1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2001-2003 Cluster File Systems, Inc.
6 * This file is part of the Lustre file system, http://www.lustre.org
7 * Lustre is a trademark of Cluster File Systems, Inc.
9 * You may have signed or agreed to another license before downloading
10 * this software. If so, you are bound by the terms and conditions
11 * of that agreement, and the following does not apply to you. See the
12 * LICENSE file included with this distribution for more information.
14 * If you did not agree to a different license, then this copy of Lustre
15 * is open source software; you can redistribute it and/or modify it
16 * under the terms of version 2 of the GNU General Public License as
17 * published by the Free Software Foundation.
19 * In either case, Lustre is distributed in the hope that it will be
20 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
21 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
22 * license text for more details.
26 # define EXPORT_SYMTAB
28 #define DEBUG_SUBSYSTEM S_MDC
31 # include <linux/module.h>
32 # include <linux/pagemap.h>
33 # include <linux/miscdevice.h>
34 # include <linux/init.h>
36 # include <liblustre.h>
39 #include <linux/lustre_acl.h>
40 #include <obd_class.h>
41 #include <lustre_dlm.h>
42 #include <lustre_fid.h>
43 #include <md_object.h>
44 #include <lprocfs_status.h>
45 #include <lustre_param.h>
46 #include "mdc_internal.h"
48 static quota_interface_t *quota_interface;
50 #define REQUEST_MINOR 244
52 static quota_interface_t *quota_interface;
53 extern quota_interface_t mdc_quota_interface;
55 static int mdc_cleanup(struct obd_device *obd);
57 static struct obd_capa *mdc_unpack_capa(struct ptlrpc_request *req,
58 const struct req_msg_field *field)
60 struct lustre_capa *capa;
63 /* swabbed already in mdc_enqueue */
64 capa = req_capsule_server_get(&req->rq_pill, field);
66 return ERR_PTR(-EPROTO);
68 oc = alloc_capa(CAPA_SITE_CLIENT);
70 CDEBUG(D_INFO, "alloc capa failed!\n");
71 return ERR_PTR(-ENOMEM);
78 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
79 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
80 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
81 struct obd_capa **pc, int level, int msg_flags)
83 struct ptlrpc_request *req;
84 struct mdt_body *body;
88 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
89 LUSTRE_MDS_VERSION, MDS_GETSTATUS);
93 mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
94 lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
95 req->rq_send_state = level;
97 ptlrpc_request_set_replen(req);
99 rc = ptlrpc_queue_wait(req);
103 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
105 GOTO(out, rc = -EPROTO);
107 if (body->valid & OBD_MD_FLMDSCAPA) {
110 oc = mdc_unpack_capa(req, &RMF_CAPA1);
112 GOTO(out, rc = PTR_ERR(oc));
116 *rootfid = body->fid1;
118 "root fid="DFID", last_committed="LPU64", last_xid="LPU64"\n",
120 lustre_msg_get_last_committed(req->rq_repmsg),
121 lustre_msg_get_last_xid(req->rq_repmsg));
124 ptlrpc_req_finished(req);
128 /* This should be mdc_get_info("rootfid") */
129 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
130 struct obd_capa **pc)
132 return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
137 * This function now is known to always saying that it will receive 4 buffers
138 * from server. Even for cases when acl_size and md_size is zero, RPC header
139 * willcontain 4 fields and RPC itself will contain zero size fields. This is
140 * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
141 * and thus zero, it shirinks it, making zero size. The same story about
142 * md_size. And this is course of problem when client waits for smaller number
143 * of fields. This issue will be fixed later when client gets awar of RPC
146 static int mdc_getattr_common(struct obd_export *exp,
147 struct ptlrpc_request *req)
149 struct req_capsule *pill = &req->rq_pill;
150 struct mdt_body *body;
155 /* Request message already built. */
156 rc = ptlrpc_queue_wait(req);
160 /* sanity check for the reply */
161 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
165 CDEBUG(D_NET, "mode: %o\n", body->mode);
167 if (body->eadatasize != 0) {
168 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
174 if (body->valid & OBD_MD_FLMODEASIZE) {
175 struct client_obd *cli = &exp->exp_obd->u.cli;
177 if (cli->cl_max_mds_easize < body->max_mdsize)
178 cli->cl_max_mds_easize = body->max_mdsize;
179 if (cli->cl_max_mds_cookiesize < body->max_cookiesize)
180 cli->cl_max_mds_cookiesize = body->max_cookiesize;
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 = current->cap_effective;
366 rec->sx_suppgid1 = suppgid;
367 rec->sx_suppgid1 = -1;
369 rec->sx_valid = valid;
370 rec->sx_size = output_size;
371 rec->sx_flags = flags;
373 mdc_pack_capa(req, &RMF_CAPA1, oc);
375 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
379 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
380 memcpy(tmp, xattr_name, xattr_namelen);
383 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
384 memcpy(tmp, input, input_size);
387 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
388 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
389 RCL_SERVER, output_size);
390 ptlrpc_request_set_replen(req);
393 if (opcode == MDS_REINT)
394 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
396 rc = ptlrpc_queue_wait(req);
398 if (opcode == MDS_REINT)
399 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
402 ptlrpc_req_finished(req);
408 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
409 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
410 const char *input, int input_size, int output_size,
411 int flags, __u32 suppgid, struct ptlrpc_request **request)
413 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
414 fid, oc, MDS_REINT, valid, xattr_name,
415 input, input_size, output_size, flags,
419 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
420 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
421 const char *input, int input_size, int output_size,
422 int flags, struct ptlrpc_request **request)
424 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
425 fid, oc, MDS_GETXATTR, valid, xattr_name,
426 input, input_size, output_size, flags,
430 #ifdef CONFIG_FS_POSIX_ACL
431 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
433 struct req_capsule *pill = &req->rq_pill;
434 struct mdt_body *body = md->body;
435 struct posix_acl *acl;
443 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
448 acl = posix_acl_from_xattr(buf, body->aclsize);
451 CERROR("convert xattr to acl: %d\n", rc);
455 rc = posix_acl_valid(acl);
457 CERROR("validate acl: %d\n", rc);
458 posix_acl_release(acl);
466 #define mdc_unpack_acl(req, md) 0
469 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
470 struct obd_export *dt_exp, struct obd_export *md_exp,
471 struct lustre_md *md)
473 struct req_capsule *pill = &req->rq_pill;
478 memset(md, 0, sizeof(*md));
480 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
481 LASSERT(md->body != NULL);
483 if (md->body->valid & OBD_MD_FLEASIZE) {
485 struct lov_mds_md *lmm;
487 if (!S_ISREG(md->body->mode)) {
488 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
489 "regular file, but is not\n");
490 GOTO(out, rc = -EPROTO);
493 if (md->body->eadatasize == 0) {
494 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
495 "but eadatasize 0\n");
496 GOTO(out, rc = -EPROTO);
498 lmmsize = md->body->eadatasize;
499 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
501 GOTO(out, rc = -EPROTO);
503 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
507 if (rc < sizeof(*md->lsm)) {
508 CDEBUG(D_INFO, "lsm size too small: "
509 "rc < sizeof (*md->lsm) (%d < %d)\n",
510 rc, (int)sizeof(*md->lsm));
511 GOTO(out, rc = -EPROTO);
514 } else if (md->body->valid & OBD_MD_FLDIREA) {
516 struct lov_mds_md *lmv;
518 if(!S_ISDIR(md->body->mode)) {
519 CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
520 "directory, but is not\n");
521 GOTO(out, rc = -EPROTO);
524 if (md->body->eadatasize == 0) {
525 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
526 "but eadatasize 0\n");
529 if (md->body->valid & OBD_MD_MEA) {
530 lmvsize = md->body->eadatasize;
531 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
534 GOTO(out, rc = -EPROTO);
536 rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
541 if (rc < sizeof(*md->mea)) {
542 CDEBUG(D_INFO, "size too small: "
543 "rc < sizeof(*md->mea) (%d < %d)\n",
544 rc, sizeof(*md->mea));
545 GOTO(out, rc = -EPROTO);
551 if (md->body->valid & OBD_MD_FLRMTPERM) {
552 /* remote permission */
553 LASSERT(client_is_remote(exp));
554 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
555 lustre_swab_mdt_remote_perm);
556 if (!md->remote_perm)
557 GOTO(out, rc = -EPROTO);
559 else if (md->body->valid & OBD_MD_FLACL) {
560 /* for ACL, it's possible that FLACL is set but aclsize is zero.
561 * only when aclsize != 0 there's an actual segment for ACL
564 if (md->body->aclsize) {
565 rc = mdc_unpack_acl(req, md);
568 #ifdef CONFIG_FS_POSIX_ACL
570 md->posix_acl = NULL;
574 if (md->body->valid & OBD_MD_FLMDSCAPA) {
575 struct obd_capa *oc = mdc_unpack_capa(req, &RMF_CAPA1);
578 GOTO(out, rc = PTR_ERR(oc));
582 if (md->body->valid & OBD_MD_FLOSSCAPA) {
583 struct obd_capa *oc = mdc_unpack_capa(req, &RMF_CAPA2);
586 GOTO(out, rc = PTR_ERR(oc));
594 free_capa(md->oss_capa);
596 free_capa(md->mds_capa);
597 #ifdef CONFIG_FS_POSIX_ACL
598 posix_acl_release(md->posix_acl);
601 obd_free_memmd(dt_exp, &md->lsm);
606 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
612 static void mdc_replay_open(struct ptlrpc_request *req)
614 struct md_open_data *mod = req->rq_cb_data;
615 struct ptlrpc_request *cur, *tmp;
616 struct obd_client_handle *och;
617 struct lustre_handle old;
618 struct mdt_body *body;
622 DEBUG_REQ(D_ERROR, req,
623 "Can't properly replay without open data.");
628 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
629 LASSERT(body != NULL);
633 struct lustre_handle *file_fh;
635 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
637 file_fh = &och->och_fh;
638 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
639 file_fh->cookie, body->handle.cookie);
641 *file_fh = body->handle;
643 list_for_each_entry_safe(cur, tmp, &mod->mod_replay_list, rq_mod_list) {
644 int opc = lustre_msg_get_opc(cur->rq_reqmsg);
645 struct mdt_epoch *epoch = NULL;
647 if (opc == MDS_CLOSE || opc == MDS_DONE_WRITING) {
648 epoch = req_capsule_client_get(&cur->rq_pill,
651 DEBUG_REQ(D_HA, cur, "updating %s body with new fh",
652 opc == MDS_CLOSE ? "CLOSE" : "DONE_WRITING");
653 } else if (opc == MDS_REINT) {
654 struct mdt_rec_setattr *rec;
656 /* Check this is REINT_SETATTR. */
657 rec = req_capsule_client_get(&cur->rq_pill,
659 LASSERT(rec && rec->sa_opcode == REINT_SETATTR);
661 epoch = req_capsule_client_get(&cur->rq_pill,
664 DEBUG_REQ(D_HA, cur, "updating REINT_SETATTR body "
669 LASSERT(!memcmp(&old, &epoch->handle,
671 epoch->handle = body->handle;
677 void mdc_commit_delayed(struct ptlrpc_request *req)
679 struct md_open_data *mod = req->rq_cb_data;
680 struct ptlrpc_request *cur, *tmp;
682 DEBUG_REQ(D_HA, req, "req committed");
687 req->rq_cb_data = NULL;
688 req->rq_commit_cb = NULL;
689 list_del_init(&req->rq_mod_list);
690 if (req->rq_sequence) {
691 list_for_each_entry_safe(cur, tmp, &mod->mod_replay_list,
693 LASSERT(cur != LP_POISON);
694 LASSERT(cur->rq_type != LI_POISON);
695 DEBUG_REQ(D_HA, cur, "req balanced");
696 LASSERT(cur->rq_transno != 0);
697 LASSERT(cur->rq_import == req->rq_import);
699 /* We no longer want to preserve this for transno-
700 * unconditional replay. */
701 spin_lock(&cur->rq_lock);
703 spin_unlock(&cur->rq_lock);
707 if (list_empty(&mod->mod_replay_list)) {
708 if (mod->mod_och != NULL)
709 mod->mod_och->och_mod = NULL;
715 int mdc_set_open_replay_data(struct obd_export *exp,
716 struct obd_client_handle *och,
717 struct ptlrpc_request *open_req)
719 struct md_open_data *mod;
720 struct mdt_rec_create *rec;
721 struct mdt_body *body;
722 struct obd_import *imp = open_req->rq_import;
725 if (!open_req->rq_replay)
728 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
729 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
730 LASSERT(rec != NULL);
731 /* Incoming message in my byte order (it's been swabbed). */
732 /* Outgoing messages always in my byte order. */
733 LASSERT(body != NULL);
735 /* Only if the import is replayable, we set replay_open data */
736 if (och && imp->imp_replayable) {
739 DEBUG_REQ(D_ERROR, open_req,
740 "Can't allocate md_open_data");
743 CFS_INIT_LIST_HEAD(&mod->mod_replay_list);
745 spin_lock(&open_req->rq_lock);
748 open_req->rq_cb_data = mod;
749 list_add_tail(&open_req->rq_mod_list, &mod->mod_replay_list);
750 open_req->rq_commit_cb = mdc_commit_delayed;
751 spin_unlock(&open_req->rq_lock);
754 rec->cr_fid2 = body->fid1;
755 rec->cr_ioepoch = body->ioepoch;
756 rec->cr_old_handle.cookie = body->handle.cookie;
757 open_req->rq_replay_cb = mdc_replay_open;
758 if (!fid_is_sane(&body->fid1)) {
759 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
764 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
768 int mdc_clear_open_replay_data(struct obd_export *exp,
769 struct obd_client_handle *och)
771 struct md_open_data *mod = och->och_mod;
775 * Don't free the structure now (it happens in mdc_commit_delayed(),
776 * after the last request is removed from its replay list),
777 * but make sure that replay doesn't poke at the och, which is about to
780 LASSERT(mod != LP_POISON);
788 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
789 struct md_open_data *mod, struct ptlrpc_request **request)
791 struct obd_device *obd = class_exp2obd(exp);
792 struct ptlrpc_request *req;
797 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_CLOSE);
801 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
803 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
805 ptlrpc_request_free(req);
809 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
810 * portal whose threads are not taking any DLM locks and are therefore
811 * always progressing */
812 /* XXX FIXME bug 249 */
813 req->rq_request_portal = MDS_READPAGE_PORTAL;
815 /* Ensure that this close's handle is fixed up during replay. */
816 if (likely(mod != NULL))
817 list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
819 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
821 mdc_close_pack(req, op_data);
823 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
824 obd->u.cli.cl_max_mds_easize);
825 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
826 obd->u.cli.cl_max_mds_cookiesize);
828 ptlrpc_request_set_replen(req);
830 req->rq_commit_cb = mdc_commit_delayed;
832 LASSERT(req->rq_cb_data == NULL);
833 req->rq_cb_data = mod;
835 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
836 rc = ptlrpc_queue_wait(req);
837 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
839 if (req->rq_repmsg == NULL) {
840 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
843 rc = req->rq_status ?: -EIO;
844 } else if (rc == 0 || rc == -EAGAIN) {
845 struct mdt_body *body;
847 rc = lustre_msg_get_status(req->rq_repmsg);
848 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
849 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
853 } else if (mod == NULL) {
854 if (req->rq_import->imp_replayable)
855 CERROR("Unexpected: can't find md_open_data,"
856 "but close succeeded with replayable imp"
857 "Please tell CFS.\n");
860 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
866 if (rc != 0 && rc != -EAGAIN && req && req->rq_commit_cb)
867 req->rq_commit_cb(req);
873 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
874 struct md_open_data *mod)
876 struct obd_device *obd = class_exp2obd(exp);
877 struct ptlrpc_request *req;
881 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
882 &RQF_MDS_DONE_WRITING);
886 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
887 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
889 ptlrpc_request_free(req);
893 /* XXX: add DONE_WRITING request to och -- when Size-on-MDS
894 * recovery will be ready. */
895 mdc_close_pack(req, op_data);
896 ptlrpc_request_set_replen(req);
898 req->rq_cb_data = mod;
899 req->rq_commit_cb = mdc_commit_delayed;
900 if (likely(mod != NULL))
901 list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
903 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
905 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
906 rc = ptlrpc_queue_wait(req);
907 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
909 /* Close the open replay sequence if an error occured or no SOM
910 * attribute update is needed. */
912 ptlrpc_close_replay_seq(req);
914 if (rc && rc != -EAGAIN && req->rq_commit_cb)
915 req->rq_commit_cb(req);
917 ptlrpc_req_finished(req);
921 #ifdef HAVE_SPLIT_SUPPORT
922 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
923 const struct page *page, int offset)
925 struct ptlrpc_request *req;
926 struct ptlrpc_bulk_desc *desc;
930 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
934 /* FIXME: capa doesn't support split yet */
935 mdc_set_capa_size(req, &RMF_CAPA1, NULL);
937 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
939 ptlrpc_request_free(req);
943 req->rq_request_portal = MDS_READPAGE_PORTAL;
945 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
947 GOTO(out, rc = -ENOMEM);
949 /* NB req now owns desc and will free it when it gets freed. */
950 ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
951 mdc_readdir_pack(req, 0, offset, fid, NULL);
953 ptlrpc_request_set_replen(req);
954 rc = ptlrpc_queue_wait(req);
957 ptlrpc_req_finished(req);
960 EXPORT_SYMBOL(mdc_sendpage);
963 int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
964 struct obd_capa *oc, __u64 offset, struct page *page,
965 struct ptlrpc_request **request)
967 struct ptlrpc_request *req;
968 struct ptlrpc_bulk_desc *desc;
973 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
977 mdc_set_capa_size(req, &RMF_CAPA1, oc);
979 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
981 ptlrpc_request_free(req);
985 /* XXX FIXME bug 249 */
986 req->rq_request_portal = MDS_READPAGE_PORTAL;
987 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
989 ptlrpc_request_free(req);
993 /* NB req now owns desc and will free it when it gets freed */
994 ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
995 mdc_readdir_pack(req, offset, CFS_PAGE_SIZE, fid, oc);
997 ptlrpc_request_set_replen(req);
998 rc = ptlrpc_queue_wait(req);
1000 ptlrpc_req_finished(req);
1004 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
1005 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1006 req->rq_bulk->bd_nob_transferred, CFS_PAGE_SIZE);
1007 ptlrpc_req_finished(req);
1015 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1016 void *karg, void *uarg)
1018 struct obd_device *obd = exp->exp_obd;
1019 struct obd_ioctl_data *data = karg;
1020 struct obd_import *imp = obd->u.cli.cl_import;
1021 struct llog_ctxt *ctxt;
1025 if (!try_module_get(THIS_MODULE)) {
1026 CERROR("Can't get module. Is it alive?");
1030 case OBD_IOC_CLIENT_RECOVER:
1031 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
1035 case IOC_OSC_SET_ACTIVE:
1036 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1038 case OBD_IOC_PARSE: {
1039 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1040 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
1041 llog_ctxt_put(ctxt);
1045 case OBD_IOC_LLOG_INFO:
1046 case OBD_IOC_LLOG_PRINT: {
1047 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1048 rc = llog_ioctl(ctxt, cmd, data);
1049 llog_ctxt_put(ctxt);
1053 case OBD_IOC_POLL_QUOTACHECK:
1054 rc = lquota_poll_check(quota_interface, exp,
1055 (struct if_quotacheck *)karg);
1058 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1059 GOTO(out, rc = -ENOTTY);
1062 module_put(THIS_MODULE);
1067 static int do_set_info_async(struct obd_export *exp,
1068 obd_count keylen, void *key,
1069 obd_count vallen, void *val,
1070 struct ptlrpc_request_set *set)
1072 struct obd_import *imp = class_exp2cliimp(exp);
1073 struct ptlrpc_request *req;
1078 if (vallen != sizeof(int))
1081 spin_lock(&imp->imp_lock);
1082 if (*((int *)val)) {
1083 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
1084 imp->imp_connect_data.ocd_connect_flags |= OBD_CONNECT_RDONLY;
1086 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
1087 imp->imp_connect_data.ocd_connect_flags &= ~OBD_CONNECT_RDONLY;
1089 spin_unlock(&imp->imp_lock);
1091 req = ptlrpc_request_alloc(imp, &RQF_MDS_SET_INFO);
1095 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
1096 RCL_CLIENT, keylen);
1097 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
1098 RCL_CLIENT, vallen);
1099 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SET_INFO);
1101 ptlrpc_request_free(req);
1105 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1106 memcpy(tmp, key, keylen);
1107 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
1108 memcpy(tmp, val, vallen);
1110 ptlrpc_request_set_replen(req);
1113 ptlrpc_set_add_req(set, req);
1114 ptlrpc_check_set(set);
1116 rc = ptlrpc_queue_wait(req);
1117 ptlrpc_req_finished(req);
1123 int mdc_set_info_async(struct obd_export *exp,
1124 obd_count keylen, void *key,
1125 obd_count vallen, void *val,
1126 struct ptlrpc_request_set *set)
1128 struct obd_import *imp = class_exp2cliimp(exp);
1132 if (KEY_IS(KEY_INIT_RECOV)) {
1133 if (vallen != sizeof(int))
1135 spin_lock(&imp->imp_lock);
1136 imp->imp_initial_recov = *(int *)val;
1137 spin_unlock(&imp->imp_lock);
1138 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
1139 exp->exp_obd->obd_name, imp->imp_initial_recov);
1142 /* Turn off initial_recov after we try all backup servers once */
1143 if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
1144 if (vallen != sizeof(int))
1146 spin_lock(&imp->imp_lock);
1147 imp->imp_initial_recov_bk = *(int *)val;
1148 if (imp->imp_initial_recov_bk)
1149 imp->imp_initial_recov = 1;
1150 spin_unlock(&imp->imp_lock);
1151 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
1152 exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
1155 if (KEY_IS(KEY_READ_ONLY)) {
1156 rc = do_set_info_async(exp, keylen, key, vallen, val, set);
1159 if (KEY_IS(KEY_FLUSH_CTX)) {
1160 sptlrpc_import_flush_my_ctx(imp);
1163 if (KEY_IS(KEY_MDS_CONN)) {
1164 struct obd_import *imp = class_exp2cliimp(exp);
1166 /* mds-mds import */
1167 spin_lock(&imp->imp_lock);
1168 imp->imp_server_timeout = 1;
1169 spin_unlock(&imp->imp_lock);
1170 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1171 CDEBUG(D_OTHER|D_WARNING, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1178 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
1179 __u32 *vallen, void *val)
1183 if (KEY_IS(KEY_MAX_EASIZE)) {
1184 int mdsize, *max_easize;
1186 if (*vallen != sizeof(int))
1188 mdsize = *(int*)val;
1189 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
1190 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
1192 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
1195 if (KEY_IS(KEY_CONN_DATA)) {
1196 struct obd_import *imp = class_exp2cliimp(exp);
1197 struct obd_connect_data *data = val;
1199 if (*vallen != sizeof(*data))
1202 *data = imp->imp_connect_data;
1209 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1210 __u64 max_age, __u32 flags)
1212 struct ptlrpc_request *req;
1213 struct obd_statfs *msfs;
1217 req = ptlrpc_request_alloc_pack(obd->u.cli.cl_import, &RQF_MDS_STATFS,
1218 LUSTRE_MDS_VERSION, MDS_STATFS);
1222 ptlrpc_request_set_replen(req);
1224 if (flags & OBD_STATFS_NODELAY) {
1225 /* procfs requests not want stay in wait for avoid deadlock */
1226 req->rq_no_resend = 1;
1227 req->rq_no_delay = 1;
1230 rc = ptlrpc_queue_wait(req);
1232 /* check connection error first */
1233 if (obd->u.cli.cl_import->imp_connect_error)
1234 rc = obd->u.cli.cl_import->imp_connect_error;
1238 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1240 GOTO(out, rc = -EPROTO);
1245 ptlrpc_req_finished(req);
1249 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
1250 struct obd_capa *oc, struct obd_client_handle *handle,
1253 struct ptlrpc_request *req;
1254 struct mdt_body *body;
1258 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
1262 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1264 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
1266 ptlrpc_request_free(req);
1270 mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
1272 ptlrpc_request_set_replen(req);
1274 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1275 rc = ptlrpc_queue_wait(req);
1276 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1278 CERROR("Pin failed: %d\n", rc);
1282 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1284 GOTO(err_out, rc = -EPROTO);
1286 handle->och_fh = body->handle;
1287 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1289 OBD_ALLOC_PTR(handle->och_mod);
1290 if (handle->och_mod == NULL) {
1291 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
1292 GOTO(err_out, rc = -ENOMEM);
1294 /* will be dropped by unpin */
1295 CFS_INIT_LIST_HEAD(&handle->och_mod->mod_replay_list);
1296 list_add_tail(&req->rq_mod_list, &handle->och_mod->mod_replay_list);
1301 ptlrpc_req_finished(req);
1305 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
1308 struct ptlrpc_request *req;
1309 struct mdt_body *body;
1313 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
1314 LUSTRE_MDS_VERSION, MDS_UNPIN);
1318 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1319 body->handle = handle->och_fh;
1322 ptlrpc_request_set_replen(req);
1324 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1325 rc = ptlrpc_queue_wait(req);
1326 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1329 CERROR("Unpin failed: %d\n", rc);
1331 ptlrpc_req_finished(req);
1333 LASSERT(!list_empty(&handle->och_mod->mod_replay_list));
1334 req = list_entry(handle->och_mod->mod_replay_list.next,
1335 typeof(*req), rq_mod_list);
1336 list_del_init(&req->rq_mod_list);
1337 ptlrpc_req_finished(req);
1338 LASSERT(list_empty(&handle->och_mod->mod_replay_list));
1340 OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1344 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
1345 struct obd_capa *oc, struct ptlrpc_request **request)
1347 struct ptlrpc_request *req;
1352 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
1356 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1358 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
1360 ptlrpc_request_free(req);
1364 mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
1366 ptlrpc_request_set_replen(req);
1368 rc = ptlrpc_queue_wait(req);
1370 ptlrpc_req_finished(req);
1376 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1377 enum obd_import_event event)
1381 LASSERT(imp->imp_obd == obd);
1384 case IMP_EVENT_DISCON: {
1386 /* XXX Pass event up to OBDs stack. used only for FLD now */
1387 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
1391 case IMP_EVENT_INACTIVE: {
1392 struct client_obd *cli = &obd->u.cli;
1394 * Flush current sequence to make client obtain new one
1395 * from server in case of disconnect/reconnect.
1396 * If range is already empty then no need to flush it.
1398 if (cli->cl_seq != NULL &&
1399 !range_is_exhausted(&cli->cl_seq->lcs_space)) {
1400 seq_client_flush(cli->cl_seq);
1403 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1406 case IMP_EVENT_INVALIDATE: {
1407 struct ldlm_namespace *ns = obd->obd_namespace;
1409 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1413 case IMP_EVENT_ACTIVE: {
1414 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1418 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
1422 CERROR("Unknown import event %x\n", event);
1428 static int mdc_fid_init(struct obd_export *exp)
1430 struct client_obd *cli = &exp->exp_obd->u.cli;
1435 OBD_ALLOC_PTR(cli->cl_seq);
1436 if (cli->cl_seq == NULL)
1439 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
1441 GOTO(out_free_seq, rc = -ENOMEM);
1443 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
1444 exp->exp_obd->obd_name);
1446 /* Init client side sequence-manager */
1447 rc = seq_client_init(cli->cl_seq, exp,
1448 LUSTRE_SEQ_METADATA,
1450 OBD_FREE(prefix, MAX_OBD_NAME + 5);
1452 GOTO(out_free_seq, rc);
1456 OBD_FREE_PTR(cli->cl_seq);
1461 static int mdc_fid_fini(struct obd_export *exp)
1463 struct client_obd *cli = &exp->exp_obd->u.cli;
1466 if (cli->cl_seq != NULL) {
1467 seq_client_fini(cli->cl_seq);
1468 OBD_FREE_PTR(cli->cl_seq);
1475 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1476 struct md_op_data *op_data)
1478 struct client_obd *cli = &exp->exp_obd->u.cli;
1479 struct lu_client_seq *seq = cli->cl_seq;
1481 RETURN(seq_client_alloc_fid(seq, fid));
1484 /* XXX This method is used only to clear current fid seq
1485 * once fld/mds insert failed */
1486 static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1488 struct client_obd *cli = &exp->exp_obd->u.cli;
1490 seq_client_flush(cli->cl_seq);
1494 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
1496 struct client_obd *cli = &obd->u.cli;
1497 struct lprocfs_static_vars lvars = { 0 };
1501 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1502 if (!cli->cl_rpc_lock)
1504 mdc_init_rpc_lock(cli->cl_rpc_lock);
1508 OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1509 if (!cli->cl_setattr_lock)
1510 GOTO(err_rpc_lock, rc = -ENOMEM);
1511 mdc_init_rpc_lock(cli->cl_setattr_lock);
1513 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1514 if (!cli->cl_close_lock)
1515 GOTO(err_setattr_lock, rc = -ENOMEM);
1516 mdc_init_rpc_lock(cli->cl_close_lock);
1518 rc = client_obd_setup(obd, cfg);
1520 GOTO(err_close_lock, rc);
1521 lprocfs_mdc_init_vars(&lvars);
1522 lprocfs_obd_setup(obd, lvars.obd_vars);
1523 sptlrpc_lprocfs_cliobd_attach(obd);
1524 ptlrpc_lprocfs_register_obd(obd);
1526 rc = obd_llog_init(obd, OBD_LLOG_GROUP, obd, 0, NULL, NULL);
1529 CERROR("failed to setup llogging subsystems\n");
1535 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1537 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1539 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1544 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
1545 * us to make MDS RPCs with large enough reply buffers to hold the
1546 * maximum-sized (= maximum striped) EA and cookie without having to
1547 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1548 int mdc_init_ea_size(struct obd_export *exp, int easize,
1549 int def_easize, int cookiesize)
1551 struct obd_device *obd = exp->exp_obd;
1552 struct client_obd *cli = &obd->u.cli;
1555 if (cli->cl_max_mds_easize < easize)
1556 cli->cl_max_mds_easize = easize;
1558 if (cli->cl_default_mds_easize < def_easize)
1559 cli->cl_default_mds_easize = def_easize;
1561 if (cli->cl_max_mds_cookiesize < cookiesize)
1562 cli->cl_max_mds_cookiesize = cookiesize;
1567 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1573 case OBD_CLEANUP_EARLY:
1574 case OBD_CLEANUP_EXPORTS:
1575 /* If we set up but never connected, the
1576 client import will not have been cleaned. */
1577 if (obd->u.cli.cl_import) {
1578 struct obd_import *imp;
1579 imp = obd->u.cli.cl_import;
1580 CERROR("client import never connected\n");
1581 ptlrpc_invalidate_import(imp);
1582 ptlrpc_free_rq_pool(imp->imp_rq_pool);
1583 class_destroy_import(imp);
1584 obd->u.cli.cl_import = NULL;
1587 case OBD_CLEANUP_SELF_EXP:
1588 rc = obd_llog_finish(obd, 0);
1590 CERROR("failed to cleanup llogging subsystems\n");
1591 case OBD_CLEANUP_OBD:
1597 static int mdc_cleanup(struct obd_device *obd)
1599 struct client_obd *cli = &obd->u.cli;
1601 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1602 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1603 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1605 ptlrpc_lprocfs_unregister_obd(obd);
1606 lprocfs_obd_cleanup(obd);
1609 return client_obd_cleanup(obd);
1613 static int mdc_llog_init(struct obd_device *obd, int group,
1614 struct obd_device *tgt, int count,
1615 struct llog_catid *logid, struct obd_uuid *uuid)
1617 struct llog_ctxt *ctxt;
1618 struct obd_llog_group *olg = &obd->obd_olg;
1622 LASSERT(group == OBD_LLOG_GROUP);
1623 LASSERT(olg->olg_group == group);
1625 rc = llog_setup(obd, olg, LLOG_LOVEA_REPL_CTXT, tgt, 0,
1626 NULL, &llog_client_ops);
1628 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1629 llog_initiator_connect(ctxt);
1630 llog_ctxt_put(ctxt);
1636 static int mdc_llog_finish(struct obd_device *obd, int count)
1638 struct llog_ctxt *ctxt;
1642 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1644 rc = llog_cleanup(ctxt);
1649 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1651 struct lustre_cfg *lcfg = buf;
1652 struct lprocfs_static_vars lvars = { 0 };
1655 lprocfs_mdc_init_vars(&lvars);
1657 switch (lcfg->lcfg_command) {
1658 case LCFG_SPTLRPC_CONF:
1659 rc = sptlrpc_cliobd_process_config(obd, lcfg);
1662 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
1670 /* get remote permission for current user on fid */
1671 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
1672 struct obd_capa *oc, __u32 suppgid,
1673 struct ptlrpc_request **request)
1675 struct ptlrpc_request *req;
1679 LASSERT(client_is_remote(exp));
1682 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1686 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1688 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1690 ptlrpc_request_free(req);
1694 mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
1696 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
1697 sizeof(struct mdt_remote_perm));
1699 ptlrpc_request_set_replen(req);
1701 rc = ptlrpc_queue_wait(req);
1703 ptlrpc_req_finished(req);
1709 static int mdc_interpret_renew_capa(struct ptlrpc_request *req, void *unused,
1712 struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
1713 renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
1714 struct mdt_body *body = NULL;
1715 struct lustre_capa *capa;
1719 GOTO(out, capa = ERR_PTR(status));
1721 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1723 GOTO(out, capa = ERR_PTR(-EFAULT));
1725 if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
1726 GOTO(out, capa = ERR_PTR(-ENOENT));
1728 capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
1730 GOTO(out, capa = ERR_PTR(-EFAULT));
1737 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
1740 struct ptlrpc_request *req;
1743 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
1744 LUSTRE_MDS_VERSION, MDS_GETATTR);
1748 /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
1749 * capa to renew is oss capa.
1751 mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
1752 ptlrpc_request_set_replen(req);
1754 req->rq_async_args.pointer_arg[0] = oc;
1755 req->rq_async_args.pointer_arg[1] = cb;
1756 req->rq_interpret_reply = mdc_interpret_renew_capa;
1757 ptlrpcd_add_req(req);
1761 static int mdc_connect(const struct lu_env *env,
1762 struct lustre_handle *dlm_handle,
1763 struct obd_device *obd, struct obd_uuid *cluuid,
1764 struct obd_connect_data *data,
1767 struct obd_import *imp = obd->u.cli.cl_import;
1769 /* mds-mds import features */
1770 if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
1771 spin_lock(&imp->imp_lock);
1772 imp->imp_server_timeout = 1;
1773 spin_unlock(&imp->imp_lock);
1774 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1775 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
1779 return client_connect_import(env, dlm_handle, obd, cluuid, data, NULL);
1782 struct obd_ops mdc_obd_ops = {
1783 .o_owner = THIS_MODULE,
1784 .o_setup = mdc_setup,
1785 .o_precleanup = mdc_precleanup,
1786 .o_cleanup = mdc_cleanup,
1787 .o_add_conn = client_import_add_conn,
1788 .o_del_conn = client_import_del_conn,
1789 .o_connect = mdc_connect,
1790 .o_disconnect = client_disconnect_export,
1791 .o_iocontrol = mdc_iocontrol,
1792 .o_set_info_async = mdc_set_info_async,
1793 .o_statfs = mdc_statfs,
1795 .o_unpin = mdc_unpin,
1796 .o_fid_init = mdc_fid_init,
1797 .o_fid_fini = mdc_fid_fini,
1798 .o_fid_alloc = mdc_fid_alloc,
1799 .o_fid_delete = mdc_fid_delete,
1800 .o_import_event = mdc_import_event,
1801 .o_llog_init = mdc_llog_init,
1802 .o_llog_finish = mdc_llog_finish,
1803 .o_get_info = mdc_get_info,
1804 .o_process_config = mdc_process_config,
1807 struct md_ops mdc_md_ops = {
1808 .m_getstatus = mdc_getstatus,
1809 .m_change_cbdata = mdc_change_cbdata,
1810 .m_close = mdc_close,
1811 .m_create = mdc_create,
1812 .m_done_writing = mdc_done_writing,
1813 .m_enqueue = mdc_enqueue,
1814 .m_getattr = mdc_getattr,
1815 .m_getattr_name = mdc_getattr_name,
1816 .m_intent_lock = mdc_intent_lock,
1818 .m_is_subdir = mdc_is_subdir,
1819 .m_rename = mdc_rename,
1820 .m_setattr = mdc_setattr,
1821 .m_setxattr = mdc_setxattr,
1822 .m_getxattr = mdc_getxattr,
1824 .m_readpage = mdc_readpage,
1825 .m_unlink = mdc_unlink,
1826 .m_cancel_unused = mdc_cancel_unused,
1827 .m_init_ea_size = mdc_init_ea_size,
1828 .m_set_lock_data = mdc_set_lock_data,
1829 .m_lock_match = mdc_lock_match,
1830 .m_get_lustre_md = mdc_get_lustre_md,
1831 .m_free_lustre_md = mdc_free_lustre_md,
1832 .m_set_open_replay_data = mdc_set_open_replay_data,
1833 .m_clear_open_replay_data = mdc_clear_open_replay_data,
1834 .m_get_remote_perm = mdc_get_remote_perm,
1835 .m_renew_capa = mdc_renew_capa
1838 extern quota_interface_t mdc_quota_interface;
1840 int __init mdc_init(void)
1843 struct lprocfs_static_vars lvars = { 0 };
1844 lprocfs_mdc_init_vars(&lvars);
1846 request_module("lquota");
1847 quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1848 init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1850 rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1851 LUSTRE_MDC_NAME, NULL);
1852 if (rc && quota_interface)
1853 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1859 static void /*__exit*/ mdc_exit(void)
1861 if (quota_interface)
1862 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1864 class_unregister_type(LUSTRE_MDC_NAME);
1867 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1868 MODULE_DESCRIPTION("Lustre Metadata Client");
1869 MODULE_LICENSE("GPL");
1871 module_init(mdc_init);
1872 module_exit(mdc_exit);