4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_MDC
39 # include <linux/module.h>
40 # include <linux/pagemap.h>
41 # include <linux/miscdevice.h>
42 # include <linux/init.h>
43 # include <linux/utsname.h>
45 #include "../include/lustre_acl.h"
46 #include "../include/obd_class.h"
47 #include "../include/lustre_fid.h"
48 #include "../include/lprocfs_status.h"
49 #include "../include/lustre_param.h"
50 #include "../include/lustre_log.h"
52 #include "mdc_internal.h"
54 #define REQUEST_MINOR 244
56 struct mdc_renew_capa_args {
57 struct obd_capa *ra_oc;
58 renew_capa_cb_t ra_cb;
61 static int mdc_cleanup(struct obd_device *obd);
63 int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
64 const struct req_msg_field *field, struct obd_capa **oc)
66 struct lustre_capa *capa;
69 /* swabbed already in mdc_enqueue */
70 capa = req_capsule_server_get(&req->rq_pill, field);
74 c = alloc_capa(CAPA_SITE_CLIENT);
76 CDEBUG(D_INFO, "alloc capa failed!\n");
85 static inline int mdc_queue_wait(struct ptlrpc_request *req)
87 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
90 /* mdc_enter_request() ensures that this client has no more
91 * than cl_max_rpcs_in_flight RPCs simultaneously inf light
93 rc = mdc_enter_request(cli);
97 rc = ptlrpc_queue_wait(req);
98 mdc_exit_request(cli);
103 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
104 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
105 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
106 struct obd_capa **pc, int level, int msg_flags)
108 struct ptlrpc_request *req;
109 struct mdt_body *body;
112 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
113 LUSTRE_MDS_VERSION, MDS_GETSTATUS);
117 mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
118 lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
119 req->rq_send_state = level;
121 ptlrpc_request_set_replen(req);
123 rc = ptlrpc_queue_wait(req);
127 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
129 GOTO(out, rc = -EPROTO);
131 if (body->valid & OBD_MD_FLMDSCAPA) {
132 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
137 *rootfid = body->fid1;
139 "root fid="DFID", last_committed=%llu\n",
141 lustre_msg_get_last_committed(req->rq_repmsg));
143 ptlrpc_req_finished(req);
147 /* This should be mdc_get_info("rootfid") */
148 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
149 struct obd_capa **pc)
151 return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
156 * This function now is known to always saying that it will receive 4 buffers
157 * from server. Even for cases when acl_size and md_size is zero, RPC header
158 * will contain 4 fields and RPC itself will contain zero size fields. This is
159 * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
160 * and thus zero, it shrinks it, making zero size. The same story about
161 * md_size. And this is course of problem when client waits for smaller number
162 * of fields. This issue will be fixed later when client gets aware of RPC
165 static int mdc_getattr_common(struct obd_export *exp,
166 struct ptlrpc_request *req)
168 struct req_capsule *pill = &req->rq_pill;
169 struct mdt_body *body;
173 /* Request message already built. */
174 rc = ptlrpc_queue_wait(req);
178 /* sanity check for the reply */
179 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
183 CDEBUG(D_NET, "mode: %o\n", body->mode);
185 if (body->eadatasize != 0) {
186 mdc_update_max_ea_from_body(exp, body);
188 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
194 if (body->valid & OBD_MD_FLRMTPERM) {
195 struct mdt_remote_perm *perm;
197 LASSERT(client_is_remote(exp));
198 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
199 lustre_swab_mdt_remote_perm);
204 if (body->valid & OBD_MD_FLMDSCAPA) {
205 struct lustre_capa *capa;
206 capa = req_capsule_server_get(pill, &RMF_CAPA1);
214 int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
215 struct ptlrpc_request **request)
217 struct ptlrpc_request *req;
220 /* Single MDS without an LMV case */
221 if (op_data->op_flags & MF_GET_MDT_IDX) {
226 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
230 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
232 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
234 ptlrpc_request_free(req);
238 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
239 op_data->op_valid, op_data->op_mode, -1, 0);
241 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
243 if (op_data->op_valid & OBD_MD_FLRMTPERM) {
244 LASSERT(client_is_remote(exp));
245 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
246 sizeof(struct mdt_remote_perm));
248 ptlrpc_request_set_replen(req);
250 rc = mdc_getattr_common(exp, req);
252 ptlrpc_req_finished(req);
258 int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
259 struct ptlrpc_request **request)
261 struct ptlrpc_request *req;
265 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
266 &RQF_MDS_GETATTR_NAME);
270 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
271 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
272 op_data->op_namelen + 1);
274 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
276 ptlrpc_request_free(req);
280 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
281 op_data->op_valid, op_data->op_mode,
282 op_data->op_suppgids[0], 0);
284 if (op_data->op_name) {
285 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
286 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
287 op_data->op_namelen);
288 memcpy(name, op_data->op_name, op_data->op_namelen);
291 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
293 ptlrpc_request_set_replen(req);
295 rc = mdc_getattr_common(exp, req);
297 ptlrpc_req_finished(req);
303 static int mdc_is_subdir(struct obd_export *exp,
304 const struct lu_fid *pfid,
305 const struct lu_fid *cfid,
306 struct ptlrpc_request **request)
308 struct ptlrpc_request *req;
312 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
313 &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
318 mdc_is_subdir_pack(req, pfid, cfid, 0);
319 ptlrpc_request_set_replen(req);
321 rc = ptlrpc_queue_wait(req);
322 if (rc && rc != -EREMOTE)
323 ptlrpc_req_finished(req);
329 static int mdc_xattr_common(struct obd_export *exp,
330 const struct req_format *fmt,
331 const struct lu_fid *fid,
332 struct obd_capa *oc, int opcode, obd_valid valid,
333 const char *xattr_name, const char *input,
334 int input_size, int output_size, int flags,
335 __u32 suppgid, struct ptlrpc_request **request)
337 struct ptlrpc_request *req;
338 int xattr_namelen = 0;
343 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
347 mdc_set_capa_size(req, &RMF_CAPA1, oc);
349 xattr_namelen = strlen(xattr_name) + 1;
350 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
355 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
359 /* Flush local XATTR locks to get rid of a possible cancel RPC */
360 if (opcode == MDS_REINT && fid_is_sane(fid) &&
361 exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
365 /* Without that packing would fail */
367 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
370 count = mdc_resource_get_unused(exp, fid,
372 MDS_INODELOCK_XATTR);
374 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
376 ptlrpc_request_free(req);
380 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
382 ptlrpc_request_free(req);
387 if (opcode == MDS_REINT) {
388 struct mdt_rec_setxattr *rec;
390 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
391 sizeof(struct mdt_rec_reint));
392 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
393 rec->sx_opcode = REINT_SETXATTR;
394 rec->sx_fsuid = from_kuid(&init_user_ns, current_fsuid());
395 rec->sx_fsgid = from_kgid(&init_user_ns, current_fsgid());
396 rec->sx_cap = cfs_curproc_cap_pack();
397 rec->sx_suppgid1 = suppgid;
398 rec->sx_suppgid2 = -1;
400 rec->sx_valid = valid | OBD_MD_FLCTIME;
401 rec->sx_time = get_seconds();
402 rec->sx_size = output_size;
403 rec->sx_flags = flags;
405 mdc_pack_capa(req, &RMF_CAPA1, oc);
407 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
411 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
412 memcpy(tmp, xattr_name, xattr_namelen);
415 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
416 memcpy(tmp, input, input_size);
419 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
420 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
421 RCL_SERVER, output_size);
422 ptlrpc_request_set_replen(req);
425 if (opcode == MDS_REINT)
426 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
428 rc = ptlrpc_queue_wait(req);
430 if (opcode == MDS_REINT)
431 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
434 ptlrpc_req_finished(req);
440 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
441 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
442 const char *input, int input_size, int output_size,
443 int flags, __u32 suppgid, struct ptlrpc_request **request)
445 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
446 fid, oc, MDS_REINT, valid, xattr_name,
447 input, input_size, output_size, flags,
451 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
452 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
453 const char *input, int input_size, int output_size,
454 int flags, struct ptlrpc_request **request)
456 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
457 fid, oc, MDS_GETXATTR, valid, xattr_name,
458 input, input_size, output_size, flags,
462 #ifdef CONFIG_FS_POSIX_ACL
463 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
465 struct req_capsule *pill = &req->rq_pill;
466 struct mdt_body *body = md->body;
467 struct posix_acl *acl;
474 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
479 acl = posix_acl_from_xattr(&init_user_ns, buf, body->aclsize);
482 CERROR("convert xattr to acl: %d\n", rc);
486 rc = posix_acl_valid(acl);
488 CERROR("validate acl: %d\n", rc);
489 posix_acl_release(acl);
497 #define mdc_unpack_acl(req, md) 0
500 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
501 struct obd_export *dt_exp, struct obd_export *md_exp,
502 struct lustre_md *md)
504 struct req_capsule *pill = &req->rq_pill;
508 memset(md, 0, sizeof(*md));
510 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
511 LASSERT(md->body != NULL);
513 if (md->body->valid & OBD_MD_FLEASIZE) {
515 struct lov_mds_md *lmm;
517 if (!S_ISREG(md->body->mode)) {
518 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
519 "regular file, but is not\n");
520 GOTO(out, rc = -EPROTO);
523 if (md->body->eadatasize == 0) {
524 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
525 "but eadatasize 0\n");
526 GOTO(out, rc = -EPROTO);
528 lmmsize = md->body->eadatasize;
529 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
531 GOTO(out, rc = -EPROTO);
533 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
537 if (rc < sizeof(*md->lsm)) {
538 CDEBUG(D_INFO, "lsm size too small: "
539 "rc < sizeof (*md->lsm) (%d < %d)\n",
540 rc, (int)sizeof(*md->lsm));
541 GOTO(out, rc = -EPROTO);
544 } else if (md->body->valid & OBD_MD_FLDIREA) {
546 struct lov_mds_md *lmv;
548 if (!S_ISDIR(md->body->mode)) {
549 CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
550 "directory, but is not\n");
551 GOTO(out, rc = -EPROTO);
554 if (md->body->eadatasize == 0) {
555 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
556 "but eadatasize 0\n");
559 if (md->body->valid & OBD_MD_MEA) {
560 lmvsize = md->body->eadatasize;
561 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
564 GOTO(out, rc = -EPROTO);
566 rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
571 if (rc < sizeof(*md->mea)) {
572 CDEBUG(D_INFO, "size too small: "
573 "rc < sizeof(*md->mea) (%d < %d)\n",
574 rc, (int)sizeof(*md->mea));
575 GOTO(out, rc = -EPROTO);
581 if (md->body->valid & OBD_MD_FLRMTPERM) {
582 /* remote permission */
583 LASSERT(client_is_remote(exp));
584 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
585 lustre_swab_mdt_remote_perm);
586 if (!md->remote_perm)
587 GOTO(out, rc = -EPROTO);
589 else if (md->body->valid & OBD_MD_FLACL) {
590 /* for ACL, it's possible that FLACL is set but aclsize is zero.
591 * only when aclsize != 0 there's an actual segment for ACL
594 if (md->body->aclsize) {
595 rc = mdc_unpack_acl(req, md);
598 #ifdef CONFIG_FS_POSIX_ACL
600 md->posix_acl = NULL;
604 if (md->body->valid & OBD_MD_FLMDSCAPA) {
605 struct obd_capa *oc = NULL;
607 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
613 if (md->body->valid & OBD_MD_FLOSSCAPA) {
614 struct obd_capa *oc = NULL;
616 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
625 capa_put(md->oss_capa);
629 capa_put(md->mds_capa);
632 #ifdef CONFIG_FS_POSIX_ACL
633 posix_acl_release(md->posix_acl);
636 obd_free_memmd(dt_exp, &md->lsm);
641 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
647 * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
650 void mdc_replay_open(struct ptlrpc_request *req)
652 struct md_open_data *mod = req->rq_cb_data;
653 struct ptlrpc_request *close_req;
654 struct obd_client_handle *och;
655 struct lustre_handle old;
656 struct mdt_body *body;
659 DEBUG_REQ(D_ERROR, req,
660 "Can't properly replay without open data.");
664 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
665 LASSERT(body != NULL);
669 struct lustre_handle *file_fh;
671 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
673 file_fh = &och->och_fh;
674 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
675 file_fh->cookie, body->handle.cookie);
677 *file_fh = body->handle;
679 close_req = mod->mod_close_req;
680 if (close_req != NULL) {
681 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
682 struct mdt_ioepoch *epoch;
684 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
685 epoch = req_capsule_client_get(&close_req->rq_pill,
690 LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
691 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
692 epoch->handle = body->handle;
696 void mdc_commit_open(struct ptlrpc_request *req)
698 struct md_open_data *mod = req->rq_cb_data;
703 * No need to touch md_open_data::mod_och, it holds a reference on
704 * \var mod and will zero references to each other, \var mod will be
705 * freed after that when md_open_data::mod_och will put the reference.
709 * Do not let open request to disappear as it still may be needed
710 * for close rpc to happen (it may happen on evict only, otherwise
711 * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
712 * called), just mark this rpc as committed to distinguish these 2
713 * cases, see mdc_close() for details. The open request reference will
714 * be put along with freeing \var mod.
716 ptlrpc_request_addref(req);
717 spin_lock(&req->rq_lock);
718 req->rq_committed = 1;
719 spin_unlock(&req->rq_lock);
720 req->rq_cb_data = NULL;
724 int mdc_set_open_replay_data(struct obd_export *exp,
725 struct obd_client_handle *och,
726 struct lookup_intent *it)
728 struct md_open_data *mod;
729 struct mdt_rec_create *rec;
730 struct mdt_body *body;
731 struct ptlrpc_request *open_req = it->d.lustre.it_data;
732 struct obd_import *imp = open_req->rq_import;
734 if (!open_req->rq_replay)
737 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
738 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
739 LASSERT(rec != NULL);
740 /* Incoming message in my byte order (it's been swabbed). */
741 /* Outgoing messages always in my byte order. */
742 LASSERT(body != NULL);
744 /* Only if the import is replayable, we set replay_open data */
745 if (och && imp->imp_replayable) {
746 mod = obd_mod_alloc();
748 DEBUG_REQ(D_ERROR, open_req,
749 "Can't allocate md_open_data");
754 * Take a reference on \var mod, to be freed on mdc_close().
755 * It protects \var mod from being freed on eviction (commit
756 * callback is called despite rq_replay flag).
757 * Another reference for \var och.
762 spin_lock(&open_req->rq_lock);
765 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
766 it_disposition(it, DISP_OPEN_STRIPE);
767 mod->mod_open_req = open_req;
768 open_req->rq_cb_data = mod;
769 open_req->rq_commit_cb = mdc_commit_open;
770 spin_unlock(&open_req->rq_lock);
773 rec->cr_fid2 = body->fid1;
774 rec->cr_ioepoch = body->ioepoch;
775 rec->cr_old_handle.cookie = body->handle.cookie;
776 open_req->rq_replay_cb = mdc_replay_open;
777 if (!fid_is_sane(&body->fid1)) {
778 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
783 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
787 static void mdc_free_open(struct md_open_data *mod)
791 if (mod->mod_is_create == 0 &&
792 imp_connect_disp_stripe(mod->mod_open_req->rq_import))
795 LASSERT(mod->mod_open_req->rq_replay == 0);
797 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
799 ptlrpc_request_committed(mod->mod_open_req, committed);
800 if (mod->mod_close_req)
801 ptlrpc_request_committed(mod->mod_close_req, committed);
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;
810 * It is possible to not have \var mod in a case of eviction between
811 * lookup and ll_file_open().
816 LASSERT(mod != LP_POISON);
817 LASSERT(mod->mod_open_req != NULL);
827 /* Prepares the request for the replay by the given reply */
828 static void mdc_close_handle_reply(struct ptlrpc_request *req,
829 struct md_op_data *op_data, int rc) {
830 struct mdt_body *repbody;
831 struct mdt_ioepoch *epoch;
833 if (req && rc == -EAGAIN) {
834 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
835 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
837 epoch->flags |= MF_SOM_AU;
838 if (repbody->valid & OBD_MD_FLGETATTRLOCK)
839 op_data->op_flags |= MF_GETATTR_LOCK;
843 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
844 struct md_open_data *mod, struct ptlrpc_request **request)
846 struct obd_device *obd = class_exp2obd(exp);
847 struct ptlrpc_request *req;
848 struct req_format *req_fmt;
853 req_fmt = &RQF_MDS_CLOSE;
854 if (op_data->op_bias & MDS_HSM_RELEASE) {
855 req_fmt = &RQF_MDS_RELEASE_CLOSE;
857 /* allocate a FID for volatile file */
858 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
860 CERROR("%s: "DFID" failed to allocate FID: %d\n",
861 obd->obd_name, PFID(&op_data->op_fid1), rc);
862 /* save the errcode and proceed to close */
868 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
872 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
874 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
876 ptlrpc_request_free(req);
880 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
881 * portal whose threads are not taking any DLM locks and are therefore
882 * always progressing */
883 req->rq_request_portal = MDS_READPAGE_PORTAL;
884 ptlrpc_at_set_req_timeout(req);
886 /* Ensure that this close's handle is fixed up during replay. */
887 if (likely(mod != NULL)) {
888 LASSERTF(mod->mod_open_req != NULL &&
889 mod->mod_open_req->rq_type != LI_POISON,
890 "POISONED open %p!\n", mod->mod_open_req);
892 mod->mod_close_req = req;
894 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
895 /* We no longer want to preserve this open for replay even
896 * though the open was committed. b=3632, b=3633 */
897 spin_lock(&mod->mod_open_req->rq_lock);
898 mod->mod_open_req->rq_replay = 0;
899 spin_unlock(&mod->mod_open_req->rq_lock);
902 "couldn't find open req; expecting close error\n");
905 mdc_close_pack(req, op_data);
907 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
908 obd->u.cli.cl_default_mds_easize);
909 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
910 obd->u.cli.cl_default_mds_cookiesize);
912 ptlrpc_request_set_replen(req);
914 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
915 rc = ptlrpc_queue_wait(req);
916 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
918 if (req->rq_repmsg == NULL) {
919 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
922 rc = req->rq_status ?: -EIO;
923 } else if (rc == 0 || rc == -EAGAIN) {
924 struct mdt_body *body;
926 rc = lustre_msg_get_status(req->rq_repmsg);
927 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
928 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
933 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
936 } else if (rc == -ESTALE) {
938 * it can be allowed error after 3633 if open was committed and
939 * server failed before close was sent. Let's check if mod
940 * exists and return no error in that case
943 DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
944 LASSERT(mod->mod_open_req != NULL);
945 if (mod->mod_open_req->rq_committed)
952 mod->mod_close_req = NULL;
953 /* Since now, mod is accessed through open_req only,
954 * thus close req does not keep a reference on mod anymore. */
958 mdc_close_handle_reply(req, op_data, rc);
959 return rc < 0 ? rc : saved_rc;
962 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
963 struct md_open_data *mod)
965 struct obd_device *obd = class_exp2obd(exp);
966 struct ptlrpc_request *req;
969 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
970 &RQF_MDS_DONE_WRITING);
974 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
975 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
977 ptlrpc_request_free(req);
982 LASSERTF(mod->mod_open_req != NULL &&
983 mod->mod_open_req->rq_type != LI_POISON,
984 "POISONED setattr %p!\n", mod->mod_open_req);
986 mod->mod_close_req = req;
987 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
988 /* We no longer want to preserve this setattr for replay even
989 * though the open was committed. b=3632, b=3633 */
990 spin_lock(&mod->mod_open_req->rq_lock);
991 mod->mod_open_req->rq_replay = 0;
992 spin_unlock(&mod->mod_open_req->rq_lock);
995 mdc_close_pack(req, op_data);
996 ptlrpc_request_set_replen(req);
998 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
999 rc = ptlrpc_queue_wait(req);
1000 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
1002 if (rc == -ESTALE) {
1004 * it can be allowed error after 3633 if open or setattr were
1005 * committed and server failed before close was sent.
1006 * Let's check if mod exists and return no error in that case
1009 LASSERT(mod->mod_open_req != NULL);
1010 if (mod->mod_open_req->rq_committed)
1017 mod->mod_close_req = NULL;
1018 LASSERT(mod->mod_open_req != NULL);
1021 /* Since now, mod is accessed through setattr req only,
1022 * thus DW req does not keep a reference on mod anymore. */
1026 mdc_close_handle_reply(req, op_data, rc);
1027 ptlrpc_req_finished(req);
1032 int mdc_readpage(struct obd_export *exp, struct md_op_data *op_data,
1033 struct page **pages, struct ptlrpc_request **request)
1035 struct ptlrpc_request *req;
1036 struct ptlrpc_bulk_desc *desc;
1038 wait_queue_head_t waitq;
1040 struct l_wait_info lwi;
1044 init_waitqueue_head(&waitq);
1047 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1051 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1053 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1055 ptlrpc_request_free(req);
1059 req->rq_request_portal = MDS_READPAGE_PORTAL;
1060 ptlrpc_at_set_req_timeout(req);
1062 desc = ptlrpc_prep_bulk_imp(req, op_data->op_npages, 1, BULK_PUT_SINK,
1065 ptlrpc_request_free(req);
1069 /* NB req now owns desc and will free it when it gets freed */
1070 for (i = 0; i < op_data->op_npages; i++)
1071 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
1073 mdc_readdir_pack(req, op_data->op_offset,
1074 PAGE_CACHE_SIZE * op_data->op_npages,
1075 &op_data->op_fid1, op_data->op_capa1);
1077 ptlrpc_request_set_replen(req);
1078 rc = ptlrpc_queue_wait(req);
1080 ptlrpc_req_finished(req);
1081 if (rc != -ETIMEDOUT)
1085 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1086 CERROR("too many resend retries, returning error\n");
1089 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends),
1091 l_wait_event(waitq, 0, &lwi);
1096 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1097 req->rq_bulk->bd_nob_transferred);
1099 ptlrpc_req_finished(req);
1103 if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1104 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1105 req->rq_bulk->bd_nob_transferred,
1106 PAGE_CACHE_SIZE * op_data->op_npages);
1107 ptlrpc_req_finished(req);
1115 static int mdc_statfs(const struct lu_env *env,
1116 struct obd_export *exp, struct obd_statfs *osfs,
1117 __u64 max_age, __u32 flags)
1119 struct obd_device *obd = class_exp2obd(exp);
1120 struct ptlrpc_request *req;
1121 struct obd_statfs *msfs;
1122 struct obd_import *imp = NULL;
1126 * Since the request might also come from lprocfs, so we need
1127 * sync this with client_disconnect_export Bug15684
1129 down_read(&obd->u.cli.cl_sem);
1130 if (obd->u.cli.cl_import)
1131 imp = class_import_get(obd->u.cli.cl_import);
1132 up_read(&obd->u.cli.cl_sem);
1136 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1137 LUSTRE_MDS_VERSION, MDS_STATFS);
1139 GOTO(output, rc = -ENOMEM);
1141 ptlrpc_request_set_replen(req);
1143 if (flags & OBD_STATFS_NODELAY) {
1144 /* procfs requests not want stay in wait for avoid deadlock */
1145 req->rq_no_resend = 1;
1146 req->rq_no_delay = 1;
1149 rc = ptlrpc_queue_wait(req);
1151 /* check connection error first */
1152 if (imp->imp_connect_error)
1153 rc = imp->imp_connect_error;
1157 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1159 GOTO(out, rc = -EPROTO);
1163 ptlrpc_req_finished(req);
1165 class_import_put(imp);
1169 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1171 __u32 keylen, vallen;
1175 if (gf->gf_pathlen > PATH_MAX)
1176 return -ENAMETOOLONG;
1177 if (gf->gf_pathlen < 2)
1180 /* Key is KEY_FID2PATH + getinfo_fid2path description */
1181 keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1182 OBD_ALLOC(key, keylen);
1185 memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1186 memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1188 CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1189 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1191 if (!fid_is_sane(&gf->gf_fid))
1192 GOTO(out, rc = -EINVAL);
1194 /* Val is struct getinfo_fid2path result plus path */
1195 vallen = sizeof(*gf) + gf->gf_pathlen;
1197 rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1198 if (rc != 0 && rc != -EREMOTE)
1201 if (vallen <= sizeof(*gf))
1202 GOTO(out, rc = -EPROTO);
1203 else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1204 GOTO(out, rc = -EOVERFLOW);
1206 CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n%s\n",
1207 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1210 OBD_FREE(key, keylen);
1214 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1215 struct hsm_progress_kernel *hpk)
1217 struct obd_import *imp = class_exp2cliimp(exp);
1218 struct hsm_progress_kernel *req_hpk;
1219 struct ptlrpc_request *req;
1222 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1223 LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1225 GOTO(out, rc = -ENOMEM);
1227 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1229 /* Copy hsm_progress struct */
1230 req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1231 if (req_hpk == NULL)
1232 GOTO(out, rc = -EPROTO);
1235 req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1237 ptlrpc_request_set_replen(req);
1239 rc = mdc_queue_wait(req);
1242 ptlrpc_req_finished(req);
1246 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1248 __u32 *archive_mask;
1249 struct ptlrpc_request *req;
1252 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1254 MDS_HSM_CT_REGISTER);
1256 GOTO(out, rc = -ENOMEM);
1258 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1260 /* Copy hsm_progress struct */
1261 archive_mask = req_capsule_client_get(&req->rq_pill,
1262 &RMF_MDS_HSM_ARCHIVE);
1263 if (archive_mask == NULL)
1264 GOTO(out, rc = -EPROTO);
1266 *archive_mask = archives;
1268 ptlrpc_request_set_replen(req);
1270 rc = mdc_queue_wait(req);
1273 ptlrpc_req_finished(req);
1277 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1278 struct md_op_data *op_data)
1280 struct hsm_current_action *hca = op_data->op_data;
1281 struct hsm_current_action *req_hca;
1282 struct ptlrpc_request *req;
1285 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1286 &RQF_MDS_HSM_ACTION);
1290 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1292 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1294 ptlrpc_request_free(req);
1298 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1299 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1301 ptlrpc_request_set_replen(req);
1303 rc = mdc_queue_wait(req);
1307 req_hca = req_capsule_server_get(&req->rq_pill,
1308 &RMF_MDS_HSM_CURRENT_ACTION);
1309 if (req_hca == NULL)
1310 GOTO(out, rc = -EPROTO);
1315 ptlrpc_req_finished(req);
1319 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1321 struct ptlrpc_request *req;
1324 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1326 MDS_HSM_CT_UNREGISTER);
1328 GOTO(out, rc = -ENOMEM);
1330 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1332 ptlrpc_request_set_replen(req);
1334 rc = mdc_queue_wait(req);
1337 ptlrpc_req_finished(req);
1341 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1342 struct md_op_data *op_data)
1344 struct hsm_user_state *hus = op_data->op_data;
1345 struct hsm_user_state *req_hus;
1346 struct ptlrpc_request *req;
1349 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1350 &RQF_MDS_HSM_STATE_GET);
1354 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1356 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1358 ptlrpc_request_free(req);
1362 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1363 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1365 ptlrpc_request_set_replen(req);
1367 rc = mdc_queue_wait(req);
1371 req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1372 if (req_hus == NULL)
1373 GOTO(out, rc = -EPROTO);
1378 ptlrpc_req_finished(req);
1382 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1383 struct md_op_data *op_data)
1385 struct hsm_state_set *hss = op_data->op_data;
1386 struct hsm_state_set *req_hss;
1387 struct ptlrpc_request *req;
1390 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1391 &RQF_MDS_HSM_STATE_SET);
1395 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1397 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1399 ptlrpc_request_free(req);
1403 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1404 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1407 req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1408 if (req_hss == NULL)
1409 GOTO(out, rc = -EPROTO);
1412 ptlrpc_request_set_replen(req);
1414 rc = mdc_queue_wait(req);
1418 ptlrpc_req_finished(req);
1422 static int mdc_ioc_hsm_request(struct obd_export *exp,
1423 struct hsm_user_request *hur)
1425 struct obd_import *imp = class_exp2cliimp(exp);
1426 struct ptlrpc_request *req;
1427 struct hsm_request *req_hr;
1428 struct hsm_user_item *req_hui;
1432 req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1434 GOTO(out, rc = -ENOMEM);
1436 req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1437 hur->hur_request.hr_itemcount
1438 * sizeof(struct hsm_user_item));
1439 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1440 hur->hur_request.hr_data_len);
1442 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1444 ptlrpc_request_free(req);
1448 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1450 /* Copy hsm_request struct */
1451 req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1453 GOTO(out, rc = -EPROTO);
1454 *req_hr = hur->hur_request;
1456 /* Copy hsm_user_item structs */
1457 req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1458 if (req_hui == NULL)
1459 GOTO(out, rc = -EPROTO);
1460 memcpy(req_hui, hur->hur_user_item,
1461 hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1463 /* Copy opaque field */
1464 req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1465 if (req_opaque == NULL)
1466 GOTO(out, rc = -EPROTO);
1467 memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1469 ptlrpc_request_set_replen(req);
1471 rc = mdc_queue_wait(req);
1475 ptlrpc_req_finished(req);
1479 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
1481 struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1483 LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1485 lh->kuc_magic = KUC_MAGIC;
1486 lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1487 lh->kuc_flags = flags;
1488 lh->kuc_msgtype = CL_RECORD;
1489 lh->kuc_msglen = len;
1493 #define D_CHANGELOG 0
1495 struct changelog_show {
1500 struct obd_device *cs_obd;
1503 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1504 struct llog_rec_hdr *hdr, void *data)
1506 struct changelog_show *cs = data;
1507 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1511 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1513 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1514 cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
1515 rec->cr.cr_type, rc);
1519 if (rec->cr.cr_index < cs->cs_startrec) {
1520 /* Skip entries earlier than what we are interested in */
1521 CDEBUG(D_CHANGELOG, "rec=%llu start=%llu\n",
1522 rec->cr.cr_index, cs->cs_startrec);
1526 CDEBUG(D_CHANGELOG, "%llu %02d%-5s %llu 0x%x t="DFID" p="DFID
1527 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
1528 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1529 rec->cr.cr_flags & CLF_FLAGMASK,
1530 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1531 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1533 len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1535 /* Set up the message */
1536 lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1537 memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1539 rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1540 CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len, rc);
1545 static int mdc_changelog_send_thread(void *csdata)
1547 struct changelog_show *cs = csdata;
1548 struct llog_ctxt *ctxt = NULL;
1549 struct llog_handle *llh = NULL;
1550 struct kuc_hdr *kuch;
1553 CDEBUG(D_CHANGELOG, "changelog to fp=%p start %llu\n",
1554 cs->cs_fp, cs->cs_startrec);
1556 OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1557 if (cs->cs_buf == NULL)
1558 GOTO(out, rc = -ENOMEM);
1560 /* Set up the remote catalog handle */
1561 ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1563 GOTO(out, rc = -ENOENT);
1564 rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1567 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1568 cs->cs_obd->obd_name, rc);
1571 rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
1573 CERROR("llog_init_handle failed %d\n", rc);
1577 rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1579 /* Send EOF no matter what our result */
1580 kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags);
1582 kuch->kuc_msgtype = CL_EOF;
1583 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1589 llog_cat_close(NULL, llh);
1591 llog_ctxt_put(ctxt);
1593 OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1598 static int mdc_ioc_changelog_send(struct obd_device *obd,
1599 struct ioc_changelog *icc)
1601 struct changelog_show *cs;
1604 /* Freed in mdc_changelog_send_thread */
1610 cs->cs_startrec = icc->icc_recno;
1611 /* matching fput in mdc_changelog_send_thread */
1612 cs->cs_fp = fget(icc->icc_id);
1613 cs->cs_flags = icc->icc_flags;
1616 * New thread because we should return to user app before
1617 * writing into our pipe
1619 rc = PTR_ERR(kthread_run(mdc_changelog_send_thread, cs,
1620 "mdc_clg_send_thread"));
1621 if (!IS_ERR_VALUE(rc)) {
1622 CDEBUG(D_CHANGELOG, "start changelog thread\n");
1626 CERROR("Failed to start changelog thread: %d\n", rc);
1631 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1632 struct lustre_kernelcomm *lk);
1634 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
1635 struct obd_quotactl *oqctl)
1637 struct client_obd *cli = &exp->exp_obd->u.cli;
1638 struct ptlrpc_request *req;
1639 struct obd_quotactl *body;
1642 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1643 &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
1648 body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1651 ptlrpc_request_set_replen(req);
1653 /* the next poll will find -ENODATA, that means quotacheck is
1655 cli->cl_qchk_stat = -ENODATA;
1656 rc = ptlrpc_queue_wait(req);
1658 cli->cl_qchk_stat = rc;
1659 ptlrpc_req_finished(req);
1663 static int mdc_quota_poll_check(struct obd_export *exp,
1664 struct if_quotacheck *qchk)
1666 struct client_obd *cli = &exp->exp_obd->u.cli;
1669 qchk->obd_uuid = cli->cl_target_uuid;
1670 memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
1672 rc = cli->cl_qchk_stat;
1673 /* the client is not the previous one */
1674 if (rc == CL_NOT_QUOTACHECKED)
1679 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1680 struct obd_quotactl *oqctl)
1682 struct ptlrpc_request *req;
1683 struct obd_quotactl *oqc;
1686 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1687 &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1692 oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1695 ptlrpc_request_set_replen(req);
1696 ptlrpc_at_set_req_timeout(req);
1697 req->rq_no_resend = 1;
1699 rc = ptlrpc_queue_wait(req);
1701 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1703 if (req->rq_repmsg) {
1704 oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1708 CERROR("Can't unpack obd_quotactl\n");
1712 CERROR("Can't unpack obd_quotactl\n");
1715 ptlrpc_req_finished(req);
1720 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1721 struct md_op_data *op_data)
1724 struct ptlrpc_request *req;
1726 struct mdc_swap_layouts *msl, *payload;
1728 msl = op_data->op_data;
1730 /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1731 * first thing it will do is to cancel the 2 layout
1732 * locks hold by this client.
1733 * So the client must cancel its layout locks on the 2 fids
1734 * with the request RPC to avoid extra RPC round trips
1736 count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1737 LCK_CR, MDS_INODELOCK_LAYOUT);
1738 count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1739 LCK_CR, MDS_INODELOCK_LAYOUT);
1741 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1742 &RQF_MDS_SWAP_LAYOUTS);
1744 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1748 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1749 mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
1751 rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1753 ptlrpc_request_free(req);
1757 mdc_swap_layouts_pack(req, op_data);
1759 payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1764 ptlrpc_request_set_replen(req);
1766 rc = ptlrpc_queue_wait(req);
1771 ptlrpc_req_finished(req);
1775 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1776 void *karg, void *uarg)
1778 struct obd_device *obd = exp->exp_obd;
1779 struct obd_ioctl_data *data = karg;
1780 struct obd_import *imp = obd->u.cli.cl_import;
1781 struct llog_ctxt *ctxt;
1784 if (!try_module_get(THIS_MODULE)) {
1785 CERROR("Can't get module. Is it alive?");
1789 case OBD_IOC_CHANGELOG_SEND:
1790 rc = mdc_ioc_changelog_send(obd, karg);
1792 case OBD_IOC_CHANGELOG_CLEAR: {
1793 struct ioc_changelog *icc = karg;
1794 struct changelog_setinfo cs =
1795 {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
1796 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
1797 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1801 case OBD_IOC_FID2PATH:
1802 rc = mdc_ioc_fid2path(exp, karg);
1804 case LL_IOC_HSM_CT_START:
1805 rc = mdc_ioc_hsm_ct_start(exp, karg);
1806 /* ignore if it was already registered on this MDS. */
1810 case LL_IOC_HSM_PROGRESS:
1811 rc = mdc_ioc_hsm_progress(exp, karg);
1813 case LL_IOC_HSM_STATE_GET:
1814 rc = mdc_ioc_hsm_state_get(exp, karg);
1816 case LL_IOC_HSM_STATE_SET:
1817 rc = mdc_ioc_hsm_state_set(exp, karg);
1819 case LL_IOC_HSM_ACTION:
1820 rc = mdc_ioc_hsm_current_action(exp, karg);
1822 case LL_IOC_HSM_REQUEST:
1823 rc = mdc_ioc_hsm_request(exp, karg);
1825 case OBD_IOC_CLIENT_RECOVER:
1826 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1830 case IOC_OSC_SET_ACTIVE:
1831 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1833 case OBD_IOC_PARSE: {
1834 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1835 rc = class_config_parse_llog(NULL, ctxt, data->ioc_inlbuf1,
1837 llog_ctxt_put(ctxt);
1840 case OBD_IOC_LLOG_INFO:
1841 case OBD_IOC_LLOG_PRINT: {
1842 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1843 rc = llog_ioctl(NULL, ctxt, cmd, data);
1844 llog_ctxt_put(ctxt);
1847 case OBD_IOC_POLL_QUOTACHECK:
1848 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
1850 case OBD_IOC_PING_TARGET:
1851 rc = ptlrpc_obd_ping(obd);
1854 * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1855 * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1856 * there'd be no LMV layer thus we might be called here. Eventually
1857 * this code should be removed.
1860 case IOC_OBD_STATFS: {
1861 struct obd_statfs stat_buf = {0};
1863 if (*((__u32 *) data->ioc_inlbuf2) != 0)
1864 GOTO(out, rc = -ENODEV);
1867 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1868 min((int) data->ioc_plen2,
1869 (int) sizeof(struct obd_uuid))))
1870 GOTO(out, rc = -EFAULT);
1872 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1873 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1878 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1879 min((int) data->ioc_plen1,
1880 (int) sizeof(stat_buf))))
1881 GOTO(out, rc = -EFAULT);
1885 case OBD_IOC_QUOTACTL: {
1886 struct if_quotactl *qctl = karg;
1887 struct obd_quotactl *oqctl;
1889 OBD_ALLOC_PTR(oqctl);
1891 GOTO(out, rc = -ENOMEM);
1893 QCTL_COPY(oqctl, qctl);
1894 rc = obd_quotactl(exp, oqctl);
1896 QCTL_COPY(qctl, oqctl);
1897 qctl->qc_valid = QC_MDTIDX;
1898 qctl->obd_uuid = obd->u.cli.cl_target_uuid;
1901 OBD_FREE_PTR(oqctl);
1904 case LL_IOC_GET_CONNECT_FLAGS:
1905 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
1906 sizeof(*exp_connect_flags_ptr(exp))))
1907 GOTO(out, rc = -EFAULT);
1910 case LL_IOC_LOV_SWAP_LAYOUTS:
1911 rc = mdc_ioc_swap_layouts(exp, karg);
1914 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
1915 GOTO(out, rc = -ENOTTY);
1918 module_put(THIS_MODULE);
1923 int mdc_get_info_rpc(struct obd_export *exp,
1924 obd_count keylen, void *key,
1925 int vallen, void *val)
1927 struct obd_import *imp = class_exp2cliimp(exp);
1928 struct ptlrpc_request *req;
1932 req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1936 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1937 RCL_CLIENT, keylen);
1938 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1939 RCL_CLIENT, sizeof(__u32));
1941 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1943 ptlrpc_request_free(req);
1947 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
1948 memcpy(tmp, key, keylen);
1949 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
1950 memcpy(tmp, &vallen, sizeof(__u32));
1952 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1953 RCL_SERVER, vallen);
1954 ptlrpc_request_set_replen(req);
1956 rc = ptlrpc_queue_wait(req);
1957 /* -EREMOTE means the get_info result is partial, and it needs to
1958 * continue on another MDT, see fid2path part in lmv_iocontrol */
1959 if (rc == 0 || rc == -EREMOTE) {
1960 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
1961 memcpy(val, tmp, vallen);
1962 if (ptlrpc_rep_need_swab(req)) {
1963 if (KEY_IS(KEY_FID2PATH))
1964 lustre_swab_fid2path(val);
1967 ptlrpc_req_finished(req);
1972 static void lustre_swab_hai(struct hsm_action_item *h)
1974 __swab32s(&h->hai_len);
1975 __swab32s(&h->hai_action);
1976 lustre_swab_lu_fid(&h->hai_fid);
1977 lustre_swab_lu_fid(&h->hai_dfid);
1978 __swab64s(&h->hai_cookie);
1979 __swab64s(&h->hai_extent.offset);
1980 __swab64s(&h->hai_extent.length);
1981 __swab64s(&h->hai_gid);
1984 static void lustre_swab_hal(struct hsm_action_list *h)
1986 struct hsm_action_item *hai;
1989 __swab32s(&h->hal_version);
1990 __swab32s(&h->hal_count);
1991 __swab32s(&h->hal_archive_id);
1992 __swab64s(&h->hal_flags);
1994 for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
1995 lustre_swab_hai(hai);
1998 static void lustre_swab_kuch(struct kuc_hdr *l)
2000 __swab16s(&l->kuc_magic);
2001 /* __u8 l->kuc_transport */
2002 __swab16s(&l->kuc_msgtype);
2003 __swab16s(&l->kuc_msglen);
2006 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2007 struct lustre_kernelcomm *lk)
2009 struct obd_import *imp = class_exp2cliimp(exp);
2010 __u32 archive = lk->lk_data;
2013 if (lk->lk_group != KUC_GRP_HSM) {
2014 CERROR("Bad copytool group %d\n", lk->lk_group);
2018 CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2019 lk->lk_uid, lk->lk_group, lk->lk_flags);
2021 if (lk->lk_flags & LK_FLG_STOP) {
2022 /* Unregister with the coordinator */
2023 rc = mdc_ioc_hsm_ct_unregister(imp);
2025 rc = mdc_ioc_hsm_ct_register(imp, archive);
2032 * Send a message to any listening copytools
2033 * @param val KUC message (kuc_hdr + hsm_action_list)
2034 * @param len total length of message
2036 static int mdc_hsm_copytool_send(int len, void *val)
2038 struct kuc_hdr *lh = (struct kuc_hdr *)val;
2039 struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
2042 if (len < sizeof(*lh) + sizeof(*hal)) {
2043 CERROR("Short HSM message %d < %d\n", len,
2044 (int) (sizeof(*lh) + sizeof(*hal)));
2047 if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2048 lustre_swab_kuch(lh);
2049 lustre_swab_hal(hal);
2050 } else if (lh->kuc_magic != KUC_MAGIC) {
2051 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2055 CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2057 lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2058 lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2060 /* Broadcast to HSM listeners */
2061 rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2067 * callback function passed to kuc for re-registering each HSM copytool
2068 * running on MDC, after MDT shutdown/recovery.
2069 * @param data archive id served by the copytool
2070 * @param cb_arg callback argument (obd_import)
2072 static int mdc_hsm_ct_reregister(__u32 data, void *cb_arg)
2074 struct obd_import *imp = (struct obd_import *)cb_arg;
2075 __u32 archive = data;
2078 CDEBUG(D_HA, "recover copytool registration to MDT (archive=%#x)\n",
2080 rc = mdc_ioc_hsm_ct_register(imp, archive);
2082 /* ignore error if the copytool is already registered */
2083 return ((rc != 0) && (rc != -EEXIST)) ? rc : 0;
2087 * Re-establish all kuc contexts with MDT
2088 * after MDT shutdown/recovery.
2090 static int mdc_kuc_reregister(struct obd_import *imp)
2092 /* re-register HSM agents */
2093 return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2097 int mdc_set_info_async(const struct lu_env *env,
2098 struct obd_export *exp,
2099 obd_count keylen, void *key,
2100 obd_count vallen, void *val,
2101 struct ptlrpc_request_set *set)
2103 struct obd_import *imp = class_exp2cliimp(exp);
2106 if (KEY_IS(KEY_READ_ONLY)) {
2107 if (vallen != sizeof(int))
2110 spin_lock(&imp->imp_lock);
2111 if (*((int *)val)) {
2112 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2113 imp->imp_connect_data.ocd_connect_flags |=
2116 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2117 imp->imp_connect_data.ocd_connect_flags &=
2118 ~OBD_CONNECT_RDONLY;
2120 spin_unlock(&imp->imp_lock);
2122 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2123 keylen, key, vallen, val, set);
2126 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2127 sptlrpc_conf_client_adapt(exp->exp_obd);
2130 if (KEY_IS(KEY_FLUSH_CTX)) {
2131 sptlrpc_import_flush_my_ctx(imp);
2134 if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2135 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2136 keylen, key, vallen, val, set);
2139 if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2140 rc = mdc_hsm_copytool_send(vallen, val);
2144 CERROR("Unknown key %s\n", (char *)key);
2148 int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2149 __u32 keylen, void *key, __u32 *vallen, void *val,
2150 struct lov_stripe_md *lsm)
2154 if (KEY_IS(KEY_MAX_EASIZE)) {
2155 int mdsize, *max_easize;
2157 if (*vallen != sizeof(int))
2159 mdsize = *(int *)val;
2160 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2161 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2163 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2165 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2166 int *default_easize;
2168 if (*vallen != sizeof(int))
2170 default_easize = val;
2171 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2173 } else if (KEY_IS(KEY_MAX_COOKIESIZE)) {
2174 int mdsize, *max_cookiesize;
2176 if (*vallen != sizeof(int))
2178 mdsize = *(int *)val;
2179 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_cookiesize)
2180 exp->exp_obd->u.cli.cl_max_mds_cookiesize = mdsize;
2181 max_cookiesize = val;
2182 *max_cookiesize = exp->exp_obd->u.cli.cl_max_mds_cookiesize;
2184 } else if (KEY_IS(KEY_DEFAULT_COOKIESIZE)) {
2185 int *default_cookiesize;
2187 if (*vallen != sizeof(int))
2189 default_cookiesize = val;
2190 *default_cookiesize =
2191 exp->exp_obd->u.cli.cl_default_mds_cookiesize;
2193 } else if (KEY_IS(KEY_CONN_DATA)) {
2194 struct obd_import *imp = class_exp2cliimp(exp);
2195 struct obd_connect_data *data = val;
2197 if (*vallen != sizeof(*data))
2200 *data = imp->imp_connect_data;
2202 } else if (KEY_IS(KEY_TGT_COUNT)) {
2207 rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2212 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
2213 struct obd_capa *oc, struct obd_client_handle *handle,
2216 struct ptlrpc_request *req;
2217 struct mdt_body *body;
2220 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
2224 mdc_set_capa_size(req, &RMF_CAPA1, oc);
2226 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
2228 ptlrpc_request_free(req);
2232 mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
2234 ptlrpc_request_set_replen(req);
2236 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2237 rc = ptlrpc_queue_wait(req);
2238 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2240 CERROR("Pin failed: %d\n", rc);
2244 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2246 GOTO(err_out, rc = -EPROTO);
2248 handle->och_fh = body->handle;
2249 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
2251 handle->och_mod = obd_mod_alloc();
2252 if (handle->och_mod == NULL) {
2253 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
2254 GOTO(err_out, rc = -ENOMEM);
2256 handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
2261 ptlrpc_req_finished(req);
2265 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
2268 struct ptlrpc_request *req;
2269 struct mdt_body *body;
2272 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
2273 LUSTRE_MDS_VERSION, MDS_UNPIN);
2277 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
2278 body->handle = handle->och_fh;
2281 ptlrpc_request_set_replen(req);
2283 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2284 rc = ptlrpc_queue_wait(req);
2285 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2288 CERROR("Unpin failed: %d\n", rc);
2290 ptlrpc_req_finished(req);
2291 ptlrpc_req_finished(handle->och_mod->mod_open_req);
2293 obd_mod_put(handle->och_mod);
2297 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
2298 struct obd_capa *oc, struct ptlrpc_request **request)
2300 struct ptlrpc_request *req;
2304 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2308 mdc_set_capa_size(req, &RMF_CAPA1, oc);
2310 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2312 ptlrpc_request_free(req);
2316 mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
2318 ptlrpc_request_set_replen(req);
2320 rc = ptlrpc_queue_wait(req);
2322 ptlrpc_req_finished(req);
2328 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2329 enum obd_import_event event)
2333 LASSERT(imp->imp_obd == obd);
2336 case IMP_EVENT_DISCON: {
2338 /* XXX Pass event up to OBDs stack. used only for FLD now */
2339 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2343 case IMP_EVENT_INACTIVE: {
2344 struct client_obd *cli = &obd->u.cli;
2346 * Flush current sequence to make client obtain new one
2347 * from server in case of disconnect/reconnect.
2349 if (cli->cl_seq != NULL)
2350 seq_client_flush(cli->cl_seq);
2352 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2355 case IMP_EVENT_INVALIDATE: {
2356 struct ldlm_namespace *ns = obd->obd_namespace;
2358 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2362 case IMP_EVENT_ACTIVE:
2363 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2364 /* redo the kuc registration after reconnecting */
2366 rc = mdc_kuc_reregister(imp);
2369 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2371 case IMP_EVENT_DEACTIVATE:
2372 case IMP_EVENT_ACTIVATE:
2375 CERROR("Unknown import event %x\n", event);
2381 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
2382 struct md_op_data *op_data)
2384 struct client_obd *cli = &exp->exp_obd->u.cli;
2385 struct lu_client_seq *seq = cli->cl_seq;
2387 return seq_client_alloc_fid(NULL, seq, fid);
2390 struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2392 struct client_obd *cli = &exp->exp_obd->u.cli;
2393 return &cli->cl_target_uuid;
2397 * Determine whether the lock can be canceled before replaying it during
2398 * recovery, non zero value will be return if the lock can be canceled,
2399 * or zero returned for not
2401 static int mdc_cancel_for_recovery(struct ldlm_lock *lock)
2403 if (lock->l_resource->lr_type != LDLM_IBITS)
2406 /* FIXME: if we ever get into a situation where there are too many
2407 * opened files with open locks on a single node, then we really
2408 * should replay these open locks to reget it */
2409 if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2415 static int mdc_resource_inode_free(struct ldlm_resource *res)
2417 if (res->lr_lvb_inode)
2418 res->lr_lvb_inode = NULL;
2423 struct ldlm_valblock_ops inode_lvbo = {
2424 .lvbo_free = mdc_resource_inode_free,
2427 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2429 struct client_obd *cli = &obd->u.cli;
2430 struct lprocfs_static_vars lvars = { NULL };
2433 OBD_ALLOC(cli->cl_rpc_lock, sizeof(*cli->cl_rpc_lock));
2434 if (!cli->cl_rpc_lock)
2436 mdc_init_rpc_lock(cli->cl_rpc_lock);
2440 OBD_ALLOC(cli->cl_close_lock, sizeof(*cli->cl_close_lock));
2441 if (!cli->cl_close_lock)
2442 GOTO(err_rpc_lock, rc = -ENOMEM);
2443 mdc_init_rpc_lock(cli->cl_close_lock);
2445 rc = client_obd_setup(obd, cfg);
2447 GOTO(err_close_lock, rc);
2448 lprocfs_mdc_init_vars(&lvars);
2449 lprocfs_obd_setup(obd, lvars.obd_vars);
2450 sptlrpc_lprocfs_cliobd_attach(obd);
2451 ptlrpc_lprocfs_register_obd(obd);
2453 ns_register_cancel(obd->obd_namespace, mdc_cancel_for_recovery);
2455 obd->obd_namespace->ns_lvbo = &inode_lvbo;
2457 rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
2460 CERROR("failed to setup llogging subsystems\n");
2466 OBD_FREE(cli->cl_close_lock, sizeof(*cli->cl_close_lock));
2468 OBD_FREE(cli->cl_rpc_lock, sizeof(*cli->cl_rpc_lock));
2473 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
2474 * us to make MDS RPCs with large enough reply buffers to hold a default
2475 * sized EA and cookie without having to calculate this (via a call into the
2476 * LOV + OSCs) each time we make an RPC. The maximum size is also tracked
2477 * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2478 * a large number of stripes is possible. If a larger reply buffer is
2479 * required it will be reallocated in the ptlrpc layer due to overflow.
2481 static int mdc_init_ea_size(struct obd_export *exp, int easize,
2482 int def_easize, int cookiesize, int def_cookiesize)
2484 struct obd_device *obd = exp->exp_obd;
2485 struct client_obd *cli = &obd->u.cli;
2487 if (cli->cl_max_mds_easize < easize)
2488 cli->cl_max_mds_easize = easize;
2490 if (cli->cl_default_mds_easize < def_easize)
2491 cli->cl_default_mds_easize = def_easize;
2493 if (cli->cl_max_mds_cookiesize < cookiesize)
2494 cli->cl_max_mds_cookiesize = cookiesize;
2496 if (cli->cl_default_mds_cookiesize < def_cookiesize)
2497 cli->cl_default_mds_cookiesize = def_cookiesize;
2502 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2507 case OBD_CLEANUP_EARLY:
2509 case OBD_CLEANUP_EXPORTS:
2510 /* Failsafe, ok if racy */
2511 if (obd->obd_type->typ_refcnt <= 1)
2512 libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2514 obd_cleanup_client_import(obd);
2515 ptlrpc_lprocfs_unregister_obd(obd);
2516 lprocfs_obd_cleanup(obd);
2518 rc = obd_llog_finish(obd, 0);
2520 CERROR("failed to cleanup llogging subsystems\n");
2526 static int mdc_cleanup(struct obd_device *obd)
2528 struct client_obd *cli = &obd->u.cli;
2530 OBD_FREE(cli->cl_rpc_lock, sizeof(*cli->cl_rpc_lock));
2531 OBD_FREE(cli->cl_close_lock, sizeof(*cli->cl_close_lock));
2535 return client_obd_cleanup(obd);
2539 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
2540 struct obd_device *tgt, int *index)
2542 struct llog_ctxt *ctxt;
2545 LASSERT(olg == &obd->obd_olg);
2547 rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt,
2552 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2553 llog_initiator_connect(ctxt);
2554 llog_ctxt_put(ctxt);
2559 static int mdc_llog_finish(struct obd_device *obd, int count)
2561 struct llog_ctxt *ctxt;
2563 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2565 llog_cleanup(NULL, ctxt);
2570 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
2572 struct lustre_cfg *lcfg = buf;
2573 struct lprocfs_static_vars lvars = { NULL };
2576 lprocfs_mdc_init_vars(&lvars);
2577 switch (lcfg->lcfg_command) {
2579 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
2589 /* get remote permission for current user on fid */
2590 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2591 struct obd_capa *oc, __u32 suppgid,
2592 struct ptlrpc_request **request)
2594 struct ptlrpc_request *req;
2597 LASSERT(client_is_remote(exp));
2600 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2604 mdc_set_capa_size(req, &RMF_CAPA1, oc);
2606 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2608 ptlrpc_request_free(req);
2612 mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2614 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2615 sizeof(struct mdt_remote_perm));
2617 ptlrpc_request_set_replen(req);
2619 rc = ptlrpc_queue_wait(req);
2621 ptlrpc_req_finished(req);
2627 static int mdc_interpret_renew_capa(const struct lu_env *env,
2628 struct ptlrpc_request *req, void *args,
2631 struct mdc_renew_capa_args *ra = args;
2632 struct mdt_body *body = NULL;
2633 struct lustre_capa *capa;
2636 GOTO(out, capa = ERR_PTR(status));
2638 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2640 GOTO(out, capa = ERR_PTR(-EFAULT));
2642 if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
2643 GOTO(out, capa = ERR_PTR(-ENOENT));
2645 capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
2647 GOTO(out, capa = ERR_PTR(-EFAULT));
2649 ra->ra_cb(ra->ra_oc, capa);
2653 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2656 struct ptlrpc_request *req;
2657 struct mdc_renew_capa_args *ra;
2659 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
2660 LUSTRE_MDS_VERSION, MDS_GETATTR);
2664 /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
2665 * capa to renew is oss capa.
2667 mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
2668 ptlrpc_request_set_replen(req);
2670 CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
2671 ra = ptlrpc_req_async_args(req);
2674 req->rq_interpret_reply = mdc_interpret_renew_capa;
2675 ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
2679 struct obd_ops mdc_obd_ops = {
2680 .o_owner = THIS_MODULE,
2681 .o_setup = mdc_setup,
2682 .o_precleanup = mdc_precleanup,
2683 .o_cleanup = mdc_cleanup,
2684 .o_add_conn = client_import_add_conn,
2685 .o_del_conn = client_import_del_conn,
2686 .o_connect = client_connect_import,
2687 .o_disconnect = client_disconnect_export,
2688 .o_iocontrol = mdc_iocontrol,
2689 .o_set_info_async = mdc_set_info_async,
2690 .o_statfs = mdc_statfs,
2692 .o_unpin = mdc_unpin,
2693 .o_fid_init = client_fid_init,
2694 .o_fid_fini = client_fid_fini,
2695 .o_fid_alloc = mdc_fid_alloc,
2696 .o_import_event = mdc_import_event,
2697 .o_llog_init = mdc_llog_init,
2698 .o_llog_finish = mdc_llog_finish,
2699 .o_get_info = mdc_get_info,
2700 .o_process_config = mdc_process_config,
2701 .o_get_uuid = mdc_get_uuid,
2702 .o_quotactl = mdc_quotactl,
2703 .o_quotacheck = mdc_quotacheck
2706 struct md_ops mdc_md_ops = {
2707 .m_getstatus = mdc_getstatus,
2708 .m_null_inode = mdc_null_inode,
2709 .m_find_cbdata = mdc_find_cbdata,
2710 .m_close = mdc_close,
2711 .m_create = mdc_create,
2712 .m_done_writing = mdc_done_writing,
2713 .m_enqueue = mdc_enqueue,
2714 .m_getattr = mdc_getattr,
2715 .m_getattr_name = mdc_getattr_name,
2716 .m_intent_lock = mdc_intent_lock,
2718 .m_is_subdir = mdc_is_subdir,
2719 .m_rename = mdc_rename,
2720 .m_setattr = mdc_setattr,
2721 .m_setxattr = mdc_setxattr,
2722 .m_getxattr = mdc_getxattr,
2724 .m_readpage = mdc_readpage,
2725 .m_unlink = mdc_unlink,
2726 .m_cancel_unused = mdc_cancel_unused,
2727 .m_init_ea_size = mdc_init_ea_size,
2728 .m_set_lock_data = mdc_set_lock_data,
2729 .m_lock_match = mdc_lock_match,
2730 .m_get_lustre_md = mdc_get_lustre_md,
2731 .m_free_lustre_md = mdc_free_lustre_md,
2732 .m_set_open_replay_data = mdc_set_open_replay_data,
2733 .m_clear_open_replay_data = mdc_clear_open_replay_data,
2734 .m_renew_capa = mdc_renew_capa,
2735 .m_unpack_capa = mdc_unpack_capa,
2736 .m_get_remote_perm = mdc_get_remote_perm,
2737 .m_intent_getattr_async = mdc_intent_getattr_async,
2738 .m_revalidate_lock = mdc_revalidate_lock
2741 int __init mdc_init(void)
2744 struct lprocfs_static_vars lvars = { NULL };
2745 lprocfs_mdc_init_vars(&lvars);
2747 rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
2748 LUSTRE_MDC_NAME, NULL);
2752 static void /*__exit*/ mdc_exit(void)
2754 class_unregister_type(LUSTRE_MDC_NAME);
2757 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2758 MODULE_DESCRIPTION("Lustre Metadata Client");
2759 MODULE_LICENSE("GPL");
2761 module_init(mdc_init);
2762 module_exit(mdc_exit);