd->u.cli;
1485 struct lu_client_seq *seq = cli->cl_seq;
1487 RETURN(seq_client_alloc_fid(seq, fid));
1490 /* XXX This method is used only to clear current fid seq
1491 * once fld/mds insert failed */
1492 static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1494 struct client_obd *cli = &exp->exp_obd->u.cli;
1496 seq_client_flush(cli->cl_seq);
1500 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
1501 struct client_obd *cli = &exp->exp_obd->u.cli;
1502 return &cli->cl_target_uuid;
1505 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
1507 struct client_obd *cli = &obd->u.cli;
1508 struct lprocfs_static_vars lvars = { 0 };
1512 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1513 if (!cli->cl_rpc_lock)
1515 mdc_init_rpc_lock(cli->cl_rpc_lock);
1519 OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1520 if (!cli->cl_setattr_lock)
1521 GOTO(err_rpc_lock, rc = -ENOMEM);
1522 mdc_init_rpc_lock(cli->cl_setattr_lock);
1524 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1525 if (!cli->cl_close_lock)
1526 GOTO(err_setattr_lock, rc = -ENOMEM);
1527 mdc_init_rpc_lock(cli->cl_close_lock);
1529 rc = client_obd_setup(obd, cfg);
1531 GOTO(err_close_lock, rc);
1532 lprocfs_mdc_init_vars(&lvars);
1533 lprocfs_obd_setup(obd, lvars.obd_vars);
1534 sptlrpc_lprocfs_cliobd_attach(obd);
1535 ptlrpc_lprocfs_register_obd(obd);
1537 rc = obd_llog_init(obd, &obd->obd_olg, obd, 0, NULL, NULL);
1540 CERROR("failed to setup llogging subsystems\n");
1546 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1548 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1550 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1555 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
1556 * us to make MDS RPCs with large enough reply buffers to hold the
1557 * maximum-sized (= maximum striped) EA and cookie without having to
1558 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1559 static int mdc_init_ea_size(struct obd_export *exp, int easize,
1560 int def_easize, int cookiesize)
1562 struct obd_device *obd = exp->exp_obd;
1563 struct client_obd *cli = &obd->u.cli;
1566 if (cli->cl_max_mds_easize < easize)
1567 cli->cl_max_mds_easize = easize;
1569 if (cli->cl_default_mds_easize < def_easize)
1570 cli->cl_default_mds_easize = def_easize;
1572 if (cli->cl_max_mds_cookiesize < cookiesize)
1573 cli->cl_max_mds_cookiesize = cookiesize;
1578 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1584 case OBD_CLEANUP_EARLY:
1585 case OBD_CLEANUP_EXPORTS:
1586 /* If we set up but never connected, the
1587 client import will not have been cleaned. */
1588 if (obd->u.cli.cl_import) {
1589 struct obd_import *imp;
1590 down_write(&obd->u.cli.cl_sem);
1591 imp = obd->u.cli.cl_import;
1592 CERROR("client import never connected\n");
1593 ptlrpc_invalidate_import(imp);
1594 class_destroy_import(imp);
1595 up_write(&obd->u.cli.cl_sem);
1596 obd->u.cli.cl_import = NULL;
1598 rc = obd_llog_finish(obd, 0);
1600 CERROR("failed to cleanup llogging subsystems\n");
1606 static int mdc_cleanup(struct obd_device *obd)
1608 struct client_obd *cli = &obd->u.cli;
1610 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1611 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1612 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1614 ptlrpc_lprocfs_unregister_obd(obd);
1615 lprocfs_obd_cleanup(obd);
1618 return client_obd_cleanup(obd);
1622 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
1623 struct obd_device *tgt, int count,
1624 struct llog_catid *logid, struct obd_uuid *uuid)
1626 struct llog_ctxt *ctxt;
1630 LASSERT(olg == &obd->obd_olg);
1632 rc = llog_setup(obd, olg, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1636 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1637 llog_initiator_connect(ctxt);
1638 llog_ctxt_put(ctxt);
1640 rc = llog_setup(obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt, 0, NULL,
1643 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
1644 llog_initiator_connect(ctxt);
1645 llog_ctxt_put(ctxt);
1651 static int mdc_llog_finish(struct obd_device *obd, int count)
1653 struct llog_ctxt *ctxt;
1657 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1659 rc = llog_cleanup(ctxt);
1661 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
1663 rc = llog_cleanup(ctxt);
1668 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1670 struct lustre_cfg *lcfg = buf;
1671 struct lprocfs_static_vars lvars = { 0 };
1674 lprocfs_mdc_init_vars(&lvars);
1675 switch (lcfg->lcfg_command) {
1677 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
1687 /* get remote permission for current user on fid */
1688 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
1689 struct obd_capa *oc, __u32 suppgid,
1690 struct ptlrpc_request **request)
1692 struct ptlrpc_request *req;
1696 LASSERT(client_is_remote(exp));
1699 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1703 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1705 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1707 ptlrpc_request_free(req);
1711 mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
1713 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
1714 sizeof(struct mdt_remote_perm));
1716 ptlrpc_request_set_replen(req);
1718 rc = ptlrpc_queue_wait(req);
1720 ptlrpc_req_finished(req);
1726 static int mdc_interpret_renew_capa(const struct lu_env *env,
1727 struct ptlrpc_request *req, void *unused,
1730 struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
1731 renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
1732 struct mdt_body *body = NULL;
1733 struct lustre_capa *capa;
1737 GOTO(out, capa = ERR_PTR(status));
1739 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1741 GOTO(out, capa = ERR_PTR(-EFAULT));
1743 if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
1744 GOTO(out, capa = ERR_PTR(-ENOENT));
1746 capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
1748 GOTO(out, capa = ERR_PTR(-EFAULT));
1755 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
1758 struct ptlrpc_request *req;
1761 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
1762 LUSTRE_MDS_VERSION, MDS_GETATTR);
1766 /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
1767 * capa to renew is oss capa.
1769 mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
1770 ptlrpc_request_set_replen(req);
1772 req->rq_async_args.pointer_arg[0] = oc;
1773 req->rq_async_args.pointer_arg[1] = cb;
1774 req->rq_interpret_reply = mdc_interpret_renew_capa;
1775 ptlrpcd_add_req(req, PSCOPE_OTHER);
1779 static int mdc_connect(const struct lu_env *env,
1780 struct obd_export **exp,
1781 struct obd_device *obd, struct obd_uuid *cluuid,
1782 struct obd_connect_data *data,
1785 struct obd_import *imp = obd->u.cli.cl_import;
1787 /* mds-mds import features */
1788 if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
1789 spin_lock(&imp->imp_lock);
1790 imp->imp_server_timeout = 1;
1791 spin_unlock(&imp->imp_lock);
1792 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1793 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
1797 return client_connect_import(env, exp, obd, cluuid, data, NULL);
1800 struct obd_ops mdc_obd_ops = {
1801 .o_owner = THIS_MODULE,
1802 .o_setup = mdc_setup,
1803 .o_precleanup = mdc_precleanup,
1804 .o_cleanup = mdc_cleanup,
1805 .o_add_conn = client_import_add_conn,
1806 .o_del_conn = client_import_del_conn,
1807 .o_connect = mdc_connect,
1808 .o_disconnect = client_disconnect_export,
1809 .o_iocontrol = mdc_iocontrol,
1810 .o_set_info_async = mdc_set_info_async,
1811 .o_statfs = mdc_statfs,
1813 .o_unpin = mdc_unpin,
1814 .o_fid_init = mdc_fid_init,
1815 .o_fid_fini = mdc_fid_fini,
1816 .o_fid_alloc = mdc_fid_alloc,
1817 .o_fid_delete = mdc_fid_delete,
1818 .o_import_event = mdc_import_event,
1819 .o_llog_init = mdc_llog_init,
1820 .o_llog_finish = mdc_llog_finish,
1821 .o_get_info = mdc_get_info,
1822 .o_process_config = mdc_process_config,
1823 .o_get_uuid = mdc_get_uuid,
1826 struct md_ops mdc_md_ops = {
1827 .m_getstatus = mdc_getstatus,
1828 .m_change_cbdata = mdc_change_cbdata,
1829 .m_close = mdc_close,
1830 .m_create = mdc_create,
1831 .m_done_writing = mdc_done_writing,
1832 .m_enqueue = mdc_enqueue,
1833 .m_getattr = mdc_getattr,
1834 .m_getattr_name = mdc_getattr_name,
1835 .m_intent_lock = mdc_intent_lock,
1837 .m_is_subdir = mdc_is_subdir,
1838 .m_rename = mdc_rename,
1839 .m_setattr = mdc_setattr,
1840 .m_setxattr = mdc_setxattr,
1841 .m_getxattr = mdc_getxattr,
1843 .m_readpage = mdc_readpage,
1844 .m_unlink = mdc_unlink,
1845 .m_cancel_unused = mdc_cancel_unused,
1846 .m_init_ea_size = mdc_init_ea_size,
1847 .m_set_lock_data = mdc_set_lock_data,
1848 .m_lock_match = mdc_lock_match,
1849 .m_get_lustre_md = mdc_get_lustre_md,
1850 .m_free_lustre_md = mdc_free_lustre_md,
1851 .m_set_open_replay_data = mdc_set_open_replay_data,
1852 .m_clear_open_replay_data = mdc_clear_open_replay_data,
1853 .m_renew_capa = mdc_renew_capa,
1854 .m_unpack_capa = mdc_unpack_capa,
1855 .m_get_remote_perm = mdc_get_remote_perm,
1856 .m_intent_getattr_async = mdc_intent_getattr_async,
1857 .m_revalidate_lock = mdc_revalidate_lock
1860 int __init mdc_init(void)
1863 struct lprocfs_static_vars lvars = { 0 };
1864 lprocfs_mdc_init_vars(&lvars);
1866 request_module("lquota");
1867 quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1868 init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1870 rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1871 LUSTRE_MDC_NAME, NULL);
1872 if (rc && quota_interface)
1873 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1879 static void /*__exit*/ mdc_exit(void)
1881 if (quota_interface)
1882 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1884 class_unregister_type(LUSTRE_MDC_NAME);
1887 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1888 MODULE_DESCRIPTION("Lustre Metadata Client");
1889 MODULE_LICENSE("GPL");
1891 module_init(mdc_init);
1892 module_exit(mdc_exit);