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 lustre_set_rep_swabbed(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(lustre_rep_swabbed(req, offset));
505 if (md->body->valid & OBD_MD_FLEASIZE) {
507 struct lov_mds_md *lmm;
509 if (!S_ISREG(md->body->mode)) {
510 CERROR("OBD_MD_FLEASIZE set, should be a regular file, "
512 GOTO(out, rc = -EPROTO);
515 if (md->body->eadatasize == 0) {
516 CERROR("OBD_MD_FLEASIZE set, but eadatasize 0\n");
517 GOTO(out, rc = -EPROTO);
519 lmmsize = md->body->eadatasize;
520 lmm = lustre_msg_buf(req->rq_repmsg, offset, lmmsize);
522 CERROR ("incorrect message: lmm == 0\n");
523 GOTO(out, rc = -EPROTO);
525 LASSERT(lustre_rep_swabbed(req, offset));
527 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
531 if (rc < sizeof(*md->lsm)) {
532 CERROR ("lsm size too small: rc < sizeof (*md->lsm) "
533 "(%d < %d)\n", rc, sizeof(*md->lsm));
534 GOTO(out, rc = -EPROTO);
538 } else if (md->body->valid & OBD_MD_FLDIREA) {
540 struct lov_mds_md *lmv;
542 if(!S_ISDIR(md->body->mode)) {
543 CERROR("OBD_MD_FLDIREA set, should be a directory, but "
545 GOTO(out, rc = -EPROTO);
548 if (md->body->eadatasize == 0) {
549 CERROR("OBD_MD_FLDIREA is set, but eadatasize 0\n");
552 if (md->body->valid & OBD_MD_MEA) {
553 lmvsize = md->body->eadatasize;
554 lmv = lustre_msg_buf(req->rq_repmsg, offset, lmvsize);
556 CERROR ("incorrect message: lmv == 0\n");
557 GOTO(out, rc = -EPROTO);
560 LASSERT(lustre_rep_swabbed(req, offset));
562 rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
567 if (rc < sizeof(*md->mea)) {
568 CERROR ("size too small: rc < sizeof(*md->mea) "
569 "(%d < %d)\n", rc, sizeof(*md->mea));
570 GOTO(out, rc = -EPROTO);
577 /* remote permission */
578 if (md->body->valid & OBD_MD_FLRMTPERM) {
579 md->remote_perm = lustre_msg_buf(req->rq_repmsg, offset++,
580 sizeof(struct mdt_remote_perm));
581 if (!md->remote_perm) {
582 CERROR ("incorrect message: remote_perm == 0\n");
583 GOTO(out, rc = -EPROTO);
587 /* for ACL, it's possible that FLACL is set but aclsize is zero. only
588 * when aclsize != 0 there's an actual segment for ACL in reply
590 else if (md->body->valid & OBD_MD_FLACL) {
591 if (md->body->aclsize) {
592 rc = mdc_unpack_acl(dt_exp, req, md, offset++);
595 #ifdef CONFIG_FS_POSIX_ACL
597 md->posix_acl = NULL;
602 if (md->body->valid & OBD_MD_FLMDSCAPA) {
603 struct obd_capa *oc = mdc_unpack_capa(req, offset++);
606 GOTO(out, rc = PTR_ERR(oc));
610 if (md->body->valid & OBD_MD_FLOSSCAPA) {
611 struct obd_capa *oc = mdc_unpack_capa(req, offset++);
614 GOTO(out, rc = PTR_ERR(oc));
622 free_capa(md->oss_capa);
624 free_capa(md->mds_capa);
625 #ifdef CONFIG_FS_POSIX_ACL
626 posix_acl_release(md->posix_acl);
629 obd_free_memmd(dt_exp, &md->lsm);
634 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
640 static void mdc_replay_open(struct ptlrpc_request *req)
642 struct md_open_data *mod = req->rq_cb_data;
643 struct ptlrpc_request *cur, *tmp;
644 struct obd_client_handle *och;
645 struct lustre_handle old;
646 struct mdt_body *body;
650 DEBUG_REQ(D_ERROR, req,
651 "Can't properly replay without open data.");
656 body = lustre_swab_repbuf(req, DLM_REPLY_REC_OFF, sizeof(*body),
657 lustre_swab_mdt_body);
658 LASSERT(body != NULL);
662 struct lustre_handle *file_fh;
664 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
666 file_fh = &och->och_fh;
667 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
668 file_fh->cookie, body->handle.cookie);
670 *file_fh = body->handle;
673 list_for_each_entry_safe(cur, tmp, &mod->mod_replay_list, rq_mod_list) {
674 int opc = lustre_msg_get_opc(cur->rq_reqmsg);
675 struct mdt_epoch *epoch = NULL;
677 if (opc == MDS_CLOSE || opc == MDS_DONE_WRITING) {
678 epoch = lustre_msg_buf(cur->rq_reqmsg,
679 REQ_REC_OFF, sizeof(*epoch));
681 DEBUG_REQ(D_HA, cur, "updating %s body with new fh",
682 opc == MDS_CLOSE ? "CLOSE" : "DONE_WRITING");
683 } else if (opc == MDS_REINT) {
684 struct mdt_rec_setattr *rec;
686 /* Check this is REINT_SETATTR. */
687 rec = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF,
689 LASSERT(rec && rec->sa_opcode == REINT_SETATTR);
691 epoch = lustre_msg_buf(cur->rq_reqmsg,
692 REQ_REC_OFF + 2, sizeof(*epoch));
694 DEBUG_REQ(D_HA, cur, "updating REINT_SETATTR body "
699 LASSERT(!memcmp(&old, &epoch->handle,
701 epoch->handle = body->handle;
707 void mdc_commit_delayed(struct ptlrpc_request *req)
709 struct md_open_data *mod = req->rq_cb_data;
710 struct ptlrpc_request *cur, *tmp;
712 DEBUG_REQ(D_HA, req, "req committed");
717 req->rq_cb_data = NULL;
718 req->rq_commit_cb = NULL;
719 list_del_init(&req->rq_mod_list);
720 if (req->rq_sequence) {
721 list_for_each_entry_safe(cur, tmp, &mod->mod_replay_list,
723 LASSERT(cur != LP_POISON);
724 LASSERT(cur->rq_type != LI_POISON);
725 DEBUG_REQ(D_HA, cur, "req balanced");
726 LASSERT(cur->rq_transno != 0);
727 LASSERT(cur->rq_import == req->rq_import);
729 /* We no longer want to preserve this for transno-
730 * unconditional replay. */
731 spin_lock(&cur->rq_lock);
733 spin_unlock(&cur->rq_lock);
737 if (list_empty(&mod->mod_replay_list)) {
738 if (mod->mod_och != NULL)
739 mod->mod_och->och_mod = NULL;
745 int mdc_set_open_replay_data(struct obd_export *exp,
746 struct obd_client_handle *och,
747 struct ptlrpc_request *open_req)
749 struct md_open_data *mod;
750 struct mdt_rec_create *rec = lustre_msg_buf(open_req->rq_reqmsg,
753 struct mdt_body *body = lustre_msg_buf(open_req->rq_repmsg,
756 struct obd_import *imp = open_req->rq_import;
759 LASSERT(rec != NULL);
761 /* Incoming message in my byte order (it's been swabbed). */
762 LASSERT(lustre_rep_swabbed(open_req, DLM_REPLY_REC_OFF));
764 /* Outgoing messages always in my byte order. */
765 LASSERT(body != NULL);
767 /*Only the import is replayable, we set replay_open data */
768 if (och && imp->imp_replayable) {
769 OBD_ALLOC(mod, sizeof(*mod));
771 DEBUG_REQ(D_ERROR, open_req,
772 "Can't allocate md_open_data");
775 CFS_INIT_LIST_HEAD(&mod->mod_replay_list);
777 spin_lock(&open_req->rq_lock);
778 if (!open_req->rq_replay) {
779 OBD_FREE(mod, sizeof(*mod));
780 spin_unlock(&open_req->rq_lock);
786 open_req->rq_cb_data = mod;
787 list_add_tail(&open_req->rq_mod_list, &mod->mod_replay_list);
788 open_req->rq_commit_cb = mdc_commit_delayed;
789 spin_unlock(&open_req->rq_lock);
792 rec->cr_fid2 = body->fid1;
793 rec->cr_ioepoch = body->ioepoch;
794 rec->cr_old_handle.cookie = body->handle.cookie;
795 open_req->rq_replay_cb = mdc_replay_open;
796 if (!fid_is_sane(&body->fid1)) {
797 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
802 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
806 int mdc_clear_open_replay_data(struct obd_export *exp,
807 struct obd_client_handle *och)
809 struct md_open_data *mod = och->och_mod;
813 * Don't free the structure now (it happens in mdc_commit_delayed(),
814 * after the last request is removed from its replay list),
815 * but make sure that replay doesn't poke at the och, which is about to
818 LASSERT(mod != LP_POISON);
826 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
827 struct md_open_data *mod, struct ptlrpc_request **request)
829 struct obd_device *obd = class_exp2obd(exp);
830 int reqsize[4] = { sizeof(struct ptlrpc_body),
831 sizeof(struct mdt_epoch),
832 sizeof(struct mdt_rec_setattr)};
833 int repsize[4] = { sizeof(struct ptlrpc_body),
834 sizeof(struct mdt_body),
835 obd->u.cli.cl_max_mds_easize,
836 obd->u.cli.cl_max_mds_cookiesize };
837 struct ptlrpc_request *req;
841 reqsize[REQ_REC_OFF + 2] = op_data->op_capa1 ?
842 sizeof(struct lustre_capa) : 0;
843 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
844 MDS_CLOSE, 4, reqsize, NULL);
846 GOTO(out, rc = -ENOMEM);
848 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
849 * portal whose threads are not taking any DLM locks and are therefore
850 * always progressing */
851 /* XXX FIXME bug 249 */
852 req->rq_request_portal = MDS_READPAGE_PORTAL;
854 /* Ensure that this close's handle is fixed up during replay. */
855 if (likely(mod != NULL))
856 list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
858 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
860 mdc_close_pack(req, REQ_REC_OFF, op_data);
861 ptlrpc_req_set_repsize(req, 4, repsize);
862 req->rq_commit_cb = mdc_commit_delayed;
864 LASSERT(req->rq_cb_data == NULL);
865 req->rq_cb_data = mod;
867 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
868 rc = ptlrpc_queue_wait(req);
869 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
871 if (req->rq_repmsg == NULL) {
872 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
875 rc = req->rq_status ? req->rq_status : -EIO;
876 } else if (rc == 0 || rc == -EAGAIN) {
877 rc = lustre_msg_get_status(req->rq_repmsg);
878 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
879 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
883 } else if (mod == NULL) {
884 if (req->rq_import->imp_replayable)
885 CERROR("Unexpected: can't find md_open_data,"
886 "but close succeeded with replayable imp"
887 "Please tell CFS.\n");
889 if (!lustre_swab_repbuf(req, REPLY_REC_OFF,
890 sizeof(struct mdt_body),
891 lustre_swab_mdt_body)) {
892 CERROR("Error unpacking mdt_body\n");
900 if (rc != 0 && rc != -EAGAIN && req && req->rq_commit_cb)
901 req->rq_commit_cb(req);
906 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
907 struct md_open_data *mod)
909 struct obd_device *obd = class_exp2obd(exp);
910 struct ptlrpc_request *req;
911 int size[4] = { sizeof(struct ptlrpc_body),
912 sizeof(struct mdt_epoch),
913 sizeof(struct mdt_rec_setattr)};
914 int repsize[2] = { sizeof(struct ptlrpc_body),
915 sizeof(struct mdt_body)};
919 if (op_data->op_capa1)
920 size[REQ_REC_OFF + 2] = sizeof(struct lustre_capa);
921 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
922 MDS_DONE_WRITING, 4, size, NULL);
926 mdc_close_pack(req, REQ_REC_OFF, op_data);
929 req->rq_cb_data = mod;
930 req->rq_commit_cb = mdc_commit_delayed;
931 if (likely(mod != NULL))
932 list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
934 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
936 ptlrpc_req_set_repsize(req, 2, repsize);
937 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
938 rc = ptlrpc_queue_wait(req);
939 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
941 /* Close the open replay sequence if an error occured or no SOM
942 * attribute update is needed. */
944 ptlrpc_close_replay_seq(req);
946 if (rc && rc != -EAGAIN && req->rq_commit_cb)
947 req->rq_commit_cb(req);
949 ptlrpc_req_finished(req);
953 #ifdef HAVE_SPLIT_SUPPORT
954 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
955 const struct page *page, int offset)
957 int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
958 struct obd_import *imp = class_exp2cliimp(exp);
959 struct ptlrpc_bulk_desc *desc = NULL;
960 struct ptlrpc_request *req = NULL;
963 CDEBUG(D_INODE, "object: "DFID"\n", PFID(fid));
965 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_WRITEPAGE, 3,
968 GOTO(out, rc = -ENOMEM);
970 req->rq_request_portal = MDS_READPAGE_PORTAL;
972 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
974 GOTO(out, rc = -ENOMEM);
976 /* NB req now owns desc and will free it when it gets freed. */
977 ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
978 mdc_readdir_pack(req, REQ_REC_OFF, 0, offset, fid, NULL);
980 ptlrpc_req_set_repsize(req, 2, size);
981 rc = ptlrpc_queue_wait(req);
985 ptlrpc_req_finished(req);
988 EXPORT_SYMBOL(mdc_sendpage);
991 int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
992 struct obd_capa *oc, __u64 offset, struct page *page,
993 struct ptlrpc_request **request)
995 int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
996 struct obd_import *imp = class_exp2cliimp(exp);
997 struct ptlrpc_bulk_desc *desc = NULL;
998 struct ptlrpc_request *req = NULL;
999 struct mdt_body *body;
1002 CDEBUG(D_INODE, "object: "DFID"\n", PFID(fid));
1004 size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1005 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE, 3, size,
1008 GOTO(out, rc = -ENOMEM);
1010 /* XXX FIXME bug 249 */
1011 req->rq_request_portal = MDS_READPAGE_PORTAL;
1013 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
1015 GOTO(out, rc = -ENOMEM);
1017 /* NB req now owns desc and will free it when it gets freed */
1018 ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
1019 mdc_readdir_pack(req, REQ_REC_OFF, offset, CFS_PAGE_SIZE, fid, oc);
1021 ptlrpc_req_set_repsize(req, 2, size);
1022 rc = ptlrpc_queue_wait(req);
1025 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1026 lustre_swab_mdt_body);
1028 CERROR("Can't unpack mdt_body\n");
1029 GOTO(out, rc = -EPROTO);
1032 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
1033 CERROR ("Unexpected # bytes transferred: %d"
1034 " (%ld expected)\n",
1035 req->rq_bulk->bd_nob_transferred,
1037 GOTO(out, rc = -EPROTO);
1047 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1048 void *karg, void *uarg)
1050 struct obd_device *obd = exp->exp_obd;
1051 struct obd_ioctl_data *data = karg;
1052 struct obd_import *imp = obd->u.cli.cl_import;
1053 struct llog_ctxt *ctxt;
1057 if (!try_module_get(THIS_MODULE)) {
1058 CERROR("Can't get module. Is it alive?");
1062 case OBD_IOC_CLIENT_RECOVER:
1063 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
1067 case IOC_OSC_SET_ACTIVE:
1068 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1070 case OBD_IOC_PARSE: {
1071 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1072 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
1076 case OBD_IOC_LLOG_INFO:
1077 case OBD_IOC_LLOG_PRINT: {
1078 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1079 rc = llog_ioctl(ctxt, cmd, data);
1084 case OBD_IOC_POLL_QUOTACHECK:
1085 rc = lquota_poll_check(quota_interface, exp,
1086 (struct if_quotacheck *)karg);
1089 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1090 GOTO(out, rc = -ENOTTY);
1093 module_put(THIS_MODULE);
1098 int mdc_set_info_async(struct obd_export *exp, obd_count keylen,
1099 void *key, obd_count vallen, void *val,
1100 struct ptlrpc_request_set *set)
1102 struct obd_import *imp = class_exp2cliimp(exp);
1105 if (KEY_IS(KEY_INIT_RECOV)) {
1106 if (vallen != sizeof(int))
1108 spin_lock(&imp->imp_lock);
1109 imp->imp_initial_recov = *(int *)val;
1110 spin_unlock(&imp->imp_lock);
1111 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
1112 exp->exp_obd->obd_name, imp->imp_initial_recov);
1115 /* Turn off initial_recov after we try all backup servers once */
1116 if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
1117 if (vallen != sizeof(int))
1119 spin_lock(&imp->imp_lock);
1120 imp->imp_initial_recov_bk = *(int *)val;
1121 if (imp->imp_initial_recov_bk)
1122 imp->imp_initial_recov = 1;
1123 spin_unlock(&imp->imp_lock);
1124 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
1125 exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
1128 if (KEY_IS(KEY_READ_ONLY)) {
1129 struct ptlrpc_request *req;
1130 int size[3] = { sizeof(struct ptlrpc_body), keylen, vallen };
1131 char *bufs[3] = { NULL, key, val };
1133 if (vallen != sizeof(int))
1136 spin_lock(&imp->imp_lock);
1137 if (*((int *)val)) {
1138 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
1139 imp->imp_connect_data.ocd_connect_flags |=
1142 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
1143 imp->imp_connect_data.ocd_connect_flags &=
1144 ~OBD_CONNECT_RDONLY;
1146 spin_unlock(&imp->imp_lock);
1148 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_SET_INFO,
1153 ptlrpc_req_set_repsize(req, 1, NULL);
1156 ptlrpc_set_add_req(set, req);
1157 ptlrpc_check_set(set);
1159 rc = ptlrpc_queue_wait(req);
1160 ptlrpc_req_finished(req);
1165 if (KEY_IS(KEY_FLUSH_CTX)) {
1166 sptlrpc_import_flush_my_ctx(imp);
1169 if (KEY_IS(KEY_MDS_CONN)) {
1170 struct obd_import *imp = class_exp2cliimp(exp);
1172 /* mds-mds import */
1173 spin_lock(&imp->imp_lock);
1174 imp->imp_server_timeout = 1;
1175 spin_unlock(&imp->imp_lock);
1176 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1177 CDEBUG(D_OTHER|D_WARNING, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1184 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
1185 __u32 *vallen, void *val)
1189 if (KEY_IS(KEY_MAX_EASIZE)) {
1190 int mdsize, *max_easize;
1192 if (*vallen != sizeof(int))
1194 mdsize = *(int*)val;
1195 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
1196 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
1198 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
1201 if (KEY_IS(KEY_CONN_DATA)) {
1202 struct obd_import *imp = class_exp2cliimp(exp);
1203 struct obd_connect_data *data = val;
1205 if (*vallen != sizeof(*data))
1208 *data = imp->imp_connect_data;
1215 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1218 struct ptlrpc_request *req;
1219 struct obd_statfs *msfs;
1220 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*msfs) };
1223 /* We could possibly pass max_age in the request (as an absolute
1224 * timestamp or a "seconds.usec ago") so the target can avoid doing
1225 * extra calls into the filesystem if that isn't necessary (e.g.
1226 * during mount that would help a bit). Having relative timestamps
1227 * is not so great if request processing is slow, while absolute
1228 * timestamps are not ideal because they need time synchronization. */
1229 req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
1230 MDS_STATFS, 1, NULL, NULL);
1234 ptlrpc_req_set_repsize(req, 2, size);
1236 rc = ptlrpc_queue_wait(req);
1239 /* check connection error first */
1240 if (obd->u.cli.cl_import->imp_connect_error)
1241 rc = obd->u.cli.cl_import->imp_connect_error;
1246 msfs = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*msfs),
1247 lustre_swab_obd_statfs);
1249 CERROR("Can't unpack obd_statfs\n");
1250 GOTO(out, rc = -EPROTO);
1253 memcpy(osfs, msfs, sizeof(*msfs));
1256 ptlrpc_req_finished(req);
1261 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
1262 struct obd_capa *oc,
1263 struct obd_client_handle *handle, int flag)
1265 struct ptlrpc_request *req;
1266 struct mdt_body *body;
1267 int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1270 size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1271 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1272 MDS_PIN, 3, size, NULL);
1276 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof (*body));
1279 mdc_pack_capa(req, REQ_REC_OFF + 1, oc);
1281 ptlrpc_req_set_repsize(req, 2, size);
1283 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1284 rc = ptlrpc_queue_wait(req);
1285 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1287 CERROR("pin failed: %d\n", rc);
1288 ptlrpc_req_finished(req);
1292 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1293 lustre_swab_mdt_body);
1295 ptlrpc_req_finished(req);
1299 memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
1300 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1302 OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
1303 if (handle->och_mod == NULL) {
1304 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
1308 /* will be dropped by unpin */
1309 CFS_INIT_LIST_HEAD(&handle->och_mod->mod_replay_list);
1310 list_add_tail(&req->rq_mod_list, &handle->och_mod->mod_replay_list);
1315 static int mdc_unpin(struct obd_export *exp,
1316 struct obd_client_handle *handle, int flag)
1318 struct ptlrpc_request *req;
1319 struct mdt_body *body;
1320 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1323 if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
1326 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1327 MDS_CLOSE, 2, size, NULL);
1331 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1332 memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
1335 ptlrpc_req_set_repsize(req, 1, NULL);
1336 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1337 rc = ptlrpc_queue_wait(req);
1338 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1341 CERROR("unpin failed: %d\n", rc);
1343 ptlrpc_req_finished(req);
1345 LASSERT(!list_empty(&handle->och_mod->mod_replay_list));
1346 req = list_entry(handle->och_mod->mod_replay_list.next,
1347 typeof(*req), rq_mod_list);
1348 list_del_init(&req->rq_mod_list);
1349 ptlrpc_req_finished(req);
1350 LASSERT(list_empty(&handle->och_mod->mod_replay_list));
1352 OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1356 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
1357 struct obd_capa *oc, struct ptlrpc_request **request)
1359 struct ptlrpc_request *req;
1360 int size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
1364 size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1365 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1366 MDS_SYNC, 3, size, NULL);
1368 RETURN(rc = -ENOMEM);
1370 mdc_pack_req_body(req, REQ_REC_OFF, 0, fid, oc, 0, 0);
1372 ptlrpc_req_set_repsize(req, 2, size);
1374 rc = ptlrpc_queue_wait(req);
1375 if (rc || request == NULL)
1376 ptlrpc_req_finished(req);
1383 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1384 enum obd_import_event event)
1388 LASSERT(imp->imp_obd == obd);
1391 case IMP_EVENT_DISCON: {
1393 /* XXX Pass event up to OBDs stack. used only for FLD now */
1394 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
1398 case IMP_EVENT_INACTIVE: {
1399 struct client_obd *cli = &obd->u.cli;
1401 * Flush current sequence to make client obtain new one
1402 * from server in case of disconnect/reconnect.
1403 * If range is already empty then no need to flush it.
1405 if (cli->cl_seq != NULL &&
1406 !range_is_exhausted(&cli->cl_seq->lcs_space)) {
1407 seq_client_flush(cli->cl_seq);
1410 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1413 case IMP_EVENT_INVALIDATE: {
1414 struct ldlm_namespace *ns = obd->obd_namespace;
1416 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1420 case IMP_EVENT_ACTIVE: {
1421 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1425 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
1429 CERROR("Unknown import event %x\n", event);
1435 static int mdc_fid_init(struct obd_export *exp)
1437 struct client_obd *cli = &exp->exp_obd->u.cli;
1442 OBD_ALLOC_PTR(cli->cl_seq);
1443 if (cli->cl_seq == NULL)
1446 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
1448 GOTO(out_free_seq, rc = -ENOMEM);
1450 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
1451 exp->exp_obd->obd_name);
1453 /* Init client side sequence-manager */
1454 rc = seq_client_init(cli->cl_seq, exp,
1455 LUSTRE_SEQ_METADATA,
1457 OBD_FREE(prefix, MAX_OBD_NAME + 5);
1459 GOTO(out_free_seq, rc);
1463 OBD_FREE_PTR(cli->cl_seq);
1468 static int mdc_fid_fini(struct obd_export *exp)
1470 struct client_obd *cli = &exp->exp_obd->u.cli;
1473 if (cli->cl_seq != NULL) {
1474 seq_client_fini(cli->cl_seq);
1475 OBD_FREE_PTR(cli->cl_seq);
1482 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1483 struct md_op_data *op_data)
1485 struct client_obd *cli = &exp->exp_obd->u.cli;
1486 struct lu_client_seq *seq = cli->cl_seq;
1488 RETURN(seq_client_alloc_fid(seq, fid));
1491 /* XXX This method is used only to clear current fid seq
1492 * once fld/mds insert failed */
1493 static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1495 struct client_obd *cli = &exp->exp_obd->u.cli;
1497 seq_client_flush(cli->cl_seq);
1501 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
1503 struct client_obd *cli = &obd->u.cli;
1504 struct lprocfs_static_vars lvars;
1508 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1509 if (!cli->cl_rpc_lock)
1511 mdc_init_rpc_lock(cli->cl_rpc_lock);
1515 OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1516 if (!cli->cl_setattr_lock)
1517 GOTO(err_rpc_lock, rc = -ENOMEM);
1518 mdc_init_rpc_lock(cli->cl_setattr_lock);
1520 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1521 if (!cli->cl_close_lock)
1522 GOTO(err_setattr_lock, rc = -ENOMEM);
1523 mdc_init_rpc_lock(cli->cl_close_lock);
1525 rc = client_obd_setup(obd, cfg);
1527 GOTO(err_close_lock, rc);
1528 lprocfs_init_vars(mdc, &lvars);
1529 lprocfs_obd_setup(obd, lvars.obd_vars);
1530 ptlrpc_lprocfs_register_obd(obd);
1532 rc = obd_llog_init(obd, NULL, obd, 0, NULL, NULL);
1535 CERROR("failed to setup llogging subsystems\n");
1541 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1543 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1545 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1550 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
1551 * us to make MDS RPCs with large enough reply buffers to hold the
1552 * maximum-sized (= maximum striped) EA and cookie without having to
1553 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1554 int mdc_init_ea_size(struct obd_export *exp, int easize,
1555 int def_easize, int cookiesize)
1557 struct obd_device *obd = exp->exp_obd;
1558 struct client_obd *cli = &obd->u.cli;
1561 if (cli->cl_max_mds_easize < easize)
1562 cli->cl_max_mds_easize = easize;
1564 if (cli->cl_default_mds_easize < def_easize)
1565 cli->cl_default_mds_easize = def_easize;
1567 if (cli->cl_max_mds_cookiesize < cookiesize)
1568 cli->cl_max_mds_cookiesize = cookiesize;
1573 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1579 case OBD_CLEANUP_EARLY:
1580 case OBD_CLEANUP_EXPORTS:
1581 /* If we set up but never connected, the
1582 client import will not have been cleaned. */
1583 if (obd->u.cli.cl_import) {
1584 struct obd_import *imp;
1585 imp = obd->u.cli.cl_import;
1586 CERROR("client import never connected\n");
1587 ptlrpc_invalidate_import(imp);
1588 ptlrpc_free_rq_pool(imp->imp_rq_pool);
1589 class_destroy_import(imp);
1590 obd->u.cli.cl_import = NULL;
1593 case OBD_CLEANUP_SELF_EXP:
1594 rc = obd_llog_finish(obd, 0);
1596 CERROR("failed to cleanup llogging subsystems\n");
1597 case OBD_CLEANUP_OBD:
1603 static int mdc_cleanup(struct obd_device *obd)
1605 struct client_obd *cli = &obd->u.cli;
1607 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1608 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1609 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1611 ptlrpc_lprocfs_unregister_obd(obd);
1612 lprocfs_obd_cleanup(obd);
1615 return client_obd_cleanup(obd);
1619 static int mdc_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1620 struct obd_device *tgt,
1621 int count, struct llog_catid *logid,
1622 struct obd_uuid *uuid)
1624 struct llog_ctxt *ctxt;
1628 rc = llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1631 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1632 ctxt->loc_imp = obd->u.cli.cl_import;
1635 rc = llog_setup(obd, llogs, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1638 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1639 ctxt->loc_imp = obd->u.cli.cl_import;
1645 static int mdc_llog_finish(struct obd_device *obd, int count)
1650 rc = llog_cleanup(llog_get_context(obd, LLOG_LOVEA_REPL_CTXT));
1652 CERROR("can not cleanup LLOG_CONFIG_REPL_CTXT rc %d\n", rc);
1654 rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
1658 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1660 struct lustre_cfg *lcfg = buf;
1661 struct lprocfs_static_vars lvars;
1664 lprocfs_init_vars(mdc, &lvars);
1666 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars, lcfg, obd);
1670 /* get remote permission for current user on fid */
1671 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
1672 struct obd_capa *oc, struct ptlrpc_request **request)
1674 struct ptlrpc_request *req;
1675 struct mdt_body *body;
1676 struct mdt_remote_perm *perm;
1677 int size[5] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1681 size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1684 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1685 MDS_GETATTR, 3, size, NULL);
1689 mdc_pack_req_body(req, REQ_REC_OFF, OBD_MD_FLRMTPERM, fid, oc, 0, 0);
1691 size[REPLY_REC_OFF + 1] = sizeof(*perm);
1692 ptlrpc_req_set_repsize(req, 5, size);
1693 rc = ptlrpc_queue_wait(req);
1695 ptlrpc_req_finished(req);
1699 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1700 lustre_swab_mdt_body);
1702 LASSERT(body->valid & OBD_MD_FLRMTPERM);
1704 perm = lustre_swab_repbuf(req, REPLY_REC_OFF + 1, sizeof(*perm),
1705 lustre_swab_mdt_remote_perm);
1712 static int mdc_interpret_renew_capa(struct ptlrpc_request *req, void *unused,
1715 struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
1716 renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
1717 struct mdt_body *body = NULL;
1718 struct lustre_capa *capa;
1722 GOTO(out, capa = ERR_PTR(status));
1724 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1725 lustre_swab_mdt_body);
1727 GOTO(out, capa = ERR_PTR(-EFAULT));
1729 if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
1730 GOTO(out, capa = ERR_PTR(-ENOENT));
1732 capa = lustre_unpack_capa(req->rq_repmsg, REPLY_REC_OFF + 1);
1734 GOTO(out, capa = ERR_PTR(-EFAULT));
1741 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
1744 struct ptlrpc_request *req;
1745 int size[5] = { sizeof(struct ptlrpc_body),
1746 sizeof(struct mdt_body),
1747 sizeof(struct lustre_capa) };
1750 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1751 MDS_GETATTR, 3, size, NULL);
1755 mdc_pack_req_body(req, REQ_REC_OFF, OBD_MD_FLOSSCAPA,
1756 &oc->c_capa.lc_fid, oc, 0, 0);
1758 ptlrpc_req_set_repsize(req, 5, size);
1759 req->rq_async_args.pointer_arg[0] = oc;
1760 req->rq_async_args.pointer_arg[1] = cb;
1761 req->rq_interpret_reply = mdc_interpret_renew_capa;
1762 ptlrpcd_add_req(req);
1767 static int mdc_connect(const struct lu_env *env,
1768 struct lustre_handle *dlm_handle,
1769 struct obd_device *obd, struct obd_uuid *cluuid,
1770 struct obd_connect_data *data) {
1771 struct obd_import *imp = obd->u.cli.cl_import;
1773 /* mds-mds import features */
1774 if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
1775 spin_lock(&imp->imp_lock);
1776 imp->imp_server_timeout = 1;
1777 spin_unlock(&imp->imp_lock);
1778 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1779 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
1783 return client_connect_import(env, dlm_handle, obd, cluuid, data);
1786 struct obd_ops mdc_obd_ops = {
1787 .o_owner = THIS_MODULE,
1788 .o_setup = mdc_setup,
1789 .o_precleanup = mdc_precleanup,
1790 .o_cleanup = mdc_cleanup,
1791 .o_add_conn = client_import_add_conn,
1792 .o_del_conn = client_import_del_conn,
1793 .o_connect = mdc_connect,
1794 .o_disconnect = client_disconnect_export,
1795 .o_iocontrol = mdc_iocontrol,
1796 .o_set_info_async = mdc_set_info_async,
1797 .o_statfs = mdc_statfs,
1799 .o_unpin = mdc_unpin,
1800 .o_fid_init = mdc_fid_init,
1801 .o_fid_fini = mdc_fid_fini,
1802 .o_fid_alloc = mdc_fid_alloc,
1803 .o_fid_delete = mdc_fid_delete,
1804 .o_import_event = mdc_import_event,
1805 .o_llog_init = mdc_llog_init,
1806 .o_llog_finish = mdc_llog_finish,
1807 .o_get_info = mdc_get_info,
1808 .o_process_config = mdc_process_config,
1811 struct md_ops mdc_md_ops = {
1812 .m_getstatus = mdc_getstatus,
1813 .m_change_cbdata = mdc_change_cbdata,
1814 .m_close = mdc_close,
1815 .m_create = mdc_create,
1816 .m_done_writing = mdc_done_writing,
1817 .m_enqueue = mdc_enqueue,
1818 .m_getattr = mdc_getattr,
1819 .m_getattr_name = mdc_getattr_name,
1820 .m_intent_lock = mdc_intent_lock,
1822 .m_is_subdir = mdc_is_subdir,
1823 .m_rename = mdc_rename,
1824 .m_setattr = mdc_setattr,
1825 .m_setxattr = mdc_setxattr,
1826 .m_getxattr = mdc_getxattr,
1828 .m_readpage = mdc_readpage,
1829 .m_unlink = mdc_unlink,
1830 .m_cancel_unused = mdc_cancel_unused,
1831 .m_init_ea_size = mdc_init_ea_size,
1832 .m_set_lock_data = mdc_set_lock_data,
1833 .m_lock_match = mdc_lock_match,
1834 .m_get_lustre_md = mdc_get_lustre_md,
1835 .m_free_lustre_md = mdc_free_lustre_md,
1836 .m_set_open_replay_data = mdc_set_open_replay_data,
1837 .m_clear_open_replay_data = mdc_clear_open_replay_data,
1838 .m_get_remote_perm = mdc_get_remote_perm,
1839 .m_renew_capa = mdc_renew_capa
1842 extern quota_interface_t mdc_quota_interface;
1844 int __init mdc_init(void)
1847 struct lprocfs_static_vars lvars;
1848 lprocfs_init_vars(mdc, &lvars);
1850 request_module("lquota");
1851 quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1852 init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1854 rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1855 LUSTRE_MDC_NAME, NULL);
1856 if (rc && quota_interface)
1857 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1863 static void /*__exit*/ mdc_exit(void)
1865 if (quota_interface)
1866 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1868 class_unregister_type(LUSTRE_MDC_NAME);
1871 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1872 MODULE_DESCRIPTION("Lustre Metadata Client");
1873 MODULE_LICENSE("GPL");
1875 module_init(mdc_init);
1876 module_exit(mdc_exit);