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,
60 struct lustre_capa *capa;
63 /* swabbed already in mdc_enqueue */
64 capa = lustre_msg_buf(req->rq_repmsg, offset, sizeof(*capa));
66 CERROR("missing capa at offset %d failed!\n", offset);
67 return ERR_PTR(-EFAULT);
70 oc = alloc_capa(CAPA_SITE_CLIENT);
72 CERROR("alloc capa failed!\n");
73 return ERR_PTR(-ENOMEM);
80 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
81 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
82 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
83 struct obd_capa **pc, int level, int msg_flags)
85 struct ptlrpc_request *req;
86 struct mdt_body *body;
87 int rc, size[3] = { sizeof(struct ptlrpc_body),
89 sizeof(struct lustre_capa) };
92 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_GETSTATUS, 2, size,
95 GOTO(out, rc = -ENOMEM);
97 req->rq_send_state = level;
98 ptlrpc_req_set_repsize(req, 3, size);
100 mdc_pack_req_body(req, REQ_REC_OFF, 0, NULL, NULL, 0, 0);
101 lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
102 rc = ptlrpc_queue_wait(req);
105 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
106 lustre_swab_mdt_body);
108 CERROR ("Can't extract mdt_body\n");
109 GOTO (out, rc = -EPROTO);
112 *rootfid = body->fid1;
114 if (body->valid & OBD_MD_FLMDSCAPA) {
117 oc = mdc_unpack_capa(req, REPLY_REC_OFF + 1);
119 GOTO(out, rc = PTR_ERR(oc));
123 CDEBUG(D_NET, "root fid="DFID", last_committed="LPU64
124 ", last_xid="LPU64"\n",
126 lustre_msg_get_last_committed(req->rq_repmsg),
127 lustre_msg_get_last_xid(req->rq_repmsg));
132 ptlrpc_req_finished(req);
136 /* This should be mdc_get_info("rootfid") */
137 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
138 struct obd_capa **pc)
140 return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
145 * This function now is known to always saying that it will receive 4 buffers
146 * from server. Even for cases when acl_size and md_size is zero, RPC header
147 * willcontain 4 fields and RPC itself will contain zero size fields. This is
148 * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
149 * and thus zero, it shirinks it, making zero size. The same story about
150 * md_size. And this is course of problem when client waits for smaller number
151 * of fields. This issue will be fixed later when client gets awar of RPC
154 static int mdc_getattr_common(struct obd_export *exp, unsigned int ea_size,
155 unsigned int acl_size, int mdscapa,
156 struct ptlrpc_request *req)
158 struct mdt_body *body;
160 int size[5] = { sizeof(struct ptlrpc_body),
164 sizeof(struct lustre_capa) };
168 /* Request message already built. */
170 CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
173 CDEBUG(D_INODE, "reserved %u bytes for ACL\n", acl_size);
175 ptlrpc_req_set_repsize(req, 5, size);
177 rc = ptlrpc_queue_wait(req);
181 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
182 lustre_swab_mdt_body);
184 CERROR ("Can't unpack mdt_body\n");
188 CDEBUG(D_NET, "mode: %o\n", body->mode);
190 offset = REPLY_REC_OFF + 1;
191 LASSERT_REPSWAB(req, offset);
192 if (body->eadatasize != 0) {
193 /* reply indicates presence of eadata; check it's there... */
194 eadata = lustre_msg_buf(req->rq_repmsg, offset++,
196 if (eadata == NULL) {
197 CERROR ("Missing/short eadata\n");
202 if (body->valid & OBD_MD_FLMODEASIZE) {
203 struct client_obd *cli = &exp->exp_obd->u.cli;
205 if (cli->cl_max_mds_easize < body->max_mdsize)
206 cli->cl_max_mds_easize = body->max_mdsize;
207 if (cli->cl_max_mds_cookiesize < body->max_cookiesize)
208 cli->cl_max_mds_cookiesize = body->max_cookiesize;
211 offset += !!body->aclsize;
213 if (body->valid & OBD_MD_FLMDSCAPA) {
214 struct lustre_capa *capa;
217 capa = lustre_unpack_capa(req->rq_repmsg, offset++);
219 CERROR("Missing/short client MDS capability\n");
227 int mdc_getattr(struct obd_export *exp, const struct lu_fid *fid,
228 struct obd_capa *oc, obd_valid valid, int ea_size,
229 struct ptlrpc_request **request)
231 struct ptlrpc_request *req;
232 int size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
233 int acl_size = 0, rc;
236 size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
239 * XXX: Do we need to make another request here? We just did a getattr
240 * to do the lookup in the first place.
242 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
243 MDS_GETATTR, 3, size, NULL);
245 GOTO(out, rc = -ENOMEM);
247 mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, oc, ea_size,
248 MDS_BFLAG_EXT_FLAGS/*request "new" flags(bug 9486)*/);
250 if (valid & OBD_MD_FLRMTPERM)
251 acl_size = sizeof(struct mdt_remote_perm);
253 /* Currently only root inode will call us with FLACL */
254 else if (valid & OBD_MD_FLACL)
255 acl_size = LUSTRE_POSIX_ACL_MAX_SIZE;
257 rc = mdc_getattr_common(exp, ea_size, acl_size,
258 !!(valid & OBD_MD_FLMDSCAPA), req);
260 ptlrpc_req_finished (req);
268 int mdc_getattr_name(struct obd_export *exp, const struct lu_fid *fid,
269 struct obd_capa *oc, const char *filename, int namelen,
270 obd_valid valid, int ea_size,
271 struct ptlrpc_request **request)
273 struct ptlrpc_request *req;
274 struct mdt_body *body;
275 int size[4] = { sizeof(struct ptlrpc_body), sizeof(*body), 0, namelen};
279 size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
281 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
282 MDS_GETATTR_NAME, 4, size, NULL);
284 GOTO(out, rc = -ENOMEM);
286 mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, oc, ea_size,
287 MDS_BFLAG_EXT_FLAGS/*request "new" flags(bug 9486)*/);
290 LASSERT(strnlen(filename, namelen) == namelen - 1);
291 memcpy(lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2, namelen),
295 rc = mdc_getattr_common(exp, ea_size, 0, !!(valid & OBD_MD_FLMDSCAPA),
298 ptlrpc_req_finished (req);
306 static int mdc_is_subdir(struct obd_export *exp, const struct lu_fid *pfid,
307 const struct lu_fid *cfid, struct ptlrpc_request **request)
309 int size[2] = { sizeof(struct ptlrpc_body),
310 sizeof(struct mdt_body) };
311 struct ptlrpc_request *req;
312 struct mdt_body *body;
316 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
317 MDS_IS_SUBDIR, 2, size, NULL);
319 GOTO(out, rc = -ENOMEM);
321 mdc_is_subdir_pack(req, REQ_REC_OFF, pfid, cfid, 0);
323 ptlrpc_req_set_repsize(req, 2, size);
324 rc = ptlrpc_queue_wait(req);
325 if (rc != 0 && rc != -EREMOTE)
328 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
329 lustre_swab_mdt_body);
331 CERROR ("Can't unpack mdt_body\n");
332 GOTO(out, rc = -EPROTO);
341 int mdc_xattr_common(struct obd_export *exp, const struct lu_fid *fid,
343 int opcode, obd_valid valid, const char *xattr_name,
344 const char *input, int input_size, int output_size,
345 int flags, struct ptlrpc_request **request)
347 struct ptlrpc_request *req;
348 int size[5] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
349 int bufcnt = 3, offset = REQ_REC_OFF + 2;
350 int rc, xattr_namelen = 0, remote_acl = 0;
354 size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
356 xattr_namelen = strlen(xattr_name) + 1;
357 size[bufcnt++] = xattr_namelen;
361 size[bufcnt++] = input_size;
364 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
365 opcode, bufcnt, size, NULL);
367 GOTO(out, rc = -ENOMEM);
370 mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, oc, output_size, flags);
374 tmp = lustre_msg_buf(req->rq_reqmsg, offset++, xattr_namelen);
375 memcpy(tmp, xattr_name, xattr_namelen);
376 if (!strcmp(xattr_name, XATTR_NAME_LUSTRE_ACL))
380 tmp = lustre_msg_buf(req->rq_reqmsg, offset++, input_size);
381 memcpy(tmp, input, input_size);
385 if (opcode == MDS_GETXATTR) {
386 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
392 /* we do this even output_size is 0, because server is doing that */
393 size[bufcnt++] = output_size;
394 ptlrpc_req_set_repsize(req, bufcnt, size);
397 /* NB: set remote acl doesn't need hold rpc lock, because it just
398 * send command to MDS, and when it's executed on mountpoint on MDS,
399 * another mdc_xattr_common() will be invoked there. */
400 if (opcode == MDS_SETXATTR && !remote_acl)
401 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
403 rc = ptlrpc_queue_wait(req);
405 if (opcode == MDS_SETXATTR && !remote_acl)
406 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
411 if (opcode == MDS_GETXATTR) {
412 struct mdt_body * body = lustre_swab_repbuf(req, REPLY_REC_OFF,
413 sizeof(*body), lustre_swab_mdt_body);
415 CERROR ("Can't unpack mdt_body\n");
416 GOTO(err_out, rc = -EPROTO);
423 ptlrpc_req_finished(req);
428 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
429 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
430 const char *input, int input_size, int output_size, int flags,
431 struct ptlrpc_request **request)
433 return mdc_xattr_common(exp, fid, oc, MDS_SETXATTR, valid, xattr_name,
434 input, input_size, output_size, flags, request);
437 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
438 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
439 const char *input, int input_size, int output_size, int flags,
440 struct ptlrpc_request **request)
442 return mdc_xattr_common(exp, fid, oc, MDS_GETXATTR, valid, xattr_name,
443 input, input_size, output_size, flags, request);
446 #ifdef CONFIG_FS_POSIX_ACL
448 int mdc_unpack_acl(struct obd_export *exp, struct ptlrpc_request *req,
449 struct lustre_md *md, unsigned int offset)
451 struct mdt_body *body = md->body;
452 struct posix_acl *acl;
459 buf = lustre_msg_buf(req->rq_repmsg, offset, body->aclsize);
461 CERROR("aclsize %u, bufcount %u, bufsize %u\n",
462 body->aclsize, lustre_msg_bufcount(req->rq_repmsg),
463 (lustre_msg_bufcount(req->rq_repmsg) <= offset) ?
464 -1 : lustre_msg_buflen(req->rq_repmsg, offset));
468 acl = posix_acl_from_xattr(buf, body->aclsize);
471 CERROR("convert xattr to acl: %d\n", rc);
475 rc = posix_acl_valid(acl);
477 CERROR("validate acl: %d\n", rc);
478 posix_acl_release(acl);
486 #define mdc_unpack_acl(exp, req, md, offset) 0
489 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
490 int offset, struct obd_export *dt_exp,
491 struct obd_export *md_exp,
492 struct lustre_md *md)
498 memset(md, 0, sizeof(*md));
500 md->body = lustre_msg_buf(req->rq_repmsg, offset, sizeof (*md->body));
501 LASSERT (md->body != NULL);
502 LASSERT_REPSWABBED(req, offset);
505 if (md->body->valid & OBD_MD_FLEASIZE) {
507 struct lov_mds_md *lmm;
509 LASSERT(S_ISREG(md->body->mode));
511 if (md->body->eadatasize == 0) {
512 CERROR("OBD_MD_FLEASIZE set, but eadatasize 0\n");
515 lmmsize = md->body->eadatasize;
516 lmm = lustre_msg_buf(req->rq_repmsg, offset, lmmsize);
517 LASSERT (lmm != NULL);
518 LASSERT_REPSWABBED(req, offset);
520 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
524 LASSERT (rc >= sizeof (*md->lsm));
526 } else if (md->body->valid & OBD_MD_FLDIREA) {
528 struct lov_mds_md *lmv;
530 LASSERT(S_ISDIR(md->body->mode));
532 if (md->body->eadatasize == 0) {
533 CERROR("OBD_MD_FLDIREA is set, but eadatasize 0\n");
536 if (md->body->valid & OBD_MD_MEA) {
537 lmvsize = md->body->eadatasize;
538 lmv = lustre_msg_buf(req->rq_repmsg, offset, lmvsize);
539 LASSERT (lmv != NULL);
540 LASSERT_REPSWABBED(req, offset);
542 rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
547 LASSERT (rc >= sizeof (*md->mea));
553 /* remote permission */
554 if (md->body->valid & OBD_MD_FLRMTPERM) {
555 md->remote_perm = lustre_msg_buf(req->rq_repmsg, offset++,
556 sizeof(struct mdt_remote_perm));
557 LASSERT(md->remote_perm);
560 /* for ACL, it's possible that FLACL is set but aclsize is zero. only
561 * when aclsize != 0 there's an actual segment for ACL in reply
563 else if (md->body->valid & OBD_MD_FLACL) {
564 if (md->body->aclsize) {
565 rc = mdc_unpack_acl(dt_exp, req, md, offset++);
568 #ifdef CONFIG_FS_POSIX_ACL
570 md->posix_acl = NULL;
575 if (md->body->valid & OBD_MD_FLMDSCAPA) {
576 struct obd_capa *oc = mdc_unpack_capa(req, offset++);
579 GOTO(out, rc = PTR_ERR(oc));
583 if (md->body->valid & OBD_MD_FLOSSCAPA) {
584 struct obd_capa *oc = mdc_unpack_capa(req, offset++);
587 GOTO(out, rc = PTR_ERR(oc));
595 free_capa(md->oss_capa);
597 free_capa(md->mds_capa);
598 #ifdef CONFIG_FS_POSIX_ACL
599 posix_acl_release(md->posix_acl);
602 obd_free_memmd(dt_exp, &md->lsm);
607 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
613 static void mdc_replay_open(struct ptlrpc_request *req)
615 struct md_open_data *mod = req->rq_cb_data;
616 struct ptlrpc_request *cur, *tmp;
617 struct obd_client_handle *och;
618 struct lustre_handle old;
619 struct mdt_body *body;
623 DEBUG_REQ(D_ERROR, req,
624 "Can't properly replay without open data.");
629 body = lustre_swab_repbuf(req, DLM_REPLY_REC_OFF, sizeof(*body),
630 lustre_swab_mdt_body);
631 LASSERT(body != NULL);
635 struct lustre_handle *file_fh;
637 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
639 file_fh = &och->och_fh;
640 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
641 file_fh->cookie, body->handle.cookie);
643 *file_fh = body->handle;
646 list_for_each_entry_safe(cur, tmp, &mod->mod_replay_list, rq_mod_list) {
647 int opc = lustre_msg_get_opc(cur->rq_reqmsg);
648 struct mdt_epoch *epoch = NULL;
650 if (opc == MDS_CLOSE || opc == MDS_DONE_WRITING) {
651 epoch = lustre_msg_buf(cur->rq_reqmsg,
652 REQ_REC_OFF, sizeof(*epoch));
654 DEBUG_REQ(D_HA, cur, "updating %s body with new fh",
655 opc == MDS_CLOSE ? "CLOSE" : "DONE_WRITING");
656 } else if (opc == MDS_REINT) {
657 struct mdt_rec_setattr *rec;
659 /* Check this is REINT_SETATTR. */
660 rec = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF,
662 LASSERT(rec && rec->sa_opcode == REINT_SETATTR);
664 epoch = lustre_msg_buf(cur->rq_reqmsg,
665 REQ_REC_OFF + 2, sizeof(*epoch));
667 DEBUG_REQ(D_HA, cur, "updating REINT_SETATTR body "
672 LASSERT(!memcmp(&old, &epoch->handle,
674 epoch->handle = body->handle;
680 void mdc_commit_delayed(struct ptlrpc_request *req)
682 struct md_open_data *mod = req->rq_cb_data;
683 struct ptlrpc_request *cur, *tmp;
685 DEBUG_REQ(D_HA, req, "req committed");
690 req->rq_cb_data = NULL;
691 req->rq_commit_cb = NULL;
692 list_del_init(&req->rq_mod_list);
693 if (req->rq_sequence) {
694 list_for_each_entry_safe(cur, tmp, &mod->mod_replay_list,
697 LASSERT(cur != LP_POISON);
698 LASSERT(cur->rq_type != LI_POISON);
699 DEBUG_REQ(D_HA, cur, "req balanced");
700 LASSERT(cur->rq_transno != 0);
701 LASSERT(cur->rq_import == req->rq_import);
703 list_del_init(&cur->rq_mod_list);
704 /* We no longer want to preserve this for transno-
705 * unconditional replay. */
706 spin_lock(&cur->rq_lock);
708 spin_unlock(&cur->rq_lock);
712 if (list_empty(&mod->mod_replay_list)) {
713 if (mod->mod_och != NULL)
714 mod->mod_och->och_mod = NULL;
720 int mdc_set_open_replay_data(struct obd_export *exp,
721 struct obd_client_handle *och,
722 struct ptlrpc_request *open_req)
724 struct md_open_data *mod;
725 struct mdt_rec_create *rec = lustre_msg_buf(open_req->rq_reqmsg,
728 struct mdt_body *body = lustre_msg_buf(open_req->rq_repmsg,
731 struct obd_import *imp = open_req->rq_import;
734 LASSERT(rec != NULL);
736 /* Incoming message in my byte order (it's been swabbed). */
737 LASSERT_REPSWABBED(open_req, DLM_REPLY_REC_OFF);
739 /* Outgoing messages always in my byte order. */
740 LASSERT(body != NULL);
742 /*Only the import is replayable, we set replay_open data */
743 if (och && imp->imp_replayable) {
744 OBD_ALLOC(mod, sizeof(*mod));
746 DEBUG_REQ(D_ERROR, open_req,
747 "Can't allocate md_open_data");
750 CFS_INIT_LIST_HEAD(&mod->mod_replay_list);
752 spin_lock(&open_req->rq_lock);
753 if (!open_req->rq_replay) {
754 OBD_FREE(mod, sizeof(*mod));
755 spin_unlock(&open_req->rq_lock);
761 open_req->rq_cb_data = mod;
762 list_add_tail(&open_req->rq_mod_list, &mod->mod_replay_list);
763 open_req->rq_commit_cb = mdc_commit_delayed;
764 spin_unlock(&open_req->rq_lock);
767 rec->cr_fid2 = body->fid1;
768 rec->cr_ioepoch = body->ioepoch;
769 rec->cr_old_handle.cookie = body->handle.cookie;
770 open_req->rq_replay_cb = mdc_replay_open;
771 if (!fid_is_sane(&body->fid1)) {
772 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
777 DEBUG_REQ(D_HA, open_req, "Set up open replay data");
781 int mdc_clear_open_replay_data(struct obd_export *exp,
782 struct obd_client_handle *och)
784 struct md_open_data *mod = och->och_mod;
788 * Don't free the structure now (it happens in mdc_commit_delayed(),
789 * after the last request is removed from its replay list),
790 * but make sure that replay doesn't poke at the och, which is about to
793 LASSERT(mod != LP_POISON);
801 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
802 struct md_open_data *mod, struct ptlrpc_request **request)
804 struct obd_device *obd = class_exp2obd(exp);
805 int reqsize[4] = { sizeof(struct ptlrpc_body),
806 sizeof(struct mdt_epoch),
807 sizeof(struct mdt_rec_setattr)};
808 int repsize[4] = { sizeof(struct ptlrpc_body),
809 sizeof(struct mdt_body),
810 obd->u.cli.cl_max_mds_easize,
811 obd->u.cli.cl_max_mds_cookiesize };
812 struct ptlrpc_request *req;
816 reqsize[REQ_REC_OFF + 2] = op_data->op_capa1 ?
817 sizeof(struct lustre_capa) : 0;
818 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
819 MDS_CLOSE, 4, reqsize, NULL);
821 GOTO(out, rc = -ENOMEM);
823 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
824 * portal whose threads are not taking any DLM locks and are therefore
825 * always progressing */
826 /* XXX FIXME bug 249 */
827 req->rq_request_portal = MDS_READPAGE_PORTAL;
829 /* Ensure that this close's handle is fixed up during replay. */
830 if (likely(mod != NULL))
831 list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
833 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
835 mdc_close_pack(req, REQ_REC_OFF, op_data);
836 ptlrpc_req_set_repsize(req, 4, repsize);
837 req->rq_commit_cb = mdc_commit_delayed;
839 LASSERT(req->rq_cb_data == NULL);
840 req->rq_cb_data = mod;
842 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
843 rc = ptlrpc_queue_wait(req);
844 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
846 if (req->rq_repmsg == NULL) {
847 CDEBUG(D_HA, "request failed to send: %p, %d\n", req,
850 rc = req->rq_status ? req->rq_status : -EIO;
851 } else if (rc == 0 || rc == -EAGAIN) {
852 rc = lustre_msg_get_status(req->rq_repmsg);
853 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
854 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
858 } else if (mod == NULL) {
859 if (req->rq_import->imp_replayable)
860 CERROR("Unexpected: can't find md_open_data,"
861 "but close succeeded with replayable imp"
862 "Please tell CFS.\n");
864 if (!lustre_swab_repbuf(req, REPLY_REC_OFF,
865 sizeof(struct mdt_body),
866 lustre_swab_mdt_body)) {
867 CERROR("Error unpacking mdt_body\n");
875 if (rc != 0 && rc != -EAGAIN && req && req->rq_commit_cb)
876 req->rq_commit_cb(req);
881 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
882 struct md_open_data *mod)
884 struct obd_device *obd = class_exp2obd(exp);
885 struct ptlrpc_request *req;
886 int size[4] = { sizeof(struct ptlrpc_body),
887 sizeof(struct mdt_epoch),
888 sizeof(struct mdt_rec_setattr)};
889 int repsize[2] = { sizeof(struct ptlrpc_body),
890 sizeof(struct mdt_body)};
894 if (op_data->op_capa1)
895 size[REQ_REC_OFF + 2] = sizeof(struct lustre_capa);
896 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
897 MDS_DONE_WRITING, 4, size, NULL);
901 mdc_close_pack(req, REQ_REC_OFF, op_data);
904 req->rq_cb_data = mod;
905 req->rq_commit_cb = mdc_commit_delayed;
906 if (likely(mod != NULL))
907 list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
909 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
911 ptlrpc_req_set_repsize(req, 2, repsize);
912 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
913 rc = ptlrpc_queue_wait(req);
914 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
916 /* Close the open replay sequence if an error occured or no SOM
917 * attribute update is needed. */
919 ptlrpc_close_replay_seq(req);
921 if (rc && rc != -EAGAIN && req->rq_commit_cb)
922 req->rq_commit_cb(req);
924 ptlrpc_req_finished(req);
928 #ifdef HAVE_SPLIT_SUPPORT
929 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
930 const struct page *page, int offset)
932 int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
933 struct obd_import *imp = class_exp2cliimp(exp);
934 struct ptlrpc_bulk_desc *desc = NULL;
935 struct ptlrpc_request *req = NULL;
938 CDEBUG(D_INODE, "object: "DFID"\n", PFID(fid));
940 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_WRITEPAGE, 3,
943 GOTO(out, rc = -ENOMEM);
945 req->rq_request_portal = MDS_READPAGE_PORTAL;
947 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
949 GOTO(out, rc = -ENOMEM);
951 /* NB req now owns desc and will free it when it gets freed. */
952 ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
953 mdc_readdir_pack(req, REQ_REC_OFF, 0, offset, fid, NULL);
955 ptlrpc_req_set_repsize(req, 2, size);
956 rc = ptlrpc_queue_wait(req);
960 ptlrpc_req_finished(req);
963 EXPORT_SYMBOL(mdc_sendpage);
966 int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
967 struct obd_capa *oc, __u64 offset, struct page *page,
968 struct ptlrpc_request **request)
970 int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
971 struct obd_import *imp = class_exp2cliimp(exp);
972 struct ptlrpc_bulk_desc *desc = NULL;
973 struct ptlrpc_request *req = NULL;
974 struct mdt_body *body;
977 CDEBUG(D_INODE, "object: "DFID"\n", PFID(fid));
979 size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
980 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE, 3, size,
983 GOTO(out, rc = -ENOMEM);
985 /* XXX FIXME bug 249 */
986 req->rq_request_portal = MDS_READPAGE_PORTAL;
988 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
990 GOTO(out, rc = -ENOMEM);
992 /* NB req now owns desc and will free it when it gets freed */
993 ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
994 mdc_readdir_pack(req, REQ_REC_OFF, offset, CFS_PAGE_SIZE, fid, oc);
996 ptlrpc_req_set_repsize(req, 2, size);
997 rc = ptlrpc_queue_wait(req);
1000 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1001 lustre_swab_mdt_body);
1003 CERROR("Can't unpack mdt_body\n");
1004 GOTO(out, rc = -EPROTO);
1007 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
1008 CERROR ("Unexpected # bytes transferred: %d"
1009 " (%ld expected)\n",
1010 req->rq_bulk->bd_nob_transferred,
1012 GOTO(out, rc = -EPROTO);
1022 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1023 void *karg, void *uarg)
1025 struct obd_device *obd = exp->exp_obd;
1026 struct obd_ioctl_data *data = karg;
1027 struct obd_import *imp = obd->u.cli.cl_import;
1028 struct llog_ctxt *ctxt;
1032 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1035 if (!try_module_get(THIS_MODULE)) {
1036 CERROR("Can't get module. Is it alive?");
1041 case OBD_IOC_CLIENT_RECOVER:
1042 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
1046 case IOC_OSC_SET_ACTIVE:
1047 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1049 case OBD_IOC_PARSE: {
1050 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1051 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
1055 case OBD_IOC_LLOG_INFO:
1056 case OBD_IOC_LLOG_PRINT: {
1057 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1058 rc = llog_ioctl(ctxt, cmd, data);
1063 case OBD_IOC_POLL_QUOTACHECK:
1064 rc = lquota_poll_check(quota_interface, exp,
1065 (struct if_quotacheck *)karg);
1068 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1069 GOTO(out, rc = -ENOTTY);
1072 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1075 module_put(THIS_MODULE);
1081 int mdc_set_info_async(struct obd_export *exp, obd_count keylen,
1082 void *key, obd_count vallen, void *val,
1083 struct ptlrpc_request_set *set)
1085 struct obd_import *imp = class_exp2cliimp(exp);
1088 if (KEY_IS(KEY_INIT_RECOV)) {
1089 if (vallen != sizeof(int))
1091 spin_lock(&imp->imp_lock);
1092 imp->imp_initial_recov = *(int *)val;
1093 spin_unlock(&imp->imp_lock);
1094 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
1095 exp->exp_obd->obd_name, imp->imp_initial_recov);
1098 /* Turn off initial_recov after we try all backup servers once */
1099 if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
1100 if (vallen != sizeof(int))
1102 spin_lock(&imp->imp_lock);
1103 imp->imp_initial_recov_bk = *(int *)val;
1104 if (imp->imp_initial_recov_bk)
1105 imp->imp_initial_recov = 1;
1106 spin_unlock(&imp->imp_lock);
1107 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
1108 exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
1111 if (KEY_IS(KEY_READ_ONLY)) {
1112 struct ptlrpc_request *req;
1113 int size[3] = { sizeof(struct ptlrpc_body), keylen, vallen };
1114 char *bufs[3] = { NULL, key, val };
1116 if (vallen != sizeof(int))
1119 spin_lock(&imp->imp_lock);
1120 if (*((int *)val)) {
1121 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
1122 imp->imp_connect_data.ocd_connect_flags |=
1125 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
1126 imp->imp_connect_data.ocd_connect_flags &=
1127 ~OBD_CONNECT_RDONLY;
1129 spin_unlock(&imp->imp_lock);
1131 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_SET_INFO,
1136 ptlrpc_req_set_repsize(req, 1, NULL);
1139 ptlrpc_set_add_req(set, req);
1140 ptlrpc_check_set(set);
1142 rc = ptlrpc_queue_wait(req);
1143 ptlrpc_req_finished(req);
1148 if (KEY_IS(KEY_FLUSH_CTX)) {
1149 sptlrpc_import_flush_my_ctx(imp);
1152 if (KEY_IS(KEY_MDS_CONN)) {
1153 struct obd_import *imp = class_exp2cliimp(exp);
1155 /* mds-mds import */
1156 spin_lock(&imp->imp_lock);
1157 imp->imp_server_timeout = 1;
1158 spin_unlock(&imp->imp_lock);
1159 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1160 CDEBUG(D_OTHER|D_WARNING, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1167 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
1168 __u32 *vallen, void *val)
1172 if (KEY_IS(KEY_MAX_EASIZE)) {
1173 int mdsize, *max_easize;
1175 if (*vallen != sizeof(int))
1177 mdsize = *(int*)val;
1178 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
1179 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
1181 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
1184 if (KEY_IS(KEY_CONN_DATA)) {
1185 struct obd_import *imp = class_exp2cliimp(exp);
1186 struct obd_connect_data *data = val;
1188 if (*vallen != sizeof(*data))
1191 *data = imp->imp_connect_data;
1198 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1201 struct ptlrpc_request *req;
1202 struct obd_statfs *msfs;
1203 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*msfs) };
1206 /* We could possibly pass max_age in the request (as an absolute
1207 * timestamp or a "seconds.usec ago") so the target can avoid doing
1208 * extra calls into the filesystem if that isn't necessary (e.g.
1209 * during mount that would help a bit). Having relative timestamps
1210 * is not so great if request processing is slow, while absolute
1211 * timestamps are not ideal because they need time synchronization. */
1212 req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
1213 MDS_STATFS, 1, NULL, NULL);
1217 ptlrpc_req_set_repsize(req, 2, size);
1219 rc = ptlrpc_queue_wait(req);
1222 /* check connection error first */
1223 if (obd->u.cli.cl_import->imp_connect_error)
1224 rc = obd->u.cli.cl_import->imp_connect_error;
1229 msfs = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*msfs),
1230 lustre_swab_obd_statfs);
1232 CERROR("Can't unpack obd_statfs\n");
1233 GOTO(out, rc = -EPROTO);
1236 memcpy(osfs, msfs, sizeof(*msfs));
1239 ptlrpc_req_finished(req);
1244 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
1245 struct obd_capa *oc,
1246 struct obd_client_handle *handle, int flag)
1248 struct ptlrpc_request *req;
1249 struct mdt_body *body;
1250 int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1253 size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1254 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1255 MDS_PIN, 3, size, NULL);
1259 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof (*body));
1262 mdc_pack_capa(req, REQ_REC_OFF + 1, oc);
1264 ptlrpc_req_set_repsize(req, 2, size);
1266 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1267 rc = ptlrpc_queue_wait(req);
1268 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1270 CERROR("pin failed: %d\n", rc);
1271 ptlrpc_req_finished(req);
1275 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1276 lustre_swab_mdt_body);
1278 ptlrpc_req_finished(req);
1282 memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
1283 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1285 OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
1286 if (handle->och_mod == NULL) {
1287 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
1291 /* will be dropped by unpin */
1292 CFS_INIT_LIST_HEAD(&handle->och_mod->mod_replay_list);
1293 list_add_tail(&req->rq_mod_list, &handle->och_mod->mod_replay_list);
1298 static int mdc_unpin(struct obd_export *exp,
1299 struct obd_client_handle *handle, int flag)
1301 struct ptlrpc_request *req;
1302 struct mdt_body *body;
1303 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1306 if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
1309 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1310 MDS_CLOSE, 2, size, NULL);
1314 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1315 memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
1318 ptlrpc_req_set_repsize(req, 1, NULL);
1319 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1320 rc = ptlrpc_queue_wait(req);
1321 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1324 CERROR("unpin failed: %d\n", rc);
1326 ptlrpc_req_finished(req);
1328 LASSERT(!list_empty(&handle->och_mod->mod_replay_list));
1329 req = list_entry(handle->och_mod->mod_replay_list.next,
1330 typeof(*req), rq_mod_list);
1331 list_del_init(&req->rq_mod_list);
1332 ptlrpc_req_finished(req);
1333 LASSERT(list_empty(&handle->och_mod->mod_replay_list));
1335 OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1339 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
1340 struct obd_capa *oc, struct ptlrpc_request **request)
1342 struct ptlrpc_request *req;
1343 int size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
1347 size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1348 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1349 MDS_SYNC, 3, size, NULL);
1351 RETURN(rc = -ENOMEM);
1353 mdc_pack_req_body(req, REQ_REC_OFF, 0, fid, oc, 0, 0);
1355 ptlrpc_req_set_repsize(req, 2, size);
1357 rc = ptlrpc_queue_wait(req);
1358 if (rc || request == NULL)
1359 ptlrpc_req_finished(req);
1366 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1367 enum obd_import_event event)
1371 LASSERT(imp->imp_obd == obd);
1374 case IMP_EVENT_DISCON: {
1376 /* XXX Pass event up to OBDs stack. used only for FLD now */
1377 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
1381 case IMP_EVENT_INACTIVE: {
1382 struct client_obd *cli = &obd->u.cli;
1384 * Flush current sequence to make client obtain new one
1385 * from server in case of disconnect/reconnect.
1386 * If range is already empty then no need to flush it.
1388 if (cli->cl_seq != NULL &&
1389 !range_is_exhausted(&cli->cl_seq->lcs_space)) {
1390 seq_client_flush(cli->cl_seq);
1393 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1396 case IMP_EVENT_INVALIDATE: {
1397 struct ldlm_namespace *ns = obd->obd_namespace;
1399 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1403 case IMP_EVENT_ACTIVE: {
1404 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1408 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
1412 CERROR("Unknown import event %x\n", event);
1418 static int mdc_fid_init(struct obd_export *exp)
1420 struct client_obd *cli = &exp->exp_obd->u.cli;
1425 OBD_ALLOC_PTR(cli->cl_seq);
1426 if (cli->cl_seq == NULL)
1429 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
1431 GOTO(out_free_seq, rc = -ENOMEM);
1433 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
1434 exp->exp_obd->obd_name);
1436 /* Init client side sequence-manager */
1437 rc = seq_client_init(cli->cl_seq, exp,
1438 LUSTRE_SEQ_METADATA,
1440 OBD_FREE(prefix, MAX_OBD_NAME + 5);
1442 GOTO(out_free_seq, rc);
1446 OBD_FREE_PTR(cli->cl_seq);
1451 static int mdc_fid_fini(struct obd_export *exp)
1453 struct client_obd *cli = &exp->exp_obd->u.cli;
1456 if (cli->cl_seq != NULL) {
1457 seq_client_fini(cli->cl_seq);
1458 OBD_FREE_PTR(cli->cl_seq);
1465 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1466 struct md_op_data *op_data)
1468 struct client_obd *cli = &exp->exp_obd->u.cli;
1469 struct lu_client_seq *seq = cli->cl_seq;
1471 RETURN(seq_client_alloc_fid(seq, fid));
1474 /* XXX This method is used only to clear current fid seq
1475 * once fld/mds insert failed */
1476 static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1478 struct client_obd *cli = &exp->exp_obd->u.cli;
1480 seq_client_flush(cli->cl_seq);
1484 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
1486 struct client_obd *cli = &obd->u.cli;
1487 struct lprocfs_static_vars lvars;
1491 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1492 if (!cli->cl_rpc_lock)
1494 mdc_init_rpc_lock(cli->cl_rpc_lock);
1498 OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1499 if (!cli->cl_setattr_lock)
1500 GOTO(err_rpc_lock, rc = -ENOMEM);
1501 mdc_init_rpc_lock(cli->cl_setattr_lock);
1503 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1504 if (!cli->cl_close_lock)
1505 GOTO(err_setattr_lock, rc = -ENOMEM);
1506 mdc_init_rpc_lock(cli->cl_close_lock);
1508 rc = client_obd_setup(obd, cfg);
1510 GOTO(err_close_lock, rc);
1511 lprocfs_init_vars(mdc, &lvars);
1512 lprocfs_obd_setup(obd, lvars.obd_vars);
1513 ptlrpc_lprocfs_register_obd(obd);
1515 rc = obd_llog_init(obd, NULL, obd, 0, NULL, NULL);
1518 CERROR("failed to setup llogging subsystems\n");
1524 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1526 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1528 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1533 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
1534 * us to make MDS RPCs with large enough reply buffers to hold the
1535 * maximum-sized (= maximum striped) EA and cookie without having to
1536 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1537 int mdc_init_ea_size(struct obd_export *exp, int easize,
1538 int def_easize, int cookiesize)
1540 struct obd_device *obd = exp->exp_obd;
1541 struct client_obd *cli = &obd->u.cli;
1544 if (cli->cl_max_mds_easize < easize)
1545 cli->cl_max_mds_easize = easize;
1547 if (cli->cl_default_mds_easize < def_easize)
1548 cli->cl_default_mds_easize = def_easize;
1550 if (cli->cl_max_mds_cookiesize < cookiesize)
1551 cli->cl_max_mds_cookiesize = cookiesize;
1556 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1562 case OBD_CLEANUP_EARLY:
1563 case OBD_CLEANUP_EXPORTS:
1564 /* If we set up but never connected, the
1565 client import will not have been cleaned. */
1566 if (obd->u.cli.cl_import) {
1567 struct obd_import *imp;
1568 imp = obd->u.cli.cl_import;
1569 CERROR("client import never connected\n");
1570 ptlrpc_invalidate_import(imp);
1571 ptlrpc_free_rq_pool(imp->imp_rq_pool);
1572 class_destroy_import(imp);
1573 obd->u.cli.cl_import = NULL;
1576 case OBD_CLEANUP_SELF_EXP:
1577 rc = obd_llog_finish(obd, 0);
1579 CERROR("failed to cleanup llogging subsystems\n");
1580 case OBD_CLEANUP_OBD:
1586 static int mdc_cleanup(struct obd_device *obd)
1588 struct client_obd *cli = &obd->u.cli;
1590 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1591 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1592 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1594 ptlrpc_lprocfs_unregister_obd(obd);
1595 lprocfs_obd_cleanup(obd);
1598 return client_obd_cleanup(obd);
1602 static int mdc_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1603 struct obd_device *tgt,
1604 int count, struct llog_catid *logid,
1605 struct obd_uuid *uuid)
1607 struct llog_ctxt *ctxt;
1611 rc = llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1614 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1615 ctxt->loc_imp = obd->u.cli.cl_import;
1618 rc = llog_setup(obd, llogs, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1621 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1622 ctxt->loc_imp = obd->u.cli.cl_import;
1628 static int mdc_llog_finish(struct obd_device *obd, int count)
1633 rc = llog_cleanup(llog_get_context(obd, LLOG_LOVEA_REPL_CTXT));
1635 CERROR("can not cleanup LLOG_CONFIG_REPL_CTXT rc %d\n", rc);
1637 rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
1641 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1643 struct lustre_cfg *lcfg = buf;
1644 struct lprocfs_static_vars lvars;
1647 lprocfs_init_vars(mdc, &lvars);
1649 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars, lcfg, obd);
1653 /* get remote permission for current user on fid */
1654 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
1655 struct obd_capa *oc, struct ptlrpc_request **request)
1657 struct ptlrpc_request *req;
1658 struct mdt_body *body;
1659 struct mdt_remote_perm *perm;
1660 int size[5] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1664 size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1667 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1668 MDS_GETATTR, 3, size, NULL);
1672 mdc_pack_req_body(req, REQ_REC_OFF, OBD_MD_FLRMTPERM, fid, oc, 0, 0);
1674 size[REPLY_REC_OFF + 1] = sizeof(*perm);
1675 ptlrpc_req_set_repsize(req, 5, size);
1676 rc = ptlrpc_queue_wait(req);
1678 ptlrpc_req_finished(req);
1682 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1683 lustre_swab_mdt_body);
1685 LASSERT(body->valid & OBD_MD_FLRMTPERM);
1687 perm = lustre_swab_repbuf(req, REPLY_REC_OFF + 1, sizeof(*perm),
1688 lustre_swab_mdt_remote_perm);
1695 static int mdc_interpret_renew_capa(struct ptlrpc_request *req, void *unused,
1698 struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
1699 renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
1700 struct mdt_body *body = NULL;
1701 struct lustre_capa *capa;
1705 GOTO(out, capa = ERR_PTR(status));
1707 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1708 lustre_swab_mdt_body);
1710 GOTO(out, capa = ERR_PTR(-EFAULT));
1712 if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
1713 GOTO(out, capa = ERR_PTR(-ENOENT));
1715 capa = lustre_unpack_capa(req->rq_repmsg, REPLY_REC_OFF + 1);
1717 GOTO(out, capa = ERR_PTR(-EFAULT));
1724 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
1727 struct ptlrpc_request *req;
1728 int size[5] = { sizeof(struct ptlrpc_body),
1729 sizeof(struct mdt_body),
1730 sizeof(struct lustre_capa) };
1733 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1734 MDS_GETATTR, 3, size, NULL);
1738 mdc_pack_req_body(req, REQ_REC_OFF, OBD_MD_FLOSSCAPA,
1739 &oc->c_capa.lc_fid, oc, 0, 0);
1741 ptlrpc_req_set_repsize(req, 5, size);
1742 req->rq_async_args.pointer_arg[0] = oc;
1743 req->rq_async_args.pointer_arg[1] = cb;
1744 req->rq_interpret_reply = mdc_interpret_renew_capa;
1745 ptlrpcd_add_req(req);
1750 static int mdc_connect(const struct lu_env *env,
1751 struct lustre_handle *dlm_handle,
1752 struct obd_device *obd, struct obd_uuid *cluuid,
1753 struct obd_connect_data *data) {
1754 struct obd_import *imp = obd->u.cli.cl_import;
1756 /* mds-mds import features */
1757 if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
1758 spin_lock(&imp->imp_lock);
1759 imp->imp_server_timeout = 1;
1760 spin_unlock(&imp->imp_lock);
1761 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1762 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
1766 return client_connect_import(env, dlm_handle, obd, cluuid, data);
1769 struct obd_ops mdc_obd_ops = {
1770 .o_owner = THIS_MODULE,
1771 .o_setup = mdc_setup,
1772 .o_precleanup = mdc_precleanup,
1773 .o_cleanup = mdc_cleanup,
1774 .o_add_conn = client_import_add_conn,
1775 .o_del_conn = client_import_del_conn,
1776 .o_connect = mdc_connect,
1777 .o_disconnect = client_disconnect_export,
1778 .o_iocontrol = mdc_iocontrol,
1779 .o_set_info_async = mdc_set_info_async,
1780 .o_statfs = mdc_statfs,
1782 .o_unpin = mdc_unpin,
1783 .o_fid_init = mdc_fid_init,
1784 .o_fid_fini = mdc_fid_fini,
1785 .o_fid_alloc = mdc_fid_alloc,
1786 .o_fid_delete = mdc_fid_delete,
1787 .o_import_event = mdc_import_event,
1788 .o_llog_init = mdc_llog_init,
1789 .o_llog_finish = mdc_llog_finish,
1790 .o_get_info = mdc_get_info,
1791 .o_process_config = mdc_process_config,
1794 struct md_ops mdc_md_ops = {
1795 .m_getstatus = mdc_getstatus,
1796 .m_change_cbdata = mdc_change_cbdata,
1797 .m_close = mdc_close,
1798 .m_create = mdc_create,
1799 .m_done_writing = mdc_done_writing,
1800 .m_enqueue = mdc_enqueue,
1801 .m_getattr = mdc_getattr,
1802 .m_getattr_name = mdc_getattr_name,
1803 .m_intent_lock = mdc_intent_lock,
1805 .m_is_subdir = mdc_is_subdir,
1806 .m_rename = mdc_rename,
1807 .m_setattr = mdc_setattr,
1808 .m_setxattr = mdc_setxattr,
1809 .m_getxattr = mdc_getxattr,
1811 .m_readpage = mdc_readpage,
1812 .m_unlink = mdc_unlink,
1813 .m_cancel_unused = mdc_cancel_unused,
1814 .m_init_ea_size = mdc_init_ea_size,
1815 .m_set_lock_data = mdc_set_lock_data,
1816 .m_lock_match = mdc_lock_match,
1817 .m_get_lustre_md = mdc_get_lustre_md,
1818 .m_free_lustre_md = mdc_free_lustre_md,
1819 .m_set_open_replay_data = mdc_set_open_replay_data,
1820 .m_clear_open_replay_data = mdc_clear_open_replay_data,
1821 .m_get_remote_perm = mdc_get_remote_perm,
1822 .m_renew_capa = mdc_renew_capa
1825 extern quota_interface_t mdc_quota_interface;
1827 int __init mdc_init(void)
1830 struct lprocfs_static_vars lvars;
1831 lprocfs_init_vars(mdc, &lvars);
1833 request_module("lquota");
1834 quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1835 init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1837 rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1838 LUSTRE_MDC_NAME, NULL);
1839 if (rc && quota_interface)
1840 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1846 static void /*__exit*/ mdc_exit(void)
1848 if (quota_interface)
1849 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1851 class_unregister_type(LUSTRE_MDC_NAME);
1854 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1855 MODULE_DESCRIPTION("Lustre Metadata Client");
1856 MODULE_LICENSE("GPL");
1858 module_init(mdc_init);
1859 module_exit(mdc_exit);