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, -1, 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, -1,
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, __u32 suppgid,
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, suppgid,
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, __u32 suppgid, 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;
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,
375 tmp = lustre_msg_buf(req->rq_reqmsg, offset++, xattr_namelen);
376 memcpy(tmp, xattr_name, xattr_namelen);
379 tmp = lustre_msg_buf(req->rq_reqmsg, offset++, input_size);
380 memcpy(tmp, input, input_size);
384 if (opcode == MDS_GETXATTR) {
385 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
391 /* we do this even output_size is 0, because server is doing that */
392 size[bufcnt++] = output_size;
393 ptlrpc_req_set_repsize(req, bufcnt, size);
396 if (opcode == MDS_SETXATTR)
397 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
399 rc = ptlrpc_queue_wait(req);
401 if (opcode == MDS_SETXATTR)
402 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
407 if (opcode == MDS_GETXATTR) {
408 struct mdt_body * body = lustre_swab_repbuf(req, REPLY_REC_OFF,
409 sizeof(*body), lustre_swab_mdt_body);
411 CERROR ("Can't unpack mdt_body\n");
412 GOTO(err_out, rc = -EPROTO);
419 ptlrpc_req_finished(req);
424 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
425 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
426 const char *input, int input_size, int output_size, int flags,
427 __u32 suppgid, struct ptlrpc_request **request)
429 return mdc_xattr_common(exp, fid, oc, MDS_SETXATTR, valid, xattr_name,
430 input, input_size, output_size, flags, suppgid,
434 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
435 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
436 const char *input, int input_size, int output_size, int flags,
437 struct ptlrpc_request **request)
439 return mdc_xattr_common(exp, fid, oc, MDS_GETXATTR, valid, xattr_name,
440 input, input_size, output_size, flags, -1,
444 #ifdef CONFIG_FS_POSIX_ACL
446 int mdc_unpack_acl(struct obd_export *exp, struct ptlrpc_request *req,
447 struct lustre_md *md, unsigned int offset)
449 struct mdt_body *body = md->body;
450 struct posix_acl *acl;
457 buf = lustre_msg_buf(req->rq_repmsg, offset, body->aclsize);
459 CERROR("aclsize %u, bufcount %u, bufsize %u\n",
460 body->aclsize, lustre_msg_bufcount(req->rq_repmsg),
461 (lustre_msg_bufcount(req->rq_repmsg) <= offset) ?
462 -1 : lustre_msg_buflen(req->rq_repmsg, offset));
466 acl = posix_acl_from_xattr(buf, body->aclsize);
469 CERROR("convert xattr to acl: %d\n", rc);
473 rc = posix_acl_valid(acl);
475 CERROR("validate acl: %d\n", rc);
476 posix_acl_release(acl);
484 #define mdc_unpack_acl(exp, req, md, offset) 0
487 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
488 int offset, struct obd_export *dt_exp,
489 struct obd_export *md_exp,
490 struct lustre_md *md)
496 memset(md, 0, sizeof(*md));
498 md->body = lustre_msg_buf(req->rq_repmsg, offset, sizeof (*md->body));
499 LASSERT (md->body != NULL);
500 LASSERT(lustre_rep_swabbed(req, offset));
503 if (md->body->valid & OBD_MD_FLEASIZE) {
505 struct lov_mds_md *lmm;
507 if (!S_ISREG(md->body->mode)) {
508 CERROR("OBD_MD_FLEASIZE set, should be a regular file, "
510 GOTO(out, rc = -EPROTO);
513 if (md->body->eadatasize == 0) {
514 CERROR("OBD_MD_FLEASIZE set, but eadatasize 0\n");
515 GOTO(out, rc = -EPROTO);
517 lmmsize = md->body->eadatasize;
518 lmm = lustre_msg_buf(req->rq_repmsg, offset, lmmsize);
520 CERROR ("incorrect message: lmm == 0\n");
521 GOTO(out, rc = -EPROTO);
523 LASSERT(lustre_rep_swabbed(req, offset));
525 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
529 if (rc < sizeof(*md->lsm)) {
530 CERROR ("lsm size too small: rc < sizeof (*md->lsm) "
531 "(%d < "LPSZ")\n", rc, sizeof(*md->lsm));
532 GOTO(out, rc = -EPROTO);
536 } else if (md->body->valid & OBD_MD_FLDIREA) {
538 struct lov_mds_md *lmv;
540 if(!S_ISDIR(md->body->mode)) {
541 CERROR("OBD_MD_FLDIREA set, should be a directory, but "
543 GOTO(out, rc = -EPROTO);
546 if (md->body->eadatasize == 0) {
547 CERROR("OBD_MD_FLDIREA is set, but eadatasize 0\n");
550 if (md->body->valid & OBD_MD_MEA) {
551 lmvsize = md->body->eadatasize;
552 lmv = lustre_msg_buf(req->rq_repmsg, offset, lmvsize);
554 CERROR ("incorrect message: lmv == 0\n");
555 GOTO(out, rc = -EPROTO);
558 LASSERT(lustre_rep_swabbed(req, offset));
560 rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
565 if (rc < sizeof(*md->mea)) {
566 CERROR ("size too small: rc < sizeof(*md->mea) "
567 "(%d < %d)\n", rc, sizeof(*md->mea));
568 GOTO(out, rc = -EPROTO);
575 /* remote permission */
576 if (md->body->valid & OBD_MD_FLRMTPERM) {
577 md->remote_perm = lustre_msg_buf(req->rq_repmsg, offset++,
578 sizeof(struct mdt_remote_perm));
579 if (!md->remote_perm) {
580 CERROR ("incorrect message: remote_perm == 0\n");
581 GOTO(out, rc = -EPROTO);
585 /* for ACL, it's possible that FLACL is set but aclsize is zero. only
586 * when aclsize != 0 there's an actual segment for ACL in reply
588 else if (md->body->valid & OBD_MD_FLACL) {
589 if (md->body->aclsize) {
590 rc = mdc_unpack_acl(dt_exp, req, md, offset++);
593 #ifdef CONFIG_FS_POSIX_ACL
595 md->posix_acl = NULL;
600 if (md->body->valid & OBD_MD_FLMDSCAPA) {
601 struct obd_capa *oc = mdc_unpack_capa(req, offset++);
604 GOTO(out, rc = PTR_ERR(oc));
608 if (md->body->valid & OBD_MD_FLOSSCAPA) {
609 struct obd_capa *oc = mdc_unpack_capa(req, offset++);
612 GOTO(out, rc = PTR_ERR(oc));
620 free_capa(md->oss_capa);
622 free_capa(md->mds_capa);
623 #ifdef CONFIG_FS_POSIX_ACL
624 posix_acl_release(md->posix_acl);
627 obd_free_memmd(dt_exp, &md->lsm);
632 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
638 static void mdc_replay_open(struct ptlrpc_request *req)
640 struct md_open_data *mod = req->rq_cb_data;
641 struct ptlrpc_request *cur, *tmp;
642 struct obd_client_handle *och;
643 struct lustre_handle old;
644 struct mdt_body *body;
648 DEBUG_REQ(D_ERROR, req,
649 "Can't properly replay without open data.");
654 body = lustre_swab_repbuf(req, DLM_REPLY_REC_OFF, sizeof(*body),
655 lustre_swab_mdt_body);
656 LASSERT(body != NULL);
660 struct lustre_handle *file_fh;
662 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
664 file_fh = &och->och_fh;
665 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
666 file_fh->cookie, body->handle.cookie);
668 *file_fh = body->handle;
671 list_for_each_entry_safe(cur, tmp, &mod->mod_replay_list, rq_mod_list) {
672 int opc = lustre_msg_get_opc(cur->rq_reqmsg);
673 struct mdt_epoch *epoch = NULL;
675 if (opc == MDS_CLOSE || opc == MDS_DONE_WRITING) {
676 epoch = lustre_msg_buf(cur->rq_reqmsg,
677 REQ_REC_OFF, sizeof(*epoch));
679 DEBUG_REQ(D_HA, cur, "updating %s body with new fh",
680 opc == MDS_CLOSE ? "CLOSE" : "DONE_WRITING");
681 } else if (opc == MDS_REINT) {
682 struct mdt_rec_setattr *rec;
684 /* Check this is REINT_SETATTR. */
685 rec = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF,
687 LASSERT(rec && rec->sa_opcode == REINT_SETATTR);
689 epoch = lustre_msg_buf(cur->rq_reqmsg,
690 REQ_REC_OFF + 2, sizeof(*epoch));
692 DEBUG_REQ(D_HA, cur, "updating REINT_SETATTR body "
697 LASSERT(!memcmp(&old, &epoch->handle,
699 epoch->handle = body->handle;
705 void mdc_commit_delayed(struct ptlrpc_request *req)
707 struct md_open_data *mod = req->rq_cb_data;
708 struct ptlrpc_request *cur, *tmp;
710 DEBUG_REQ(D_HA, req, "req committed");
715 req->rq_cb_data = NULL;
716 req->rq_commit_cb = NULL;
717 list_del_init(&req->rq_mod_list);
718 if (req->rq_sequence) {
719 list_for_each_entry_safe(cur, tmp, &mod->mod_replay_list,
721 LASSERT(cur != LP_POISON);
722 LASSERT(cur->rq_type != LI_POISON);
723 DEBUG_REQ(D_HA, cur, "req balanced");
724 LASSERT(cur->rq_transno != 0);
725 LASSERT(cur->rq_import == req->rq_import);
727 /* We no longer want to preserve this for transno-
728 * unconditional replay. */
729 spin_lock(&cur->rq_lock);
731 spin_unlock(&cur->rq_lock);
735 if (list_empty(&mod->mod_replay_list)) {
736 if (mod->mod_och != NULL)
737 mod->mod_och->och_mod = NULL;
743 int mdc_set_open_replay_data(struct obd_export *exp,
744 struct obd_client_handle *och,
745 struct ptlrpc_request *open_req)
747 struct md_open_data *mod;
748 struct mdt_rec_create *rec = lustre_msg_buf(open_req->rq_reqmsg,
751 struct mdt_body *body = lustre_msg_buf(open_req->rq_repmsg,
754 struct obd_import *imp = open_req->rq_import;
757 LASSERT(rec != NULL);
759 /* Incoming message in my byte order (it's been swabbed). */
760 LASSERT(lustre_rep_swabbed(open_req, DLM_REPLY_REC_OFF));
762 /* Outgoing messages always in my byte order. */
763 LASSERT(body != NULL);
765 /*Only the import is replayable, we set replay_open data */
766 if (och && imp->imp_replayable) {
767 OBD_ALLOC(mod, sizeof(*mod));
769 DEBUG_REQ(D_ERROR, open_req,
770 "Can't allocate md_open_data");
773 CFS_INIT_LIST_HEAD(&mod->mod_replay_list);
775 spin_lock(&open_req->rq_lock);
776 if (!open_req->rq_replay) {
777 OBD_FREE(mod, sizeof(*mod));
778 spin_unlock(&open_req->rq_lock);
784 open_req->rq_cb_data = mod;
785 list_add_tail(&open_req->rq_mod_list, &mod->mod_replay_list);
786 open_req->rq_commit_cb = mdc_commit_delayed;
787 spin_unlock(&open_req->rq_lock);
790 rec->cr_fid2 = body->fid1;
791 rec->cr_ioepoch = body->ioepoch;
792 rec->cr_old_handle.cookie = body->handle.cookie;
793 open_req->rq_replay_cb = mdc_replay_open;
794 if (!fid_is_sane(&body->fid1)) {
795 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
800 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
804 int mdc_clear_open_replay_data(struct obd_export *exp,
805 struct obd_client_handle *och)
807 struct md_open_data *mod = och->och_mod;
811 * Don't free the structure now (it happens in mdc_commit_delayed(),
812 * after the last request is removed from its replay list),
813 * but make sure that replay doesn't poke at the och, which is about to
816 LASSERT(mod != LP_POISON);
824 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
825 struct md_open_data *mod, struct ptlrpc_request **request)
827 struct obd_device *obd = class_exp2obd(exp);
828 int reqsize[4] = { sizeof(struct ptlrpc_body),
829 sizeof(struct mdt_epoch),
830 sizeof(struct mdt_rec_setattr)};
831 int repsize[4] = { sizeof(struct ptlrpc_body),
832 sizeof(struct mdt_body),
833 obd->u.cli.cl_max_mds_easize,
834 obd->u.cli.cl_max_mds_cookiesize };
835 struct ptlrpc_request *req;
839 reqsize[REQ_REC_OFF + 2] = op_data->op_capa1 ?
840 sizeof(struct lustre_capa) : 0;
841 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
842 MDS_CLOSE, 4, reqsize, NULL);
844 GOTO(out, rc = -ENOMEM);
846 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
847 * portal whose threads are not taking any DLM locks and are therefore
848 * always progressing */
849 /* XXX FIXME bug 249 */
850 req->rq_request_portal = MDS_READPAGE_PORTAL;
852 /* Ensure that this close's handle is fixed up during replay. */
853 if (likely(mod != NULL))
854 list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
856 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
858 mdc_close_pack(req, REQ_REC_OFF, op_data);
859 ptlrpc_req_set_repsize(req, 4, repsize);
860 req->rq_commit_cb = mdc_commit_delayed;
862 LASSERT(req->rq_cb_data == NULL);
863 req->rq_cb_data = mod;
865 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
866 rc = ptlrpc_queue_wait(req);
867 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
869 if (req->rq_repmsg == NULL) {
870 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
873 rc = req->rq_status ? req->rq_status : -EIO;
874 } else if (rc == 0 || rc == -EAGAIN) {
875 rc = lustre_msg_get_status(req->rq_repmsg);
876 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
877 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
881 } else if (mod == NULL) {
882 if (req->rq_import->imp_replayable)
883 CERROR("Unexpected: can't find md_open_data,"
884 "but close succeeded with replayable imp"
885 "Please tell CFS.\n");
887 if (!lustre_swab_repbuf(req, REPLY_REC_OFF,
888 sizeof(struct mdt_body),
889 lustre_swab_mdt_body)) {
890 CERROR("Error unpacking mdt_body\n");
898 if (rc != 0 && rc != -EAGAIN && req && req->rq_commit_cb)
899 req->rq_commit_cb(req);
904 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
905 struct md_open_data *mod)
907 struct obd_device *obd = class_exp2obd(exp);
908 struct ptlrpc_request *req;
909 int size[4] = { sizeof(struct ptlrpc_body),
910 sizeof(struct mdt_epoch),
911 sizeof(struct mdt_rec_setattr)};
912 int repsize[2] = { sizeof(struct ptlrpc_body),
913 sizeof(struct mdt_body)};
917 if (op_data->op_capa1)
918 size[REQ_REC_OFF + 2] = sizeof(struct lustre_capa);
919 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
920 MDS_DONE_WRITING, 4, size, NULL);
924 mdc_close_pack(req, REQ_REC_OFF, op_data);
927 req->rq_cb_data = mod;
928 req->rq_commit_cb = mdc_commit_delayed;
929 if (likely(mod != NULL))
930 list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
932 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
934 ptlrpc_req_set_repsize(req, 2, repsize);
935 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
936 rc = ptlrpc_queue_wait(req);
937 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
939 /* Close the open replay sequence if an error occured or no SOM
940 * attribute update is needed. */
942 ptlrpc_close_replay_seq(req);
944 if (rc && rc != -EAGAIN && req->rq_commit_cb)
945 req->rq_commit_cb(req);
947 ptlrpc_req_finished(req);
951 #ifdef HAVE_SPLIT_SUPPORT
952 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
953 const struct page *page, int offset)
955 int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
956 struct obd_import *imp = class_exp2cliimp(exp);
957 struct ptlrpc_bulk_desc *desc = NULL;
958 struct ptlrpc_request *req = NULL;
961 CDEBUG(D_INODE, "object: "DFID"\n", PFID(fid));
963 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_WRITEPAGE, 3,
966 GOTO(out, rc = -ENOMEM);
968 req->rq_request_portal = MDS_READPAGE_PORTAL;
970 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
972 GOTO(out, rc = -ENOMEM);
974 /* NB req now owns desc and will free it when it gets freed. */
975 ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
976 mdc_readdir_pack(req, REQ_REC_OFF, 0, offset, fid, NULL);
978 ptlrpc_req_set_repsize(req, 2, size);
979 rc = ptlrpc_queue_wait(req);
983 ptlrpc_req_finished(req);
986 EXPORT_SYMBOL(mdc_sendpage);
989 int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
990 struct obd_capa *oc, __u64 offset, struct page *page,
991 struct ptlrpc_request **request)
993 int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
994 struct obd_import *imp = class_exp2cliimp(exp);
995 struct ptlrpc_bulk_desc *desc = NULL;
996 struct ptlrpc_request *req = NULL;
997 struct mdt_body *body;
1000 CDEBUG(D_INODE, "object: "DFID"\n", PFID(fid));
1002 size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1003 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE, 3, size,
1006 GOTO(out, rc = -ENOMEM);
1008 /* XXX FIXME bug 249 */
1009 req->rq_request_portal = MDS_READPAGE_PORTAL;
1011 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
1013 GOTO(out, rc = -ENOMEM);
1015 /* NB req now owns desc and will free it when it gets freed */
1016 ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
1017 mdc_readdir_pack(req, REQ_REC_OFF, offset, CFS_PAGE_SIZE, fid, oc);
1019 ptlrpc_req_set_repsize(req, 2, size);
1020 rc = ptlrpc_queue_wait(req);
1023 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1024 lustre_swab_mdt_body);
1026 CERROR("Can't unpack mdt_body\n");
1027 GOTO(out, rc = -EPROTO);
1030 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
1031 CERROR ("Unexpected # bytes transferred: %d"
1032 " (%ld expected)\n",
1033 req->rq_bulk->bd_nob_transferred,
1035 GOTO(out, rc = -EPROTO);
1045 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1046 void *karg, void *uarg)
1048 struct obd_device *obd = exp->exp_obd;
1049 struct obd_ioctl_data *data = karg;
1050 struct obd_import *imp = obd->u.cli.cl_import;
1051 struct llog_ctxt *ctxt;
1055 if (!try_module_get(THIS_MODULE)) {
1056 CERROR("Can't get module. Is it alive?");
1060 case OBD_IOC_CLIENT_RECOVER:
1061 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
1065 case IOC_OSC_SET_ACTIVE:
1066 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1068 case OBD_IOC_PARSE: {
1069 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1070 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
1074 case OBD_IOC_LLOG_INFO:
1075 case OBD_IOC_LLOG_PRINT: {
1076 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1077 rc = llog_ioctl(ctxt, cmd, data);
1082 case OBD_IOC_POLL_QUOTACHECK:
1083 rc = lquota_poll_check(quota_interface, exp,
1084 (struct if_quotacheck *)karg);
1087 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1088 GOTO(out, rc = -ENOTTY);
1091 module_put(THIS_MODULE);
1096 int mdc_set_info_async(struct obd_export *exp, obd_count keylen,
1097 void *key, obd_count vallen, void *val,
1098 struct ptlrpc_request_set *set)
1100 struct obd_import *imp = class_exp2cliimp(exp);
1103 if (KEY_IS(KEY_INIT_RECOV)) {
1104 if (vallen != sizeof(int))
1106 spin_lock(&imp->imp_lock);
1107 imp->imp_initial_recov = *(int *)val;
1108 spin_unlock(&imp->imp_lock);
1109 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
1110 exp->exp_obd->obd_name, imp->imp_initial_recov);
1113 /* Turn off initial_recov after we try all backup servers once */
1114 if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
1115 if (vallen != sizeof(int))
1117 spin_lock(&imp->imp_lock);
1118 imp->imp_initial_recov_bk = *(int *)val;
1119 if (imp->imp_initial_recov_bk)
1120 imp->imp_initial_recov = 1;
1121 spin_unlock(&imp->imp_lock);
1122 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
1123 exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
1126 if (KEY_IS(KEY_READ_ONLY)) {
1127 struct ptlrpc_request *req;
1128 int size[3] = { sizeof(struct ptlrpc_body), keylen, vallen };
1129 char *bufs[3] = { NULL, key, val };
1131 if (vallen != sizeof(int))
1134 spin_lock(&imp->imp_lock);
1135 if (*((int *)val)) {
1136 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
1137 imp->imp_connect_data.ocd_connect_flags |=
1140 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
1141 imp->imp_connect_data.ocd_connect_flags &=
1142 ~OBD_CONNECT_RDONLY;
1144 spin_unlock(&imp->imp_lock);
1146 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_SET_INFO,
1151 ptlrpc_req_set_repsize(req, 1, NULL);
1154 ptlrpc_set_add_req(set, req);
1155 ptlrpc_check_set(set);
1157 rc = ptlrpc_queue_wait(req);
1158 ptlrpc_req_finished(req);
1163 if (KEY_IS(KEY_FLUSH_CTX)) {
1164 sptlrpc_import_flush_my_ctx(imp);
1167 if (KEY_IS(KEY_MDS_CONN)) {
1168 struct obd_import *imp = class_exp2cliimp(exp);
1170 /* mds-mds import */
1171 spin_lock(&imp->imp_lock);
1172 imp->imp_server_timeout = 1;
1173 spin_unlock(&imp->imp_lock);
1174 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1175 CDEBUG(D_OTHER|D_WARNING, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1182 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
1183 __u32 *vallen, void *val)
1187 if (KEY_IS(KEY_MAX_EASIZE)) {
1188 int mdsize, *max_easize;
1190 if (*vallen != sizeof(int))
1192 mdsize = *(int*)val;
1193 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
1194 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
1196 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
1199 if (KEY_IS(KEY_CONN_DATA)) {
1200 struct obd_import *imp = class_exp2cliimp(exp);
1201 struct obd_connect_data *data = val;
1203 if (*vallen != sizeof(*data))
1206 *data = imp->imp_connect_data;
1213 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1216 struct ptlrpc_request *req;
1217 struct obd_statfs *msfs;
1218 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*msfs) };
1221 /* We could possibly pass max_age in the request (as an absolute
1222 * timestamp or a "seconds.usec ago") so the target can avoid doing
1223 * extra calls into the filesystem if that isn't necessary (e.g.
1224 * during mount that would help a bit). Having relative timestamps
1225 * is not so great if request processing is slow, while absolute
1226 * timestamps are not ideal because they need time synchronization. */
1227 req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
1228 MDS_STATFS, 1, NULL, NULL);
1232 ptlrpc_req_set_repsize(req, 2, size);
1234 rc = ptlrpc_queue_wait(req);
1237 /* check connection error first */
1238 if (obd->u.cli.cl_import->imp_connect_error)
1239 rc = obd->u.cli.cl_import->imp_connect_error;
1244 msfs = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*msfs),
1245 lustre_swab_obd_statfs);
1247 CERROR("Can't unpack obd_statfs\n");
1248 GOTO(out, rc = -EPROTO);
1251 memcpy(osfs, msfs, sizeof(*msfs));
1254 ptlrpc_req_finished(req);
1259 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
1260 struct obd_capa *oc,
1261 struct obd_client_handle *handle, int flag)
1263 struct ptlrpc_request *req;
1264 struct mdt_body *body;
1265 int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1268 size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1269 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1270 MDS_PIN, 3, size, NULL);
1274 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof (*body));
1277 mdc_pack_capa(req, REQ_REC_OFF + 1, oc);
1279 ptlrpc_req_set_repsize(req, 2, size);
1281 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1282 rc = ptlrpc_queue_wait(req);
1283 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1285 CERROR("pin failed: %d\n", rc);
1286 ptlrpc_req_finished(req);
1290 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1291 lustre_swab_mdt_body);
1293 ptlrpc_req_finished(req);
1297 memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
1298 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1300 OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
1301 if (handle->och_mod == NULL) {
1302 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
1306 /* will be dropped by unpin */
1307 CFS_INIT_LIST_HEAD(&handle->och_mod->mod_replay_list);
1308 list_add_tail(&req->rq_mod_list, &handle->och_mod->mod_replay_list);
1313 static int mdc_unpin(struct obd_export *exp,
1314 struct obd_client_handle *handle, int flag)
1316 struct ptlrpc_request *req;
1317 struct mdt_body *body;
1318 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1321 if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
1324 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1325 MDS_CLOSE, 2, size, NULL);
1329 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1330 memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
1333 ptlrpc_req_set_repsize(req, 1, NULL);
1334 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1335 rc = ptlrpc_queue_wait(req);
1336 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1339 CERROR("unpin failed: %d\n", rc);
1341 ptlrpc_req_finished(req);
1343 LASSERT(!list_empty(&handle->och_mod->mod_replay_list));
1344 req = list_entry(handle->och_mod->mod_replay_list.next,
1345 typeof(*req), rq_mod_list);
1346 list_del_init(&req->rq_mod_list);
1347 ptlrpc_req_finished(req);
1348 LASSERT(list_empty(&handle->och_mod->mod_replay_list));
1350 OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1354 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
1355 struct obd_capa *oc, struct ptlrpc_request **request)
1357 struct ptlrpc_request *req;
1358 int size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
1362 size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1363 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1364 MDS_SYNC, 3, size, NULL);
1366 RETURN(rc = -ENOMEM);
1368 mdc_pack_req_body(req, REQ_REC_OFF, 0, fid, oc, 0, -1, 0);
1370 ptlrpc_req_set_repsize(req, 2, size);
1372 rc = ptlrpc_queue_wait(req);
1373 if (rc || request == NULL)
1374 ptlrpc_req_finished(req);
1381 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1382 enum obd_import_event event)
1386 LASSERT(imp->imp_obd == obd);
1389 case IMP_EVENT_DISCON: {
1391 /* XXX Pass event up to OBDs stack. used only for FLD now */
1392 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
1396 case IMP_EVENT_INACTIVE: {
1397 struct client_obd *cli = &obd->u.cli;
1399 * Flush current sequence to make client obtain new one
1400 * from server in case of disconnect/reconnect.
1401 * If range is already empty then no need to flush it.
1403 if (cli->cl_seq != NULL &&
1404 !range_is_exhausted(&cli->cl_seq->lcs_space)) {
1405 seq_client_flush(cli->cl_seq);
1408 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1411 case IMP_EVENT_INVALIDATE: {
1412 struct ldlm_namespace *ns = obd->obd_namespace;
1414 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1418 case IMP_EVENT_ACTIVE: {
1419 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1423 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
1427 CERROR("Unknown import event %x\n", event);
1433 static int mdc_fid_init(struct obd_export *exp)
1435 struct client_obd *cli = &exp->exp_obd->u.cli;
1440 OBD_ALLOC_PTR(cli->cl_seq);
1441 if (cli->cl_seq == NULL)
1444 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
1446 GOTO(out_free_seq, rc = -ENOMEM);
1448 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
1449 exp->exp_obd->obd_name);
1451 /* Init client side sequence-manager */
1452 rc = seq_client_init(cli->cl_seq, exp,
1453 LUSTRE_SEQ_METADATA,
1455 OBD_FREE(prefix, MAX_OBD_NAME + 5);
1457 GOTO(out_free_seq, rc);
1461 OBD_FREE_PTR(cli->cl_seq);
1466 static int mdc_fid_fini(struct obd_export *exp)
1468 struct client_obd *cli = &exp->exp_obd->u.cli;
1471 if (cli->cl_seq != NULL) {
1472 seq_client_fini(cli->cl_seq);
1473 OBD_FREE_PTR(cli->cl_seq);
1480 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1481 struct md_op_data *op_data)
1483 struct client_obd *cli = &exp->exp_obd->u.cli;
1484 struct lu_client_seq *seq = cli->cl_seq;
1486 RETURN(seq_client_alloc_fid(seq, fid));
1489 /* XXX This method is used only to clear current fid seq
1490 * once fld/mds insert failed */
1491 static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1493 struct client_obd *cli = &exp->exp_obd->u.cli;
1495 seq_client_flush(cli->cl_seq);
1499 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
1501 struct client_obd *cli = &obd->u.cli;
1502 struct lprocfs_static_vars lvars = { 0 };
1506 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1507 if (!cli->cl_rpc_lock)
1509 mdc_init_rpc_lock(cli->cl_rpc_lock);
1513 OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1514 if (!cli->cl_setattr_lock)
1515 GOTO(err_rpc_lock, rc = -ENOMEM);
1516 mdc_init_rpc_lock(cli->cl_setattr_lock);
1518 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1519 if (!cli->cl_close_lock)
1520 GOTO(err_setattr_lock, rc = -ENOMEM);
1521 mdc_init_rpc_lock(cli->cl_close_lock);
1523 rc = client_obd_setup(obd, cfg);
1525 GOTO(err_close_lock, rc);
1526 lprocfs_mdc_init_vars(&lvars);
1527 lprocfs_obd_setup(obd, lvars.obd_vars);
1528 sptlrpc_lprocfs_cliobd_attach(obd);
1529 ptlrpc_lprocfs_register_obd(obd);
1531 rc = obd_llog_init(obd, NULL, obd, 0, NULL, NULL);
1534 CERROR("failed to setup llogging subsystems\n");
1540 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1542 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1544 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1549 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
1550 * us to make MDS RPCs with large enough reply buffers to hold the
1551 * maximum-sized (= maximum striped) EA and cookie without having to
1552 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1553 int mdc_init_ea_size(struct obd_export *exp, int easize,
1554 int def_easize, int cookiesize)
1556 struct obd_device *obd = exp->exp_obd;
1557 struct client_obd *cli = &obd->u.cli;
1560 if (cli->cl_max_mds_easize < easize)
1561 cli->cl_max_mds_easize = easize;
1563 if (cli->cl_default_mds_easize < def_easize)
1564 cli->cl_default_mds_easize = def_easize;
1566 if (cli->cl_max_mds_cookiesize < cookiesize)
1567 cli->cl_max_mds_cookiesize = cookiesize;
1572 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1578 case OBD_CLEANUP_EARLY:
1579 case OBD_CLEANUP_EXPORTS:
1580 /* If we set up but never connected, the
1581 client import will not have been cleaned. */
1582 if (obd->u.cli.cl_import) {
1583 struct obd_import *imp;
1584 imp = obd->u.cli.cl_import;
1585 CERROR("client import never connected\n");
1586 ptlrpc_invalidate_import(imp);
1587 ptlrpc_free_rq_pool(imp->imp_rq_pool);
1588 class_destroy_import(imp);
1589 obd->u.cli.cl_import = NULL;
1592 case OBD_CLEANUP_SELF_EXP:
1593 rc = obd_llog_finish(obd, 0);
1595 CERROR("failed to cleanup llogging subsystems\n");
1596 case OBD_CLEANUP_OBD:
1602 static int mdc_cleanup(struct obd_device *obd)
1604 struct client_obd *cli = &obd->u.cli;
1606 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1607 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1608 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1610 ptlrpc_lprocfs_unregister_obd(obd);
1611 lprocfs_obd_cleanup(obd);
1614 return client_obd_cleanup(obd);
1618 static int mdc_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1619 struct obd_device *tgt,
1620 int count, struct llog_catid *logid,
1621 struct obd_uuid *uuid)
1623 struct llog_ctxt *ctxt;
1627 rc = llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1630 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1631 ctxt->loc_imp = obd->u.cli.cl_import;
1634 rc = llog_setup(obd, llogs, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1637 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1638 ctxt->loc_imp = obd->u.cli.cl_import;
1644 static int mdc_llog_finish(struct obd_device *obd, int count)
1649 rc = llog_cleanup(llog_get_context(obd, LLOG_LOVEA_REPL_CTXT));
1651 CERROR("can not cleanup LLOG_CONFIG_REPL_CTXT rc %d\n", rc);
1653 rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
1657 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1659 struct lustre_cfg *lcfg = buf;
1660 struct lprocfs_static_vars lvars = { 0 };
1663 lprocfs_mdc_init_vars(&lvars);
1665 switch (lcfg->lcfg_command) {
1666 case LCFG_SPTLRPC_CONF:
1667 rc = sptlrpc_cliobd_process_config(obd, lcfg);
1670 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
1677 /* get remote permission for current user on fid */
1678 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
1679 struct obd_capa *oc, __u32 suppgid,
1680 struct ptlrpc_request **request)
1682 struct ptlrpc_request *req;
1683 struct mdt_body *body;
1684 struct mdt_remote_perm *perm;
1685 int size[5] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1689 size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
1692 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1693 MDS_GETATTR, 3, size, NULL);
1697 mdc_pack_req_body(req, REQ_REC_OFF, OBD_MD_FLRMTPERM, fid, oc, 0,
1700 size[REPLY_REC_OFF + 1] = sizeof(*perm);
1701 ptlrpc_req_set_repsize(req, 5, size);
1702 rc = ptlrpc_queue_wait(req);
1704 ptlrpc_req_finished(req);
1708 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1709 lustre_swab_mdt_body);
1711 LASSERT(body->valid & OBD_MD_FLRMTPERM);
1713 perm = lustre_swab_repbuf(req, REPLY_REC_OFF + 1, sizeof(*perm),
1714 lustre_swab_mdt_remote_perm);
1721 static int mdc_interpret_renew_capa(struct ptlrpc_request *req, void *unused,
1724 struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
1725 renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
1726 struct mdt_body *body = NULL;
1727 struct lustre_capa *capa;
1731 GOTO(out, capa = ERR_PTR(status));
1733 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1734 lustre_swab_mdt_body);
1736 GOTO(out, capa = ERR_PTR(-EFAULT));
1738 if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
1739 GOTO(out, capa = ERR_PTR(-ENOENT));
1741 capa = lustre_unpack_capa(req->rq_repmsg, REPLY_REC_OFF + 1);
1743 GOTO(out, capa = ERR_PTR(-EFAULT));
1750 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
1753 struct ptlrpc_request *req;
1754 int size[5] = { sizeof(struct ptlrpc_body),
1755 sizeof(struct mdt_body),
1756 sizeof(struct lustre_capa) };
1759 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1760 MDS_GETATTR, 3, size, NULL);
1764 mdc_pack_req_body(req, REQ_REC_OFF, OBD_MD_FLOSSCAPA,
1765 &oc->c_capa.lc_fid, oc, 0, -1, 0);
1767 ptlrpc_req_set_repsize(req, 5, size);
1768 req->rq_async_args.pointer_arg[0] = oc;
1769 req->rq_async_args.pointer_arg[1] = cb;
1770 req->rq_interpret_reply = mdc_interpret_renew_capa;
1771 ptlrpcd_add_req(req);
1776 static int mdc_connect(const struct lu_env *env,
1777 struct lustre_handle *dlm_handle,
1778 struct obd_device *obd, struct obd_uuid *cluuid,
1779 struct obd_connect_data *data) {
1780 struct obd_import *imp = obd->u.cli.cl_import;
1782 /* mds-mds import features */
1783 if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
1784 spin_lock(&imp->imp_lock);
1785 imp->imp_server_timeout = 1;
1786 spin_unlock(&imp->imp_lock);
1787 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1788 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
1792 return client_connect_import(env, dlm_handle, obd, cluuid, data);
1795 struct obd_ops mdc_obd_ops = {
1796 .o_owner = THIS_MODULE,
1797 .o_setup = mdc_setup,
1798 .o_precleanup = mdc_precleanup,
1799 .o_cleanup = mdc_cleanup,
1800 .o_add_conn = client_import_add_conn,
1801 .o_del_conn = client_import_del_conn,
1802 .o_connect = mdc_connect,
1803 .o_disconnect = client_disconnect_export,
1804 .o_iocontrol = mdc_iocontrol,
1805 .o_set_info_async = mdc_set_info_async,
1806 .o_statfs = mdc_statfs,
1808 .o_unpin = mdc_unpin,
1809 .o_fid_init = mdc_fid_init,
1810 .o_fid_fini = mdc_fid_fini,
1811 .o_fid_alloc = mdc_fid_alloc,
1812 .o_fid_delete = mdc_fid_delete,
1813 .o_import_event = mdc_import_event,
1814 .o_llog_init = mdc_llog_init,
1815 .o_llog_finish = mdc_llog_finish,
1816 .o_get_info = mdc_get_info,
1817 .o_process_config = mdc_process_config,
1820 struct md_ops mdc_md_ops = {
1821 .m_getstatus = mdc_getstatus,
1822 .m_change_cbdata = mdc_change_cbdata,
1823 .m_close = mdc_close,
1824 .m_create = mdc_create,
1825 .m_done_writing = mdc_done_writing,
1826 .m_enqueue = mdc_enqueue,
1827 .m_getattr = mdc_getattr,
1828 .m_getattr_name = mdc_getattr_name,
1829 .m_intent_lock = mdc_intent_lock,
1831 .m_is_subdir = mdc_is_subdir,
1832 .m_rename = mdc_rename,
1833 .m_setattr = mdc_setattr,
1834 .m_setxattr = mdc_setxattr,
1835 .m_getxattr = mdc_getxattr,
1837 .m_readpage = mdc_readpage,
1838 .m_unlink = mdc_unlink,
1839 .m_cancel_unused = mdc_cancel_unused,
1840 .m_init_ea_size = mdc_init_ea_size,
1841 .m_set_lock_data = mdc_set_lock_data,
1842 .m_lock_match = mdc_lock_match,
1843 .m_get_lustre_md = mdc_get_lustre_md,
1844 .m_free_lustre_md = mdc_free_lustre_md,
1845 .m_set_open_replay_data = mdc_set_open_replay_data,
1846 .m_clear_open_replay_data = mdc_clear_open_replay_data,
1847 .m_get_remote_perm = mdc_get_remote_perm,
1848 .m_renew_capa = mdc_renew_capa
1851 extern quota_interface_t mdc_quota_interface;
1853 int __init mdc_init(void)
1856 struct lprocfs_static_vars lvars = { 0 };
1857 lprocfs_mdc_init_vars(&lvars);
1859 request_module("lquota");
1860 quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1861 init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1863 rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1864 LUSTRE_MDC_NAME, NULL);
1865 if (rc && quota_interface)
1866 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1872 static void /*__exit*/ mdc_exit(void)
1874 if (quota_interface)
1875 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1877 class_unregister_type(LUSTRE_MDC_NAME);
1880 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1881 MODULE_DESCRIPTION("Lustre Metadata Client");
1882 MODULE_LICENSE("GPL");
1884 module_init(mdc_init);
1885 module_exit(mdc_exit);