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.
687 req->rq_cb_data = NULL;
688 mod->mod_open_req = NULL;
692 int mdc_set_open_replay_data(struct obd_export *exp,
693 struct obd_client_handle *och,
694 struct ptlrpc_request *open_req)
696 struct md_open_data *mod;
697 struct mdt_rec_create *rec;
698 struct mdt_body *body;
699 struct obd_import *imp = open_req->rq_import;
702 if (!open_req->rq_replay)
705 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
706 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
707 LASSERT(rec != NULL);
708 /* Incoming message in my byte order (it's been swabbed). */
709 /* Outgoing messages always in my byte order. */
710 LASSERT(body != NULL);
712 /* Only if the import is replayable, we set replay_open data */
713 if (och && imp->imp_replayable) {
714 mod = obd_mod_alloc();
716 DEBUG_REQ(D_ERROR, open_req,
717 "Can't allocate md_open_data");
721 spin_lock(&open_req->rq_lock);
725 mod->mod_open_req = open_req;
726 open_req->rq_cb_data = mod;
727 open_req->rq_commit_cb = mdc_commit_open;
728 spin_unlock(&open_req->rq_lock);
731 rec->cr_fid2 = body->fid1;
732 rec->cr_ioepoch = body->ioepoch;
733 rec->cr_old_handle.cookie = body->handle.cookie;
734 open_req->rq_replay_cb = mdc_replay_open;
735 if (!fid_is_sane(&body->fid1)) {
736 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
741 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
745 int mdc_clear_open_replay_data(struct obd_export *exp,
746 struct obd_client_handle *och)
748 struct md_open_data *mod = och->och_mod;
751 LASSERT(mod != LP_POISON && mod != NULL);
760 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
761 struct md_open_data *mod, struct ptlrpc_request **request)
763 struct obd_device *obd = class_exp2obd(exp);
764 struct ptlrpc_request *req;
769 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_CLOSE);
773 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
775 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
777 ptlrpc_request_free(req);
781 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
782 * portal whose threads are not taking any DLM locks and are therefore
783 * always progressing */
784 req->rq_request_portal = MDS_READPAGE_PORTAL;
785 ptlrpc_at_set_req_timeout(req);
787 /* Ensure that this close's handle is fixed up during replay. */
788 if (likely(mod != NULL)) {
789 LASSERTF(mod->mod_open_req->rq_type != LI_POISON,
790 "POISONED open %p!\n", mod->mod_open_req);
792 mod->mod_close_req = req;
795 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
796 /* We no longer want to preserve this open for replay even
797 * though the open was committed. b=3632, b=3633 */
798 spin_lock(&mod->mod_open_req->rq_lock);
799 mod->mod_open_req->rq_replay = 0;
800 spin_unlock(&mod->mod_open_req->rq_lock);
802 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
805 mdc_close_pack(req, op_data);
807 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
808 obd->u.cli.cl_max_mds_easize);
809 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
810 obd->u.cli.cl_max_mds_cookiesize);
812 ptlrpc_request_set_replen(req);
814 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
815 rc = ptlrpc_queue_wait(req);
816 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
818 if (req->rq_repmsg == NULL) {
819 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
822 rc = req->rq_status ?: -EIO;
823 } else if (rc == 0 || rc == -EAGAIN) {
824 struct mdt_body *body;
826 rc = lustre_msg_get_status(req->rq_repmsg);
827 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
828 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
833 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
836 } else if (rc == -ESTALE) {
838 * it can be allowed error after 3633 if open was committed and
839 * server failed before close was sent. Let's check if mod
840 * exists and return no error in that case
842 if (mod && (mod->mod_open_req == NULL))
848 mod->mod_close_req = NULL;
849 /* Since now, mod is accessed through open_req only,
850 * thus close req does not keep a reference on mod anymore. */
857 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
858 struct md_open_data *mod)
860 struct obd_device *obd = class_exp2obd(exp);
861 struct ptlrpc_request *req;
865 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
866 &RQF_MDS_DONE_WRITING);
870 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
871 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
873 ptlrpc_request_free(req);
878 LASSERTF(mod->mod_open_req->rq_type != LI_POISON,
879 "POISONED setattr %p!\n", mod->mod_open_req);
881 mod->mod_close_req = req;
883 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
884 /* We no longer want to preserve this setattr for replay even
885 * though the open was committed. b=3632, b=3633 */
886 spin_lock(&mod->mod_open_req->rq_lock);
887 mod->mod_open_req->rq_replay = 0;
888 spin_unlock(&mod->mod_open_req->rq_lock);
891 mdc_close_pack(req, op_data);
892 ptlrpc_request_set_replen(req);
894 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
895 rc = ptlrpc_queue_wait(req);
896 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
900 * it can be allowed error after 3633 if open or setattr were
901 * committed and server failed before close was sent.
902 * Let's check if mod exists and return no error in that case
904 if (mod && (mod->mod_open_req == NULL))
910 mod->mod_close_req = NULL;
911 /* Since now, mod is accessed through setattr req only,
912 * thus DW req does not keep a reference on mod anymore. */
915 ptlrpc_req_finished(req);
919 #ifdef HAVE_SPLIT_SUPPORT
920 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
921 const struct page *page, int offset)
923 struct ptlrpc_request *req;
924 struct ptlrpc_bulk_desc *desc;
928 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
932 /* FIXME: capa doesn't support split yet */
933 mdc_set_capa_size(req, &RMF_CAPA1, NULL);
935 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
937 ptlrpc_request_free(req);
941 req->rq_request_portal = MDS_READPAGE_PORTAL;
942 ptlrpc_at_set_req_timeout(req);
944 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
946 GOTO(out, rc = -ENOMEM);
948 /* NB req now owns desc and will free it when it gets freed. */
949 ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
950 mdc_readdir_pack(req, 0, offset, fid, NULL);
952 ptlrpc_request_set_replen(req);
953 rc = ptlrpc_queue_wait(req);
957 rc = sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk);
959 ptlrpc_req_finished(req);
962 EXPORT_SYMBOL(mdc_sendpage);
965 int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
966 struct obd_capa *oc, __u64 offset, struct page *page,
967 struct ptlrpc_request **request)
969 struct ptlrpc_request *req;
970 struct ptlrpc_bulk_desc *desc;
975 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
979 mdc_set_capa_size(req, &RMF_CAPA1, oc);
981 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
983 ptlrpc_request_free(req);
987 req->rq_request_portal = MDS_READPAGE_PORTAL;
988 ptlrpc_at_set_req_timeout(req);
990 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
992 ptlrpc_request_free(req);
996 /* NB req now owns desc and will free it when it gets freed */
997 ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
998 mdc_readdir_pack(req, offset, CFS_PAGE_SIZE, fid, oc);
1000 ptlrpc_request_set_replen(req);
1001 rc = ptlrpc_queue_wait(req);
1003 ptlrpc_req_finished(req);
1007 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1008 req->rq_bulk->bd_nob_transferred);
1010 ptlrpc_req_finished(req);
1014 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
1015 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1016 req->rq_bulk->bd_nob_transferred, CFS_PAGE_SIZE);
1017 ptlrpc_req_finished(req);
1025 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1027 __u32 keylen, vallen;
1031 if (gf->gf_pathlen > PATH_MAX)
1032 RETURN(-ENAMETOOLONG);
1033 if (gf->gf_pathlen < 2)
1036 /* Key is KEY_FID2PATH + getinfo_fid2path description */
1037 keylen = size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1038 OBD_ALLOC(key, keylen);
1041 memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1042 memcpy(key + size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1044 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1045 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1047 if (!fid_is_sane(&gf->gf_fid))
1048 GOTO(out, rc = -EINVAL);
1050 /* Val is struct getinfo_fid2path result plus path */
1051 vallen = sizeof(*gf) + gf->gf_pathlen;
1053 rc = obd_get_info(exp, keylen, key, &vallen, gf, NULL);
1057 if (vallen <= sizeof(*gf))
1058 GOTO(out, rc = -EPROTO);
1059 else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1060 GOTO(out, rc = -EOVERFLOW);
1062 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1063 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1066 OBD_FREE(key, keylen);
1070 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1071 void *karg, void *uarg)
1073 struct obd_device *obd = exp->exp_obd;
1074 struct obd_ioctl_data *data = karg;
1075 struct obd_import *imp = obd->u.cli.cl_import;
1076 struct llog_ctxt *ctxt;
1080 if (!try_module_get(THIS_MODULE)) {
1081 CERROR("Can't get module. Is it alive?");
1085 case OBD_IOC_CHANGELOG_CLEAR: {
1086 struct ioc_changelog_clear *icc = karg;
1087 struct changelog_setinfo cs =
1088 {icc->icc_recno, icc->icc_id};
1089 rc = obd_set_info_async(exp, strlen(KEY_CHANGELOG_CLEAR),
1090 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1094 case OBD_IOC_FID2PATH: {
1095 rc = mdc_ioc_fid2path(exp, karg);
1098 case OBD_IOC_CLIENT_RECOVER:
1099 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
1103 case IOC_OSC_SET_ACTIVE:
1104 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1106 case OBD_IOC_PARSE: {
1107 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1108 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
1109 llog_ctxt_put(ctxt);
1113 case OBD_IOC_LLOG_INFO:
1114 case OBD_IOC_LLOG_PRINT: {
1115 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1116 rc = llog_ioctl(ctxt, cmd, data);
1117 llog_ctxt_put(ctxt);
1121 case OBD_IOC_POLL_QUOTACHECK:
1122 rc = lquota_poll_check(quota_interface, exp,
1123 (struct if_quotacheck *)karg);
1125 case OBD_IOC_PING_TARGET:
1126 rc = ptlrpc_obd_ping(obd);
1129 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1130 GOTO(out, rc = -ENOTTY);
1133 module_put(THIS_MODULE);
1138 int mdc_get_info_rpc(struct obd_export *exp,
1139 obd_count keylen, void *key,
1140 int vallen, void *val)
1142 struct obd_import *imp = class_exp2cliimp(exp);
1143 struct ptlrpc_request *req;
1148 req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1152 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1153 RCL_CLIENT, keylen);
1154 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1155 RCL_CLIENT, sizeof(__u32));
1157 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1159 ptlrpc_request_free(req);
1163 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
1164 memcpy(tmp, key, keylen);
1165 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
1166 memcpy(tmp, &vallen, sizeof(__u32));
1168 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1169 RCL_SERVER, vallen);
1170 ptlrpc_request_set_replen(req);
1172 rc = ptlrpc_queue_wait(req);
1174 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
1175 memcpy(val, tmp, vallen);
1176 if (ptlrpc_rep_need_swab(req)) {
1177 if (KEY_IS(KEY_FID2PATH)) {
1178 lustre_swab_fid2path(val);
1182 ptlrpc_req_finished(req);
1187 static void lustre_swab_hai(struct hsm_action_item *h)
1189 __swab32s(&h->hai_len);
1190 __swab32s(&h->hai_action);
1191 lustre_swab_lu_fid(&h->hai_fid);
1192 __swab64s(&h->hai_cookie);
1193 __swab64s(&h->hai_extent_start);
1194 __swab64s(&h->hai_extent_end);
1195 __swab64s(&h->hai_gid);
1198 static void lustre_swab_hal(struct hsm_action_list *h)
1200 struct hsm_action_item *hai;
1203 __swab32s(&h->hal_version);
1204 __swab32s(&h->hal_count);
1205 __swab32s(&h->hal_archive_num);
1207 for (i = 0; i < h->hal_count; i++) {
1208 lustre_swab_hai(hai);
1209 hai = hai_next(hai);
1214 * Send a message to any listening copytools, nonblocking
1215 * @param val LNL message (lnl_hdr + hsm_action_list)
1216 * @param len total length of message
1218 static int mdc_hsm_copytool_send(int len, void *val)
1220 struct lnl_hdr *lh = (struct lnl_hdr *)val;
1221 struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
1225 if (len < sizeof(*lh) + sizeof(*hal)) {
1226 CERROR("Short HSM message %d < %d\n", len,
1227 (int) (sizeof(*lh) + sizeof(*hal)));
1230 if (lh->lnl_magic == __swab16(LNL_MAGIC)) {
1231 lustre_swab_lnlh(lh);
1232 lustre_swab_hal(hal);
1233 } else if (lh->lnl_magic != LNL_MAGIC) {
1234 CERROR("Bad magic %x!=%x\n", lh->lnl_magic, LNL_MAGIC);
1238 CDEBUG(D_IOCTL, " Received message mg=%x t=%d m=%d l=%d actions=%d\n",
1239 lh->lnl_magic, lh->lnl_transport, lh->lnl_msgtype,
1240 lh->lnl_msglen, hal->hal_count);
1242 /* Broadcast to HSM listeners */
1243 rc = libcfs_klnl_msg_put(0, LNL_GRP_HSM, lh);
1248 int mdc_set_info_async(struct obd_export *exp,
1249 obd_count keylen, void *key,
1250 obd_count vallen, void *val,
1251 struct ptlrpc_request_set *set)
1253 struct obd_import *imp = class_exp2cliimp(exp);
1257 if (KEY_IS(KEY_INIT_RECOV)) {
1258 if (vallen != sizeof(int))
1260 spin_lock(&imp->imp_lock);
1261 imp->imp_initial_recov = *(int *)val;
1262 spin_unlock(&imp->imp_lock);
1263 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
1264 exp->exp_obd->obd_name, imp->imp_initial_recov);
1267 /* Turn off initial_recov after we try all backup servers once */
1268 if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
1269 if (vallen != sizeof(int))
1271 spin_lock(&imp->imp_lock);
1272 imp->imp_initial_recov_bk = *(int *)val;
1273 if (imp->imp_initial_recov_bk)
1274 imp->imp_initial_recov = 1;
1275 spin_unlock(&imp->imp_lock);
1276 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
1277 exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
1280 if (KEY_IS(KEY_READ_ONLY)) {
1281 if (vallen != sizeof(int))
1284 spin_lock(&imp->imp_lock);
1285 if (*((int *)val)) {
1286 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
1287 imp->imp_connect_data.ocd_connect_flags |= OBD_CONNECT_RDONLY;
1289 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
1290 imp->imp_connect_data.ocd_connect_flags &= ~OBD_CONNECT_RDONLY;
1292 spin_unlock(&imp->imp_lock);
1294 rc = target_set_info_rpc(imp, MDS_SET_INFO,
1295 keylen, key, vallen, val, set);
1298 if (KEY_IS(KEY_SPTLRPC_CONF)) {
1299 sptlrpc_conf_client_adapt(exp->exp_obd);
1302 if (KEY_IS(KEY_FLUSH_CTX)) {
1303 sptlrpc_import_flush_my_ctx(imp);
1306 if (KEY_IS(KEY_MDS_CONN)) {
1307 /* mds-mds import */
1308 spin_lock(&imp->imp_lock);
1309 imp->imp_server_timeout = 1;
1310 spin_unlock(&imp->imp_lock);
1311 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1312 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1315 if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1316 rc = target_set_info_rpc(imp, MDS_SET_INFO,
1317 keylen, key, vallen, val, set);
1320 if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
1321 rc = mdc_hsm_copytool_send(vallen, val);
1328 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
1329 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1333 if (KEY_IS(KEY_MAX_EASIZE)) {
1334 int mdsize, *max_easize;
1336 if (*vallen != sizeof(int))
1338 mdsize = *(int*)val;
1339 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
1340 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
1342 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
1345 if (KEY_IS(KEY_CONN_DATA)) {
1346 struct obd_import *imp = class_exp2cliimp(exp);
1347 struct obd_connect_data *data = val;
1349 if (*vallen != sizeof(*data))
1352 *data = imp->imp_connect_data;
1356 rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
1361 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1362 __u64 max_age, __u32 flags)
1364 struct ptlrpc_request *req;
1365 struct obd_statfs *msfs;
1366 struct obd_import *imp = NULL;
1371 /*Since the request might also come from lprocfs, so we need
1372 *sync this with client_disconnect_export Bug15684*/
1373 down_read(&obd->u.cli.cl_sem);
1374 if (obd->u.cli.cl_import)
1375 imp = class_import_get(obd->u.cli.cl_import);
1376 up_read(&obd->u.cli.cl_sem);
1380 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1381 LUSTRE_MDS_VERSION, MDS_STATFS);
1383 GOTO(output, rc = -ENOMEM);
1385 ptlrpc_request_set_replen(req);
1387 if (flags & OBD_STATFS_NODELAY) {
1388 /* procfs requests not want stay in wait for avoid deadlock */
1389 req->rq_no_resend = 1;
1390 req->rq_no_delay = 1;
1393 rc = ptlrpc_queue_wait(req);
1395 /* check connection error first */
1396 if (imp->imp_connect_error)
1397 rc = imp->imp_connect_error;
1401 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1403 GOTO(out, rc = -EPROTO);
1408 ptlrpc_req_finished(req);
1410 class_import_put(imp);
1414 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
1415 struct obd_capa *oc, struct obd_client_handle *handle,
1418 struct ptlrpc_request *req;
1419 struct mdt_body *body;
1423 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
1427 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1429 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
1431 ptlrpc_request_free(req);
1435 mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
1437 ptlrpc_request_set_replen(req);
1439 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1440 rc = ptlrpc_queue_wait(req);
1441 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1443 CERROR("Pin failed: %d\n", rc);
1447 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1449 GOTO(err_out, rc = -EPROTO);
1451 handle->och_fh = body->handle;
1452 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1454 handle->och_mod = obd_mod_alloc();
1455 if (handle->och_mod == NULL) {
1456 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
1457 GOTO(err_out, rc = -ENOMEM);
1459 handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
1464 ptlrpc_req_finished(req);
1468 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
1471 struct ptlrpc_request *req;
1472 struct mdt_body *body;
1476 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
1477 LUSTRE_MDS_VERSION, MDS_UNPIN);
1481 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1482 body->handle = handle->och_fh;
1485 ptlrpc_request_set_replen(req);
1487 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1488 rc = ptlrpc_queue_wait(req);
1489 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1492 CERROR("Unpin failed: %d\n", rc);
1494 ptlrpc_req_finished(req);
1495 ptlrpc_req_finished(handle->och_mod->mod_open_req);
1497 obd_mod_put(handle->och_mod);
1501 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
1502 struct obd_capa *oc, struct ptlrpc_request **request)
1504 struct ptlrpc_request *req;
1509 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
1513 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1515 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
1517 ptlrpc_request_free(req);
1521 mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
1523 ptlrpc_request_set_replen(req);
1525 rc = ptlrpc_queue_wait(req);
1527 ptlrpc_req_finished(req);
1533 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1534 enum obd_import_event event)
1538 LASSERT(imp->imp_obd == obd);
1541 case IMP_EVENT_DISCON: {
1543 /* XXX Pass event up to OBDs stack. used only for FLD now */
1544 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
1548 case IMP_EVENT_INACTIVE: {
1549 struct client_obd *cli = &obd->u.cli;
1551 * Flush current sequence to make client obtain new one
1552 * from server in case of disconnect/reconnect.
1553 * If range is already empty then no need to flush it.
1555 if (cli->cl_seq != NULL &&
1556 !range_is_exhausted(&cli->cl_seq->lcs_space)) {
1557 seq_client_flush(cli->cl_seq);
1560 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1563 case IMP_EVENT_INVALIDATE: {
1564 struct ldlm_namespace *ns = obd->obd_namespace;
1566 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1570 case IMP_EVENT_ACTIVE: {
1571 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1575 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
1579 CERROR("Unknown import event %x\n", event);
1585 static int mdc_fid_init(struct obd_export *exp)
1587 struct client_obd *cli = &exp->exp_obd->u.cli;
1592 OBD_ALLOC_PTR(cli->cl_seq);
1593 if (cli->cl_seq == NULL)
1596 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
1598 GOTO(out_free_seq, rc = -ENOMEM);
1600 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
1601 exp->exp_obd->obd_name);
1603 /* Init client side sequence-manager */
1604 rc = seq_client_init(cli->cl_seq, exp,
1605 LUSTRE_SEQ_METADATA,
1607 OBD_FREE(prefix, MAX_OBD_NAME + 5);
1609 GOTO(out_free_seq, rc);
1613 OBD_FREE_PTR(cli->cl_seq);
1618 static int mdc_fid_fini(struct obd_export *exp)
1620 struct client_obd *cli = &exp->exp_obd->u.cli;
1623 if (cli->cl_seq != NULL) {
1624 seq_client_fini(cli->cl_seq);
1625 OBD_FREE_PTR(cli->cl_seq);
1632 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1633 struct md_op_data *op_data)
1635 struct client_obd *cli = &exp->exp_obd->u.cli;
1636 struct lu_client_seq *seq = cli->cl_seq;
1638 RETURN(seq_client_alloc_fid(seq, fid));
1641 /* XXX This method is used only to clear current fid seq
1642 * once fld/mds insert failed */
1643 static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1645 struct client_obd *cli = &exp->exp_obd->u.cli;
1647 seq_client_flush(cli->cl_seq);
1651 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
1652 struct client_obd *cli = &exp->exp_obd->u.cli;
1653 return &cli->cl_target_uuid;
1656 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
1658 struct client_obd *cli = &obd->u.cli;
1659 struct lprocfs_static_vars lvars = { 0 };
1663 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1664 if (!cli->cl_rpc_lock)
1666 mdc_init_rpc_lock(cli->cl_rpc_lock);
1670 OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1671 if (!cli->cl_setattr_lock)
1672 GOTO(err_rpc_lock, rc = -ENOMEM);
1673 mdc_init_rpc_lock(cli->cl_setattr_lock);
1675 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1676 if (!cli->cl_close_lock)
1677 GOTO(err_setattr_lock, rc = -ENOMEM);
1678 mdc_init_rpc_lock(cli->cl_close_lock);
1680 rc = client_obd_setup(obd, cfg);
1682 GOTO(err_close_lock, rc);
1683 lprocfs_mdc_init_vars(&lvars);
1684 lprocfs_obd_setup(obd, lvars.obd_vars);
1685 sptlrpc_lprocfs_cliobd_attach(obd);
1686 ptlrpc_lprocfs_register_obd(obd);
1688 rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
1691 CERROR("failed to setup llogging subsystems\n");
1695 libcfs_klnl_start(LNL_TRANSPORT_HSM);
1696 libcfs_klnl_start(LNL_TRANSPORT_CHANGELOG);
1701 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1703 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1705 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1710 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
1711 * us to make MDS RPCs with large enough reply buffers to hold the
1712 * maximum-sized (= maximum striped) EA and cookie without having to
1713 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1714 static int mdc_init_ea_size(struct obd_export *exp, int easize,
1715 int def_easize, int cookiesize)
1717 struct obd_device *obd = exp->exp_obd;
1718 struct client_obd *cli = &obd->u.cli;
1721 if (cli->cl_max_mds_easize < easize)
1722 cli->cl_max_mds_easize = easize;
1724 if (cli->cl_default_mds_easize < def_easize)
1725 cli->cl_default_mds_easize = def_easize;
1727 if (cli->cl_max_mds_cookiesize < cookiesize)
1728 cli->cl_max_mds_cookiesize = cookiesize;
1733 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1739 case OBD_CLEANUP_EARLY:
1740 case OBD_CLEANUP_EXPORTS:
1741 /* If we set up but never connected, the
1742 client import will not have been cleaned. */
1743 if (obd->u.cli.cl_import) {
1744 struct obd_import *imp;
1745 down_write(&obd->u.cli.cl_sem);
1746 imp = obd->u.cli.cl_import;
1747 CERROR("client import never connected\n");
1748 ptlrpc_invalidate_import(imp);
1749 class_destroy_import(imp);
1750 up_write(&obd->u.cli.cl_sem);
1751 obd->u.cli.cl_import = NULL;
1753 rc = obd_llog_finish(obd, 0);
1755 CERROR("failed to cleanup llogging subsystems\n");
1761 static int mdc_cleanup(struct obd_device *obd)
1763 struct client_obd *cli = &obd->u.cli;
1765 libcfs_klnl_stop(LNL_TRANSPORT_HSM, LNL_GRP_HSM);
1766 libcfs_klnl_stop(LNL_TRANSPORT_CHANGELOG, 0);
1768 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1769 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1770 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1772 ptlrpc_lprocfs_unregister_obd(obd);
1773 lprocfs_obd_cleanup(obd);
1776 return client_obd_cleanup(obd);
1780 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
1781 struct obd_device *tgt, int *index)
1783 struct llog_ctxt *ctxt;
1787 LASSERT(olg == &obd->obd_olg);
1789 rc = llog_setup(obd, olg, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1794 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1795 llog_initiator_connect(ctxt);
1796 llog_ctxt_put(ctxt);
1798 rc = llog_setup(obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt, 0, NULL,
1801 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
1802 llog_initiator_connect(ctxt);
1803 llog_ctxt_put(ctxt);
1809 static int mdc_llog_finish(struct obd_device *obd, int count)
1811 struct llog_ctxt *ctxt;
1815 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1817 rc = llog_cleanup(ctxt);
1819 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
1821 rc = llog_cleanup(ctxt);
1826 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1828 struct lustre_cfg *lcfg = buf;
1829 struct lprocfs_static_vars lvars = { 0 };
1832 lprocfs_mdc_init_vars(&lvars);
1833 switch (lcfg->lcfg_command) {
1835 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
1845 /* get remote permission for current user on fid */
1846 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
1847 struct obd_capa *oc, __u32 suppgid,
1848 struct ptlrpc_request **request)
1850 struct ptlrpc_request *req;
1854 LASSERT(client_is_remote(exp));
1857 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1861 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1863 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1865 ptlrpc_request_free(req);
1869 mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
1871 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
1872 sizeof(struct mdt_remote_perm));
1874 ptlrpc_request_set_replen(req);
1876 rc = ptlrpc_queue_wait(req);
1878 ptlrpc_req_finished(req);
1884 static int mdc_interpret_renew_capa(const struct lu_env *env,
1885 struct ptlrpc_request *req, void *unused,
1888 struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
1889 renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
1890 struct mdt_body *body = NULL;
1891 struct lustre_capa *capa;
1895 GOTO(out, capa = ERR_PTR(status));
1897 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1899 GOTO(out, capa = ERR_PTR(-EFAULT));
1901 if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
1902 GOTO(out, capa = ERR_PTR(-ENOENT));
1904 capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
1906 GOTO(out, capa = ERR_PTR(-EFAULT));
1913 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
1916 struct ptlrpc_request *req;
1919 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
1920 LUSTRE_MDS_VERSION, MDS_GETATTR);
1924 /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
1925 * capa to renew is oss capa.
1927 mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
1928 ptlrpc_request_set_replen(req);
1930 req->rq_async_args.pointer_arg[0] = oc;
1931 req->rq_async_args.pointer_arg[1] = cb;
1932 req->rq_interpret_reply = mdc_interpret_renew_capa;
1933 ptlrpcd_add_req(req, PSCOPE_OTHER);
1937 static int mdc_connect(const struct lu_env *env,
1938 struct obd_export **exp,
1939 struct obd_device *obd, struct obd_uuid *cluuid,
1940 struct obd_connect_data *data,
1943 struct obd_import *imp = obd->u.cli.cl_import;
1945 /* mds-mds import features */
1946 if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
1947 spin_lock(&imp->imp_lock);
1948 imp->imp_server_timeout = 1;
1949 spin_unlock(&imp->imp_lock);
1950 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1951 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
1955 return client_connect_import(env, exp, obd, cluuid, data, NULL);
1958 struct obd_ops mdc_obd_ops = {
1959 .o_owner = THIS_MODULE,
1960 .o_setup = mdc_setup,
1961 .o_precleanup = mdc_precleanup,
1962 .o_cleanup = mdc_cleanup,
1963 .o_add_conn = client_import_add_conn,
1964 .o_del_conn = client_import_del_conn,
1965 .o_connect = mdc_connect,
1966 .o_disconnect = client_disconnect_export,
1967 .o_iocontrol = mdc_iocontrol,
1968 .o_set_info_async = mdc_set_info_async,
1969 .o_statfs = mdc_statfs,
1971 .o_unpin = mdc_unpin,
1972 .o_fid_init = mdc_fid_init,
1973 .o_fid_fini = mdc_fid_fini,
1974 .o_fid_alloc = mdc_fid_alloc,
1975 .o_fid_delete = mdc_fid_delete,
1976 .o_import_event = mdc_import_event,
1977 .o_llog_init = mdc_llog_init,
1978 .o_llog_finish = mdc_llog_finish,
1979 .o_get_info = mdc_get_info,
1980 .o_process_config = mdc_process_config,
1981 .o_get_uuid = mdc_get_uuid,
1984 struct md_ops mdc_md_ops = {
1985 .m_getstatus = mdc_getstatus,
1986 .m_change_cbdata = mdc_change_cbdata,
1987 .m_close = mdc_close,
1988 .m_create = mdc_create,
1989 .m_done_writing = mdc_done_writing,
1990 .m_enqueue = mdc_enqueue,
1991 .m_getattr = mdc_getattr,
1992 .m_getattr_name = mdc_getattr_name,
1993 .m_intent_lock = mdc_intent_lock,
1995 .m_is_subdir = mdc_is_subdir,
1996 .m_rename = mdc_rename,
1997 .m_setattr = mdc_setattr,
1998 .m_setxattr = mdc_setxattr,
1999 .m_getxattr = mdc_getxattr,
2001 .m_readpage = mdc_readpage,
2002 .m_unlink = mdc_unlink,
2003 .m_cancel_unused = mdc_cancel_unused,
2004 .m_init_ea_size = mdc_init_ea_size,
2005 .m_set_lock_data = mdc_set_lock_data,
2006 .m_lock_match = mdc_lock_match,
2007 .m_get_lustre_md = mdc_get_lustre_md,
2008 .m_free_lustre_md = mdc_free_lustre_md,
2009 .m_set_open_replay_data = mdc_set_open_replay_data,
2010 .m_clear_open_replay_data = mdc_clear_open_replay_data,
2011 .m_renew_capa = mdc_renew_capa,
2012 .m_unpack_capa = mdc_unpack_capa,
2013 .m_get_remote_perm = mdc_get_remote_perm,
2014 .m_intent_getattr_async = mdc_intent_getattr_async,
2015 .m_revalidate_lock = mdc_revalidate_lock
2018 int __init mdc_init(void)
2021 struct lprocfs_static_vars lvars = { 0 };
2022 lprocfs_mdc_init_vars(&lvars);
2024 request_module("lquota");
2025 quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
2026 init_obd_quota_ops(quota_interface, &mdc_obd_ops);
2028 rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
2029 LUSTRE_MDC_NAME, NULL);
2030 if (rc && quota_interface)
2031 PORTAL_SYMBOL_PUT(mdc_quota_interface);
2037 static void /*__exit*/ mdc_exit(void)
2039 if (quota_interface)
2040 PORTAL_SYMBOL_PUT(mdc_quota_interface);
2042 class_unregister_type(LUSTRE_MDC_NAME);
2045 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2046 MODULE_DESCRIPTION("Lustre Metadata Client");
2047 MODULE_LICENSE("GPL");
2049 module_init(mdc_init);
2050 module_exit(mdc_exit);