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 mdc_update_max_ea_from_body(exp, body);
180 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
186 if (body->valid & OBD_MD_FLRMTPERM) {
187 struct mdt_remote_perm *perm;
189 LASSERT(client_is_remote(exp));
190 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
191 lustre_swab_mdt_remote_perm);
196 if (body->valid & OBD_MD_FLMDSCAPA) {
197 struct lustre_capa *capa;
198 capa = req_capsule_server_get(pill, &RMF_CAPA1);
206 int mdc_getattr(struct obd_export *exp, const struct lu_fid *fid,
207 struct obd_capa *oc, obd_valid valid, int ea_size,
208 struct ptlrpc_request **request)
210 struct ptlrpc_request *req;
215 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
219 mdc_set_capa_size(req, &RMF_CAPA1, oc);
221 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
223 ptlrpc_request_free(req);
227 /* MDS_BFLAG_EXT_FLAGS: request "new" flags(bug 9486) */
228 mdc_pack_body(req, fid, oc, valid, ea_size, -1, MDS_BFLAG_EXT_FLAGS);
230 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, ea_size);
231 if (valid & OBD_MD_FLRMTPERM) {
232 LASSERT(client_is_remote(exp));
233 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
234 sizeof(struct mdt_remote_perm));
236 ptlrpc_request_set_replen(req);
238 rc = mdc_getattr_common(exp, req);
240 ptlrpc_req_finished(req);
246 int mdc_getattr_name(struct obd_export *exp, const struct lu_fid *fid,
247 struct obd_capa *oc, const char *filename, int namelen,
248 obd_valid valid, int ea_size, __u32 suppgid,
249 struct ptlrpc_request **request)
251 struct ptlrpc_request *req;
256 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
257 &RQF_MDS_GETATTR_NAME);
261 mdc_set_capa_size(req, &RMF_CAPA1, oc);
262 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, namelen);
264 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
266 ptlrpc_request_free(req);
270 /* MDS_BFLAG_EXT_FLAGS: request "new" flags(bug 9486) */
271 mdc_pack_body(req, fid, oc, valid, ea_size, suppgid,
272 MDS_BFLAG_EXT_FLAGS);
275 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
276 LASSERT(strnlen(filename, namelen) == namelen - 1);
277 memcpy(name, filename, namelen);
280 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, ea_size);
281 ptlrpc_request_set_replen(req);
283 rc = mdc_getattr_common(exp, req);
285 ptlrpc_req_finished(req);
291 static int mdc_is_subdir(struct obd_export *exp,
292 const struct lu_fid *pfid,
293 const struct lu_fid *cfid,
294 struct ptlrpc_request **request)
296 struct ptlrpc_request *req;
302 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
303 &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
308 mdc_is_subdir_pack(req, pfid, cfid, 0);
309 ptlrpc_request_set_replen(req);
311 rc = ptlrpc_queue_wait(req);
312 if (rc && rc != -EREMOTE)
313 ptlrpc_req_finished(req);
319 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
320 const struct lu_fid *fid,
321 struct obd_capa *oc, int opcode, obd_valid valid,
322 const char *xattr_name, const char *input,
323 int input_size, int output_size, int flags,
324 __u32 suppgid, struct ptlrpc_request **request)
326 struct ptlrpc_request *req;
327 int xattr_namelen = 0;
333 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
337 mdc_set_capa_size(req, &RMF_CAPA1, oc);
339 xattr_namelen = strlen(xattr_name) + 1;
340 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
345 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
349 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
351 ptlrpc_request_free(req);
355 if (opcode == MDS_REINT) {
356 struct mdt_rec_setxattr *rec;
358 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
359 sizeof(struct mdt_rec_reint));
360 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
361 rec->sx_opcode = REINT_SETXATTR;
363 * cfs_curproc_fs{u,g}id() should replace
364 * current->fs{u,g}id for portability.
366 rec->sx_fsuid = current->fsuid;
367 rec->sx_fsgid = current->fsgid;
368 rec->sx_cap = cfs_curproc_cap_pack();
369 rec->sx_suppgid1 = suppgid;
370 rec->sx_suppgid2 = -1;
372 rec->sx_valid = valid | OBD_MD_FLCTIME;
373 rec->sx_time = cfs_time_current_sec();
374 rec->sx_size = output_size;
375 rec->sx_flags = flags;
377 mdc_pack_capa(req, &RMF_CAPA1, oc);
379 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
383 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
384 memcpy(tmp, xattr_name, xattr_namelen);
387 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
388 memcpy(tmp, input, input_size);
391 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
392 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
393 RCL_SERVER, output_size);
394 ptlrpc_request_set_replen(req);
397 if (opcode == MDS_REINT)
398 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
400 rc = ptlrpc_queue_wait(req);
402 if (opcode == MDS_REINT)
403 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
406 ptlrpc_req_finished(req);
412 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
413 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
414 const char *input, int input_size, int output_size,
415 int flags, __u32 suppgid, struct ptlrpc_request **request)
417 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
418 fid, oc, MDS_REINT, valid, xattr_name,
419 input, input_size, output_size, flags,
423 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
424 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
425 const char *input, int input_size, int output_size,
426 int flags, struct ptlrpc_request **request)
428 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
429 fid, oc, MDS_GETXATTR, valid, xattr_name,
430 input, input_size, output_size, flags,
434 #ifdef CONFIG_FS_POSIX_ACL
435 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
437 struct req_capsule *pill = &req->rq_pill;
438 struct mdt_body *body = md->body;
439 struct posix_acl *acl;
447 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
452 acl = posix_acl_from_xattr(buf, body->aclsize);
455 CERROR("convert xattr to acl: %d\n", rc);
459 rc = posix_acl_valid(acl);
461 CERROR("validate acl: %d\n", rc);
462 posix_acl_release(acl);
470 #define mdc_unpack_acl(req, md) 0
473 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
474 struct obd_export *dt_exp, struct obd_export *md_exp,
475 struct lustre_md *md)
477 struct req_capsule *pill = &req->rq_pill;
482 memset(md, 0, sizeof(*md));
484 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
485 LASSERT(md->body != NULL);
487 if (md->body->valid & OBD_MD_FLEASIZE) {
489 struct lov_mds_md *lmm;
491 if (!S_ISREG(md->body->mode)) {
492 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
493 "regular file, but is not\n");
494 GOTO(out, rc = -EPROTO);
497 if (md->body->eadatasize == 0) {
498 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
499 "but eadatasize 0\n");
500 GOTO(out, rc = -EPROTO);
502 lmmsize = md->body->eadatasize;
503 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
505 GOTO(out, rc = -EPROTO);
507 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
511 if (rc < sizeof(*md->lsm)) {
512 CDEBUG(D_INFO, "lsm size too small: "
513 "rc < sizeof (*md->lsm) (%d < %d)\n",
514 rc, (int)sizeof(*md->lsm));
515 GOTO(out, rc = -EPROTO);
518 } else if (md->body->valid & OBD_MD_FLDIREA) {
520 struct lov_mds_md *lmv;
522 if(!S_ISDIR(md->body->mode)) {
523 CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
524 "directory, but is not\n");
525 GOTO(out, rc = -EPROTO);
528 if (md->body->eadatasize == 0) {
529 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
530 "but eadatasize 0\n");
533 if (md->body->valid & OBD_MD_MEA) {
534 lmvsize = md->body->eadatasize;
535 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
538 GOTO(out, rc = -EPROTO);
540 rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
545 if (rc < sizeof(*md->mea)) {
546 CDEBUG(D_INFO, "size too small: "
547 "rc < sizeof(*md->mea) (%d < %d)\n",
548 rc, (int)sizeof(*md->mea));
549 GOTO(out, rc = -EPROTO);
555 if (md->body->valid & OBD_MD_FLRMTPERM) {
556 /* remote permission */
557 LASSERT(client_is_remote(exp));
558 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
559 lustre_swab_mdt_remote_perm);
560 if (!md->remote_perm)
561 GOTO(out, rc = -EPROTO);
563 else if (md->body->valid & OBD_MD_FLACL) {
564 /* for ACL, it's possible that FLACL is set but aclsize is zero.
565 * only when aclsize != 0 there's an actual segment for ACL
568 if (md->body->aclsize) {
569 rc = mdc_unpack_acl(req, md);
572 #ifdef CONFIG_FS_POSIX_ACL
574 md->posix_acl = NULL;
578 if (md->body->valid & OBD_MD_FLMDSCAPA) {
579 struct obd_capa *oc = NULL;
581 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
587 if (md->body->valid & OBD_MD_FLOSSCAPA) {
588 struct obd_capa *oc = NULL;
590 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
600 capa_put(md->oss_capa);
604 capa_put(md->mds_capa);
607 #ifdef CONFIG_FS_POSIX_ACL
608 posix_acl_release(md->posix_acl);
611 obd_free_memmd(dt_exp, &md->lsm);
616 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
623 * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
626 void mdc_replay_open(struct ptlrpc_request *req)
628 struct md_open_data *mod = req->rq_cb_data;
629 struct ptlrpc_request *close_req;
630 struct obd_client_handle *och;
631 struct lustre_handle old;
632 struct mdt_body *body;
636 DEBUG_REQ(D_ERROR, req,
637 "Can't properly replay without open data.");
642 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
643 LASSERT(body != NULL);
647 struct lustre_handle *file_fh;
649 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
651 file_fh = &och->och_fh;
652 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
653 file_fh->cookie, body->handle.cookie);
655 *file_fh = body->handle;
657 close_req = mod->mod_close_req;
658 if (close_req != NULL) {
659 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
660 struct mdt_epoch *epoch;
662 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
663 epoch = req_capsule_client_get(&close_req->rq_pill,
668 LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
669 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
670 epoch->handle = body->handle;
675 void mdc_commit_open(struct ptlrpc_request *req)
677 struct md_open_data *mod = req->rq_cb_data;
682 * No need to touch md_open_data::mod_och, it holds a reference on
683 * \var mod and will zero references to each other, \var mod will be
684 * freed after that when md_open_data::mod_och will put the reference.
688 * Do not let open request to disappear as it still may be needed
689 * for close rpc to happen (it may happen on evict only, otherwise
690 * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
691 * called), just mark this rpc as committed to distinguish these 2
692 * cases, see mdc_close() for details. The open request reference will
693 * be put along with freeing \var mod.
695 ptlrpc_request_addref(req);
696 spin_lock(&req->rq_lock);
697 req->rq_committed = 1;
698 spin_unlock(&req->rq_lock);
699 req->rq_cb_data = NULL;
703 int mdc_set_open_replay_data(struct obd_export *exp,
704 struct obd_client_handle *och,
705 struct ptlrpc_request *open_req)
707 struct md_open_data *mod;
708 struct mdt_rec_create *rec;
709 struct mdt_body *body;
710 struct obd_import *imp = open_req->rq_import;
713 if (!open_req->rq_replay)
716 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
717 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
718 LASSERT(rec != NULL);
719 /* Incoming message in my byte order (it's been swabbed). */
720 /* Outgoing messages always in my byte order. */
721 LASSERT(body != NULL);
723 /* Only if the import is replayable, we set replay_open data */
724 if (och && imp->imp_replayable) {
725 mod = obd_mod_alloc();
727 DEBUG_REQ(D_ERROR, open_req,
728 "Can't allocate md_open_data");
733 * Take a reference on \var mod, to be freed on mdc_close().
734 * It protects \var mod from being freed on eviction (commit
735 * callback is called despite rq_replay flag).
736 * Another reference for \var och.
741 spin_lock(&open_req->rq_lock);
744 mod->mod_open_req = open_req;
745 open_req->rq_cb_data = mod;
746 open_req->rq_commit_cb = mdc_commit_open;
747 spin_unlock(&open_req->rq_lock);
750 rec->cr_fid2 = body->fid1;
751 rec->cr_ioepoch = body->ioepoch;
752 rec->cr_old_handle.cookie = body->handle.cookie;
753 open_req->rq_replay_cb = mdc_replay_open;
754 if (!fid_is_sane(&body->fid1)) {
755 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
760 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
764 int mdc_clear_open_replay_data(struct obd_export *exp,
765 struct obd_client_handle *och)
767 struct md_open_data *mod = och->och_mod;
770 LASSERT(mod != LP_POISON && mod != NULL);
779 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
780 struct md_open_data *mod, struct ptlrpc_request **request)
782 struct obd_device *obd = class_exp2obd(exp);
783 struct ptlrpc_request *req;
788 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_CLOSE);
792 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
794 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
796 ptlrpc_request_free(req);
800 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
801 * portal whose threads are not taking any DLM locks and are therefore
802 * always progressing */
803 req->rq_request_portal = MDS_READPAGE_PORTAL;
804 ptlrpc_at_set_req_timeout(req);
806 /* Ensure that this close's handle is fixed up during replay. */
807 if (likely(mod != NULL)) {
808 LASSERTF(mod->mod_open_req != NULL &&
809 mod->mod_open_req->rq_type != LI_POISON,
810 "POISONED open %p!\n", mod->mod_open_req);
812 mod->mod_close_req = req;
814 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
815 /* We no longer want to preserve this open for replay even
816 * though the open was committed. b=3632, b=3633 */
817 spin_lock(&mod->mod_open_req->rq_lock);
818 mod->mod_open_req->rq_replay = 0;
819 spin_unlock(&mod->mod_open_req->rq_lock);
821 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
824 mdc_close_pack(req, op_data);
826 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
827 obd->u.cli.cl_max_mds_easize);
828 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
829 obd->u.cli.cl_max_mds_cookiesize);
831 ptlrpc_request_set_replen(req);
833 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
834 rc = ptlrpc_queue_wait(req);
835 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
837 if (req->rq_repmsg == NULL) {
838 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
841 rc = req->rq_status ?: -EIO;
842 } else if (rc == 0 || rc == -EAGAIN) {
843 struct mdt_body *body;
845 rc = lustre_msg_get_status(req->rq_repmsg);
846 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
847 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
852 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
855 } else if (rc == -ESTALE) {
857 * it can be allowed error after 3633 if open was committed and
858 * server failed before close was sent. Let's check if mod
859 * exists and return no error in that case
862 LASSERT(mod->mod_open_req != NULL);
863 if (mod->mod_open_req->rq_committed)
870 mod->mod_close_req = NULL;
871 /* Since now, mod is accessed through open_req only,
872 * thus close req does not keep a reference on mod anymore. */
879 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
880 struct md_open_data *mod)
882 struct obd_device *obd = class_exp2obd(exp);
883 struct ptlrpc_request *req;
887 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
888 &RQF_MDS_DONE_WRITING);
892 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
893 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
895 ptlrpc_request_free(req);
900 LASSERTF(mod->mod_open_req != NULL &&
901 mod->mod_open_req->rq_type != LI_POISON,
902 "POISONED setattr %p!\n", mod->mod_open_req);
904 mod->mod_close_req = req;
905 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
906 /* We no longer want to preserve this setattr for replay even
907 * though the open was committed. b=3632, b=3633 */
908 spin_lock(&mod->mod_open_req->rq_lock);
909 mod->mod_open_req->rq_replay = 0;
910 spin_unlock(&mod->mod_open_req->rq_lock);
913 mdc_close_pack(req, op_data);
914 ptlrpc_request_set_replen(req);
916 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
917 rc = ptlrpc_queue_wait(req);
918 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
922 * it can be allowed error after 3633 if open or setattr were
923 * committed and server failed before close was sent.
924 * Let's check if mod exists and return no error in that case
927 LASSERT(mod->mod_open_req != NULL);
928 if (mod->mod_open_req->rq_committed)
935 mod->mod_close_req = NULL;
936 /* Since now, mod is accessed through setattr req only,
937 * thus DW req does not keep a reference on mod anymore. */
940 ptlrpc_req_finished(req);
944 #ifdef HAVE_SPLIT_SUPPORT
945 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
946 const struct page *page, int offset)
948 struct ptlrpc_request *req;
949 struct ptlrpc_bulk_desc *desc;
953 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
957 /* FIXME: capa doesn't support split yet */
958 mdc_set_capa_size(req, &RMF_CAPA1, NULL);
960 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
962 ptlrpc_request_free(req);
966 req->rq_request_portal = MDS_READPAGE_PORTAL;
967 ptlrpc_at_set_req_timeout(req);
969 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
971 GOTO(out, rc = -ENOMEM);
973 /* NB req now owns desc and will free it when it gets freed. */
974 ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
975 mdc_readdir_pack(req, 0, offset, fid, NULL);
977 ptlrpc_request_set_replen(req);
978 rc = ptlrpc_queue_wait(req);
982 rc = sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk);
984 ptlrpc_req_finished(req);
987 EXPORT_SYMBOL(mdc_sendpage);
990 int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
991 struct obd_capa *oc, __u64 offset, struct page *page,
992 struct ptlrpc_request **request)
994 struct ptlrpc_request *req;
995 struct ptlrpc_bulk_desc *desc;
1000 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1004 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1006 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1008 ptlrpc_request_free(req);
1012 req->rq_request_portal = MDS_READPAGE_PORTAL;
1013 ptlrpc_at_set_req_timeout(req);
1015 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
1017 ptlrpc_request_free(req);
1021 /* NB req now owns desc and will free it when it gets freed */
1022 ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
1023 mdc_readdir_pack(req, offset, CFS_PAGE_SIZE, fid, oc);
1025 ptlrpc_request_set_replen(req);
1026 rc = ptlrpc_queue_wait(req);
1028 ptlrpc_req_finished(req);
1032 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1033 req->rq_bulk->bd_nob_transferred);
1035 ptlrpc_req_finished(req);
1039 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
1040 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1041 req->rq_bulk->bd_nob_transferred, CFS_PAGE_SIZE);
1042 ptlrpc_req_finished(req);
1050 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1052 __u32 keylen, vallen;
1056 if (gf->gf_pathlen > PATH_MAX)
1057 RETURN(-ENAMETOOLONG);
1058 if (gf->gf_pathlen < 2)
1061 /* Key is KEY_FID2PATH + getinfo_fid2path description */
1062 keylen = size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1063 OBD_ALLOC(key, keylen);
1066 memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1067 memcpy(key + size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1069 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1070 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1072 if (!fid_is_sane(&gf->gf_fid))
1073 GOTO(out, rc = -EINVAL);
1075 /* Val is struct getinfo_fid2path result plus path */
1076 vallen = sizeof(*gf) + gf->gf_pathlen;
1078 rc = obd_get_info(exp, keylen, key, &vallen, gf, NULL);
1082 if (vallen <= sizeof(*gf))
1083 GOTO(out, rc = -EPROTO);
1084 else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1085 GOTO(out, rc = -EOVERFLOW);
1087 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1088 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1091 OBD_FREE(key, keylen);
1095 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1096 void *karg, void *uarg)
1098 struct obd_device *obd = exp->exp_obd;
1099 struct obd_ioctl_data *data = karg;
1100 struct obd_import *imp = obd->u.cli.cl_import;
1101 struct llog_ctxt *ctxt;
1105 if (!try_module_get(THIS_MODULE)) {
1106 CERROR("Can't get module. Is it alive?");
1110 case OBD_IOC_CHANGELOG_CLEAR: {
1111 struct ioc_changelog_clear *icc = karg;
1112 struct changelog_setinfo cs =
1113 {icc->icc_recno, icc->icc_id};
1114 rc = obd_set_info_async(exp, strlen(KEY_CHANGELOG_CLEAR),
1115 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1119 case OBD_IOC_FID2PATH: {
1120 rc = mdc_ioc_fid2path(exp, karg);
1123 case OBD_IOC_CLIENT_RECOVER:
1124 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
1128 case IOC_OSC_SET_ACTIVE:
1129 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1131 case OBD_IOC_PARSE: {
1132 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1133 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
1134 llog_ctxt_put(ctxt);
1138 case OBD_IOC_LLOG_INFO:
1139 case OBD_IOC_LLOG_PRINT: {
1140 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1141 rc = llog_ioctl(ctxt, cmd, data);
1142 llog_ctxt_put(ctxt);
1146 case OBD_IOC_POLL_QUOTACHECK:
1147 rc = lquota_poll_check(quota_interface, exp,
1148 (struct if_quotacheck *)karg);
1150 case OBD_IOC_PING_TARGET:
1151 rc = ptlrpc_obd_ping(obd);
1154 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1155 GOTO(out, rc = -ENOTTY);
1158 module_put(THIS_MODULE);
1163 int mdc_get_info_rpc(struct obd_export *exp,
1164 obd_count keylen, void *key,
1165 int vallen, void *val)
1167 struct obd_import *imp = class_exp2cliimp(exp);
1168 struct ptlrpc_request *req;
1173 req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1177 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1178 RCL_CLIENT, keylen);
1179 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1180 RCL_CLIENT, sizeof(__u32));
1182 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1184 ptlrpc_request_free(req);
1188 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
1189 memcpy(tmp, key, keylen);
1190 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
1191 memcpy(tmp, &vallen, sizeof(__u32));
1193 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1194 RCL_SERVER, vallen);
1195 ptlrpc_request_set_replen(req);
1197 rc = ptlrpc_queue_wait(req);
1199 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
1200 memcpy(val, tmp, vallen);
1201 if (ptlrpc_rep_need_swab(req)) {
1202 if (KEY_IS(KEY_FID2PATH)) {
1203 lustre_swab_fid2path(val);
1207 ptlrpc_req_finished(req);
1212 static void lustre_swab_hai(struct hsm_action_item *h)
1214 __swab32s(&h->hai_len);
1215 __swab32s(&h->hai_action);
1216 lustre_swab_lu_fid(&h->hai_fid);
1217 __swab64s(&h->hai_cookie);
1218 __swab64s(&h->hai_extent_start);
1219 __swab64s(&h->hai_extent_end);
1220 __swab64s(&h->hai_gid);
1223 static void lustre_swab_hal(struct hsm_action_list *h)
1225 struct hsm_action_item *hai;
1228 __swab32s(&h->hal_version);
1229 __swab32s(&h->hal_count);
1230 __swab32s(&h->hal_archive_num);
1232 for (i = 0; i < h->hal_count; i++) {
1233 lustre_swab_hai(hai);
1234 hai = hai_next(hai);
1239 * Send a message to any listening copytools, nonblocking
1240 * @param val LNL message (lnl_hdr + hsm_action_list)
1241 * @param len total length of message
1243 static int mdc_hsm_copytool_send(int len, void *val)
1245 struct lnl_hdr *lh = (struct lnl_hdr *)val;
1246 struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
1250 if (len < sizeof(*lh) + sizeof(*hal)) {
1251 CERROR("Short HSM message %d < %d\n", len,
1252 (int) (sizeof(*lh) + sizeof(*hal)));
1255 if (lh->lnl_magic == __swab16(LNL_MAGIC)) {
1256 lustre_swab_lnlh(lh);
1257 lustre_swab_hal(hal);
1258 } else if (lh->lnl_magic != LNL_MAGIC) {
1259 CERROR("Bad magic %x!=%x\n", lh->lnl_magic, LNL_MAGIC);
1263 CDEBUG(D_IOCTL, " Received message mg=%x t=%d m=%d l=%d actions=%d\n",
1264 lh->lnl_magic, lh->lnl_transport, lh->lnl_msgtype,
1265 lh->lnl_msglen, hal->hal_count);
1267 /* Broadcast to HSM listeners */
1268 rc = libcfs_klnl_msg_put(0, LNL_GRP_HSM, lh);
1273 int mdc_set_info_async(struct obd_export *exp,
1274 obd_count keylen, void *key,
1275 obd_count vallen, void *val,
1276 struct ptlrpc_request_set *set)
1278 struct obd_import *imp = class_exp2cliimp(exp);
1282 if (KEY_IS(KEY_INIT_RECOV)) {
1283 if (vallen != sizeof(int))
1285 spin_lock(&imp->imp_lock);
1286 imp->imp_initial_recov = *(int *)val;
1287 spin_unlock(&imp->imp_lock);
1288 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
1289 exp->exp_obd->obd_name, imp->imp_initial_recov);
1292 /* Turn off initial_recov after we try all backup servers once */
1293 if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
1294 if (vallen != sizeof(int))
1296 spin_lock(&imp->imp_lock);
1297 imp->imp_initial_recov_bk = *(int *)val;
1298 if (imp->imp_initial_recov_bk)
1299 imp->imp_initial_recov = 1;
1300 spin_unlock(&imp->imp_lock);
1301 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
1302 exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
1305 if (KEY_IS(KEY_READ_ONLY)) {
1306 if (vallen != sizeof(int))
1309 spin_lock(&imp->imp_lock);
1310 if (*((int *)val)) {
1311 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
1312 imp->imp_connect_data.ocd_connect_flags |= OBD_CONNECT_RDONLY;
1314 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
1315 imp->imp_connect_data.ocd_connect_flags &= ~OBD_CONNECT_RDONLY;
1317 spin_unlock(&imp->imp_lock);
1319 rc = target_set_info_rpc(imp, MDS_SET_INFO,
1320 keylen, key, vallen, val, set);
1323 if (KEY_IS(KEY_SPTLRPC_CONF)) {
1324 sptlrpc_conf_client_adapt(exp->exp_obd);
1327 if (KEY_IS(KEY_FLUSH_CTX)) {
1328 sptlrpc_import_flush_my_ctx(imp);
1331 if (KEY_IS(KEY_MDS_CONN)) {
1332 /* mds-mds import */
1333 spin_lock(&imp->imp_lock);
1334 imp->imp_server_timeout = 1;
1335 spin_unlock(&imp->imp_lock);
1336 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1337 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1340 if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1341 rc = target_set_info_rpc(imp, MDS_SET_INFO,
1342 keylen, key, vallen, val, set);
1345 if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
1346 rc = mdc_hsm_copytool_send(vallen, val);
1353 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
1354 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1358 if (KEY_IS(KEY_MAX_EASIZE)) {
1359 int mdsize, *max_easize;
1361 if (*vallen != sizeof(int))
1363 mdsize = *(int*)val;
1364 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
1365 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
1367 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
1370 if (KEY_IS(KEY_CONN_DATA)) {
1371 struct obd_import *imp = class_exp2cliimp(exp);
1372 struct obd_connect_data *data = val;
1374 if (*vallen != sizeof(*data))
1377 *data = imp->imp_connect_data;
1381 rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
1386 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1387 __u64 max_age, __u32 flags)
1389 struct ptlrpc_request *req;
1390 struct obd_statfs *msfs;
1391 struct obd_import *imp = NULL;
1396 /*Since the request might also come from lprocfs, so we need
1397 *sync this with client_disconnect_export Bug15684*/
1398 down_read(&obd->u.cli.cl_sem);
1399 if (obd->u.cli.cl_import)
1400 imp = class_import_get(obd->u.cli.cl_import);
1401 up_read(&obd->u.cli.cl_sem);
1405 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1406 LUSTRE_MDS_VERSION, MDS_STATFS);
1408 GOTO(output, rc = -ENOMEM);
1410 ptlrpc_request_set_replen(req);
1412 if (flags & OBD_STATFS_NODELAY) {
1413 /* procfs requests not want stay in wait for avoid deadlock */
1414 req->rq_no_resend = 1;
1415 req->rq_no_delay = 1;
1418 rc = ptlrpc_queue_wait(req);
1420 /* check connection error first */
1421 if (imp->imp_connect_error)
1422 rc = imp->imp_connect_error;
1426 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1428 GOTO(out, rc = -EPROTO);
1433 ptlrpc_req_finished(req);
1435 class_import_put(imp);
1439 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
1440 struct obd_capa *oc, struct obd_client_handle *handle,
1443 struct ptlrpc_request *req;
1444 struct mdt_body *body;
1448 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
1452 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1454 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
1456 ptlrpc_request_free(req);
1460 mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
1462 ptlrpc_request_set_replen(req);
1464 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1465 rc = ptlrpc_queue_wait(req);
1466 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1468 CERROR("Pin failed: %d\n", rc);
1472 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1474 GOTO(err_out, rc = -EPROTO);
1476 handle->och_fh = body->handle;
1477 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1479 handle->och_mod = obd_mod_alloc();
1480 if (handle->och_mod == NULL) {
1481 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
1482 GOTO(err_out, rc = -ENOMEM);
1484 handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
1489 ptlrpc_req_finished(req);
1493 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
1496 struct ptlrpc_request *req;
1497 struct mdt_body *body;
1501 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
1502 LUSTRE_MDS_VERSION, MDS_UNPIN);
1506 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1507 body->handle = handle->och_fh;
1510 ptlrpc_request_set_replen(req);
1512 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1513 rc = ptlrpc_queue_wait(req);
1514 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1517 CERROR("Unpin failed: %d\n", rc);
1519 ptlrpc_req_finished(req);
1520 ptlrpc_req_finished(handle->och_mod->mod_open_req);
1522 obd_mod_put(handle->och_mod);
1526 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
1527 struct obd_capa *oc, struct ptlrpc_request **request)
1529 struct ptlrpc_request *req;
1534 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
1538 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1540 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
1542 ptlrpc_request_free(req);
1546 mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
1548 ptlrpc_request_set_replen(req);
1550 rc = ptlrpc_queue_wait(req);
1552 ptlrpc_req_finished(req);
1558 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1559 enum obd_import_event event)
1563 LASSERT(imp->imp_obd == obd);
1566 case IMP_EVENT_DISCON: {
1568 /* XXX Pass event up to OBDs stack. used only for FLD now */
1569 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
1573 case IMP_EVENT_INACTIVE: {
1574 struct client_obd *cli = &obd->u.cli;
1576 * Flush current sequence to make client obtain new one
1577 * from server in case of disconnect/reconnect.
1578 * If range is already empty then no need to flush it.
1580 if (cli->cl_seq != NULL &&
1581 !range_is_exhausted(&cli->cl_seq->lcs_space)) {
1582 seq_client_flush(cli->cl_seq);
1585 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1588 case IMP_EVENT_INVALIDATE: {
1589 struct ldlm_namespace *ns = obd->obd_namespace;
1591 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1595 case IMP_EVENT_ACTIVE: {
1596 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1600 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
1604 CERROR("Unknown import event %x\n", event);
1610 static int mdc_fid_init(struct obd_export *exp)
1612 struct client_obd *cli = &exp->exp_obd->u.cli;
1617 OBD_ALLOC_PTR(cli->cl_seq);
1618 if (cli->cl_seq == NULL)
1621 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
1623 GOTO(out_free_seq, rc = -ENOMEM);
1625 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
1626 exp->exp_obd->obd_name);
1628 /* Init client side sequence-manager */
1629 rc = seq_client_init(cli->cl_seq, exp,
1630 LUSTRE_SEQ_METADATA,
1632 OBD_FREE(prefix, MAX_OBD_NAME + 5);
1634 GOTO(out_free_seq, rc);
1638 OBD_FREE_PTR(cli->cl_seq);
1643 static int mdc_fid_fini(struct obd_export *exp)
1645 struct client_obd *cli = &exp->exp_obd->u.cli;
1648 if (cli->cl_seq != NULL) {
1649 seq_client_fini(cli->cl_seq);
1650 OBD_FREE_PTR(cli->cl_seq);
1657 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1658 struct md_op_data *op_data)
1660 struct client_obd *cli = &exp->exp_obd->u.cli;
1661 struct lu_client_seq *seq = cli->cl_seq;
1663 RETURN(seq_client_alloc_fid(seq, fid));
1666 /* XXX This method is used only to clear current fid seq
1667 * once fld/mds insert failed */
1668 static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1670 struct client_obd *cli = &exp->exp_obd->u.cli;
1672 seq_client_flush(cli->cl_seq);
1676 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
1677 struct client_obd *cli = &exp->exp_obd->u.cli;
1678 return &cli->cl_target_uuid;
1681 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
1683 struct client_obd *cli = &obd->u.cli;
1684 struct lprocfs_static_vars lvars = { 0 };
1688 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1689 if (!cli->cl_rpc_lock)
1691 mdc_init_rpc_lock(cli->cl_rpc_lock);
1695 OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1696 if (!cli->cl_setattr_lock)
1697 GOTO(err_rpc_lock, rc = -ENOMEM);
1698 mdc_init_rpc_lock(cli->cl_setattr_lock);
1700 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1701 if (!cli->cl_close_lock)
1702 GOTO(err_setattr_lock, rc = -ENOMEM);
1703 mdc_init_rpc_lock(cli->cl_close_lock);
1705 rc = client_obd_setup(obd, cfg);
1707 GOTO(err_close_lock, rc);
1708 lprocfs_mdc_init_vars(&lvars);
1709 lprocfs_obd_setup(obd, lvars.obd_vars);
1710 sptlrpc_lprocfs_cliobd_attach(obd);
1711 ptlrpc_lprocfs_register_obd(obd);
1713 rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
1716 CERROR("failed to setup llogging subsystems\n");
1720 libcfs_klnl_start(LNL_TRANSPORT_HSM);
1721 libcfs_klnl_start(LNL_TRANSPORT_CHANGELOG);
1726 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1728 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1730 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1735 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
1736 * us to make MDS RPCs with large enough reply buffers to hold the
1737 * maximum-sized (= maximum striped) EA and cookie without having to
1738 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1739 static int mdc_init_ea_size(struct obd_export *exp, int easize,
1740 int def_easize, int cookiesize)
1742 struct obd_device *obd = exp->exp_obd;
1743 struct client_obd *cli = &obd->u.cli;
1746 if (cli->cl_max_mds_easize < easize)
1747 cli->cl_max_mds_easize = easize;
1749 if (cli->cl_default_mds_easize < def_easize)
1750 cli->cl_default_mds_easize = def_easize;
1752 if (cli->cl_max_mds_cookiesize < cookiesize)
1753 cli->cl_max_mds_cookiesize = cookiesize;
1758 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1764 case OBD_CLEANUP_EARLY:
1765 case OBD_CLEANUP_EXPORTS:
1766 /* If we set up but never connected, the
1767 client import will not have been cleaned. */
1768 if (obd->u.cli.cl_import) {
1769 struct obd_import *imp;
1770 down_write(&obd->u.cli.cl_sem);
1771 imp = obd->u.cli.cl_import;
1772 CERROR("client import never connected\n");
1773 ptlrpc_invalidate_import(imp);
1774 class_destroy_import(imp);
1775 up_write(&obd->u.cli.cl_sem);
1776 obd->u.cli.cl_import = NULL;
1778 rc = obd_llog_finish(obd, 0);
1780 CERROR("failed to cleanup llogging subsystems\n");
1786 static int mdc_cleanup(struct obd_device *obd)
1788 struct client_obd *cli = &obd->u.cli;
1790 libcfs_klnl_stop(LNL_TRANSPORT_HSM, LNL_GRP_HSM);
1791 libcfs_klnl_stop(LNL_TRANSPORT_CHANGELOG, 0);
1793 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1794 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1795 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1797 ptlrpc_lprocfs_unregister_obd(obd);
1798 lprocfs_obd_cleanup(obd);
1801 return client_obd_cleanup(obd);
1805 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
1806 struct obd_device *tgt, int *index)
1808 struct llog_ctxt *ctxt;
1812 LASSERT(olg == &obd->obd_olg);
1814 rc = llog_setup(obd, olg, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1819 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1820 llog_initiator_connect(ctxt);
1821 llog_ctxt_put(ctxt);
1823 rc = llog_setup(obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt, 0, NULL,
1826 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
1827 llog_initiator_connect(ctxt);
1828 llog_ctxt_put(ctxt);
1834 static int mdc_llog_finish(struct obd_device *obd, int count)
1836 struct llog_ctxt *ctxt;
1840 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1842 rc = llog_cleanup(ctxt);
1844 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
1846 rc = llog_cleanup(ctxt);
1851 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1853 struct lustre_cfg *lcfg = buf;
1854 struct lprocfs_static_vars lvars = { 0 };
1857 lprocfs_mdc_init_vars(&lvars);
1858 switch (lcfg->lcfg_command) {
1860 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
1870 /* get remote permission for current user on fid */
1871 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
1872 struct obd_capa *oc, __u32 suppgid,
1873 struct ptlrpc_request **request)
1875 struct ptlrpc_request *req;
1879 LASSERT(client_is_remote(exp));
1882 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1886 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1888 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1890 ptlrpc_request_free(req);
1894 mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
1896 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
1897 sizeof(struct mdt_remote_perm));
1899 ptlrpc_request_set_replen(req);
1901 rc = ptlrpc_queue_wait(req);
1903 ptlrpc_req_finished(req);
1909 static int mdc_interpret_renew_capa(const struct lu_env *env,
1910 struct ptlrpc_request *req, void *unused,
1913 struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
1914 renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
1915 struct mdt_body *body = NULL;
1916 struct lustre_capa *capa;
1920 GOTO(out, capa = ERR_PTR(status));
1922 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1924 GOTO(out, capa = ERR_PTR(-EFAULT));
1926 if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
1927 GOTO(out, capa = ERR_PTR(-ENOENT));
1929 capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
1931 GOTO(out, capa = ERR_PTR(-EFAULT));
1938 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
1941 struct ptlrpc_request *req;
1944 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
1945 LUSTRE_MDS_VERSION, MDS_GETATTR);
1949 /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
1950 * capa to renew is oss capa.
1952 mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
1953 ptlrpc_request_set_replen(req);
1955 req->rq_async_args.pointer_arg[0] = oc;
1956 req->rq_async_args.pointer_arg[1] = cb;
1957 req->rq_interpret_reply = mdc_interpret_renew_capa;
1958 ptlrpcd_add_req(req, PSCOPE_OTHER);
1962 static int mdc_connect(const struct lu_env *env,
1963 struct obd_export **exp,
1964 struct obd_device *obd, struct obd_uuid *cluuid,
1965 struct obd_connect_data *data,
1968 struct obd_import *imp = obd->u.cli.cl_import;
1970 /* mds-mds import features */
1971 if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
1972 spin_lock(&imp->imp_lock);
1973 imp->imp_server_timeout = 1;
1974 spin_unlock(&imp->imp_lock);
1975 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1976 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
1980 return client_connect_import(env, exp, obd, cluuid, data, NULL);
1983 struct obd_ops mdc_obd_ops = {
1984 .o_owner = THIS_MODULE,
1985 .o_setup = mdc_setup,
1986 .o_precleanup = mdc_precleanup,
1987 .o_cleanup = mdc_cleanup,
1988 .o_add_conn = client_import_add_conn,
1989 .o_del_conn = client_import_del_conn,
1990 .o_connect = mdc_connect,
1991 .o_disconnect = client_disconnect_export,
1992 .o_iocontrol = mdc_iocontrol,
1993 .o_set_info_async = mdc_set_info_async,
1994 .o_statfs = mdc_statfs,
1996 .o_unpin = mdc_unpin,
1997 .o_fid_init = mdc_fid_init,
1998 .o_fid_fini = mdc_fid_fini,
1999 .o_fid_alloc = mdc_fid_alloc,
2000 .o_fid_delete = mdc_fid_delete,
2001 .o_import_event = mdc_import_event,
2002 .o_llog_init = mdc_llog_init,
2003 .o_llog_finish = mdc_llog_finish,
2004 .o_get_info = mdc_get_info,
2005 .o_process_config = mdc_process_config,
2006 .o_get_uuid = mdc_get_uuid,
2009 struct md_ops mdc_md_ops = {
2010 .m_getstatus = mdc_getstatus,
2011 .m_change_cbdata = mdc_change_cbdata,
2012 .m_close = mdc_close,
2013 .m_create = mdc_create,
2014 .m_done_writing = mdc_done_writing,
2015 .m_enqueue = mdc_enqueue,
2016 .m_getattr = mdc_getattr,
2017 .m_getattr_name = mdc_getattr_name,
2018 .m_intent_lock = mdc_intent_lock,
2020 .m_is_subdir = mdc_is_subdir,
2021 .m_rename = mdc_rename,
2022 .m_setattr = mdc_setattr,
2023 .m_setxattr = mdc_setxattr,
2024 .m_getxattr = mdc_getxattr,
2026 .m_readpage = mdc_readpage,
2027 .m_unlink = mdc_unlink,
2028 .m_cancel_unused = mdc_cancel_unused,
2029 .m_init_ea_size = mdc_init_ea_size,
2030 .m_set_lock_data = mdc_set_lock_data,
2031 .m_lock_match = mdc_lock_match,
2032 .m_get_lustre_md = mdc_get_lustre_md,
2033 .m_free_lustre_md = mdc_free_lustre_md,
2034 .m_set_open_replay_data = mdc_set_open_replay_data,
2035 .m_clear_open_replay_data = mdc_clear_open_replay_data,
2036 .m_renew_capa = mdc_renew_capa,
2037 .m_unpack_capa = mdc_unpack_capa,
2038 .m_get_remote_perm = mdc_get_remote_perm,
2039 .m_intent_getattr_async = mdc_intent_getattr_async,
2040 .m_revalidate_lock = mdc_revalidate_lock
2043 int __init mdc_init(void)
2046 struct lprocfs_static_vars lvars = { 0 };
2047 lprocfs_mdc_init_vars(&lvars);
2049 request_module("lquota");
2050 quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
2051 init_obd_quota_ops(quota_interface, &mdc_obd_ops);
2053 rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
2054 LUSTRE_MDC_NAME, NULL);
2055 if (rc && quota_interface)
2056 PORTAL_SYMBOL_PUT(mdc_quota_interface);
2062 static void /*__exit*/ mdc_exit(void)
2064 if (quota_interface)
2065 PORTAL_SYMBOL_PUT(mdc_quota_interface);
2067 class_unregister_type(LUSTRE_MDC_NAME);
2070 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2071 MODULE_DESCRIPTION("Lustre Metadata Client");
2072 MODULE_LICENSE("GPL");
2074 module_init(mdc_init);
2075 module_exit(mdc_exit);