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;
207 capa = req_capsule_server_get(pill, &RMF_CAPA1);
215 int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
216 struct ptlrpc_request **request)
218 struct ptlrpc_request *req;
221 /* Single MDS without an LMV case */
222 if (op_data->op_flags & MF_GET_MDT_IDX) {
227 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
231 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
233 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
235 ptlrpc_request_free(req);
239 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
240 op_data->op_valid, op_data->op_mode, -1, 0);
242 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
244 if (op_data->op_valid & OBD_MD_FLRMTPERM) {
245 LASSERT(client_is_remote(exp));
246 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
247 sizeof(struct mdt_remote_perm));
249 ptlrpc_request_set_replen(req);
251 rc = mdc_getattr_common(exp, req);
253 ptlrpc_req_finished(req);
259 int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
260 struct ptlrpc_request **request)
262 struct ptlrpc_request *req;
266 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
267 &RQF_MDS_GETATTR_NAME);
271 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
272 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
273 op_data->op_namelen + 1);
275 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
277 ptlrpc_request_free(req);
281 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
282 op_data->op_valid, op_data->op_mode,
283 op_data->op_suppgids[0], 0);
285 if (op_data->op_name) {
286 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
288 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
289 op_data->op_namelen);
290 memcpy(name, op_data->op_name, op_data->op_namelen);
293 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
295 ptlrpc_request_set_replen(req);
297 rc = mdc_getattr_common(exp, req);
299 ptlrpc_req_finished(req);
305 static int mdc_is_subdir(struct obd_export *exp,
306 const struct lu_fid *pfid,
307 const struct lu_fid *cfid,
308 struct ptlrpc_request **request)
310 struct ptlrpc_request *req;
314 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
315 &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
320 mdc_is_subdir_pack(req, pfid, cfid, 0);
321 ptlrpc_request_set_replen(req);
323 rc = ptlrpc_queue_wait(req);
324 if (rc && rc != -EREMOTE)
325 ptlrpc_req_finished(req);
331 static int mdc_xattr_common(struct obd_export *exp,
332 const struct req_format *fmt,
333 const struct lu_fid *fid,
334 struct obd_capa *oc, int opcode, u64 valid,
335 const char *xattr_name, const char *input,
336 int input_size, int output_size, int flags,
337 __u32 suppgid, struct ptlrpc_request **request)
339 struct ptlrpc_request *req;
340 int xattr_namelen = 0;
345 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
349 mdc_set_capa_size(req, &RMF_CAPA1, oc);
351 xattr_namelen = strlen(xattr_name) + 1;
352 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
357 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
361 /* Flush local XATTR locks to get rid of a possible cancel RPC */
362 if (opcode == MDS_REINT && fid_is_sane(fid) &&
363 exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
367 /* Without that packing would fail */
369 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
372 count = mdc_resource_get_unused(exp, fid,
374 MDS_INODELOCK_XATTR);
376 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
378 ptlrpc_request_free(req);
382 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
384 ptlrpc_request_free(req);
389 if (opcode == MDS_REINT) {
390 struct mdt_rec_setxattr *rec;
392 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
393 sizeof(struct mdt_rec_reint));
394 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
395 rec->sx_opcode = REINT_SETXATTR;
396 rec->sx_fsuid = from_kuid(&init_user_ns, current_fsuid());
397 rec->sx_fsgid = from_kgid(&init_user_ns, current_fsgid());
398 rec->sx_cap = cfs_curproc_cap_pack();
399 rec->sx_suppgid1 = suppgid;
400 rec->sx_suppgid2 = -1;
402 rec->sx_valid = valid | OBD_MD_FLCTIME;
403 rec->sx_time = get_seconds();
404 rec->sx_size = output_size;
405 rec->sx_flags = flags;
407 mdc_pack_capa(req, &RMF_CAPA1, oc);
409 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
413 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
414 memcpy(tmp, xattr_name, xattr_namelen);
417 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
418 memcpy(tmp, input, input_size);
421 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
422 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
423 RCL_SERVER, output_size);
424 ptlrpc_request_set_replen(req);
427 if (opcode == MDS_REINT)
428 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
430 rc = ptlrpc_queue_wait(req);
432 if (opcode == MDS_REINT)
433 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
436 ptlrpc_req_finished(req);
442 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
443 struct obd_capa *oc, u64 valid, const char *xattr_name,
444 const char *input, int input_size, int output_size,
445 int flags, __u32 suppgid, struct ptlrpc_request **request)
447 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
448 fid, oc, MDS_REINT, valid, xattr_name,
449 input, input_size, output_size, flags,
453 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
454 struct obd_capa *oc, u64 valid, const char *xattr_name,
455 const char *input, int input_size, int output_size,
456 int flags, struct ptlrpc_request **request)
458 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
459 fid, oc, MDS_GETXATTR, valid, xattr_name,
460 input, input_size, output_size, flags,
464 #ifdef CONFIG_FS_POSIX_ACL
465 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
467 struct req_capsule *pill = &req->rq_pill;
468 struct mdt_body *body = md->body;
469 struct posix_acl *acl;
476 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
481 acl = posix_acl_from_xattr(&init_user_ns, buf, body->aclsize);
484 CERROR("convert xattr to acl: %d\n", rc);
488 rc = posix_acl_valid(acl);
490 CERROR("validate acl: %d\n", rc);
491 posix_acl_release(acl);
499 #define mdc_unpack_acl(req, md) 0
502 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
503 struct obd_export *dt_exp, struct obd_export *md_exp,
504 struct lustre_md *md)
506 struct req_capsule *pill = &req->rq_pill;
510 memset(md, 0, sizeof(*md));
512 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
513 LASSERT(md->body != NULL);
515 if (md->body->valid & OBD_MD_FLEASIZE) {
517 struct lov_mds_md *lmm;
519 if (!S_ISREG(md->body->mode)) {
521 "OBD_MD_FLEASIZE set, should be a regular file, but is not\n");
522 GOTO(out, rc = -EPROTO);
525 if (md->body->eadatasize == 0) {
527 "OBD_MD_FLEASIZE set, but eadatasize 0\n");
528 GOTO(out, rc = -EPROTO);
530 lmmsize = md->body->eadatasize;
531 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
533 GOTO(out, rc = -EPROTO);
535 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
539 if (rc < sizeof(*md->lsm)) {
541 "lsm size too small: rc < sizeof (*md->lsm) (%d < %d)\n",
542 rc, (int)sizeof(*md->lsm));
543 GOTO(out, rc = -EPROTO);
546 } else if (md->body->valid & OBD_MD_FLDIREA) {
548 struct lov_mds_md *lmv;
550 if (!S_ISDIR(md->body->mode)) {
552 "OBD_MD_FLDIREA set, should be a directory, but is not\n");
553 GOTO(out, rc = -EPROTO);
556 if (md->body->eadatasize == 0) {
558 "OBD_MD_FLDIREA is set, but eadatasize 0\n");
561 if (md->body->valid & OBD_MD_MEA) {
562 lmvsize = md->body->eadatasize;
563 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
566 GOTO(out, rc = -EPROTO);
568 rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
573 if (rc < sizeof(*md->mea)) {
575 "size too small: rc < sizeof(*md->mea) (%d < %d)\n",
576 rc, (int)sizeof(*md->mea));
577 GOTO(out, rc = -EPROTO);
583 if (md->body->valid & OBD_MD_FLRMTPERM) {
584 /* remote permission */
585 LASSERT(client_is_remote(exp));
586 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
587 lustre_swab_mdt_remote_perm);
588 if (!md->remote_perm)
589 GOTO(out, rc = -EPROTO);
590 } else if (md->body->valid & OBD_MD_FLACL) {
591 /* for ACL, it's possible that FLACL is set but aclsize is zero.
592 * only when aclsize != 0 there's an actual segment for ACL
595 if (md->body->aclsize) {
596 rc = mdc_unpack_acl(req, md);
599 #ifdef CONFIG_FS_POSIX_ACL
601 md->posix_acl = NULL;
605 if (md->body->valid & OBD_MD_FLMDSCAPA) {
606 struct obd_capa *oc = NULL;
608 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
614 if (md->body->valid & OBD_MD_FLOSSCAPA) {
615 struct obd_capa *oc = NULL;
617 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
626 capa_put(md->oss_capa);
630 capa_put(md->mds_capa);
633 #ifdef CONFIG_FS_POSIX_ACL
634 posix_acl_release(md->posix_acl);
637 obd_free_memmd(dt_exp, &md->lsm);
642 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
648 * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
651 void mdc_replay_open(struct ptlrpc_request *req)
653 struct md_open_data *mod = req->rq_cb_data;
654 struct ptlrpc_request *close_req;
655 struct obd_client_handle *och;
656 struct lustre_handle old;
657 struct mdt_body *body;
660 DEBUG_REQ(D_ERROR, req,
661 "Can't properly replay without open data.");
665 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
666 LASSERT(body != NULL);
670 struct lustre_handle *file_fh;
672 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
674 file_fh = &och->och_fh;
675 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
676 file_fh->cookie, body->handle.cookie);
678 *file_fh = body->handle;
680 close_req = mod->mod_close_req;
681 if (close_req != NULL) {
682 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
683 struct mdt_ioepoch *epoch;
685 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
686 epoch = req_capsule_client_get(&close_req->rq_pill,
691 LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
692 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
693 epoch->handle = body->handle;
697 void mdc_commit_open(struct ptlrpc_request *req)
699 struct md_open_data *mod = req->rq_cb_data;
705 * No need to touch md_open_data::mod_och, it holds a reference on
706 * \var mod and will zero references to each other, \var mod will be
707 * freed after that when md_open_data::mod_och will put the reference.
711 * Do not let open request to disappear as it still may be needed
712 * for close rpc to happen (it may happen on evict only, otherwise
713 * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
714 * called), just mark this rpc as committed to distinguish these 2
715 * cases, see mdc_close() for details. The open request reference will
716 * be put along with freeing \var mod.
718 ptlrpc_request_addref(req);
719 spin_lock(&req->rq_lock);
720 req->rq_committed = 1;
721 spin_unlock(&req->rq_lock);
722 req->rq_cb_data = NULL;
726 int mdc_set_open_replay_data(struct obd_export *exp,
727 struct obd_client_handle *och,
728 struct lookup_intent *it)
730 struct md_open_data *mod;
731 struct mdt_rec_create *rec;
732 struct mdt_body *body;
733 struct ptlrpc_request *open_req = it->d.lustre.it_data;
734 struct obd_import *imp = open_req->rq_import;
736 if (!open_req->rq_replay)
739 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
740 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
741 LASSERT(rec != NULL);
742 /* Incoming message in my byte order (it's been swabbed). */
743 /* Outgoing messages always in my byte order. */
744 LASSERT(body != NULL);
746 /* Only if the import is replayable, we set replay_open data */
747 if (och && imp->imp_replayable) {
748 mod = obd_mod_alloc();
750 DEBUG_REQ(D_ERROR, open_req,
751 "Can't allocate md_open_data");
756 * Take a reference on \var mod, to be freed on mdc_close().
757 * It protects \var mod from being freed on eviction (commit
758 * callback is called despite rq_replay flag).
759 * Another reference for \var och.
764 spin_lock(&open_req->rq_lock);
767 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
768 it_disposition(it, DISP_OPEN_STRIPE);
769 mod->mod_open_req = open_req;
770 open_req->rq_cb_data = mod;
771 open_req->rq_commit_cb = mdc_commit_open;
772 spin_unlock(&open_req->rq_lock);
775 rec->cr_fid2 = body->fid1;
776 rec->cr_ioepoch = body->ioepoch;
777 rec->cr_old_handle.cookie = body->handle.cookie;
778 open_req->rq_replay_cb = mdc_replay_open;
779 if (!fid_is_sane(&body->fid1)) {
780 DEBUG_REQ(D_ERROR, open_req,
781 "Saving replay request with insane fid");
785 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
789 static void mdc_free_open(struct md_open_data *mod)
793 if (mod->mod_is_create == 0 &&
794 imp_connect_disp_stripe(mod->mod_open_req->rq_import))
797 LASSERT(mod->mod_open_req->rq_replay == 0);
799 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
801 ptlrpc_request_committed(mod->mod_open_req, committed);
802 if (mod->mod_close_req)
803 ptlrpc_request_committed(mod->mod_close_req, committed);
806 int mdc_clear_open_replay_data(struct obd_export *exp,
807 struct obd_client_handle *och)
809 struct md_open_data *mod = och->och_mod;
812 * It is possible to not have \var mod in a case of eviction between
813 * lookup and ll_file_open().
818 LASSERT(mod != LP_POISON);
819 LASSERT(mod->mod_open_req != NULL);
829 /* Prepares the request for the replay by the given reply */
830 static void mdc_close_handle_reply(struct ptlrpc_request *req,
831 struct md_op_data *op_data, int rc) {
832 struct mdt_body *repbody;
833 struct mdt_ioepoch *epoch;
835 if (req && rc == -EAGAIN) {
836 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
837 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
839 epoch->flags |= MF_SOM_AU;
840 if (repbody->valid & OBD_MD_FLGETATTRLOCK)
841 op_data->op_flags |= MF_GETATTR_LOCK;
845 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
846 struct md_open_data *mod, struct ptlrpc_request **request)
848 struct obd_device *obd = class_exp2obd(exp);
849 struct ptlrpc_request *req;
850 struct req_format *req_fmt;
855 req_fmt = &RQF_MDS_CLOSE;
856 if (op_data->op_bias & MDS_HSM_RELEASE) {
857 req_fmt = &RQF_MDS_RELEASE_CLOSE;
859 /* allocate a FID for volatile file */
860 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
862 CERROR("%s: "DFID" failed to allocate FID: %d\n",
863 obd->obd_name, PFID(&op_data->op_fid1), rc);
864 /* save the errcode and proceed to close */
870 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
874 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
876 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
878 ptlrpc_request_free(req);
882 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
883 * portal whose threads are not taking any DLM locks and are therefore
884 * always progressing */
885 req->rq_request_portal = MDS_READPAGE_PORTAL;
886 ptlrpc_at_set_req_timeout(req);
888 /* Ensure that this close's handle is fixed up during replay. */
889 if (likely(mod != NULL)) {
890 LASSERTF(mod->mod_open_req != NULL &&
891 mod->mod_open_req->rq_type != LI_POISON,
892 "POISONED open %p!\n", mod->mod_open_req);
894 mod->mod_close_req = req;
896 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
897 /* We no longer want to preserve this open for replay even
898 * though the open was committed. b=3632, b=3633 */
899 spin_lock(&mod->mod_open_req->rq_lock);
900 mod->mod_open_req->rq_replay = 0;
901 spin_unlock(&mod->mod_open_req->rq_lock);
904 "couldn't find open req; expecting close error\n");
907 mdc_close_pack(req, op_data);
909 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
910 obd->u.cli.cl_default_mds_easize);
911 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
912 obd->u.cli.cl_default_mds_cookiesize);
914 ptlrpc_request_set_replen(req);
916 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
917 rc = ptlrpc_queue_wait(req);
918 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
920 if (req->rq_repmsg == NULL) {
921 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
924 rc = req->rq_status ?: -EIO;
925 } else if (rc == 0 || rc == -EAGAIN) {
926 struct mdt_body *body;
928 rc = lustre_msg_get_status(req->rq_repmsg);
929 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
930 DEBUG_REQ(D_ERROR, req,
931 "type == PTL_RPC_MSG_ERR, err = %d", rc);
935 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
938 } else if (rc == -ESTALE) {
940 * it can be allowed error after 3633 if open was committed and
941 * server failed before close was sent. Let's check if mod
942 * exists and return no error in that case
945 DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
946 LASSERT(mod->mod_open_req != NULL);
947 if (mod->mod_open_req->rq_committed)
954 mod->mod_close_req = NULL;
955 /* Since now, mod is accessed through open_req only,
956 * thus close req does not keep a reference on mod anymore. */
960 mdc_close_handle_reply(req, op_data, rc);
961 return rc < 0 ? rc : saved_rc;
964 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
965 struct md_open_data *mod)
967 struct obd_device *obd = class_exp2obd(exp);
968 struct ptlrpc_request *req;
971 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
972 &RQF_MDS_DONE_WRITING);
976 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
977 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
979 ptlrpc_request_free(req);
984 LASSERTF(mod->mod_open_req != NULL &&
985 mod->mod_open_req->rq_type != LI_POISON,
986 "POISONED setattr %p!\n", mod->mod_open_req);
988 mod->mod_close_req = req;
989 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
990 /* We no longer want to preserve this setattr for replay even
991 * though the open was committed. b=3632, b=3633 */
992 spin_lock(&mod->mod_open_req->rq_lock);
993 mod->mod_open_req->rq_replay = 0;
994 spin_unlock(&mod->mod_open_req->rq_lock);
997 mdc_close_pack(req, op_data);
998 ptlrpc_request_set_replen(req);
1000 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
1001 rc = ptlrpc_queue_wait(req);
1002 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
1004 if (rc == -ESTALE) {
1006 * it can be allowed error after 3633 if open or setattr were
1007 * committed and server failed before close was sent.
1008 * Let's check if mod exists and return no error in that case
1011 LASSERT(mod->mod_open_req != NULL);
1012 if (mod->mod_open_req->rq_committed)
1019 mod->mod_close_req = NULL;
1020 LASSERT(mod->mod_open_req != NULL);
1023 /* Since now, mod is accessed through setattr req only,
1024 * thus DW req does not keep a reference on mod anymore. */
1028 mdc_close_handle_reply(req, op_data, rc);
1029 ptlrpc_req_finished(req);
1034 int mdc_readpage(struct obd_export *exp, struct md_op_data *op_data,
1035 struct page **pages, struct ptlrpc_request **request)
1037 struct ptlrpc_request *req;
1038 struct ptlrpc_bulk_desc *desc;
1040 wait_queue_head_t waitq;
1042 struct l_wait_info lwi;
1046 init_waitqueue_head(&waitq);
1049 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1053 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1055 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1057 ptlrpc_request_free(req);
1061 req->rq_request_portal = MDS_READPAGE_PORTAL;
1062 ptlrpc_at_set_req_timeout(req);
1064 desc = ptlrpc_prep_bulk_imp(req, op_data->op_npages, 1, BULK_PUT_SINK,
1067 ptlrpc_request_free(req);
1071 /* NB req now owns desc and will free it when it gets freed */
1072 for (i = 0; i < op_data->op_npages; i++)
1073 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
1075 mdc_readdir_pack(req, op_data->op_offset,
1076 PAGE_CACHE_SIZE * op_data->op_npages,
1077 &op_data->op_fid1, op_data->op_capa1);
1079 ptlrpc_request_set_replen(req);
1080 rc = ptlrpc_queue_wait(req);
1082 ptlrpc_req_finished(req);
1083 if (rc != -ETIMEDOUT)
1087 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1088 CERROR("too many resend retries, returning error\n");
1091 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends),
1093 l_wait_event(waitq, 0, &lwi);
1098 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1099 req->rq_bulk->bd_nob_transferred);
1101 ptlrpc_req_finished(req);
1105 if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1106 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1107 req->rq_bulk->bd_nob_transferred,
1108 PAGE_CACHE_SIZE * op_data->op_npages);
1109 ptlrpc_req_finished(req);
1117 static int mdc_statfs(const struct lu_env *env,
1118 struct obd_export *exp, struct obd_statfs *osfs,
1119 __u64 max_age, __u32 flags)
1121 struct obd_device *obd = class_exp2obd(exp);
1122 struct ptlrpc_request *req;
1123 struct obd_statfs *msfs;
1124 struct obd_import *imp = NULL;
1128 * Since the request might also come from lprocfs, so we need
1129 * sync this with client_disconnect_export Bug15684
1131 down_read(&obd->u.cli.cl_sem);
1132 if (obd->u.cli.cl_import)
1133 imp = class_import_get(obd->u.cli.cl_import);
1134 up_read(&obd->u.cli.cl_sem);
1138 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1139 LUSTRE_MDS_VERSION, MDS_STATFS);
1141 GOTO(output, rc = -ENOMEM);
1143 ptlrpc_request_set_replen(req);
1145 if (flags & OBD_STATFS_NODELAY) {
1146 /* procfs requests not want stay in wait for avoid deadlock */
1147 req->rq_no_resend = 1;
1148 req->rq_no_delay = 1;
1151 rc = ptlrpc_queue_wait(req);
1153 /* check connection error first */
1154 if (imp->imp_connect_error)
1155 rc = imp->imp_connect_error;
1159 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1161 GOTO(out, rc = -EPROTO);
1165 ptlrpc_req_finished(req);
1167 class_import_put(imp);
1171 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1173 __u32 keylen, vallen;
1177 if (gf->gf_pathlen > PATH_MAX)
1178 return -ENAMETOOLONG;
1179 if (gf->gf_pathlen < 2)
1182 /* Key is KEY_FID2PATH + getinfo_fid2path description */
1183 keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1184 OBD_ALLOC(key, keylen);
1187 memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1188 memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1190 CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1191 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1193 if (!fid_is_sane(&gf->gf_fid))
1194 GOTO(out, rc = -EINVAL);
1196 /* Val is struct getinfo_fid2path result plus path */
1197 vallen = sizeof(*gf) + gf->gf_pathlen;
1199 rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1200 if (rc != 0 && rc != -EREMOTE)
1203 if (vallen <= sizeof(*gf))
1204 GOTO(out, rc = -EPROTO);
1205 else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1206 GOTO(out, rc = -EOVERFLOW);
1208 CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n%s\n",
1209 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1212 OBD_FREE(key, keylen);
1216 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1217 struct hsm_progress_kernel *hpk)
1219 struct obd_import *imp = class_exp2cliimp(exp);
1220 struct hsm_progress_kernel *req_hpk;
1221 struct ptlrpc_request *req;
1224 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1225 LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1227 GOTO(out, rc = -ENOMEM);
1229 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1231 /* Copy hsm_progress struct */
1232 req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1233 if (req_hpk == NULL)
1234 GOTO(out, rc = -EPROTO);
1237 req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1239 ptlrpc_request_set_replen(req);
1241 rc = mdc_queue_wait(req);
1244 ptlrpc_req_finished(req);
1248 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1250 __u32 *archive_mask;
1251 struct ptlrpc_request *req;
1254 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1256 MDS_HSM_CT_REGISTER);
1258 GOTO(out, rc = -ENOMEM);
1260 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1262 /* Copy hsm_progress struct */
1263 archive_mask = req_capsule_client_get(&req->rq_pill,
1264 &RMF_MDS_HSM_ARCHIVE);
1265 if (archive_mask == NULL)
1266 GOTO(out, rc = -EPROTO);
1268 *archive_mask = archives;
1270 ptlrpc_request_set_replen(req);
1272 rc = mdc_queue_wait(req);
1275 ptlrpc_req_finished(req);
1279 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1280 struct md_op_data *op_data)
1282 struct hsm_current_action *hca = op_data->op_data;
1283 struct hsm_current_action *req_hca;
1284 struct ptlrpc_request *req;
1287 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1288 &RQF_MDS_HSM_ACTION);
1292 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1294 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1296 ptlrpc_request_free(req);
1300 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1301 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1303 ptlrpc_request_set_replen(req);
1305 rc = mdc_queue_wait(req);
1309 req_hca = req_capsule_server_get(&req->rq_pill,
1310 &RMF_MDS_HSM_CURRENT_ACTION);
1311 if (req_hca == NULL)
1312 GOTO(out, rc = -EPROTO);
1317 ptlrpc_req_finished(req);
1321 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1323 struct ptlrpc_request *req;
1326 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1328 MDS_HSM_CT_UNREGISTER);
1330 GOTO(out, rc = -ENOMEM);
1332 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1334 ptlrpc_request_set_replen(req);
1336 rc = mdc_queue_wait(req);
1339 ptlrpc_req_finished(req);
1343 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1344 struct md_op_data *op_data)
1346 struct hsm_user_state *hus = op_data->op_data;
1347 struct hsm_user_state *req_hus;
1348 struct ptlrpc_request *req;
1351 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1352 &RQF_MDS_HSM_STATE_GET);
1356 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1358 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1360 ptlrpc_request_free(req);
1364 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1365 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1367 ptlrpc_request_set_replen(req);
1369 rc = mdc_queue_wait(req);
1373 req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1374 if (req_hus == NULL)
1375 GOTO(out, rc = -EPROTO);
1380 ptlrpc_req_finished(req);
1384 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1385 struct md_op_data *op_data)
1387 struct hsm_state_set *hss = op_data->op_data;
1388 struct hsm_state_set *req_hss;
1389 struct ptlrpc_request *req;
1392 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1393 &RQF_MDS_HSM_STATE_SET);
1397 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1399 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1401 ptlrpc_request_free(req);
1405 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1406 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1409 req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1410 if (req_hss == NULL)
1411 GOTO(out, rc = -EPROTO);
1414 ptlrpc_request_set_replen(req);
1416 rc = mdc_queue_wait(req);
1420 ptlrpc_req_finished(req);
1424 static int mdc_ioc_hsm_request(struct obd_export *exp,
1425 struct hsm_user_request *hur)
1427 struct obd_import *imp = class_exp2cliimp(exp);
1428 struct ptlrpc_request *req;
1429 struct hsm_request *req_hr;
1430 struct hsm_user_item *req_hui;
1434 req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1436 GOTO(out, rc = -ENOMEM);
1438 req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1439 hur->hur_request.hr_itemcount
1440 * sizeof(struct hsm_user_item));
1441 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1442 hur->hur_request.hr_data_len);
1444 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1446 ptlrpc_request_free(req);
1450 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1452 /* Copy hsm_request struct */
1453 req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1455 GOTO(out, rc = -EPROTO);
1456 *req_hr = hur->hur_request;
1458 /* Copy hsm_user_item structs */
1459 req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1460 if (req_hui == NULL)
1461 GOTO(out, rc = -EPROTO);
1462 memcpy(req_hui, hur->hur_user_item,
1463 hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1465 /* Copy opaque field */
1466 req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1467 if (req_opaque == NULL)
1468 GOTO(out, rc = -EPROTO);
1469 memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1471 ptlrpc_request_set_replen(req);
1473 rc = mdc_queue_wait(req);
1477 ptlrpc_req_finished(req);
1481 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
1483 struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1485 LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1487 lh->kuc_magic = KUC_MAGIC;
1488 lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1489 lh->kuc_flags = flags;
1490 lh->kuc_msgtype = CL_RECORD;
1491 lh->kuc_msglen = len;
1495 #define D_CHANGELOG 0
1497 struct changelog_show {
1502 struct obd_device *cs_obd;
1505 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1506 struct llog_rec_hdr *hdr, void *data)
1508 struct changelog_show *cs = data;
1509 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1513 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1515 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1516 cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
1517 rec->cr.cr_type, rc);
1521 if (rec->cr.cr_index < cs->cs_startrec) {
1522 /* Skip entries earlier than what we are interested in */
1523 CDEBUG(D_CHANGELOG, "rec=%llu start=%llu\n",
1524 rec->cr.cr_index, cs->cs_startrec);
1528 CDEBUG(D_CHANGELOG, "%llu %02d%-5s %llu 0x%x t="DFID" p="DFID
1529 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
1530 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1531 rec->cr.cr_flags & CLF_FLAGMASK,
1532 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1533 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1535 len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1537 /* Set up the message */
1538 lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1539 memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1541 rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1542 CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len, rc);
1547 static int mdc_changelog_send_thread(void *csdata)
1549 struct changelog_show *cs = csdata;
1550 struct llog_ctxt *ctxt = NULL;
1551 struct llog_handle *llh = NULL;
1552 struct kuc_hdr *kuch;
1555 CDEBUG(D_CHANGELOG, "changelog to fp=%p start %llu\n",
1556 cs->cs_fp, cs->cs_startrec);
1558 OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1559 if (cs->cs_buf == NULL)
1560 GOTO(out, rc = -ENOMEM);
1562 /* Set up the remote catalog handle */
1563 ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1565 GOTO(out, rc = -ENOENT);
1566 rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1569 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1570 cs->cs_obd->obd_name, rc);
1573 rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
1575 CERROR("llog_init_handle failed %d\n", rc);
1579 rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1581 /* Send EOF no matter what our result */
1582 kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags);
1584 kuch->kuc_msgtype = CL_EOF;
1585 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1591 llog_cat_close(NULL, llh);
1593 llog_ctxt_put(ctxt);
1595 OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1600 static int mdc_ioc_changelog_send(struct obd_device *obd,
1601 struct ioc_changelog *icc)
1603 struct changelog_show *cs;
1606 /* Freed in mdc_changelog_send_thread */
1612 cs->cs_startrec = icc->icc_recno;
1613 /* matching fput in mdc_changelog_send_thread */
1614 cs->cs_fp = fget(icc->icc_id);
1615 cs->cs_flags = icc->icc_flags;
1618 * New thread because we should return to user app before
1619 * writing into our pipe
1621 rc = PTR_ERR(kthread_run(mdc_changelog_send_thread, cs,
1622 "mdc_clg_send_thread"));
1623 if (!IS_ERR_VALUE(rc)) {
1624 CDEBUG(D_CHANGELOG, "start changelog thread\n");
1628 CERROR("Failed to start changelog thread: %d\n", rc);
1633 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1634 struct lustre_kernelcomm *lk);
1636 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
1637 struct obd_quotactl *oqctl)
1639 struct client_obd *cli = &exp->exp_obd->u.cli;
1640 struct ptlrpc_request *req;
1641 struct obd_quotactl *body;
1644 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1645 &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
1650 body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1653 ptlrpc_request_set_replen(req);
1655 /* the next poll will find -ENODATA, that means quotacheck is
1657 cli->cl_qchk_stat = -ENODATA;
1658 rc = ptlrpc_queue_wait(req);
1660 cli->cl_qchk_stat = rc;
1661 ptlrpc_req_finished(req);
1665 static int mdc_quota_poll_check(struct obd_export *exp,
1666 struct if_quotacheck *qchk)
1668 struct client_obd *cli = &exp->exp_obd->u.cli;
1671 qchk->obd_uuid = cli->cl_target_uuid;
1672 memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
1674 rc = cli->cl_qchk_stat;
1675 /* the client is not the previous one */
1676 if (rc == CL_NOT_QUOTACHECKED)
1681 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1682 struct obd_quotactl *oqctl)
1684 struct ptlrpc_request *req;
1685 struct obd_quotactl *oqc;
1688 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1689 &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1694 oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1697 ptlrpc_request_set_replen(req);
1698 ptlrpc_at_set_req_timeout(req);
1699 req->rq_no_resend = 1;
1701 rc = ptlrpc_queue_wait(req);
1703 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1705 if (req->rq_repmsg) {
1706 oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1710 CERROR("Can't unpack obd_quotactl\n");
1714 CERROR("Can't unpack obd_quotactl\n");
1717 ptlrpc_req_finished(req);
1722 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1723 struct md_op_data *op_data)
1726 struct ptlrpc_request *req;
1728 struct mdc_swap_layouts *msl, *payload;
1730 msl = op_data->op_data;
1732 /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1733 * first thing it will do is to cancel the 2 layout
1734 * locks hold by this client.
1735 * So the client must cancel its layout locks on the 2 fids
1736 * with the request RPC to avoid extra RPC round trips
1738 count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1739 LCK_CR, MDS_INODELOCK_LAYOUT);
1740 count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1741 LCK_CR, MDS_INODELOCK_LAYOUT);
1743 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1744 &RQF_MDS_SWAP_LAYOUTS);
1746 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1750 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1751 mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
1753 rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1755 ptlrpc_request_free(req);
1759 mdc_swap_layouts_pack(req, op_data);
1761 payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1766 ptlrpc_request_set_replen(req);
1768 rc = ptlrpc_queue_wait(req);
1773 ptlrpc_req_finished(req);
1777 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1778 void *karg, void *uarg)
1780 struct obd_device *obd = exp->exp_obd;
1781 struct obd_ioctl_data *data = karg;
1782 struct obd_import *imp = obd->u.cli.cl_import;
1783 struct llog_ctxt *ctxt;
1786 if (!try_module_get(THIS_MODULE)) {
1787 CERROR("Can't get module. Is it alive?");
1791 case OBD_IOC_CHANGELOG_SEND:
1792 rc = mdc_ioc_changelog_send(obd, karg);
1794 case OBD_IOC_CHANGELOG_CLEAR: {
1795 struct ioc_changelog *icc = karg;
1796 struct changelog_setinfo cs = {
1797 .cs_recno = icc->icc_recno,
1798 .cs_id = icc->icc_id
1801 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
1802 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1806 case OBD_IOC_FID2PATH:
1807 rc = mdc_ioc_fid2path(exp, karg);
1809 case LL_IOC_HSM_CT_START:
1810 rc = mdc_ioc_hsm_ct_start(exp, karg);
1811 /* ignore if it was already registered on this MDS. */
1815 case LL_IOC_HSM_PROGRESS:
1816 rc = mdc_ioc_hsm_progress(exp, karg);
1818 case LL_IOC_HSM_STATE_GET:
1819 rc = mdc_ioc_hsm_state_get(exp, karg);
1821 case LL_IOC_HSM_STATE_SET:
1822 rc = mdc_ioc_hsm_state_set(exp, karg);
1824 case LL_IOC_HSM_ACTION:
1825 rc = mdc_ioc_hsm_current_action(exp, karg);
1827 case LL_IOC_HSM_REQUEST:
1828 rc = mdc_ioc_hsm_request(exp, karg);
1830 case OBD_IOC_CLIENT_RECOVER:
1831 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1835 case IOC_OSC_SET_ACTIVE:
1836 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1838 case OBD_IOC_PARSE: {
1839 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1840 rc = class_config_parse_llog(NULL, ctxt, data->ioc_inlbuf1,
1842 llog_ctxt_put(ctxt);
1845 case OBD_IOC_LLOG_INFO:
1846 case OBD_IOC_LLOG_PRINT: {
1847 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1848 rc = llog_ioctl(NULL, ctxt, cmd, data);
1849 llog_ctxt_put(ctxt);
1852 case OBD_IOC_POLL_QUOTACHECK:
1853 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
1855 case OBD_IOC_PING_TARGET:
1856 rc = ptlrpc_obd_ping(obd);
1859 * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1860 * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1861 * there'd be no LMV layer thus we might be called here. Eventually
1862 * this code should be removed.
1865 case IOC_OBD_STATFS: {
1866 struct obd_statfs stat_buf = {0};
1868 if (*((__u32 *) data->ioc_inlbuf2) != 0)
1869 GOTO(out, rc = -ENODEV);
1872 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1873 min((int) data->ioc_plen2,
1874 (int) sizeof(struct obd_uuid))))
1875 GOTO(out, rc = -EFAULT);
1877 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1878 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1883 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1884 min((int) data->ioc_plen1,
1885 (int) sizeof(stat_buf))))
1886 GOTO(out, rc = -EFAULT);
1890 case OBD_IOC_QUOTACTL: {
1891 struct if_quotactl *qctl = karg;
1892 struct obd_quotactl *oqctl;
1894 OBD_ALLOC_PTR(oqctl);
1896 GOTO(out, rc = -ENOMEM);
1898 QCTL_COPY(oqctl, qctl);
1899 rc = obd_quotactl(exp, oqctl);
1901 QCTL_COPY(qctl, oqctl);
1902 qctl->qc_valid = QC_MDTIDX;
1903 qctl->obd_uuid = obd->u.cli.cl_target_uuid;
1906 OBD_FREE_PTR(oqctl);
1909 case LL_IOC_GET_CONNECT_FLAGS:
1910 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
1911 sizeof(*exp_connect_flags_ptr(exp))))
1912 GOTO(out, rc = -EFAULT);
1915 case LL_IOC_LOV_SWAP_LAYOUTS:
1916 rc = mdc_ioc_swap_layouts(exp, karg);
1919 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
1920 GOTO(out, rc = -ENOTTY);
1923 module_put(THIS_MODULE);
1928 int mdc_get_info_rpc(struct obd_export *exp,
1929 u32 keylen, void *key,
1930 int vallen, void *val)
1932 struct obd_import *imp = class_exp2cliimp(exp);
1933 struct ptlrpc_request *req;
1937 req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1941 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1942 RCL_CLIENT, keylen);
1943 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1944 RCL_CLIENT, sizeof(__u32));
1946 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1948 ptlrpc_request_free(req);
1952 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
1953 memcpy(tmp, key, keylen);
1954 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
1955 memcpy(tmp, &vallen, sizeof(__u32));
1957 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1958 RCL_SERVER, vallen);
1959 ptlrpc_request_set_replen(req);
1961 rc = ptlrpc_queue_wait(req);
1962 /* -EREMOTE means the get_info result is partial, and it needs to
1963 * continue on another MDT, see fid2path part in lmv_iocontrol */
1964 if (rc == 0 || rc == -EREMOTE) {
1965 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
1966 memcpy(val, tmp, vallen);
1967 if (ptlrpc_rep_need_swab(req)) {
1968 if (KEY_IS(KEY_FID2PATH))
1969 lustre_swab_fid2path(val);
1972 ptlrpc_req_finished(req);
1977 static void lustre_swab_hai(struct hsm_action_item *h)
1979 __swab32s(&h->hai_len);
1980 __swab32s(&h->hai_action);
1981 lustre_swab_lu_fid(&h->hai_fid);
1982 lustre_swab_lu_fid(&h->hai_dfid);
1983 __swab64s(&h->hai_cookie);
1984 __swab64s(&h->hai_extent.offset);
1985 __swab64s(&h->hai_extent.length);
1986 __swab64s(&h->hai_gid);
1989 static void lustre_swab_hal(struct hsm_action_list *h)
1991 struct hsm_action_item *hai;
1994 __swab32s(&h->hal_version);
1995 __swab32s(&h->hal_count);
1996 __swab32s(&h->hal_archive_id);
1997 __swab64s(&h->hal_flags);
1999 for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2000 lustre_swab_hai(hai);
2003 static void lustre_swab_kuch(struct kuc_hdr *l)
2005 __swab16s(&l->kuc_magic);
2006 /* __u8 l->kuc_transport */
2007 __swab16s(&l->kuc_msgtype);
2008 __swab16s(&l->kuc_msglen);
2011 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2012 struct lustre_kernelcomm *lk)
2014 struct obd_import *imp = class_exp2cliimp(exp);
2015 __u32 archive = lk->lk_data;
2018 if (lk->lk_group != KUC_GRP_HSM) {
2019 CERROR("Bad copytool group %d\n", lk->lk_group);
2023 CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2024 lk->lk_uid, lk->lk_group, lk->lk_flags);
2026 if (lk->lk_flags & LK_FLG_STOP) {
2027 /* Unregister with the coordinator */
2028 rc = mdc_ioc_hsm_ct_unregister(imp);
2030 rc = mdc_ioc_hsm_ct_register(imp, archive);
2037 * Send a message to any listening copytools
2038 * @param val KUC message (kuc_hdr + hsm_action_list)
2039 * @param len total length of message
2041 static int mdc_hsm_copytool_send(int len, void *val)
2043 struct kuc_hdr *lh = (struct kuc_hdr *)val;
2044 struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
2047 if (len < sizeof(*lh) + sizeof(*hal)) {
2048 CERROR("Short HSM message %d < %d\n", len,
2049 (int) (sizeof(*lh) + sizeof(*hal)));
2052 if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2053 lustre_swab_kuch(lh);
2054 lustre_swab_hal(hal);
2055 } else if (lh->kuc_magic != KUC_MAGIC) {
2056 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2061 "Received message mg=%x t=%d m=%d l=%d actions=%d on %s\n",
2062 lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2063 lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2065 /* Broadcast to HSM listeners */
2066 rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2072 * callback function passed to kuc for re-registering each HSM copytool
2073 * running on MDC, after MDT shutdown/recovery.
2074 * @param data archive id served by the copytool
2075 * @param cb_arg callback argument (obd_import)
2077 static int mdc_hsm_ct_reregister(__u32 data, void *cb_arg)
2079 struct obd_import *imp = (struct obd_import *)cb_arg;
2080 __u32 archive = data;
2083 CDEBUG(D_HA, "recover copytool registration to MDT (archive=%#x)\n",
2085 rc = mdc_ioc_hsm_ct_register(imp, archive);
2087 /* ignore error if the copytool is already registered */
2088 return ((rc != 0) && (rc != -EEXIST)) ? rc : 0;
2092 * Re-establish all kuc contexts with MDT
2093 * after MDT shutdown/recovery.
2095 static int mdc_kuc_reregister(struct obd_import *imp)
2097 /* re-register HSM agents */
2098 return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2102 int mdc_set_info_async(const struct lu_env *env,
2103 struct obd_export *exp,
2104 u32 keylen, void *key,
2105 u32 vallen, void *val,
2106 struct ptlrpc_request_set *set)
2108 struct obd_import *imp = class_exp2cliimp(exp);
2111 if (KEY_IS(KEY_READ_ONLY)) {
2112 if (vallen != sizeof(int))
2115 spin_lock(&imp->imp_lock);
2116 if (*((int *)val)) {
2117 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2118 imp->imp_connect_data.ocd_connect_flags |=
2121 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2122 imp->imp_connect_data.ocd_connect_flags &=
2123 ~OBD_CONNECT_RDONLY;
2125 spin_unlock(&imp->imp_lock);
2127 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2128 keylen, key, vallen, val, set);
2131 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2132 sptlrpc_conf_client_adapt(exp->exp_obd);
2135 if (KEY_IS(KEY_FLUSH_CTX)) {
2136 sptlrpc_import_flush_my_ctx(imp);
2139 if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2140 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2141 keylen, key, vallen, val, set);
2144 if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2145 rc = mdc_hsm_copytool_send(vallen, val);
2149 CERROR("Unknown key %s\n", (char *)key);
2153 int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2154 __u32 keylen, void *key, __u32 *vallen, void *val,
2155 struct lov_stripe_md *lsm)
2159 if (KEY_IS(KEY_MAX_EASIZE)) {
2160 int mdsize, *max_easize;
2162 if (*vallen != sizeof(int))
2164 mdsize = *(int *)val;
2165 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2166 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2168 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2170 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2171 int *default_easize;
2173 if (*vallen != sizeof(int))
2175 default_easize = val;
2176 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2178 } else if (KEY_IS(KEY_MAX_COOKIESIZE)) {
2179 int mdsize, *max_cookiesize;
2181 if (*vallen != sizeof(int))
2183 mdsize = *(int *)val;
2184 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_cookiesize)
2185 exp->exp_obd->u.cli.cl_max_mds_cookiesize = mdsize;
2186 max_cookiesize = val;
2187 *max_cookiesize = exp->exp_obd->u.cli.cl_max_mds_cookiesize;
2189 } else if (KEY_IS(KEY_DEFAULT_COOKIESIZE)) {
2190 int *default_cookiesize;
2192 if (*vallen != sizeof(int))
2194 default_cookiesize = val;
2195 *default_cookiesize =
2196 exp->exp_obd->u.cli.cl_default_mds_cookiesize;
2198 } else if (KEY_IS(KEY_CONN_DATA)) {
2199 struct obd_import *imp = class_exp2cliimp(exp);
2200 struct obd_connect_data *data = val;
2202 if (*vallen != sizeof(*data))
2205 *data = imp->imp_connect_data;
2207 } else if (KEY_IS(KEY_TGT_COUNT)) {
2212 rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2217 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
2218 struct obd_capa *oc, struct obd_client_handle *handle,
2221 struct ptlrpc_request *req;
2222 struct mdt_body *body;
2225 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
2229 mdc_set_capa_size(req, &RMF_CAPA1, oc);
2231 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
2233 ptlrpc_request_free(req);
2237 mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
2239 ptlrpc_request_set_replen(req);
2241 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2242 rc = ptlrpc_queue_wait(req);
2243 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2245 CERROR("Pin failed: %d\n", rc);
2249 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2251 GOTO(err_out, rc = -EPROTO);
2253 handle->och_fh = body->handle;
2254 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
2256 handle->och_mod = obd_mod_alloc();
2257 if (handle->och_mod == NULL) {
2258 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
2259 GOTO(err_out, rc = -ENOMEM);
2261 handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
2266 ptlrpc_req_finished(req);
2270 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
2273 struct ptlrpc_request *req;
2274 struct mdt_body *body;
2277 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
2278 LUSTRE_MDS_VERSION, MDS_UNPIN);
2282 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
2283 body->handle = handle->och_fh;
2286 ptlrpc_request_set_replen(req);
2288 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2289 rc = ptlrpc_queue_wait(req);
2290 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2293 CERROR("Unpin failed: %d\n", rc);
2295 ptlrpc_req_finished(req);
2296 ptlrpc_req_finished(handle->och_mod->mod_open_req);
2298 obd_mod_put(handle->och_mod);
2302 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
2303 struct obd_capa *oc, struct ptlrpc_request **request)
2305 struct ptlrpc_request *req;
2309 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2313 mdc_set_capa_size(req, &RMF_CAPA1, oc);
2315 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2317 ptlrpc_request_free(req);
2321 mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
2323 ptlrpc_request_set_replen(req);
2325 rc = ptlrpc_queue_wait(req);
2327 ptlrpc_req_finished(req);
2333 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2334 enum obd_import_event event)
2338 LASSERT(imp->imp_obd == obd);
2341 case IMP_EVENT_DISCON: {
2343 /* XXX Pass event up to OBDs stack. used only for FLD now */
2344 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2348 case IMP_EVENT_INACTIVE: {
2349 struct client_obd *cli = &obd->u.cli;
2351 * Flush current sequence to make client obtain new one
2352 * from server in case of disconnect/reconnect.
2354 if (cli->cl_seq != NULL)
2355 seq_client_flush(cli->cl_seq);
2357 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2360 case IMP_EVENT_INVALIDATE: {
2361 struct ldlm_namespace *ns = obd->obd_namespace;
2363 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2367 case IMP_EVENT_ACTIVE:
2368 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2369 /* redo the kuc registration after reconnecting */
2371 rc = mdc_kuc_reregister(imp);
2374 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2376 case IMP_EVENT_DEACTIVATE:
2377 case IMP_EVENT_ACTIVATE:
2380 CERROR("Unknown import event %x\n", event);
2386 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
2387 struct md_op_data *op_data)
2389 struct client_obd *cli = &exp->exp_obd->u.cli;
2390 struct lu_client_seq *seq = cli->cl_seq;
2392 return seq_client_alloc_fid(NULL, seq, fid);
2395 struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2397 struct client_obd *cli = &exp->exp_obd->u.cli;
2399 return &cli->cl_target_uuid;
2403 * Determine whether the lock can be canceled before replaying it during
2404 * recovery, non zero value will be return if the lock can be canceled,
2405 * or zero returned for not
2407 static int mdc_cancel_for_recovery(struct ldlm_lock *lock)
2409 if (lock->l_resource->lr_type != LDLM_IBITS)
2412 /* FIXME: if we ever get into a situation where there are too many
2413 * opened files with open locks on a single node, then we really
2414 * should replay these open locks to reget it */
2415 if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2421 static int mdc_resource_inode_free(struct ldlm_resource *res)
2423 if (res->lr_lvb_inode)
2424 res->lr_lvb_inode = NULL;
2429 struct ldlm_valblock_ops inode_lvbo = {
2430 .lvbo_free = mdc_resource_inode_free,
2433 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2435 struct client_obd *cli = &obd->u.cli;
2436 struct lprocfs_static_vars lvars = { NULL };
2439 OBD_ALLOC(cli->cl_rpc_lock, sizeof(*cli->cl_rpc_lock));
2440 if (!cli->cl_rpc_lock)
2442 mdc_init_rpc_lock(cli->cl_rpc_lock);
2446 OBD_ALLOC(cli->cl_close_lock, sizeof(*cli->cl_close_lock));
2447 if (!cli->cl_close_lock)
2448 GOTO(err_rpc_lock, rc = -ENOMEM);
2449 mdc_init_rpc_lock(cli->cl_close_lock);
2451 rc = client_obd_setup(obd, cfg);
2453 GOTO(err_close_lock, rc);
2454 lprocfs_mdc_init_vars(&lvars);
2455 lprocfs_obd_setup(obd, lvars.obd_vars);
2456 sptlrpc_lprocfs_cliobd_attach(obd);
2457 ptlrpc_lprocfs_register_obd(obd);
2459 ns_register_cancel(obd->obd_namespace, mdc_cancel_for_recovery);
2461 obd->obd_namespace->ns_lvbo = &inode_lvbo;
2463 rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
2466 CERROR("failed to setup llogging subsystems\n");
2472 OBD_FREE(cli->cl_close_lock, sizeof(*cli->cl_close_lock));
2474 OBD_FREE(cli->cl_rpc_lock, sizeof(*cli->cl_rpc_lock));
2479 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
2480 * us to make MDS RPCs with large enough reply buffers to hold a default
2481 * sized EA and cookie without having to calculate this (via a call into the
2482 * LOV + OSCs) each time we make an RPC. The maximum size is also tracked
2483 * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2484 * a large number of stripes is possible. If a larger reply buffer is
2485 * required it will be reallocated in the ptlrpc layer due to overflow.
2487 static int mdc_init_ea_size(struct obd_export *exp, int easize,
2488 int def_easize, int cookiesize, int def_cookiesize)
2490 struct obd_device *obd = exp->exp_obd;
2491 struct client_obd *cli = &obd->u.cli;
2493 if (cli->cl_max_mds_easize < easize)
2494 cli->cl_max_mds_easize = easize;
2496 if (cli->cl_default_mds_easize < def_easize)
2497 cli->cl_default_mds_easize = def_easize;
2499 if (cli->cl_max_mds_cookiesize < cookiesize)
2500 cli->cl_max_mds_cookiesize = cookiesize;
2502 if (cli->cl_default_mds_cookiesize < def_cookiesize)
2503 cli->cl_default_mds_cookiesize = def_cookiesize;
2508 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2513 case OBD_CLEANUP_EARLY:
2515 case OBD_CLEANUP_EXPORTS:
2516 /* Failsafe, ok if racy */
2517 if (obd->obd_type->typ_refcnt <= 1)
2518 libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2520 obd_cleanup_client_import(obd);
2521 ptlrpc_lprocfs_unregister_obd(obd);
2522 lprocfs_obd_cleanup(obd);
2524 rc = obd_llog_finish(obd, 0);
2526 CERROR("failed to cleanup llogging subsystems\n");
2532 static int mdc_cleanup(struct obd_device *obd)
2534 struct client_obd *cli = &obd->u.cli;
2536 OBD_FREE(cli->cl_rpc_lock, sizeof(*cli->cl_rpc_lock));
2537 OBD_FREE(cli->cl_close_lock, sizeof(*cli->cl_close_lock));
2541 return client_obd_cleanup(obd);
2545 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
2546 struct obd_device *tgt, int *index)
2548 struct llog_ctxt *ctxt;
2551 LASSERT(olg == &obd->obd_olg);
2553 rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt,
2558 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2559 llog_initiator_connect(ctxt);
2560 llog_ctxt_put(ctxt);
2565 static int mdc_llog_finish(struct obd_device *obd, int count)
2567 struct llog_ctxt *ctxt;
2569 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2571 llog_cleanup(NULL, ctxt);
2576 static int mdc_process_config(struct obd_device *obd, u32 len, void *buf)
2578 struct lustre_cfg *lcfg = buf;
2579 struct lprocfs_static_vars lvars = { NULL };
2582 lprocfs_mdc_init_vars(&lvars);
2583 switch (lcfg->lcfg_command) {
2585 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
2595 /* get remote permission for current user on fid */
2596 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2597 struct obd_capa *oc, __u32 suppgid,
2598 struct ptlrpc_request **request)
2600 struct ptlrpc_request *req;
2603 LASSERT(client_is_remote(exp));
2606 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2610 mdc_set_capa_size(req, &RMF_CAPA1, oc);
2612 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2614 ptlrpc_request_free(req);
2618 mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2620 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2621 sizeof(struct mdt_remote_perm));
2623 ptlrpc_request_set_replen(req);
2625 rc = ptlrpc_queue_wait(req);
2627 ptlrpc_req_finished(req);
2633 static int mdc_interpret_renew_capa(const struct lu_env *env,
2634 struct ptlrpc_request *req, void *args,
2637 struct mdc_renew_capa_args *ra = args;
2638 struct mdt_body *body = NULL;
2639 struct lustre_capa *capa;
2642 GOTO(out, capa = ERR_PTR(status));
2644 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2646 GOTO(out, capa = ERR_PTR(-EFAULT));
2648 if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
2649 GOTO(out, capa = ERR_PTR(-ENOENT));
2651 capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
2653 GOTO(out, capa = ERR_PTR(-EFAULT));
2655 ra->ra_cb(ra->ra_oc, capa);
2659 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2662 struct ptlrpc_request *req;
2663 struct mdc_renew_capa_args *ra;
2665 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
2666 LUSTRE_MDS_VERSION, MDS_GETATTR);
2670 /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
2671 * capa to renew is oss capa.
2673 mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
2674 ptlrpc_request_set_replen(req);
2676 CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
2677 ra = ptlrpc_req_async_args(req);
2680 req->rq_interpret_reply = mdc_interpret_renew_capa;
2681 ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
2685 struct obd_ops mdc_obd_ops = {
2686 .o_owner = THIS_MODULE,
2687 .o_setup = mdc_setup,
2688 .o_precleanup = mdc_precleanup,
2689 .o_cleanup = mdc_cleanup,
2690 .o_add_conn = client_import_add_conn,
2691 .o_del_conn = client_import_del_conn,
2692 .o_connect = client_connect_import,
2693 .o_disconnect = client_disconnect_export,
2694 .o_iocontrol = mdc_iocontrol,
2695 .o_set_info_async = mdc_set_info_async,
2696 .o_statfs = mdc_statfs,
2698 .o_unpin = mdc_unpin,
2699 .o_fid_init = client_fid_init,
2700 .o_fid_fini = client_fid_fini,
2701 .o_fid_alloc = mdc_fid_alloc,
2702 .o_import_event = mdc_import_event,
2703 .o_llog_init = mdc_llog_init,
2704 .o_llog_finish = mdc_llog_finish,
2705 .o_get_info = mdc_get_info,
2706 .o_process_config = mdc_process_config,
2707 .o_get_uuid = mdc_get_uuid,
2708 .o_quotactl = mdc_quotactl,
2709 .o_quotacheck = mdc_quotacheck
2712 struct md_ops mdc_md_ops = {
2713 .m_getstatus = mdc_getstatus,
2714 .m_null_inode = mdc_null_inode,
2715 .m_find_cbdata = mdc_find_cbdata,
2716 .m_close = mdc_close,
2717 .m_create = mdc_create,
2718 .m_done_writing = mdc_done_writing,
2719 .m_enqueue = mdc_enqueue,
2720 .m_getattr = mdc_getattr,
2721 .m_getattr_name = mdc_getattr_name,
2722 .m_intent_lock = mdc_intent_lock,
2724 .m_is_subdir = mdc_is_subdir,
2725 .m_rename = mdc_rename,
2726 .m_setattr = mdc_setattr,
2727 .m_setxattr = mdc_setxattr,
2728 .m_getxattr = mdc_getxattr,
2730 .m_readpage = mdc_readpage,
2731 .m_unlink = mdc_unlink,
2732 .m_cancel_unused = mdc_cancel_unused,
2733 .m_init_ea_size = mdc_init_ea_size,
2734 .m_set_lock_data = mdc_set_lock_data,
2735 .m_lock_match = mdc_lock_match,
2736 .m_get_lustre_md = mdc_get_lustre_md,
2737 .m_free_lustre_md = mdc_free_lustre_md,
2738 .m_set_open_replay_data = mdc_set_open_replay_data,
2739 .m_clear_open_replay_data = mdc_clear_open_replay_data,
2740 .m_renew_capa = mdc_renew_capa,
2741 .m_unpack_capa = mdc_unpack_capa,
2742 .m_get_remote_perm = mdc_get_remote_perm,
2743 .m_intent_getattr_async = mdc_intent_getattr_async,
2744 .m_revalidate_lock = mdc_revalidate_lock
2747 int __init mdc_init(void)
2750 struct lprocfs_static_vars lvars = { NULL };
2752 lprocfs_mdc_init_vars(&lvars);
2754 rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
2755 LUSTRE_MDC_NAME, NULL);
2759 static void /*__exit*/ mdc_exit(void)
2761 class_unregister_type(LUSTRE_MDC_NAME);
2764 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2765 MODULE_DESCRIPTION("Lustre Metadata Client");
2766 MODULE_LICENSE("GPL");
2768 module_init(mdc_init);
2769 module_exit(mdc_exit);