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 quota_interface_t *quota_interface;
62 #define REQUEST_MINOR 244
64 extern quota_interface_t mdc_quota_interface;
66 static int mdc_cleanup(struct obd_device *obd);
68 static struct obd_capa *mdc_unpack_capa(struct ptlrpc_request *req,
69 const struct req_msg_field *field)
71 struct lustre_capa *capa;
74 /* swabbed already in mdc_enqueue */
75 capa = req_capsule_server_get(&req->rq_pill, field);
77 return ERR_PTR(-EPROTO);
79 oc = alloc_capa(CAPA_SITE_CLIENT);
81 CDEBUG(D_INFO, "alloc capa failed!\n");
82 return ERR_PTR(-ENOMEM);
89 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
90 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
91 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
92 struct obd_capa **pc, int level, int msg_flags)
94 struct ptlrpc_request *req;
95 struct mdt_body *body;
99 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
100 LUSTRE_MDS_VERSION, MDS_GETSTATUS);
104 mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
105 lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
106 req->rq_send_state = level;
108 ptlrpc_request_set_replen(req);
110 rc = ptlrpc_queue_wait(req);
114 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
116 GOTO(out, rc = -EPROTO);
118 if (body->valid & OBD_MD_FLMDSCAPA) {
121 oc = mdc_unpack_capa(req, &RMF_CAPA1);
123 GOTO(out, rc = PTR_ERR(oc));
127 *rootfid = body->fid1;
129 "root fid="DFID", last_committed="LPU64", last_xid="LPU64"\n",
131 lustre_msg_get_last_committed(req->rq_repmsg),
132 lustre_msg_get_last_xid(req->rq_repmsg));
135 ptlrpc_req_finished(req);
139 /* This should be mdc_get_info("rootfid") */
140 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
141 struct obd_capa **pc)
143 return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
148 * This function now is known to always saying that it will receive 4 buffers
149 * from server. Even for cases when acl_size and md_size is zero, RPC header
150 * will contain 4 fields and RPC itself will contain zero size fields. This is
151 * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
152 * and thus zero, it shrinks it, making zero size. The same story about
153 * md_size. And this is course of problem when client waits for smaller number
154 * of fields. This issue will be fixed later when client gets aware of RPC
157 static int mdc_getattr_common(struct obd_export *exp,
158 struct ptlrpc_request *req)
160 struct req_capsule *pill = &req->rq_pill;
161 struct mdt_body *body;
166 /* Request message already built. */
167 rc = ptlrpc_queue_wait(req);
171 /* sanity check for the reply */
172 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
176 CDEBUG(D_NET, "mode: %o\n", body->mode);
178 if (body->eadatasize != 0) {
179 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
185 if (body->valid & OBD_MD_FLMODEASIZE) {
186 struct client_obd *cli = &exp->exp_obd->u.cli;
188 if (cli->cl_max_mds_easize < body->max_mdsize)
189 cli->cl_max_mds_easize = body->max_mdsize;
190 if (cli->cl_max_mds_cookiesize < body->max_cookiesize)
191 cli->cl_max_mds_cookiesize = body->max_cookiesize;
194 if (body->valid & OBD_MD_FLRMTPERM) {
195 struct mdt_remote_perm *perm;
197 LASSERT(client_is_remote(exp));
198 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
199 lustre_swab_mdt_remote_perm);
204 if (body->valid & OBD_MD_FLMDSCAPA) {
205 struct lustre_capa *capa;
206 capa = req_capsule_server_get(pill, &RMF_CAPA1);
214 int mdc_getattr(struct obd_export *exp, const struct lu_fid *fid,
215 struct obd_capa *oc, obd_valid valid, int ea_size,
216 struct ptlrpc_request **request)
218 struct ptlrpc_request *req;
223 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
227 mdc_set_capa_size(req, &RMF_CAPA1, oc);
229 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
231 ptlrpc_request_free(req);
235 /* MDS_BFLAG_EXT_FLAGS: request "new" flags(bug 9486) */
236 mdc_pack_body(req, fid, oc, valid, ea_size, -1, MDS_BFLAG_EXT_FLAGS);
238 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, ea_size);
239 if (valid & OBD_MD_FLRMTPERM) {
240 LASSERT(client_is_remote(exp));
241 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
242 sizeof(struct mdt_remote_perm));
244 ptlrpc_request_set_replen(req);
246 rc = mdc_getattr_common(exp, req);
248 ptlrpc_req_finished(req);
254 int mdc_getattr_name(struct obd_export *exp, const struct lu_fid *fid,
255 struct obd_capa *oc, const char *filename, int namelen,
256 obd_valid valid, int ea_size, __u32 suppgid,
257 struct ptlrpc_request **request)
259 struct ptlrpc_request *req;
264 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
265 &RQF_MDS_GETATTR_NAME);
269 mdc_set_capa_size(req, &RMF_CAPA1, oc);
270 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, namelen);
272 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
274 ptlrpc_request_free(req);
278 /* MDS_BFLAG_EXT_FLAGS: request "new" flags(bug 9486) */
279 mdc_pack_body(req, fid, oc, valid, ea_size, suppgid,
280 MDS_BFLAG_EXT_FLAGS);
283 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
284 LASSERT(strnlen(filename, namelen) == namelen - 1);
285 memcpy(name, filename, namelen);
288 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, ea_size);
289 ptlrpc_request_set_replen(req);
291 rc = mdc_getattr_common(exp, req);
293 ptlrpc_req_finished(req);
299 static int mdc_is_subdir(struct obd_export *exp,
300 const struct lu_fid *pfid,
301 const struct lu_fid *cfid,
302 struct ptlrpc_request **request)
304 struct ptlrpc_request *req;
310 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
311 &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
316 mdc_is_subdir_pack(req, pfid, cfid, 0);
317 ptlrpc_request_set_replen(req);
319 rc = ptlrpc_queue_wait(req);
320 if (rc && rc != -EREMOTE)
321 ptlrpc_req_finished(req);
327 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
328 const struct lu_fid *fid,
329 struct obd_capa *oc, int opcode, obd_valid valid,
330 const char *xattr_name, const char *input,
331 int input_size, int output_size, int flags,
332 __u32 suppgid, struct ptlrpc_request **request)
334 struct ptlrpc_request *req;
335 int xattr_namelen = 0;
341 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
345 mdc_set_capa_size(req, &RMF_CAPA1, oc);
347 xattr_namelen = strlen(xattr_name) + 1;
348 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
353 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
357 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
359 ptlrpc_request_free(req);
363 if (opcode == MDS_REINT) {
364 struct mdt_rec_setxattr *rec;
366 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
367 sizeof(struct mdt_rec_reint));
368 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
369 rec->sx_opcode = REINT_SETXATTR;
371 * cfs_curproc_fs{u,g}id() should replace
372 * current->fs{u,g}id for portability.
374 rec->sx_fsuid = current->fsuid;
375 rec->sx_fsgid = current->fsgid;
376 rec->sx_cap = cfs_curproc_cap_pack();
377 rec->sx_suppgid1 = suppgid;
378 rec->sx_suppgid2 = -1;
380 rec->sx_valid = valid | OBD_MD_FLCTIME;
381 rec->sx_time = cfs_time_current_sec();
382 rec->sx_size = output_size;
383 rec->sx_flags = flags;
385 mdc_pack_capa(req, &RMF_CAPA1, oc);
387 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
391 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
392 memcpy(tmp, xattr_name, xattr_namelen);
395 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
396 memcpy(tmp, input, input_size);
399 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
400 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
401 RCL_SERVER, output_size);
402 ptlrpc_request_set_replen(req);
405 if (opcode == MDS_REINT)
406 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
408 rc = ptlrpc_queue_wait(req);
410 if (opcode == MDS_REINT)
411 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
414 ptlrpc_req_finished(req);
420 int mdc_setxattr(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, __u32 suppgid, struct ptlrpc_request **request)
425 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
426 fid, oc, MDS_REINT, valid, xattr_name,
427 input, input_size, output_size, flags,
431 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
432 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
433 const char *input, int input_size, int output_size,
434 int flags, struct ptlrpc_request **request)
436 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
437 fid, oc, MDS_GETXATTR, valid, xattr_name,
438 input, input_size, output_size, flags,
442 #ifdef CONFIG_FS_POSIX_ACL
443 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
445 struct req_capsule *pill = &req->rq_pill;
446 struct mdt_body *body = md->body;
447 struct posix_acl *acl;
455 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
460 acl = posix_acl_from_xattr(buf, body->aclsize);
463 CERROR("convert xattr to acl: %d\n", rc);
467 rc = posix_acl_valid(acl);
469 CERROR("validate acl: %d\n", rc);
470 posix_acl_release(acl);
478 #define mdc_unpack_acl(req, md) 0
481 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
482 struct obd_export *dt_exp, struct obd_export *md_exp,
483 struct lustre_md *md)
485 struct req_capsule *pill = &req->rq_pill;
490 memset(md, 0, sizeof(*md));
492 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
493 LASSERT(md->body != NULL);
495 if (md->body->valid & OBD_MD_FLEASIZE) {
497 struct lov_mds_md *lmm;
499 if (!S_ISREG(md->body->mode)) {
500 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
501 "regular file, but is not\n");
502 GOTO(out, rc = -EPROTO);
505 if (md->body->eadatasize == 0) {
506 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
507 "but eadatasize 0\n");
508 GOTO(out, rc = -EPROTO);
510 lmmsize = md->body->eadatasize;
511 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
513 GOTO(out, rc = -EPROTO);
515 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
519 if (rc < sizeof(*md->lsm)) {
520 CDEBUG(D_INFO, "lsm size too small: "
521 "rc < sizeof (*md->lsm) (%d < %d)\n",
522 rc, (int)sizeof(*md->lsm));
523 GOTO(out, rc = -EPROTO);
526 } else if (md->body->valid & OBD_MD_FLDIREA) {
528 struct lov_mds_md *lmv;
530 if(!S_ISDIR(md->body->mode)) {
531 CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
532 "directory, but is not\n");
533 GOTO(out, rc = -EPROTO);
536 if (md->body->eadatasize == 0) {
537 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
538 "but eadatasize 0\n");
541 if (md->body->valid & OBD_MD_MEA) {
542 lmvsize = md->body->eadatasize;
543 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
546 GOTO(out, rc = -EPROTO);
548 rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
553 if (rc < sizeof(*md->mea)) {
554 CDEBUG(D_INFO, "size too small: "
555 "rc < sizeof(*md->mea) (%d < %d)\n",
556 rc, (int)sizeof(*md->mea));
557 GOTO(out, rc = -EPROTO);
563 if (md->body->valid & OBD_MD_FLRMTPERM) {
564 /* remote permission */
565 LASSERT(client_is_remote(exp));
566 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
567 lustre_swab_mdt_remote_perm);
568 if (!md->remote_perm)
569 GOTO(out, rc = -EPROTO);
571 else if (md->body->valid & OBD_MD_FLACL) {
572 /* for ACL, it's possible that FLACL is set but aclsize is zero.
573 * only when aclsize != 0 there's an actual segment for ACL
576 if (md->body->aclsize) {
577 rc = mdc_unpack_acl(req, md);
580 #ifdef CONFIG_FS_POSIX_ACL
582 md->posix_acl = NULL;
586 if (md->body->valid & OBD_MD_FLMDSCAPA) {
587 struct obd_capa *oc = mdc_unpack_capa(req, &RMF_CAPA1);
590 GOTO(out, rc = PTR_ERR(oc));
594 if (md->body->valid & OBD_MD_FLOSSCAPA) {
595 struct obd_capa *oc = mdc_unpack_capa(req, &RMF_CAPA2);
598 GOTO(out, rc = PTR_ERR(oc));
606 free_capa(md->oss_capa);
608 free_capa(md->mds_capa);
609 #ifdef CONFIG_FS_POSIX_ACL
610 posix_acl_release(md->posix_acl);
613 obd_free_memmd(dt_exp, &md->lsm);
618 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
624 static void mdc_replay_open(struct ptlrpc_request *req)
626 struct md_open_data *mod = req->rq_cb_data;
627 struct ptlrpc_request *cur, *tmp;
628 struct obd_client_handle *och;
629 struct lustre_handle old;
630 struct mdt_body *body;
634 DEBUG_REQ(D_ERROR, req,
635 "Can't properly replay without open data.");
640 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
641 LASSERT(body != NULL);
645 struct lustre_handle *file_fh;
647 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
649 file_fh = &och->och_fh;
650 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
651 file_fh->cookie, body->handle.cookie);
653 *file_fh = body->handle;
655 list_for_each_entry_safe(cur, tmp, &mod->mod_replay_list, rq_mod_list) {
656 int opc = lustre_msg_get_opc(cur->rq_reqmsg);
657 struct mdt_epoch *epoch = NULL;
659 if (opc == MDS_CLOSE || opc == MDS_DONE_WRITING) {
660 epoch = req_capsule_client_get(&cur->rq_pill,
663 DEBUG_REQ(D_HA, cur, "updating %s body with new fh",
664 opc == MDS_CLOSE ? "CLOSE" : "DONE_WRITING");
665 } else if (opc == MDS_REINT) {
666 struct mdt_rec_setattr *rec;
668 /* Check this is REINT_SETATTR. */
669 rec = req_capsule_client_get(&cur->rq_pill,
671 LASSERT(rec && rec->sa_opcode == REINT_SETATTR);
673 epoch = req_capsule_client_get(&cur->rq_pill,
676 DEBUG_REQ(D_HA, cur, "updating REINT_SETATTR body "
681 LASSERT(!memcmp(&old, &epoch->handle,
683 epoch->handle = body->handle;
689 void mdc_commit_delayed(struct ptlrpc_request *req)
691 struct md_open_data *mod = req->rq_cb_data;
692 struct ptlrpc_request *cur, *tmp;
694 DEBUG_REQ(D_HA, req, "req committed");
699 req->rq_cb_data = NULL;
700 req->rq_commit_cb = NULL;
701 list_del_init(&req->rq_mod_list);
702 if (req->rq_sequence) {
703 list_for_each_entry_safe(cur, tmp, &mod->mod_replay_list,
705 LASSERT(cur != LP_POISON);
706 LASSERT(cur->rq_type != LI_POISON);
707 DEBUG_REQ(D_HA, cur, "req balanced");
708 LASSERT(cur->rq_transno != 0);
709 LASSERT(cur->rq_import == req->rq_import);
711 /* We no longer want to preserve this for transno-
712 * unconditional replay. */
713 spin_lock(&cur->rq_lock);
715 spin_unlock(&cur->rq_lock);
719 if (list_empty(&mod->mod_replay_list)) {
720 if (mod->mod_och != NULL)
721 mod->mod_och->och_mod = NULL;
727 int mdc_set_open_replay_data(struct obd_export *exp,
728 struct obd_client_handle *och,
729 struct ptlrpc_request *open_req)
731 struct md_open_data *mod;
732 struct mdt_rec_create *rec;
733 struct mdt_body *body;
734 struct obd_import *imp = open_req->rq_import;
737 if (!open_req->rq_replay)
740 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
741 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
742 LASSERT(rec != NULL);
743 /* Incoming message in my byte order (it's been swabbed). */
744 /* Outgoing messages always in my byte order. */
745 LASSERT(body != NULL);
747 /* Only if the import is replayable, we set replay_open data */
748 if (och && imp->imp_replayable) {
751 DEBUG_REQ(D_ERROR, open_req,
752 "Can't allocate md_open_data");
755 CFS_INIT_LIST_HEAD(&mod->mod_replay_list);
757 spin_lock(&open_req->rq_lock);
760 open_req->rq_cb_data = mod;
761 list_add_tail(&open_req->rq_mod_list, &mod->mod_replay_list);
762 open_req->rq_commit_cb = mdc_commit_delayed;
763 spin_unlock(&open_req->rq_lock);
766 rec->cr_fid2 = body->fid1;
767 rec->cr_ioepoch = body->ioepoch;
768 rec->cr_old_handle.cookie = body->handle.cookie;
769 open_req->rq_replay_cb = mdc_replay_open;
770 if (!fid_is_sane(&body->fid1)) {
771 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
776 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
780 int mdc_clear_open_replay_data(struct obd_export *exp,
781 struct obd_client_handle *och)
783 struct md_open_data *mod = och->och_mod;
787 * Don't free the structure now (it happens in mdc_commit_delayed(),
788 * after the last request is removed from its replay list),
789 * but make sure that replay doesn't poke at the och, which is about to
792 LASSERT(mod != LP_POISON);
800 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
801 struct md_open_data *mod, struct ptlrpc_request **request)
803 struct obd_device *obd = class_exp2obd(exp);
804 struct ptlrpc_request *req;
809 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_CLOSE);
813 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
815 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
817 ptlrpc_request_free(req);
821 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
822 * portal whose threads are not taking any DLM locks and are therefore
823 * always progressing */
824 req->rq_request_portal = MDS_READPAGE_PORTAL;
825 ptlrpc_at_set_req_timeout(req);
827 /* Ensure that this close's handle is fixed up during replay. */
828 if (likely(mod != NULL))
829 list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
831 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
833 mdc_close_pack(req, op_data);
835 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
836 obd->u.cli.cl_max_mds_easize);
837 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
838 obd->u.cli.cl_max_mds_cookiesize);
840 ptlrpc_request_set_replen(req);
842 req->rq_commit_cb = mdc_commit_delayed;
844 LASSERT(req->rq_cb_data == NULL);
845 req->rq_cb_data = mod;
847 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
848 rc = ptlrpc_queue_wait(req);
849 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
851 if (req->rq_repmsg == NULL) {
852 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
855 rc = req->rq_status ?: -EIO;
856 } else if (rc == 0 || rc == -EAGAIN) {
857 struct mdt_body *body;
859 rc = lustre_msg_get_status(req->rq_repmsg);
860 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
861 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
865 } else if (mod == NULL) {
866 if (req->rq_import->imp_replayable)
867 CERROR("Unexpected: can't find md_open_data,"
868 "but close succeeded with replayable imp"
870 "http://bugzilla.lustre.org/\n");
873 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
879 if (rc != 0 && rc != -EAGAIN && req && req->rq_commit_cb)
880 req->rq_commit_cb(req);
886 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
887 struct md_open_data *mod)
889 struct obd_device *obd = class_exp2obd(exp);
890 struct ptlrpc_request *req;
894 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
895 &RQF_MDS_DONE_WRITING);
899 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
900 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
902 ptlrpc_request_free(req);
906 /* XXX: add DONE_WRITING request to och -- when Size-on-MDS
907 * recovery will be ready. */
908 mdc_close_pack(req, op_data);
909 ptlrpc_request_set_replen(req);
911 req->rq_cb_data = mod;
912 req->rq_commit_cb = mdc_commit_delayed;
913 if (likely(mod != NULL))
914 list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
916 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
918 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
919 rc = ptlrpc_queue_wait(req);
920 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
922 /* Close the open replay sequence if an error occured or no SOM
923 * attribute update is needed. */
925 ptlrpc_close_replay_seq(req);
927 if (rc && rc != -EAGAIN && req->rq_commit_cb)
928 req->rq_commit_cb(req);
930 ptlrpc_req_finished(req);
934 #ifdef HAVE_SPLIT_SUPPORT
935 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
936 const struct page *page, int offset)
938 struct ptlrpc_request *req;
939 struct ptlrpc_bulk_desc *desc;
943 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
947 /* FIXME: capa doesn't support split yet */
948 mdc_set_capa_size(req, &RMF_CAPA1, NULL);
950 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
952 ptlrpc_request_free(req);
956 req->rq_request_portal = MDS_READPAGE_PORTAL;
957 ptlrpc_at_set_req_timeout(req);
959 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
961 GOTO(out, rc = -ENOMEM);
963 /* NB req now owns desc and will free it when it gets freed. */
964 ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
965 mdc_readdir_pack(req, 0, offset, fid, NULL);
967 ptlrpc_request_set_replen(req);
968 rc = ptlrpc_queue_wait(req);
971 ptlrpc_req_finished(req);
974 EXPORT_SYMBOL(mdc_sendpage);
977 int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
978 struct obd_capa *oc, __u64 offset, struct page *page,
979 struct ptlrpc_request **request)
981 struct ptlrpc_request *req;
982 struct ptlrpc_bulk_desc *desc;
987 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
991 mdc_set_capa_size(req, &RMF_CAPA1, oc);
993 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
995 ptlrpc_request_free(req);
999 req->rq_request_portal = MDS_READPAGE_PORTAL;
1000 ptlrpc_at_set_req_timeout(req);
1002 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
1004 ptlrpc_request_free(req);
1008 /* NB req now owns desc and will free it when it gets freed */
1009 ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
1010 mdc_readdir_pack(req, offset, CFS_PAGE_SIZE, fid, oc);
1012 ptlrpc_request_set_replen(req);
1013 rc = ptlrpc_queue_wait(req);
1015 ptlrpc_req_finished(req);
1019 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
1020 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1021 req->rq_bulk->bd_nob_transferred, CFS_PAGE_SIZE);
1022 ptlrpc_req_finished(req);
1030 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1031 void *karg, void *uarg)
1033 struct obd_device *obd = exp->exp_obd;
1034 struct obd_ioctl_data *data = karg;
1035 struct obd_import *imp = obd->u.cli.cl_import;
1036 struct llog_ctxt *ctxt;
1040 if (!try_module_get(THIS_MODULE)) {
1041 CERROR("Can't get module. Is it alive?");
1045 case OBD_IOC_CLIENT_RECOVER:
1046 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
1050 case IOC_OSC_SET_ACTIVE:
1051 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1053 case OBD_IOC_PARSE: {
1054 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1055 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
1056 llog_ctxt_put(ctxt);
1060 case OBD_IOC_LLOG_INFO:
1061 case OBD_IOC_LLOG_PRINT: {
1062 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1063 rc = llog_ioctl(ctxt, cmd, data);
1064 llog_ctxt_put(ctxt);
1068 case OBD_IOC_POLL_QUOTACHECK:
1069 rc = lquota_poll_check(quota_interface, exp,
1070 (struct if_quotacheck *)karg);
1073 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1074 GOTO(out, rc = -ENOTTY);
1077 module_put(THIS_MODULE);
1082 static int do_set_info_async(struct obd_export *exp,
1083 obd_count keylen, void *key,
1084 obd_count vallen, void *val,
1085 struct ptlrpc_request_set *set)
1087 struct obd_import *imp = class_exp2cliimp(exp);
1088 struct ptlrpc_request *req;
1093 if (vallen != sizeof(int))
1096 spin_lock(&imp->imp_lock);
1097 if (*((int *)val)) {
1098 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
1099 imp->imp_connect_data.ocd_connect_flags |= OBD_CONNECT_RDONLY;
1101 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
1102 imp->imp_connect_data.ocd_connect_flags &= ~OBD_CONNECT_RDONLY;
1104 spin_unlock(&imp->imp_lock);
1106 req = ptlrpc_request_alloc(imp, &RQF_MDS_SET_INFO);
1110 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
1111 RCL_CLIENT, keylen);
1112 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
1113 RCL_CLIENT, vallen);
1114 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SET_INFO);
1116 ptlrpc_request_free(req);
1120 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1121 memcpy(tmp, key, keylen);
1122 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
1123 memcpy(tmp, val, vallen);
1125 ptlrpc_request_set_replen(req);
1128 ptlrpc_set_add_req(set, req);
1129 ptlrpc_check_set(NULL, set);
1131 rc = ptlrpc_queue_wait(req);
1132 ptlrpc_req_finished(req);
1138 int mdc_set_info_async(struct obd_export *exp,
1139 obd_count keylen, void *key,
1140 obd_count vallen, void *val,
1141 struct ptlrpc_request_set *set)
1143 struct obd_import *imp = class_exp2cliimp(exp);
1147 if (KEY_IS(KEY_INIT_RECOV)) {
1148 if (vallen != sizeof(int))
1150 spin_lock(&imp->imp_lock);
1151 imp->imp_initial_recov = *(int *)val;
1152 spin_unlock(&imp->imp_lock);
1153 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
1154 exp->exp_obd->obd_name, imp->imp_initial_recov);
1157 /* Turn off initial_recov after we try all backup servers once */
1158 if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
1159 if (vallen != sizeof(int))
1161 spin_lock(&imp->imp_lock);
1162 imp->imp_initial_recov_bk = *(int *)val;
1163 if (imp->imp_initial_recov_bk)
1164 imp->imp_initial_recov = 1;
1165 spin_unlock(&imp->imp_lock);
1166 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
1167 exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
1170 if (KEY_IS(KEY_READ_ONLY)) {
1171 rc = do_set_info_async(exp, keylen, key, vallen, val, set);
1174 if (KEY_IS(KEY_FLUSH_CTX)) {
1175 sptlrpc_import_flush_my_ctx(imp);
1178 if (KEY_IS(KEY_MDS_CONN)) {
1179 struct obd_import *imp = class_exp2cliimp(exp);
1181 /* mds-mds import */
1182 spin_lock(&imp->imp_lock);
1183 imp->imp_server_timeout = 1;
1184 spin_unlock(&imp->imp_lock);
1185 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1186 CDEBUG(D_OTHER|D_WARNING, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1193 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
1194 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1198 if (KEY_IS(KEY_MAX_EASIZE)) {
1199 int mdsize, *max_easize;
1201 if (*vallen != sizeof(int))
1203 mdsize = *(int*)val;
1204 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
1205 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
1207 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
1210 if (KEY_IS(KEY_CONN_DATA)) {
1211 struct obd_import *imp = class_exp2cliimp(exp);
1212 struct obd_connect_data *data = val;
1214 if (*vallen != sizeof(*data))
1217 *data = imp->imp_connect_data;
1224 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1225 __u64 max_age, __u32 flags)
1227 struct ptlrpc_request *req;
1228 struct obd_statfs *msfs;
1229 struct obd_import *imp = NULL;
1234 /*Since the request might also come from lprocfs, so we need
1235 *sync this with client_disconnect_export Bug15684*/
1236 down_read(&obd->u.cli.cl_sem);
1237 if (obd->u.cli.cl_import)
1238 imp = class_import_get(obd->u.cli.cl_import);
1239 up_read(&obd->u.cli.cl_sem);
1243 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1244 LUSTRE_MDS_VERSION, MDS_STATFS);
1246 GOTO(output, rc = -ENOMEM);
1248 ptlrpc_request_set_replen(req);
1250 if (flags & OBD_STATFS_NODELAY) {
1251 /* procfs requests not want stay in wait for avoid deadlock */
1252 req->rq_no_resend = 1;
1253 req->rq_no_delay = 1;
1256 rc = ptlrpc_queue_wait(req);
1258 /* check connection error first */
1259 if (imp->imp_connect_error)
1260 rc = imp->imp_connect_error;
1264 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1266 GOTO(out, rc = -EPROTO);
1271 ptlrpc_req_finished(req);
1273 class_import_put(imp);
1277 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
1278 struct obd_capa *oc, struct obd_client_handle *handle,
1281 struct ptlrpc_request *req;
1282 struct mdt_body *body;
1286 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
1290 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1292 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
1294 ptlrpc_request_free(req);
1298 mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
1300 ptlrpc_request_set_replen(req);
1302 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1303 rc = ptlrpc_queue_wait(req);
1304 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1306 CERROR("Pin failed: %d\n", rc);
1310 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1312 GOTO(err_out, rc = -EPROTO);
1314 handle->och_fh = body->handle;
1315 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1317 OBD_ALLOC_PTR(handle->och_mod);
1318 if (handle->och_mod == NULL) {
1319 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
1320 GOTO(err_out, rc = -ENOMEM);
1322 /* will be dropped by unpin */
1323 CFS_INIT_LIST_HEAD(&handle->och_mod->mod_replay_list);
1324 list_add_tail(&req->rq_mod_list, &handle->och_mod->mod_replay_list);
1329 ptlrpc_req_finished(req);
1333 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
1336 struct ptlrpc_request *req;
1337 struct mdt_body *body;
1341 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
1342 LUSTRE_MDS_VERSION, MDS_UNPIN);
1346 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1347 body->handle = handle->och_fh;
1350 ptlrpc_request_set_replen(req);
1352 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1353 rc = ptlrpc_queue_wait(req);
1354 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1357 CERROR("Unpin failed: %d\n", rc);
1359 ptlrpc_req_finished(req);
1361 LASSERT(!list_empty(&handle->och_mod->mod_replay_list));
1362 req = list_entry(handle->och_mod->mod_replay_list.next,
1363 typeof(*req), rq_mod_list);
1364 list_del_init(&req->rq_mod_list);
1365 ptlrpc_req_finished(req);
1366 LASSERT(list_empty(&handle->och_mod->mod_replay_list));
1368 OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1372 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
1373 struct obd_capa *oc, struct ptlrpc_request **request)
1375 struct ptlrpc_request *req;
1380 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
1384 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1386 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
1388 ptlrpc_request_free(req);
1392 mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
1394 ptlrpc_request_set_replen(req);
1396 rc = ptlrpc_queue_wait(req);
1398 ptlrpc_req_finished(req);
1404 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1405 enum obd_import_event event)
1409 LASSERT(imp->imp_obd == obd);
1412 case IMP_EVENT_DISCON: {
1414 /* XXX Pass event up to OBDs stack. used only for FLD now */
1415 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
1419 case IMP_EVENT_INACTIVE: {
1420 struct client_obd *cli = &obd->u.cli;
1422 * Flush current sequence to make client obtain new one
1423 * from server in case of disconnect/reconnect.
1424 * If range is already empty then no need to flush it.
1426 if (cli->cl_seq != NULL &&
1427 !range_is_exhausted(&cli->cl_seq->lcs_space)) {
1428 seq_client_flush(cli->cl_seq);
1431 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1434 case IMP_EVENT_INVALIDATE: {
1435 struct ldlm_namespace *ns = obd->obd_namespace;
1437 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1441 case IMP_EVENT_ACTIVE: {
1442 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1446 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
1450 CERROR("Unknown import event %x\n", event);
1456 static int mdc_fid_init(struct obd_export *exp)
1458 struct client_obd *cli = &exp->exp_obd->u.cli;
1463 OBD_ALLOC_PTR(cli->cl_seq);
1464 if (cli->cl_seq == NULL)
1467 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
1469 GOTO(out_free_seq, rc = -ENOMEM);
1471 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
1472 exp->exp_obd->obd_name);
1474 /* Init client side sequence-manager */
1475 rc = seq_client_init(cli->cl_seq, exp,
1476 LUSTRE_SEQ_METADATA,
1478 OBD_FREE(prefix, MAX_OBD_NAME + 5);
1480 GOTO(out_free_seq, rc);
1484 OBD_FREE_PTR(cli->cl_seq);
1489 static int mdc_fid_fini(struct obd_export *exp)
1491 struct client_obd *cli = &exp->exp_obd->u.cli;
1494 if (cli->cl_seq != NULL) {
1495 seq_client_fini(cli->cl_seq);
1496 OBD_FREE_PTR(cli->cl_seq);
1503 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1504 struct md_op_data *op_data)
1506 struct client_obd *cli = &exp->exp_obd->u.cli;
1507 struct lu_client_seq *seq = cli->cl_seq;
1509 RETURN(seq_client_alloc_fid(seq, fid));
1512 /* XXX This method is used only to clear current fid seq
1513 * once fld/mds insert failed */
1514 static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1516 struct client_obd *cli = &exp->exp_obd->u.cli;
1518 seq_client_flush(cli->cl_seq);
1522 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
1523 struct client_obd *cli = &exp->exp_obd->u.cli;
1524 return &cli->cl_target_uuid;
1527 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
1529 struct client_obd *cli = &obd->u.cli;
1530 struct lprocfs_static_vars lvars = { 0 };
1534 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1535 if (!cli->cl_rpc_lock)
1537 mdc_init_rpc_lock(cli->cl_rpc_lock);
1541 OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1542 if (!cli->cl_setattr_lock)
1543 GOTO(err_rpc_lock, rc = -ENOMEM);
1544 mdc_init_rpc_lock(cli->cl_setattr_lock);
1546 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1547 if (!cli->cl_close_lock)
1548 GOTO(err_setattr_lock, rc = -ENOMEM);
1549 mdc_init_rpc_lock(cli->cl_close_lock);
1551 rc = client_obd_setup(obd, cfg);
1553 GOTO(err_close_lock, rc);
1554 lprocfs_mdc_init_vars(&lvars);
1555 lprocfs_obd_setup(obd, lvars.obd_vars);
1556 sptlrpc_lprocfs_cliobd_attach(obd);
1557 ptlrpc_lprocfs_register_obd(obd);
1559 rc = obd_llog_init(obd, &obd->obd_olg, obd, 0, NULL, NULL);
1562 CERROR("failed to setup llogging subsystems\n");
1568 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1570 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1572 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1577 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
1578 * us to make MDS RPCs with large enough reply buffers to hold the
1579 * maximum-sized (= maximum striped) EA and cookie without having to
1580 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1581 static int mdc_init_ea_size(struct obd_export *exp, int easize,
1582 int def_easize, int cookiesize)
1584 struct obd_device *obd = exp->exp_obd;
1585 struct client_obd *cli = &obd->u.cli;
1588 if (cli->cl_max_mds_easize < easize)
1589 cli->cl_max_mds_easize = easize;
1591 if (cli->cl_default_mds_easize < def_easize)
1592 cli->cl_default_mds_easize = def_easize;
1594 if (cli->cl_max_mds_cookiesize < cookiesize)
1595 cli->cl_max_mds_cookiesize = cookiesize;
1600 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1606 case OBD_CLEANUP_EARLY:
1607 case OBD_CLEANUP_EXPORTS:
1608 /* If we set up but never connected, the
1609 client import will not have been cleaned. */
1610 if (obd->u.cli.cl_import) {
1611 struct obd_import *imp;
1612 down_write(&obd->u.cli.cl_sem);
1613 imp = obd->u.cli.cl_import;
1614 CERROR("client import never connected\n");
1615 ptlrpc_invalidate_import(imp);
1616 class_destroy_import(imp);
1617 up_write(&obd->u.cli.cl_sem);
1618 obd->u.cli.cl_import = NULL;
1620 rc = obd_llog_finish(obd, 0);
1622 CERROR("failed to cleanup llogging subsystems\n");
1628 static int mdc_cleanup(struct obd_device *obd)
1630 struct client_obd *cli = &obd->u.cli;
1632 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1633 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1634 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1636 ptlrpc_lprocfs_unregister_obd(obd);
1637 lprocfs_obd_cleanup(obd);
1640 return client_obd_cleanup(obd);
1644 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
1645 struct obd_device *tgt, int count,
1646 struct llog_catid *logid, struct obd_uuid *uuid)
1648 struct llog_ctxt *ctxt;
1652 LASSERT(olg == &obd->obd_olg);
1654 rc = llog_setup(obd, olg, LLOG_LOVEA_REPL_CTXT, tgt, 0,
1655 NULL, &llog_client_ops);
1657 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1658 llog_initiator_connect(ctxt);
1659 llog_ctxt_put(ctxt);
1665 static int mdc_llog_finish(struct obd_device *obd, int count)
1667 struct llog_ctxt *ctxt;
1671 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1673 rc = llog_cleanup(ctxt);
1678 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1680 struct lustre_cfg *lcfg = buf;
1681 struct lprocfs_static_vars lvars = { 0 };
1684 lprocfs_mdc_init_vars(&lvars);
1685 switch (lcfg->lcfg_command) {
1686 case LCFG_SPTLRPC_CONF:
1687 rc = sptlrpc_cliobd_process_config(obd, lcfg);
1690 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
1698 /* get remote permission for current user on fid */
1699 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
1700 struct obd_capa *oc, __u32 suppgid,
1701 struct ptlrpc_request **request)
1703 struct ptlrpc_request *req;
1707 LASSERT(client_is_remote(exp));
1710 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1714 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1716 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1718 ptlrpc_request_free(req);
1722 mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
1724 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
1725 sizeof(struct mdt_remote_perm));
1727 ptlrpc_request_set_replen(req);
1729 rc = ptlrpc_queue_wait(req);
1731 ptlrpc_req_finished(req);
1737 static int mdc_interpret_renew_capa(const struct lu_env *env,
1738 struct ptlrpc_request *req, void *unused,
1741 struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
1742 renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
1743 struct mdt_body *body = NULL;
1744 struct lustre_capa *capa;
1748 GOTO(out, capa = ERR_PTR(status));
1750 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1752 GOTO(out, capa = ERR_PTR(-EFAULT));
1754 if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
1755 GOTO(out, capa = ERR_PTR(-ENOENT));
1757 capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
1759 GOTO(out, capa = ERR_PTR(-EFAULT));
1766 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
1769 struct ptlrpc_request *req;
1772 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
1773 LUSTRE_MDS_VERSION, MDS_GETATTR);
1777 /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
1778 * capa to renew is oss capa.
1780 mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
1781 ptlrpc_request_set_replen(req);
1783 req->rq_async_args.pointer_arg[0] = oc;
1784 req->rq_async_args.pointer_arg[1] = cb;
1785 req->rq_interpret_reply = mdc_interpret_renew_capa;
1786 ptlrpcd_add_req(req, PSCOPE_OTHER);
1790 static int mdc_connect(const struct lu_env *env,
1791 struct lustre_handle *dlm_handle,
1792 struct obd_device *obd, struct obd_uuid *cluuid,
1793 struct obd_connect_data *data,
1796 struct obd_import *imp = obd->u.cli.cl_import;
1798 /* mds-mds import features */
1799 if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
1800 spin_lock(&imp->imp_lock);
1801 imp->imp_server_timeout = 1;
1802 spin_unlock(&imp->imp_lock);
1803 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1804 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
1808 return client_connect_import(env, dlm_handle, obd, cluuid, data, NULL);
1811 struct obd_ops mdc_obd_ops = {
1812 .o_owner = THIS_MODULE,
1813 .o_setup = mdc_setup,
1814 .o_precleanup = mdc_precleanup,
1815 .o_cleanup = mdc_cleanup,
1816 .o_add_conn = client_import_add_conn,
1817 .o_del_conn = client_import_del_conn,
1818 .o_connect = mdc_connect,
1819 .o_disconnect = client_disconnect_export,
1820 .o_iocontrol = mdc_iocontrol,
1821 .o_set_info_async = mdc_set_info_async,
1822 .o_statfs = mdc_statfs,
1824 .o_unpin = mdc_unpin,
1825 .o_fid_init = mdc_fid_init,
1826 .o_fid_fini = mdc_fid_fini,
1827 .o_fid_alloc = mdc_fid_alloc,
1828 .o_fid_delete = mdc_fid_delete,
1829 .o_import_event = mdc_import_event,
1830 .o_llog_init = mdc_llog_init,
1831 .o_llog_finish = mdc_llog_finish,
1832 .o_get_info = mdc_get_info,
1833 .o_process_config = mdc_process_config,
1834 .o_get_uuid = mdc_get_uuid,
1837 struct md_ops mdc_md_ops = {
1838 .m_getstatus = mdc_getstatus,
1839 .m_change_cbdata = mdc_change_cbdata,
1840 .m_close = mdc_close,
1841 .m_create = mdc_create,
1842 .m_done_writing = mdc_done_writing,
1843 .m_enqueue = mdc_enqueue,
1844 .m_getattr = mdc_getattr,
1845 .m_getattr_name = mdc_getattr_name,
1846 .m_intent_lock = mdc_intent_lock,
1848 .m_is_subdir = mdc_is_subdir,
1849 .m_rename = mdc_rename,
1850 .m_setattr = mdc_setattr,
1851 .m_setxattr = mdc_setxattr,
1852 .m_getxattr = mdc_getxattr,
1854 .m_readpage = mdc_readpage,
1855 .m_unlink = mdc_unlink,
1856 .m_cancel_unused = mdc_cancel_unused,
1857 .m_init_ea_size = mdc_init_ea_size,
1858 .m_set_lock_data = mdc_set_lock_data,
1859 .m_lock_match = mdc_lock_match,
1860 .m_get_lustre_md = mdc_get_lustre_md,
1861 .m_free_lustre_md = mdc_free_lustre_md,
1862 .m_set_open_replay_data = mdc_set_open_replay_data,
1863 .m_clear_open_replay_data = mdc_clear_open_replay_data,
1864 .m_renew_capa = mdc_renew_capa,
1865 .m_get_remote_perm = mdc_get_remote_perm,
1866 .m_intent_getattr_async = mdc_intent_getattr_async,
1867 .m_revalidate_lock = mdc_revalidate_lock
1870 extern quota_interface_t mdc_quota_interface;
1872 int __init mdc_init(void)
1875 struct lprocfs_static_vars lvars = { 0 };
1876 lprocfs_mdc_init_vars(&lvars);
1878 request_module("lquota");
1879 quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1880 init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1882 rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1883 LUSTRE_MDC_NAME, NULL);
1884 if (rc && quota_interface)
1885 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1891 static void /*__exit*/ mdc_exit(void)
1893 if (quota_interface)
1894 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1896 class_unregister_type(LUSTRE_MDC_NAME);
1899 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
1900 MODULE_DESCRIPTION("Lustre Metadata Client");
1901 MODULE_LICENSE("GPL");
1903 module_init(mdc_init);
1904 module_exit(mdc_exit);