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, 2013, 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
40 # include <linux/module.h>
41 # include <linux/pagemap.h>
42 # include <linux/miscdevice.h>
43 # include <linux/init.h>
44 # include <linux/utsname.h>
46 # include <liblustre.h>
49 #include <lustre_acl.h>
50 #include <lustre_ioctl.h>
51 #include <obd_class.h>
52 #include <lustre_lmv.h>
53 #include <lustre_fid.h>
54 #include <lprocfs_status.h>
55 #include <lustre_param.h>
56 #include <lustre_log.h>
57 #include <cl_object.h>
60 #include "mdc_internal.h"
62 #define REQUEST_MINOR 244
64 struct mdc_renew_capa_args {
65 struct obd_capa *ra_oc;
66 renew_capa_cb_t ra_cb;
69 static int mdc_cleanup(struct obd_device *obd);
71 int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
72 const struct req_msg_field *field, struct obd_capa **oc)
74 struct lustre_capa *capa;
78 /* swabbed already in mdc_enqueue */
79 capa = req_capsule_server_get(&req->rq_pill, field);
83 c = alloc_capa(CAPA_SITE_CLIENT);
85 CDEBUG(D_INFO, "alloc capa failed!\n");
94 static inline int mdc_queue_wait(struct ptlrpc_request *req)
96 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
99 /* obd_get_request_slot() ensures that this client has no more
100 * than cl_max_rpcs_in_flight RPCs simultaneously inf light
102 rc = obd_get_request_slot(cli);
106 rc = ptlrpc_queue_wait(req);
107 obd_put_request_slot(cli);
112 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
113 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
114 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
115 struct obd_capa **pc, int level, int msg_flags)
117 struct ptlrpc_request *req;
118 struct mdt_body *body;
122 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
123 LUSTRE_MDS_VERSION, MDS_GETSTATUS);
127 mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
128 lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
129 req->rq_send_state = level;
131 ptlrpc_request_set_replen(req);
133 rc = ptlrpc_queue_wait(req);
137 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
139 GOTO(out, rc = -EPROTO);
141 if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
142 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
147 *rootfid = body->mbo_fid1;
148 CDEBUG(D_NET, "root fid="DFID", last_committed="LPU64"\n",
149 PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
152 ptlrpc_req_finished(req);
156 /* This should be mdc_get_info("rootfid") */
157 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
158 struct obd_capa **pc)
160 return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
165 * This function now is known to always saying that it will receive 4 buffers
166 * from server. Even for cases when acl_size and md_size is zero, RPC header
167 * will contain 4 fields and RPC itself will contain zero size fields. This is
168 * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
169 * and thus zero, it shrinks it, making zero size. The same story about
170 * md_size. And this is course of problem when client waits for smaller number
171 * of fields. This issue will be fixed later when client gets aware of RPC
174 static int mdc_getattr_common(struct obd_export *exp,
175 struct ptlrpc_request *req)
177 struct req_capsule *pill = &req->rq_pill;
178 struct mdt_body *body;
183 /* Request message already built. */
184 rc = ptlrpc_queue_wait(req);
188 /* sanity check for the reply */
189 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
193 CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
195 mdc_update_max_ea_from_body(exp, body);
196 if (body->mbo_eadatasize != 0) {
197 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
198 body->mbo_eadatasize);
203 if (body->mbo_valid & OBD_MD_FLRMTPERM) {
204 struct mdt_remote_perm *perm;
206 LASSERT(client_is_remote(exp));
207 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
208 lustre_swab_mdt_remote_perm);
213 if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
214 struct lustre_capa *capa;
215 capa = req_capsule_server_get(pill, &RMF_CAPA1);
223 int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
224 struct ptlrpc_request **request)
226 struct ptlrpc_request *req;
230 /* Single MDS without an LMV case */
231 if (op_data->op_flags & MF_GET_MDT_IDX) {
236 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
240 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
242 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
244 ptlrpc_request_free(req);
248 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
249 op_data->op_valid, op_data->op_mode, -1, 0);
251 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
253 if (op_data->op_valid & OBD_MD_FLRMTPERM) {
254 LASSERT(client_is_remote(exp));
255 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
256 sizeof(struct mdt_remote_perm));
258 ptlrpc_request_set_replen(req);
260 rc = mdc_getattr_common(exp, req);
262 ptlrpc_req_finished(req);
268 int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
269 struct ptlrpc_request **request)
271 struct ptlrpc_request *req;
276 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
277 &RQF_MDS_GETATTR_NAME);
281 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
282 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
283 op_data->op_namelen + 1);
285 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
287 ptlrpc_request_free(req);
291 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
292 op_data->op_valid, op_data->op_mode,
293 op_data->op_suppgids[0], 0);
295 if (op_data->op_name) {
296 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
297 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
298 op_data->op_namelen);
299 memcpy(name, op_data->op_name, op_data->op_namelen);
302 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
304 ptlrpc_request_set_replen(req);
306 rc = mdc_getattr_common(exp, req);
308 ptlrpc_req_finished(req);
314 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
315 const struct lu_fid *fid,
316 struct obd_capa *oc, int opcode, obd_valid valid,
317 const char *xattr_name, const char *input,
318 int input_size, int output_size, int flags,
319 __u32 suppgid, struct ptlrpc_request **request)
321 struct ptlrpc_request *req;
322 int xattr_namelen = 0;
328 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
332 mdc_set_capa_size(req, &RMF_CAPA1, oc);
334 xattr_namelen = strlen(xattr_name) + 1;
335 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
340 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
344 /* Flush local XATTR locks to get rid of a possible cancel RPC */
345 if (opcode == MDS_REINT && fid_is_sane(fid) &&
346 exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
347 struct list_head cancels = LIST_HEAD_INIT(cancels);
350 /* Without that packing would fail */
352 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
355 count = mdc_resource_get_unused(exp, fid,
357 MDS_INODELOCK_XATTR);
359 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
361 ptlrpc_request_free(req);
365 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
367 ptlrpc_request_free(req);
372 if (opcode == MDS_REINT) {
373 struct mdt_rec_setxattr *rec;
375 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
376 sizeof(struct mdt_rec_reint));
377 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
378 rec->sx_opcode = REINT_SETXATTR;
379 rec->sx_fsuid = from_kuid(&init_user_ns, current_fsuid());
380 rec->sx_fsgid = from_kgid(&init_user_ns, current_fsgid());
381 rec->sx_cap = cfs_curproc_cap_pack();
382 rec->sx_suppgid1 = suppgid;
383 rec->sx_suppgid2 = -1;
385 rec->sx_valid = valid | OBD_MD_FLCTIME;
386 rec->sx_time = cfs_time_current_sec();
387 rec->sx_size = output_size;
388 rec->sx_flags = flags;
390 mdc_pack_capa(req, &RMF_CAPA1, oc);
392 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
396 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
397 memcpy(tmp, xattr_name, xattr_namelen);
400 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
401 memcpy(tmp, input, input_size);
404 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
405 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
406 RCL_SERVER, output_size);
407 ptlrpc_request_set_replen(req);
410 if (opcode == MDS_REINT)
411 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
413 rc = ptlrpc_queue_wait(req);
415 if (opcode == MDS_REINT)
416 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
419 ptlrpc_req_finished(req);
425 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
426 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
427 const char *input, int input_size, int output_size,
428 int flags, __u32 suppgid, struct ptlrpc_request **request)
430 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
431 fid, oc, MDS_REINT, valid, xattr_name,
432 input, input_size, output_size, flags,
436 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
437 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
438 const char *input, int input_size, int output_size,
439 int flags, struct ptlrpc_request **request)
441 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
442 fid, oc, MDS_GETXATTR, valid, xattr_name,
443 input, input_size, output_size, flags,
447 #ifdef CONFIG_FS_POSIX_ACL
448 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
450 struct req_capsule *pill = &req->rq_pill;
451 struct mdt_body *body = md->body;
452 struct posix_acl *acl;
457 if (!body->mbo_aclsize)
460 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
465 acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
468 CERROR("convert xattr to acl: %d\n", rc);
472 rc = posix_acl_valid(acl);
474 CERROR("validate acl: %d\n", rc);
475 posix_acl_release(acl);
483 #define mdc_unpack_acl(req, md) 0
486 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
487 struct obd_export *dt_exp, struct obd_export *md_exp,
488 struct lustre_md *md)
490 struct req_capsule *pill = &req->rq_pill;
495 memset(md, 0, sizeof(*md));
497 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
498 LASSERT(md->body != NULL);
500 if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
502 struct lov_mds_md *lmm;
504 if (!S_ISREG(md->body->mbo_mode)) {
505 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
506 "regular file, but is not\n");
507 GOTO(out, rc = -EPROTO);
510 if (md->body->mbo_eadatasize == 0) {
511 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
512 "but eadatasize 0\n");
513 GOTO(out, rc = -EPROTO);
516 lmmsize = md->body->mbo_eadatasize;
517 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
519 GOTO(out, rc = -EPROTO);
521 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
525 if (rc < sizeof(*md->lsm)) {
526 CDEBUG(D_INFO, "lsm size too small: "
527 "rc < sizeof (*md->lsm) (%d < %d)\n",
528 rc, (int)sizeof(*md->lsm));
529 GOTO(out, rc = -EPROTO);
532 } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
534 struct lov_mds_md *lmv;
536 if (!S_ISDIR(md->body->mbo_mode)) {
537 CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
538 "directory, but is not\n");
539 GOTO(out, rc = -EPROTO);
542 if (md->body->mbo_eadatasize == 0) {
543 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
544 "but eadatasize 0\n");
548 if (md->body->mbo_valid & OBD_MD_MEA) {
549 lmvsize = md->body->mbo_eadatasize;
550 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
553 GOTO(out, rc = -EPROTO);
555 rc = obd_unpackmd(md_exp, (void *)&md->lmv, lmv,
560 if (rc < sizeof(*md->lmv)) {
561 CDEBUG(D_INFO, "size too small: "
562 "rc < sizeof(*md->lmv) (%d < %d)\n",
563 rc, (int)sizeof(*md->lmv));
564 GOTO(out, rc = -EPROTO);
570 if (md->body->mbo_valid & OBD_MD_FLRMTPERM) {
571 /* remote permission */
572 LASSERT(client_is_remote(exp));
573 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
574 lustre_swab_mdt_remote_perm);
575 if (!md->remote_perm)
576 GOTO(out, rc = -EPROTO);
577 } else if (md->body->mbo_valid & OBD_MD_FLACL) {
578 /* for ACL, it's possible that FLACL is set but aclsize is zero.
579 * only when aclsize != 0 there's an actual segment for ACL
582 if (md->body->mbo_aclsize) {
583 rc = mdc_unpack_acl(req, md);
586 #ifdef CONFIG_FS_POSIX_ACL
588 md->posix_acl = NULL;
592 if (md->body->mbo_valid & OBD_MD_FLMDSCAPA) {
593 struct obd_capa *oc = NULL;
595 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
601 if (md->body->mbo_valid & OBD_MD_FLOSSCAPA) {
602 struct obd_capa *oc = NULL;
604 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
614 capa_put(md->oss_capa);
618 capa_put(md->mds_capa);
621 #ifdef CONFIG_FS_POSIX_ACL
622 posix_acl_release(md->posix_acl);
625 obd_free_memmd(dt_exp, &md->lsm);
630 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
637 * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
640 void mdc_replay_open(struct ptlrpc_request *req)
642 struct md_open_data *mod = req->rq_cb_data;
643 struct ptlrpc_request *close_req;
644 struct obd_client_handle *och;
645 struct lustre_handle old;
646 struct mdt_body *body;
650 DEBUG_REQ(D_ERROR, req,
651 "Can't properly replay without open data.");
656 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
657 LASSERT(body != NULL);
661 struct lustre_handle *file_fh;
663 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
665 file_fh = &och->och_fh;
666 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
667 file_fh->cookie, body->mbo_handle.cookie);
669 *file_fh = body->mbo_handle;
671 close_req = mod->mod_close_req;
672 if (close_req != NULL) {
673 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
674 struct mdt_ioepoch *epoch;
676 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
677 epoch = req_capsule_client_get(&close_req->rq_pill,
682 LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
683 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
684 epoch->handle = body->mbo_handle;
689 void mdc_commit_open(struct ptlrpc_request *req)
691 struct md_open_data *mod = req->rq_cb_data;
696 * No need to touch md_open_data::mod_och, it holds a reference on
697 * \var mod and will zero references to each other, \var mod will be
698 * freed after that when md_open_data::mod_och will put the reference.
702 * Do not let open request to disappear as it still may be needed
703 * for close rpc to happen (it may happen on evict only, otherwise
704 * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
705 * called), just mark this rpc as committed to distinguish these 2
706 * cases, see mdc_close() for details. The open request reference will
707 * be put along with freeing \var mod.
709 ptlrpc_request_addref(req);
710 spin_lock(&req->rq_lock);
711 req->rq_committed = 1;
712 spin_unlock(&req->rq_lock);
713 req->rq_cb_data = NULL;
717 int mdc_set_open_replay_data(struct obd_export *exp,
718 struct obd_client_handle *och,
719 struct lookup_intent *it)
721 struct md_open_data *mod;
722 struct mdt_rec_create *rec;
723 struct mdt_body *body;
724 struct ptlrpc_request *open_req = it->d.lustre.it_data;
725 struct obd_import *imp = open_req->rq_import;
728 if (!open_req->rq_replay)
731 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
732 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
733 LASSERT(rec != NULL);
734 /* Incoming message in my byte order (it's been swabbed). */
735 /* Outgoing messages always in my byte order. */
736 LASSERT(body != NULL);
738 /* Only if the import is replayable, we set replay_open data */
739 if (och && imp->imp_replayable) {
740 mod = obd_mod_alloc();
742 DEBUG_REQ(D_ERROR, open_req,
743 "Can't allocate md_open_data");
748 * Take a reference on \var mod, to be freed on mdc_close().
749 * It protects \var mod from being freed on eviction (commit
750 * callback is called despite rq_replay flag).
751 * Another reference for \var och.
756 spin_lock(&open_req->rq_lock);
759 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
760 it_disposition(it, DISP_OPEN_STRIPE);
761 mod->mod_open_req = open_req;
762 open_req->rq_cb_data = mod;
763 open_req->rq_commit_cb = mdc_commit_open;
764 spin_unlock(&open_req->rq_lock);
767 rec->cr_fid2 = body->mbo_fid1;
768 rec->cr_ioepoch = body->mbo_ioepoch;
769 rec->cr_old_handle.cookie = body->mbo_handle.cookie;
770 open_req->rq_replay_cb = mdc_replay_open;
771 if (!fid_is_sane(&body->mbo_fid1)) {
772 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
777 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
781 static void mdc_free_open(struct md_open_data *mod)
785 if (mod->mod_is_create == 0 &&
786 imp_connect_disp_stripe(mod->mod_open_req->rq_import))
789 LASSERT(mod->mod_open_req->rq_replay == 0);
791 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
793 ptlrpc_request_committed(mod->mod_open_req, committed);
794 if (mod->mod_close_req)
795 ptlrpc_request_committed(mod->mod_close_req, committed);
798 int mdc_clear_open_replay_data(struct obd_export *exp,
799 struct obd_client_handle *och)
801 struct md_open_data *mod = och->och_mod;
805 * It is possible to not have \var mod in a case of eviction between
806 * lookup and ll_file_open().
811 LASSERT(mod != LP_POISON);
812 LASSERT(mod->mod_open_req != NULL);
822 /* Prepares the request for the replay by the given reply */
823 static void mdc_close_handle_reply(struct ptlrpc_request *req,
824 struct md_op_data *op_data, int rc) {
825 struct mdt_body *repbody;
826 struct mdt_ioepoch *epoch;
828 if (req && rc == -EAGAIN) {
829 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
830 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
832 epoch->flags |= MF_SOM_AU;
833 if (repbody->mbo_valid & OBD_MD_FLGETATTRLOCK)
834 op_data->op_flags |= MF_GETATTR_LOCK;
838 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
839 struct md_open_data *mod, struct ptlrpc_request **request)
841 struct obd_device *obd = class_exp2obd(exp);
842 struct ptlrpc_request *req;
843 struct req_format *req_fmt;
848 req_fmt = &RQF_MDS_CLOSE;
849 if (op_data->op_bias & MDS_HSM_RELEASE) {
850 req_fmt = &RQF_MDS_RELEASE_CLOSE;
852 /* allocate a FID for volatile file */
853 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
855 CERROR("%s: "DFID" failed to allocate FID: %d\n",
856 obd->obd_name, PFID(&op_data->op_fid1), rc);
857 /* save the errcode and proceed to close */
863 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
867 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
869 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
871 ptlrpc_request_free(req);
875 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
876 * portal whose threads are not taking any DLM locks and are therefore
877 * always progressing */
878 req->rq_request_portal = MDS_READPAGE_PORTAL;
879 ptlrpc_at_set_req_timeout(req);
881 /* Ensure that this close's handle is fixed up during replay. */
882 if (likely(mod != NULL)) {
883 LASSERTF(mod->mod_open_req != NULL &&
884 mod->mod_open_req->rq_type != LI_POISON,
885 "POISONED open %p!\n", mod->mod_open_req);
887 mod->mod_close_req = req;
889 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
890 /* We no longer want to preserve this open for replay even
891 * though the open was committed. b=3632, b=3633 */
892 spin_lock(&mod->mod_open_req->rq_lock);
893 mod->mod_open_req->rq_replay = 0;
894 spin_unlock(&mod->mod_open_req->rq_lock);
896 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
899 mdc_close_pack(req, op_data);
901 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
902 obd->u.cli.cl_default_mds_easize);
903 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
904 obd->u.cli.cl_default_mds_cookiesize);
906 ptlrpc_request_set_replen(req);
908 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
909 rc = ptlrpc_queue_wait(req);
910 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
912 if (req->rq_repmsg == NULL) {
913 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
916 rc = req->rq_status ?: -EIO;
917 } else if (rc == 0 || rc == -EAGAIN) {
918 struct mdt_body *body;
920 rc = lustre_msg_get_status(req->rq_repmsg);
921 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
922 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
927 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
930 } else if (rc == -ESTALE) {
932 * it can be allowed error after 3633 if open was committed and
933 * server failed before close was sent. Let's check if mod
934 * exists and return no error in that case
937 DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
938 LASSERT(mod->mod_open_req != NULL);
939 if (mod->mod_open_req->rq_committed)
946 mod->mod_close_req = NULL;
947 /* Since now, mod is accessed through open_req only,
948 * thus close req does not keep a reference on mod anymore. */
952 mdc_close_handle_reply(req, op_data, rc);
953 RETURN(rc < 0 ? rc : saved_rc);
956 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
957 struct md_open_data *mod)
959 struct obd_device *obd = class_exp2obd(exp);
960 struct ptlrpc_request *req;
964 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
965 &RQF_MDS_DONE_WRITING);
969 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
970 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
972 ptlrpc_request_free(req);
977 LASSERTF(mod->mod_open_req != NULL &&
978 mod->mod_open_req->rq_type != LI_POISON,
979 "POISONED setattr %p!\n", mod->mod_open_req);
981 mod->mod_close_req = req;
982 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
983 /* We no longer want to preserve this setattr for replay even
984 * though the open was committed. b=3632, b=3633 */
985 spin_lock(&mod->mod_open_req->rq_lock);
986 mod->mod_open_req->rq_replay = 0;
987 spin_unlock(&mod->mod_open_req->rq_lock);
990 mdc_close_pack(req, op_data);
991 ptlrpc_request_set_replen(req);
993 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
994 rc = ptlrpc_queue_wait(req);
995 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
999 * it can be allowed error after 3633 if open or setattr were
1000 * committed and server failed before close was sent.
1001 * Let's check if mod exists and return no error in that case
1004 LASSERT(mod->mod_open_req != NULL);
1005 if (mod->mod_open_req->rq_committed)
1012 mod->mod_close_req = NULL;
1013 LASSERT(mod->mod_open_req != NULL);
1016 /* Since now, mod is accessed through setattr req only,
1017 * thus DW req does not keep a reference on mod anymore. */
1021 mdc_close_handle_reply(req, op_data, rc);
1022 ptlrpc_req_finished(req);
1026 #ifdef HAVE_SPLIT_SUPPORT
1027 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
1028 const struct page *page, int offset)
1030 struct ptlrpc_request *req;
1031 struct ptlrpc_bulk_desc *desc;
1035 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
1039 /* FIXME: capa doesn't support split yet */
1040 mdc_set_capa_size(req, &RMF_CAPA1, NULL);
1042 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
1044 ptlrpc_request_free(req);
1048 req->rq_request_portal = MDS_READPAGE_PORTAL;
1049 ptlrpc_at_set_req_timeout(req);
1051 desc = ptlrpc_prep_bulk_imp(req, 1, 1,BULK_GET_SOURCE, MDS_BULK_PORTAL);
1053 GOTO(out, rc = -ENOMEM);
1055 /* NB req now owns desc and will free it when it gets freed. */
1056 ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
1057 mdc_readdir_pack(req, 0, offset, fid, NULL);
1059 ptlrpc_request_set_replen(req);
1060 rc = ptlrpc_queue_wait(req);
1064 rc = sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk);
1066 ptlrpc_req_finished(req);
1069 EXPORT_SYMBOL(mdc_sendpage);
1072 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
1073 __u64 offset, struct obd_capa *oc,
1074 struct page **pages, int npages,
1075 struct ptlrpc_request **request)
1077 struct ptlrpc_request *req;
1078 struct ptlrpc_bulk_desc *desc;
1080 wait_queue_head_t waitq;
1082 struct l_wait_info lwi;
1087 init_waitqueue_head(&waitq);
1090 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1094 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1096 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1098 ptlrpc_request_free(req);
1102 req->rq_request_portal = MDS_READPAGE_PORTAL;
1103 ptlrpc_at_set_req_timeout(req);
1105 desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK,
1108 ptlrpc_request_free(req);
1112 /* NB req now owns desc and will free it when it gets freed */
1113 for (i = 0; i < npages; i++)
1114 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
1116 mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE * npages, fid, oc);
1118 ptlrpc_request_set_replen(req);
1119 rc = ptlrpc_queue_wait(req);
1121 ptlrpc_req_finished(req);
1122 if (rc != -ETIMEDOUT)
1126 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1127 CERROR("%s: too many resend retries: rc = %d\n",
1128 exp->exp_obd->obd_name, -EIO);
1131 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
1133 l_wait_event(waitq, 0, &lwi);
1138 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1139 req->rq_bulk->bd_nob_transferred);
1141 ptlrpc_req_finished(req);
1145 if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1146 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1147 exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1148 PAGE_CACHE_SIZE * npages);
1149 ptlrpc_req_finished(req);
1158 static void mdc_release_page(struct page *page, int remove)
1162 if (likely(page->mapping != NULL))
1163 truncate_complete_page(page->mapping, page);
1166 page_cache_release(page);
1169 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1170 __u64 *start, __u64 *end, int hash64)
1173 * Complement of hash is used as an index so that
1174 * radix_tree_gang_lookup() can be used to find a page with starting
1175 * hash _smaller_ than one we are looking for.
1177 unsigned long offset = hash_x_index(*hash, hash64);
1181 spin_lock_irq(&mapping->tree_lock);
1182 found = radix_tree_gang_lookup(&mapping->page_tree,
1183 (void **)&page, offset, 1);
1184 if (found > 0 && !radix_tree_exceptional_entry(page)) {
1185 struct lu_dirpage *dp;
1187 page_cache_get(page);
1188 spin_unlock_irq(&mapping->tree_lock);
1190 * In contrast to find_lock_page() we are sure that directory
1191 * page cannot be truncated (while DLM lock is held) and,
1192 * hence, can avoid restart.
1194 * In fact, page cannot be locked here at all, because
1195 * mdc_read_page_remote does synchronous io.
1197 wait_on_page_locked(page);
1198 if (PageUptodate(page)) {
1200 if (BITS_PER_LONG == 32 && hash64) {
1201 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1202 *end = le64_to_cpu(dp->ldp_hash_end) >> 32;
1203 *hash = *hash >> 32;
1205 *start = le64_to_cpu(dp->ldp_hash_start);
1206 *end = le64_to_cpu(dp->ldp_hash_end);
1208 if (unlikely(*start == 1 && *hash == 0))
1211 LASSERTF(*start <= *hash, "start = "LPX64
1212 ",end = "LPX64",hash = "LPX64"\n",
1213 *start, *end, *hash);
1214 CDEBUG(D_VFSTRACE, "offset %lx ["LPX64" "LPX64"],"
1215 " hash "LPX64"\n", offset, *start, *end, *hash);
1218 mdc_release_page(page, 0);
1220 } else if (*end != *start && *hash == *end) {
1222 * upon hash collision, remove this page,
1223 * otherwise put page reference, and
1224 * ll_get_dir_page() will issue RPC to fetch
1228 mdc_release_page(page,
1229 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1233 page_cache_release(page);
1234 page = ERR_PTR(-EIO);
1237 spin_unlock_irq(&mapping->tree_lock);
1244 * Adjust a set of pages, each page containing an array of lu_dirpages,
1245 * so that each page can be used as a single logical lu_dirpage.
1247 * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1248 * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1249 * struct lu_dirent. It has size up to LU_PAGE_SIZE. The ldp_hash_end
1250 * value is used as a cookie to request the next lu_dirpage in a
1251 * directory listing that spans multiple pages (two in this example):
1254 * .|--------v------- -----.
1255 * |s|e|f|p|ent|ent| ... |ent|
1256 * '--|-------------- -----' Each CFS_PAGE contains a single
1257 * '------. lu_dirpage.
1258 * .---------v------- -----.
1259 * |s|e|f|p|ent| 0 | ... | 0 |
1260 * '----------------- -----'
1262 * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
1263 * larger than LU_PAGE_SIZE, a single host page may contain multiple
1264 * lu_dirpages. After reading the lu_dirpages from the MDS, the
1265 * ldp_hash_end of the first lu_dirpage refers to the one immediately
1266 * after it in the same CFS_PAGE (arrows simplified for brevity, but
1267 * in general e0==s1, e1==s2, etc.):
1269 * .-------------------- -----.
1270 * |s0|e0|f0|p|ent|ent| ... |ent|
1271 * |---v---------------- -----|
1272 * |s1|e1|f1|p|ent|ent| ... |ent|
1273 * |---v---------------- -----| Here, each CFS_PAGE contains
1274 * ... multiple lu_dirpages.
1275 * |---v---------------- -----|
1276 * |s'|e'|f'|p|ent|ent| ... |ent|
1277 * '---|---------------- -----'
1279 * .----------------------------.
1282 * This structure is transformed into a single logical lu_dirpage as follows:
1284 * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1285 * labeled 'next CFS_PAGE'.
1287 * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1288 * a hash collision with the next page exists.
1290 * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1291 * to the first entry of the next lu_dirpage.
1293 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
1294 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1298 for (i = 0; i < cfs_pgs; i++) {
1299 struct lu_dirpage *dp = kmap(pages[i]);
1300 struct lu_dirpage *first = dp;
1301 struct lu_dirent *end_dirent = NULL;
1302 struct lu_dirent *ent;
1303 __u64 hash_end = le64_to_cpu(dp->ldp_hash_end);
1304 __u32 flags = le32_to_cpu(dp->ldp_flags);
1306 while (--lu_pgs > 0) {
1307 ent = lu_dirent_start(dp);
1308 for (end_dirent = ent; ent != NULL;
1309 end_dirent = ent, ent = lu_dirent_next(ent));
1311 /* Advance dp to next lu_dirpage. */
1312 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1314 /* Check if we've reached the end of the CFS_PAGE. */
1315 if (!((unsigned long)dp & ~CFS_PAGE_MASK))
1318 /* Save the hash and flags of this lu_dirpage. */
1319 hash_end = le64_to_cpu(dp->ldp_hash_end);
1320 flags = le32_to_cpu(dp->ldp_flags);
1322 /* Check if lu_dirpage contains no entries. */
1323 if (end_dirent == NULL)
1326 /* Enlarge the end entry lde_reclen from 0 to
1327 * first entry of next lu_dirpage. */
1328 LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1329 end_dirent->lde_reclen =
1330 cpu_to_le16((char *)(dp->ldp_entries) -
1331 (char *)end_dirent);
1334 first->ldp_hash_end = hash_end;
1335 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1336 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1340 LASSERTF(lu_pgs == 0, "left = %d", lu_pgs);
1343 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1344 #endif /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
1346 /* parameters for readdir page */
1347 struct readpage_param {
1348 struct md_op_data *rp_mod;
1351 struct obd_export *rp_exp;
1352 struct md_callback *rp_cb;
1356 * Read pages from server.
1358 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1359 * a header lu_dirpage which describes the start/end hash, and whether this
1360 * page is empty (contains no dir entry) or hash collide with next page.
1361 * After client receives reply, several pages will be integrated into dir page
1362 * in CFS_PAGE_SIZE (if CFS_PAGE_SIZE greater than LU_PAGE_SIZE), and the
1363 * lu_dirpage for this integrated page will be adjusted.
1365 static int mdc_read_page_remote(void *data, struct page *page0)
1367 struct readpage_param *rp = data;
1368 struct page **page_pool;
1370 struct lu_dirpage *dp;
1371 int rd_pgs = 0; /* number of pages read actually */
1373 struct md_op_data *op_data = rp->rp_mod;
1374 struct ptlrpc_request *req;
1375 int max_pages = op_data->op_max_pages;
1376 struct inode *inode;
1382 LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
1383 if (op_data->op_mea1 != NULL) {
1384 __u32 index = op_data->op_stripe_offset;
1386 inode = op_data->op_mea1->lsm_md_oinfo[index].lmo_root;
1387 fid = &op_data->op_mea1->lsm_md_oinfo[index].lmo_fid;
1389 inode = op_data->op_data;
1390 fid = &op_data->op_fid1;
1392 LASSERT(inode != NULL);
1394 OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1395 if (page_pool != NULL) {
1396 page_pool[0] = page0;
1402 for (npages = 1; npages < max_pages; npages++) {
1403 page = page_cache_alloc_cold(inode->i_mapping);
1406 page_pool[npages] = page;
1409 rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, op_data->op_capa1,
1410 page_pool, npages, &req);
1414 rd_pgs = (req->rq_bulk->bd_nob_transferred +
1415 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1416 lu_pgs = req->rq_bulk->bd_nob_transferred >>
1418 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1420 CDEBUG(D_INODE, "read %d(%d)/%d pages\n", rd_pgs, lu_pgs,
1421 op_data->op_npages);
1423 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1425 SetPageUptodate(page0);
1429 ptlrpc_req_finished(req);
1430 CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1431 for (i = 1; i < npages; i++) {
1432 unsigned long offset;
1436 page = page_pool[i];
1438 if (rc < 0 || i >= rd_pgs) {
1439 page_cache_release(page);
1443 SetPageUptodate(page);
1446 hash = le64_to_cpu(dp->ldp_hash_start);
1449 offset = hash_x_index(hash, rp->rp_hash64);
1451 prefetchw(&page->flags);
1452 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1457 CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1458 " rc = %d\n", offset, ret);
1459 page_cache_release(page);
1462 if (page_pool != &page0)
1463 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1469 * Read dir page from cache first, if it can not find it, read it from
1470 * server and add into the cache.
1472 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1473 struct md_callback *cb_op, struct page **ppage)
1475 struct lookup_intent it = { .it_op = IT_READDIR };
1477 struct inode *dir = op_data->op_data;
1478 struct address_space *mapping;
1479 struct lu_dirpage *dp;
1482 struct lustre_handle lockh;
1483 struct ptlrpc_request *enq_req = NULL;
1484 struct readpage_param rp_param;
1491 LASSERT(dir != NULL);
1492 mapping = dir->i_mapping;
1494 rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1495 cb_op->md_blocking_ast, 0);
1496 if (enq_req != NULL)
1497 ptlrpc_req_finished(enq_req);
1500 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1501 exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1506 mdc_set_lock_data(exp, &it.d.lustre.it_lock_handle, dir, NULL);
1508 rp_param.rp_off = op_data->op_hash_offset;
1509 rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1510 page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1511 rp_param.rp_hash64);
1513 CERROR("%s: dir page locate: "DFID" at "LPU64": rc %ld\n",
1514 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1515 rp_param.rp_off, PTR_ERR(page));
1516 GOTO(out_unlock, rc = PTR_ERR(page));
1517 } else if (page != NULL) {
1519 * XXX nikita: not entirely correct handling of a corner case:
1520 * suppose hash chain of entries with hash value HASH crosses
1521 * border between pages P0 and P1. First both P0 and P1 are
1522 * cached, seekdir() is called for some entry from the P0 part
1523 * of the chain. Later P0 goes out of cache. telldir(HASH)
1524 * happens and finds P1, as it starts with matching hash
1525 * value. Remaining entries from P0 part of the chain are
1526 * skipped. (Is that really a bug?)
1528 * Possible solutions: 0. don't cache P1 is such case, handle
1529 * it as an "overflow" page. 1. invalidate all pages at
1530 * once. 2. use HASH|1 as an index for P1.
1532 GOTO(hash_collision, page);
1535 rp_param.rp_exp = exp;
1536 rp_param.rp_mod = op_data;
1537 page = read_cache_page(mapping,
1538 hash_x_index(rp_param.rp_off,
1539 rp_param.rp_hash64),
1540 mdc_read_page_remote, &rp_param);
1542 CERROR("%s: read cache page: "DFID" at "LPU64": rc %ld\n",
1543 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1544 rp_param.rp_off, PTR_ERR(page));
1545 GOTO(out_unlock, rc = PTR_ERR(page));
1548 wait_on_page_locked(page);
1550 if (!PageUptodate(page)) {
1551 CERROR("%s: page not updated: "DFID" at "LPU64": rc %d\n",
1552 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1553 rp_param.rp_off, -5);
1556 if (!PageChecked(page))
1557 SetPageChecked(page);
1558 if (PageError(page)) {
1559 CERROR("%s: page error: "DFID" at "LPU64": rc %d\n",
1560 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1561 rp_param.rp_off, -5);
1566 dp = page_address(page);
1567 if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1568 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1569 end = le64_to_cpu(dp->ldp_hash_end) >> 32;
1570 rp_param.rp_off = op_data->op_hash_offset >> 32;
1572 start = le64_to_cpu(dp->ldp_hash_start);
1573 end = le64_to_cpu(dp->ldp_hash_end);
1574 rp_param.rp_off = op_data->op_hash_offset;
1577 LASSERT(start == rp_param.rp_off);
1578 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1579 #if BITS_PER_LONG == 32
1580 CWARN("Real page-wide hash collision at ["LPU64" "LPU64"] with "
1581 "hash "LPU64"\n", le64_to_cpu(dp->ldp_hash_start),
1582 le64_to_cpu(dp->ldp_hash_end), op_data->op_hash_offset);
1586 * Fetch whole overflow chain...
1594 lockh.cookie = it.d.lustre.it_lock_handle;
1595 ldlm_lock_decref(&lockh, it.d.lustre.it_lock_mode);
1596 it.d.lustre.it_lock_handle = 0;
1600 mdc_release_page(page, 1);
1606 * Read one directory entry from the cache.
1608 int mdc_read_entry(struct obd_export *exp, struct md_op_data *op_data,
1609 struct md_callback *cb_op, struct lu_dirent **entp,
1610 struct page **ppage)
1612 struct page *page = NULL;
1613 struct lu_dirpage *dp;
1614 struct lu_dirent *ent;
1616 __u32 same_hash_count;
1617 __u64 hash_offset = op_data->op_hash_offset;
1620 CDEBUG(D_INFO, DFID " offset = "LPU64", flags %#x\n",
1621 PFID(&op_data->op_fid1), op_data->op_hash_offset,
1622 op_data->op_cli_flags);
1627 if (op_data->op_hash_offset == MDS_DIR_END_OFF)
1630 rc = mdc_read_page(exp, op_data, cb_op, &page);
1634 /* same_hash_count means how many entries with this
1635 * hash value has been read */
1636 same_hash_count = op_data->op_same_hash_offset + 1;
1637 dp = page_address(page);
1638 for (ent = lu_dirent_start(dp); ent != NULL;
1639 ent = lu_dirent_next(ent)) {
1640 /* Skip dummy entry */
1641 if (le16_to_cpu(ent->lde_namelen) == 0)
1644 if (le64_to_cpu(ent->lde_hash) <
1645 op_data->op_hash_offset)
1648 if (unlikely(le64_to_cpu(ent->lde_hash) ==
1649 op_data->op_hash_offset)) {
1650 /* If it is not for next entry, which usually from
1651 * ll_dir_entry_start, return this entry. */
1652 if (!(op_data->op_cli_flags & CLI_NEXT_ENTRY))
1655 /* Keep reading until all of entries being read are
1657 if (same_hash_count > 0) {
1665 /* If it can not find entry in current page, try next page. */
1667 if (le64_to_cpu(dp->ldp_hash_end) == MDS_DIR_END_OFF) {
1668 op_data->op_same_hash_offset = 0;
1669 mdc_release_page(page,
1670 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1674 op_data->op_hash_offset = le64_to_cpu(dp->ldp_hash_end);
1675 mdc_release_page(page,
1676 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1677 rc = mdc_read_page(exp, op_data, cb_op, &page);
1682 dp = page_address(page);
1683 ent = lu_dirent_start(dp);
1687 /* If the next hash is the same as the current hash, increase
1688 * the op_same_hash_offset to resolve the same hash conflict */
1689 if (ent != NULL && op_data->op_cli_flags & CLI_NEXT_ENTRY) {
1690 if (unlikely(le64_to_cpu(ent->lde_hash) == hash_offset))
1691 op_data->op_same_hash_offset++;
1693 op_data->op_same_hash_offset = 0;
1701 #else /* __KERNEL__ */
1704 *mdc_read_page_remote(struct obd_export *exp, const struct lmv_oinfo *lmo,
1705 const __u64 hash, struct obd_capa *oc)
1707 struct ptlrpc_request *req = NULL;
1711 OBD_PAGE_ALLOC(page, 0);
1713 return ERR_PTR(-ENOMEM);
1715 rc = mdc_getpage(exp, &lmo->lmo_fid, hash, oc, &page, 1, &req);
1717 ptlrpc_req_finished(req);
1720 OBD_PAGE_FREE(page);
1727 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1728 struct md_callback *cb_op,
1729 struct page **ppage)
1732 struct lmv_oinfo *lmo;
1735 /* No local cache for liblustre, always read entry remotely */
1736 lmo = &op_data->op_mea1->lsm_md_oinfo[op_data->op_stripe_offset];
1737 page = mdc_read_page_remote(exp, lmo, op_data->op_hash_offset,
1740 return PTR_ERR(page);
1747 int mdc_read_entry(struct obd_export *exp, struct md_op_data *op_data,
1748 struct md_callback *cb_op, struct lu_dirent **entp,
1749 struct page **ppage)
1751 struct page *page = NULL;
1752 struct lu_dirpage *dp;
1753 struct lu_dirent *ent;
1757 rc = mdc_read_page(exp, op_data, cb_op, &page);
1761 dp = page_address(page);
1762 if (le64_to_cpu(dp->ldp_hash_end) < op_data->op_hash_offset)
1763 GOTO(out, *entp = NULL);
1765 for (ent = lu_dirent_start(dp); ent != NULL;
1766 ent = lu_dirent_next(ent))
1767 if (le64_to_cpu(ent->lde_hash) >= op_data->op_hash_offset)
1772 OBD_PAGE_FREE(page);
1778 static int mdc_statfs(const struct lu_env *env,
1779 struct obd_export *exp, struct obd_statfs *osfs,
1780 __u64 max_age, __u32 flags)
1782 struct obd_device *obd = class_exp2obd(exp);
1783 struct ptlrpc_request *req;
1784 struct obd_statfs *msfs;
1785 struct obd_import *imp = NULL;
1790 * Since the request might also come from lprocfs, so we need
1791 * sync this with client_disconnect_export Bug15684
1793 down_read(&obd->u.cli.cl_sem);
1794 if (obd->u.cli.cl_import)
1795 imp = class_import_get(obd->u.cli.cl_import);
1796 up_read(&obd->u.cli.cl_sem);
1800 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1801 LUSTRE_MDS_VERSION, MDS_STATFS);
1803 GOTO(output, rc = -ENOMEM);
1805 ptlrpc_request_set_replen(req);
1807 if (flags & OBD_STATFS_NODELAY) {
1808 /* procfs requests not want stay in wait for avoid deadlock */
1809 req->rq_no_resend = 1;
1810 req->rq_no_delay = 1;
1813 rc = ptlrpc_queue_wait(req);
1815 /* check connection error first */
1816 if (imp->imp_connect_error)
1817 rc = imp->imp_connect_error;
1821 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1823 GOTO(out, rc = -EPROTO);
1828 ptlrpc_req_finished(req);
1830 class_import_put(imp);
1834 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1836 __u32 keylen, vallen;
1840 if (gf->gf_pathlen > PATH_MAX)
1841 RETURN(-ENAMETOOLONG);
1842 if (gf->gf_pathlen < 2)
1845 /* Key is KEY_FID2PATH + getinfo_fid2path description */
1846 keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1847 OBD_ALLOC(key, keylen);
1850 memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1851 memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1853 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1854 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1856 if (!fid_is_sane(&gf->gf_fid))
1857 GOTO(out, rc = -EINVAL);
1859 /* Val is struct getinfo_fid2path result plus path */
1860 vallen = sizeof(*gf) + gf->gf_pathlen;
1862 rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1863 if (rc != 0 && rc != -EREMOTE)
1866 if (vallen <= sizeof(*gf))
1867 GOTO(out, rc = -EPROTO);
1868 else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1869 GOTO(out, rc = -EOVERFLOW);
1871 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1872 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1875 OBD_FREE(key, keylen);
1879 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1880 struct hsm_progress_kernel *hpk)
1882 struct obd_import *imp = class_exp2cliimp(exp);
1883 struct hsm_progress_kernel *req_hpk;
1884 struct ptlrpc_request *req;
1888 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1889 LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1891 GOTO(out, rc = -ENOMEM);
1893 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1895 /* Copy hsm_progress struct */
1896 req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1897 if (req_hpk == NULL)
1898 GOTO(out, rc = -EPROTO);
1901 req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1903 ptlrpc_request_set_replen(req);
1905 rc = mdc_queue_wait(req);
1908 ptlrpc_req_finished(req);
1912 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1914 __u32 *archive_mask;
1915 struct ptlrpc_request *req;
1919 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1921 MDS_HSM_CT_REGISTER);
1923 GOTO(out, rc = -ENOMEM);
1925 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1927 /* Copy hsm_progress struct */
1928 archive_mask = req_capsule_client_get(&req->rq_pill,
1929 &RMF_MDS_HSM_ARCHIVE);
1930 if (archive_mask == NULL)
1931 GOTO(out, rc = -EPROTO);
1933 *archive_mask = archives;
1935 ptlrpc_request_set_replen(req);
1937 rc = mdc_queue_wait(req);
1940 ptlrpc_req_finished(req);
1944 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1945 struct md_op_data *op_data)
1947 struct hsm_current_action *hca = op_data->op_data;
1948 struct hsm_current_action *req_hca;
1949 struct ptlrpc_request *req;
1953 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1954 &RQF_MDS_HSM_ACTION);
1958 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1960 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1962 ptlrpc_request_free(req);
1966 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1967 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1969 ptlrpc_request_set_replen(req);
1971 rc = mdc_queue_wait(req);
1975 req_hca = req_capsule_server_get(&req->rq_pill,
1976 &RMF_MDS_HSM_CURRENT_ACTION);
1977 if (req_hca == NULL)
1978 GOTO(out, rc = -EPROTO);
1984 ptlrpc_req_finished(req);
1988 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1990 struct ptlrpc_request *req;
1994 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1996 MDS_HSM_CT_UNREGISTER);
1998 GOTO(out, rc = -ENOMEM);
2000 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
2002 ptlrpc_request_set_replen(req);
2004 rc = mdc_queue_wait(req);
2007 ptlrpc_req_finished(req);
2011 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
2012 struct md_op_data *op_data)
2014 struct hsm_user_state *hus = op_data->op_data;
2015 struct hsm_user_state *req_hus;
2016 struct ptlrpc_request *req;
2020 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2021 &RQF_MDS_HSM_STATE_GET);
2025 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2027 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
2029 ptlrpc_request_free(req);
2033 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
2034 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
2036 ptlrpc_request_set_replen(req);
2038 rc = mdc_queue_wait(req);
2042 req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
2043 if (req_hus == NULL)
2044 GOTO(out, rc = -EPROTO);
2050 ptlrpc_req_finished(req);
2054 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
2055 struct md_op_data *op_data)
2057 struct hsm_state_set *hss = op_data->op_data;
2058 struct hsm_state_set *req_hss;
2059 struct ptlrpc_request *req;
2063 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2064 &RQF_MDS_HSM_STATE_SET);
2068 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2070 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
2072 ptlrpc_request_free(req);
2076 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
2077 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
2080 req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
2081 if (req_hss == NULL)
2082 GOTO(out, rc = -EPROTO);
2085 ptlrpc_request_set_replen(req);
2087 rc = mdc_queue_wait(req);
2092 ptlrpc_req_finished(req);
2096 static int mdc_ioc_hsm_request(struct obd_export *exp,
2097 struct hsm_user_request *hur)
2099 struct obd_import *imp = class_exp2cliimp(exp);
2100 struct ptlrpc_request *req;
2101 struct hsm_request *req_hr;
2102 struct hsm_user_item *req_hui;
2107 req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
2109 GOTO(out, rc = -ENOMEM);
2111 req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
2112 hur->hur_request.hr_itemcount
2113 * sizeof(struct hsm_user_item));
2114 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
2115 hur->hur_request.hr_data_len);
2117 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
2119 ptlrpc_request_free(req);
2123 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
2125 /* Copy hsm_request struct */
2126 req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
2128 GOTO(out, rc = -EPROTO);
2129 *req_hr = hur->hur_request;
2131 /* Copy hsm_user_item structs */
2132 req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
2133 if (req_hui == NULL)
2134 GOTO(out, rc = -EPROTO);
2135 memcpy(req_hui, hur->hur_user_item,
2136 hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
2138 /* Copy opaque field */
2139 req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
2140 if (req_opaque == NULL)
2141 GOTO(out, rc = -EPROTO);
2142 memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
2144 ptlrpc_request_set_replen(req);
2146 rc = mdc_queue_wait(req);
2150 ptlrpc_req_finished(req);
2154 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
2156 struct kuc_hdr *lh = (struct kuc_hdr *)buf;
2158 LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
2160 lh->kuc_magic = KUC_MAGIC;
2161 lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
2162 lh->kuc_flags = flags;
2163 lh->kuc_msgtype = CL_RECORD;
2164 lh->kuc_msglen = len;
2168 #define D_CHANGELOG 0
2170 struct changelog_show {
2175 struct obd_device *cs_obd;
2178 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
2179 struct llog_rec_hdr *hdr, void *data)
2181 struct changelog_show *cs = data;
2182 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
2187 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
2189 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
2190 cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
2191 rec->cr.cr_type, rc);
2195 if (rec->cr.cr_index < cs->cs_startrec) {
2196 /* Skip entries earlier than what we are interested in */
2197 CDEBUG(D_CHANGELOG, "rec="LPU64" start="LPU64"\n",
2198 rec->cr.cr_index, cs->cs_startrec);
2202 CDEBUG(D_CHANGELOG, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID
2203 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
2204 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
2205 rec->cr.cr_flags & CLF_FLAGMASK,
2206 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
2207 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
2209 len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
2211 /* Set up the message */
2212 lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
2213 memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
2215 rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
2216 CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len,rc);
2221 static int mdc_changelog_send_thread(void *csdata)
2223 struct changelog_show *cs = csdata;
2224 struct llog_ctxt *ctxt = NULL;
2225 struct llog_handle *llh = NULL;
2226 struct kuc_hdr *kuch;
2229 CDEBUG(D_CHANGELOG, "changelog to fp=%p start "LPU64"\n",
2230 cs->cs_fp, cs->cs_startrec);
2232 OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
2233 if (cs->cs_buf == NULL)
2234 GOTO(out, rc = -ENOMEM);
2236 /* Set up the remote catalog handle */
2237 ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
2239 GOTO(out, rc = -ENOENT);
2240 rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
2243 CERROR("%s: fail to open changelog catalog: rc = %d\n",
2244 cs->cs_obd->obd_name, rc);
2247 rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
2249 CERROR("llog_init_handle failed %d\n", rc);
2253 rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
2255 /* Send EOF no matter what our result */
2256 if ((kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch),
2258 kuch->kuc_msgtype = CL_EOF;
2259 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
2265 llog_cat_close(NULL, llh);
2267 llog_ctxt_put(ctxt);
2269 OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
2274 static int mdc_ioc_changelog_send(struct obd_device *obd,
2275 struct ioc_changelog *icc)
2277 struct changelog_show *cs;
2278 struct task_struct *task;
2281 /* Freed in mdc_changelog_send_thread */
2287 cs->cs_startrec = icc->icc_recno;
2288 /* matching fput in mdc_changelog_send_thread */
2289 cs->cs_fp = fget(icc->icc_id);
2290 cs->cs_flags = icc->icc_flags;
2293 * New thread because we should return to user app before
2294 * writing into our pipe
2296 task = kthread_run(mdc_changelog_send_thread, cs,
2297 "mdc_clg_send_thread");
2300 CERROR("%s: cannot start changelog thread: rc = %d\n",
2305 CDEBUG(D_CHANGELOG, "%s: started changelog thread\n",
2312 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2313 struct lustre_kernelcomm *lk);
2315 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
2316 struct obd_quotactl *oqctl)
2318 struct client_obd *cli = &exp->exp_obd->u.cli;
2319 struct ptlrpc_request *req;
2320 struct obd_quotactl *body;
2324 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2325 &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
2330 body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2333 ptlrpc_request_set_replen(req);
2335 /* the next poll will find -ENODATA, that means quotacheck is
2337 cli->cl_qchk_stat = -ENODATA;
2338 rc = ptlrpc_queue_wait(req);
2340 cli->cl_qchk_stat = rc;
2341 ptlrpc_req_finished(req);
2345 static int mdc_quota_poll_check(struct obd_export *exp,
2346 struct if_quotacheck *qchk)
2348 struct client_obd *cli = &exp->exp_obd->u.cli;
2352 qchk->obd_uuid = cli->cl_target_uuid;
2353 memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
2355 rc = cli->cl_qchk_stat;
2356 /* the client is not the previous one */
2357 if (rc == CL_NOT_QUOTACHECKED)
2362 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
2363 struct obd_quotactl *oqctl)
2365 struct ptlrpc_request *req;
2366 struct obd_quotactl *oqc;
2370 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2371 &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
2376 oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2379 ptlrpc_request_set_replen(req);
2380 ptlrpc_at_set_req_timeout(req);
2381 req->rq_no_resend = 1;
2383 rc = ptlrpc_queue_wait(req);
2385 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
2387 if (req->rq_repmsg &&
2388 (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2391 CERROR ("Can't unpack obd_quotactl\n");
2394 ptlrpc_req_finished(req);
2399 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2400 struct md_op_data *op_data)
2402 struct list_head cancels = LIST_HEAD_INIT(cancels);
2403 struct ptlrpc_request *req;
2405 struct mdc_swap_layouts *msl, *payload;
2408 msl = op_data->op_data;
2410 /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2411 * first thing it will do is to cancel the 2 layout
2412 * locks hold by this client.
2413 * So the client must cancel its layout locks on the 2 fids
2414 * with the request RPC to avoid extra RPC round trips
2416 count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2417 LCK_EX, MDS_INODELOCK_LAYOUT |
2418 MDS_INODELOCK_XATTR);
2419 count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2420 LCK_EX, MDS_INODELOCK_LAYOUT |
2421 MDS_INODELOCK_XATTR);
2423 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2424 &RQF_MDS_SWAP_LAYOUTS);
2426 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2430 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2431 mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
2433 rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2435 ptlrpc_request_free(req);
2439 mdc_swap_layouts_pack(req, op_data);
2441 payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2446 ptlrpc_request_set_replen(req);
2448 rc = ptlrpc_queue_wait(req);
2454 ptlrpc_req_finished(req);
2458 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2459 void *karg, void *uarg)
2461 struct obd_device *obd = exp->exp_obd;
2462 struct obd_ioctl_data *data = karg;
2463 struct obd_import *imp = obd->u.cli.cl_import;
2467 if (!try_module_get(THIS_MODULE)) {
2468 CERROR("Can't get module. Is it alive?");
2472 case OBD_IOC_CHANGELOG_SEND:
2473 rc = mdc_ioc_changelog_send(obd, karg);
2475 case OBD_IOC_CHANGELOG_CLEAR: {
2476 struct ioc_changelog *icc = karg;
2477 struct changelog_setinfo cs =
2478 {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
2479 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
2480 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
2484 case OBD_IOC_FID2PATH:
2485 rc = mdc_ioc_fid2path(exp, karg);
2487 case LL_IOC_HSM_CT_START:
2488 rc = mdc_ioc_hsm_ct_start(exp, karg);
2489 /* ignore if it was already registered on this MDS. */
2493 case LL_IOC_HSM_PROGRESS:
2494 rc = mdc_ioc_hsm_progress(exp, karg);
2496 case LL_IOC_HSM_STATE_GET:
2497 rc = mdc_ioc_hsm_state_get(exp, karg);
2499 case LL_IOC_HSM_STATE_SET:
2500 rc = mdc_ioc_hsm_state_set(exp, karg);
2502 case LL_IOC_HSM_ACTION:
2503 rc = mdc_ioc_hsm_current_action(exp, karg);
2505 case LL_IOC_HSM_REQUEST:
2506 rc = mdc_ioc_hsm_request(exp, karg);
2508 case OBD_IOC_CLIENT_RECOVER:
2509 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2513 case IOC_OSC_SET_ACTIVE:
2514 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2516 case OBD_IOC_POLL_QUOTACHECK:
2517 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
2519 case OBD_IOC_PING_TARGET:
2520 rc = ptlrpc_obd_ping(obd);
2523 * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2524 * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2525 * there'd be no LMV layer thus we might be called here. Eventually
2526 * this code should be removed.
2529 case IOC_OBD_STATFS: {
2530 struct obd_statfs stat_buf = {0};
2532 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2533 GOTO(out, rc = -ENODEV);
2536 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2537 min((int)data->ioc_plen2,
2538 (int)sizeof(struct obd_uuid))))
2539 GOTO(out, rc = -EFAULT);
2541 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2542 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2547 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2548 min((int) data->ioc_plen1,
2549 (int) sizeof(stat_buf))))
2550 GOTO(out, rc = -EFAULT);
2554 case OBD_IOC_QUOTACTL: {
2555 struct if_quotactl *qctl = karg;
2556 struct obd_quotactl *oqctl;
2558 OBD_ALLOC_PTR(oqctl);
2560 GOTO(out, rc = -ENOMEM);
2562 QCTL_COPY(oqctl, qctl);
2563 rc = obd_quotactl(exp, oqctl);
2565 QCTL_COPY(qctl, oqctl);
2566 qctl->qc_valid = QC_MDTIDX;
2567 qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2570 OBD_FREE_PTR(oqctl);
2573 case LL_IOC_GET_CONNECT_FLAGS:
2574 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2575 sizeof(*exp_connect_flags_ptr(exp))))
2576 GOTO(out, rc = -EFAULT);
2579 case LL_IOC_LOV_SWAP_LAYOUTS:
2580 rc = mdc_ioc_swap_layouts(exp, karg);
2583 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2584 GOTO(out, rc = -ENOTTY);
2587 module_put(THIS_MODULE);
2592 int mdc_get_info_rpc(struct obd_export *exp,
2593 obd_count keylen, void *key,
2594 int vallen, void *val)
2596 struct obd_import *imp = class_exp2cliimp(exp);
2597 struct ptlrpc_request *req;
2602 req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2606 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2607 RCL_CLIENT, keylen);
2608 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2609 RCL_CLIENT, sizeof(__u32));
2611 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2613 ptlrpc_request_free(req);
2617 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2618 memcpy(tmp, key, keylen);
2619 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2620 memcpy(tmp, &vallen, sizeof(__u32));
2622 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2623 RCL_SERVER, vallen);
2624 ptlrpc_request_set_replen(req);
2626 rc = ptlrpc_queue_wait(req);
2627 /* -EREMOTE means the get_info result is partial, and it needs to
2628 * continue on another MDT, see fid2path part in lmv_iocontrol */
2629 if (rc == 0 || rc == -EREMOTE) {
2630 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2631 memcpy(val, tmp, vallen);
2632 if (ptlrpc_rep_need_swab(req)) {
2633 if (KEY_IS(KEY_FID2PATH))
2634 lustre_swab_fid2path(val);
2637 ptlrpc_req_finished(req);
2642 static void lustre_swab_hai(struct hsm_action_item *h)
2644 __swab32s(&h->hai_len);
2645 __swab32s(&h->hai_action);
2646 lustre_swab_lu_fid(&h->hai_fid);
2647 lustre_swab_lu_fid(&h->hai_dfid);
2648 __swab64s(&h->hai_cookie);
2649 __swab64s(&h->hai_extent.offset);
2650 __swab64s(&h->hai_extent.length);
2651 __swab64s(&h->hai_gid);
2654 static void lustre_swab_hal(struct hsm_action_list *h)
2656 struct hsm_action_item *hai;
2659 __swab32s(&h->hal_version);
2660 __swab32s(&h->hal_count);
2661 __swab32s(&h->hal_archive_id);
2662 __swab64s(&h->hal_flags);
2664 for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2665 lustre_swab_hai(hai);
2668 static void lustre_swab_kuch(struct kuc_hdr *l)
2670 __swab16s(&l->kuc_magic);
2671 /* __u8 l->kuc_transport */
2672 __swab16s(&l->kuc_msgtype);
2673 __swab16s(&l->kuc_msglen);
2676 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2677 struct lustre_kernelcomm *lk)
2679 struct obd_import *imp = class_exp2cliimp(exp);
2680 __u32 archive = lk->lk_data;
2683 if (lk->lk_group != KUC_GRP_HSM) {
2684 CERROR("Bad copytool group %d\n", lk->lk_group);
2688 CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2689 lk->lk_uid, lk->lk_group, lk->lk_flags);
2691 if (lk->lk_flags & LK_FLG_STOP) {
2692 /* Unregister with the coordinator */
2693 rc = mdc_ioc_hsm_ct_unregister(imp);
2695 rc = mdc_ioc_hsm_ct_register(imp, archive);
2702 * Send a message to any listening copytools
2703 * @param val KUC message (kuc_hdr + hsm_action_list)
2704 * @param len total length of message
2706 static int mdc_hsm_copytool_send(int len, void *val)
2708 struct kuc_hdr *lh = (struct kuc_hdr *)val;
2709 struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
2713 if (len < sizeof(*lh) + sizeof(*hal)) {
2714 CERROR("Short HSM message %d < %d\n", len,
2715 (int) (sizeof(*lh) + sizeof(*hal)));
2718 if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2719 lustre_swab_kuch(lh);
2720 lustre_swab_hal(hal);
2721 } else if (lh->kuc_magic != KUC_MAGIC) {
2722 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2726 CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2728 lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2729 lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2731 /* Broadcast to HSM listeners */
2732 rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2738 * callback function passed to kuc for re-registering each HSM copytool
2739 * running on MDC, after MDT shutdown/recovery.
2740 * @param data copytool registration data
2741 * @param cb_arg callback argument (obd_import)
2743 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2745 struct kkuc_ct_data *kcd = data;
2746 struct obd_import *imp = (struct obd_import *)cb_arg;
2749 if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2752 if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2755 CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2756 imp->imp_obd->obd_name, kcd->kcd_archive);
2757 rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2759 /* ignore error if the copytool is already registered */
2760 return (rc == -EEXIST) ? 0 : rc;
2764 * Re-establish all kuc contexts with MDT
2765 * after MDT shutdown/recovery.
2767 static int mdc_kuc_reregister(struct obd_import *imp)
2769 /* re-register HSM agents */
2770 return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2774 int mdc_set_info_async(const struct lu_env *env,
2775 struct obd_export *exp,
2776 obd_count keylen, void *key,
2777 obd_count vallen, void *val,
2778 struct ptlrpc_request_set *set)
2780 struct obd_import *imp = class_exp2cliimp(exp);
2784 if (KEY_IS(KEY_READ_ONLY)) {
2785 if (vallen != sizeof(int))
2788 spin_lock(&imp->imp_lock);
2789 if (*((int *)val)) {
2790 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2791 imp->imp_connect_data.ocd_connect_flags |=
2794 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2795 imp->imp_connect_data.ocd_connect_flags &=
2796 ~OBD_CONNECT_RDONLY;
2798 spin_unlock(&imp->imp_lock);
2800 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2801 keylen, key, vallen, val, set);
2804 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2805 sptlrpc_conf_client_adapt(exp->exp_obd);
2808 if (KEY_IS(KEY_FLUSH_CTX)) {
2809 sptlrpc_import_flush_my_ctx(imp);
2812 if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2813 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2814 keylen, key, vallen, val, set);
2817 if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2818 rc = mdc_hsm_copytool_send(vallen, val);
2822 CERROR("Unknown key %s\n", (char *)key);
2826 int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2827 __u32 keylen, void *key, __u32 *vallen, void *val,
2828 struct lov_stripe_md *lsm)
2832 if (KEY_IS(KEY_MAX_EASIZE)) {
2833 int mdsize, *max_easize;
2835 if (*vallen != sizeof(int))
2837 mdsize = *(int *)val;
2838 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2839 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2841 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2843 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2844 int *default_easize;
2846 if (*vallen != sizeof(int))
2848 default_easize = val;
2849 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2851 } else if (KEY_IS(KEY_MAX_COOKIESIZE)) {
2852 int mdsize, *max_cookiesize;
2854 if (*vallen != sizeof(int))
2856 mdsize = *(int *)val;
2857 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_cookiesize)
2858 exp->exp_obd->u.cli.cl_max_mds_cookiesize = mdsize;
2859 max_cookiesize = val;
2860 *max_cookiesize = exp->exp_obd->u.cli.cl_max_mds_cookiesize;
2862 } else if (KEY_IS(KEY_DEFAULT_COOKIESIZE)) {
2863 int *default_cookiesize;
2865 if (*vallen != sizeof(int))
2867 default_cookiesize = val;
2868 *default_cookiesize =
2869 exp->exp_obd->u.cli.cl_default_mds_cookiesize;
2871 } else if (KEY_IS(KEY_CONN_DATA)) {
2872 struct obd_import *imp = class_exp2cliimp(exp);
2873 struct obd_connect_data *data = val;
2875 if (*vallen != sizeof(*data))
2878 *data = imp->imp_connect_data;
2880 } else if (KEY_IS(KEY_TGT_COUNT)) {
2885 rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2890 int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2891 struct obd_capa *oc, struct ptlrpc_request **request)
2893 struct ptlrpc_request *req;
2898 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2902 mdc_set_capa_size(req, &RMF_CAPA1, oc);
2904 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2906 ptlrpc_request_free(req);
2910 mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
2912 ptlrpc_request_set_replen(req);
2914 rc = ptlrpc_queue_wait(req);
2916 ptlrpc_req_finished(req);
2922 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2923 enum obd_import_event event)
2927 LASSERT(imp->imp_obd == obd);
2930 case IMP_EVENT_DISCON: {
2932 /* XXX Pass event up to OBDs stack. used only for FLD now */
2933 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2937 case IMP_EVENT_INACTIVE: {
2938 struct client_obd *cli = &obd->u.cli;
2940 * Flush current sequence to make client obtain new one
2941 * from server in case of disconnect/reconnect.
2943 if (cli->cl_seq != NULL)
2944 seq_client_flush(cli->cl_seq);
2946 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2949 case IMP_EVENT_INVALIDATE: {
2950 struct ldlm_namespace *ns = obd->obd_namespace;
2952 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2956 case IMP_EVENT_ACTIVE:
2957 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2958 /* redo the kuc registration after reconnecting */
2960 rc = mdc_kuc_reregister(imp);
2963 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2965 case IMP_EVENT_DEACTIVATE:
2966 case IMP_EVENT_ACTIVATE:
2969 CERROR("Unknown import event %x\n", event);
2975 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2976 struct lu_fid *fid, struct md_op_data *op_data)
2978 struct client_obd *cli = &exp->exp_obd->u.cli;
2979 struct lu_client_seq *seq = cli->cl_seq;
2981 RETURN(seq_client_alloc_fid(env, seq, fid));
2984 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
2985 struct client_obd *cli = &exp->exp_obd->u.cli;
2986 return &cli->cl_target_uuid;
2990 * Determine whether the lock can be canceled before replaying it during
2991 * recovery, non zero value will be return if the lock can be canceled,
2992 * or zero returned for not
2994 static int mdc_cancel_weight(struct ldlm_lock *lock)
2996 if (lock->l_resource->lr_type != LDLM_IBITS)
2999 /* FIXME: if we ever get into a situation where there are too many
3000 * opened files with open locks on a single node, then we really
3001 * should replay these open locks to reget it */
3002 if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
3008 static int mdc_resource_inode_free(struct ldlm_resource *res)
3010 if (res->lr_lvb_inode)
3011 res->lr_lvb_inode = NULL;
3016 struct ldlm_valblock_ops inode_lvbo = {
3017 .lvbo_free = mdc_resource_inode_free
3020 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
3022 struct client_obd *cli = &obd->u.cli;
3026 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3027 if (!cli->cl_rpc_lock)
3029 mdc_init_rpc_lock(cli->cl_rpc_lock);
3031 rc = ptlrpcd_addref();
3033 GOTO(err_rpc_lock, rc);
3035 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3036 if (!cli->cl_close_lock)
3037 GOTO(err_ptlrpcd_decref, rc = -ENOMEM);
3038 mdc_init_rpc_lock(cli->cl_close_lock);
3040 rc = client_obd_setup(obd, cfg);
3042 GOTO(err_close_lock, rc);
3044 obd->obd_vars = lprocfs_mdc_obd_vars;
3045 lprocfs_seq_obd_setup(obd);
3046 lprocfs_alloc_md_stats(obd, 0);
3048 sptlrpc_lprocfs_cliobd_attach(obd);
3049 ptlrpc_lprocfs_register_obd(obd);
3051 ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
3053 obd->obd_namespace->ns_lvbo = &inode_lvbo;
3055 rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
3058 CERROR("failed to setup llogging subsystems\n");
3064 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3068 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3072 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
3073 * us to make MDS RPCs with large enough reply buffers to hold a default
3074 * sized EA and cookie without having to calculate this (via a call into the
3075 * LOV + OSCs) each time we make an RPC. The maximum size is also tracked
3076 * but not used to avoid wastefully vmalloc()'ing large reply buffers when
3077 * a large number of stripes is possible. If a larger reply buffer is
3078 * required it will be reallocated in the ptlrpc layer due to overflow.
3080 static int mdc_init_ea_size(struct obd_export *exp, int easize,
3081 int def_easize, int cookiesize, int def_cookiesize)
3083 struct obd_device *obd = exp->exp_obd;
3084 struct client_obd *cli = &obd->u.cli;
3087 if (cli->cl_max_mds_easize < easize)
3088 cli->cl_max_mds_easize = easize;
3090 if (cli->cl_default_mds_easize < def_easize)
3091 cli->cl_default_mds_easize = def_easize;
3093 if (cli->cl_max_mds_cookiesize < cookiesize)
3094 cli->cl_max_mds_cookiesize = cookiesize;
3096 if (cli->cl_default_mds_cookiesize < def_cookiesize)
3097 cli->cl_default_mds_cookiesize = def_cookiesize;
3102 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
3108 case OBD_CLEANUP_EARLY:
3110 case OBD_CLEANUP_EXPORTS:
3111 /* Failsafe, ok if racy */
3112 if (obd->obd_type->typ_refcnt <= 1)
3113 libcfs_kkuc_group_rem(0, KUC_GRP_HSM, NULL);
3115 obd_cleanup_client_import(obd);
3116 ptlrpc_lprocfs_unregister_obd(obd);
3117 lprocfs_obd_cleanup(obd);
3118 lprocfs_free_md_stats(obd);
3120 rc = obd_llog_finish(obd, 0);
3122 CERROR("failed to cleanup llogging subsystems\n");
3128 static int mdc_cleanup(struct obd_device *obd)
3130 struct client_obd *cli = &obd->u.cli;
3132 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3133 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3137 return client_obd_cleanup(obd);
3141 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
3142 struct obd_device *tgt, int *index)
3144 struct llog_ctxt *ctxt;
3149 LASSERT(olg == &obd->obd_olg);
3151 rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt,
3156 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
3157 llog_initiator_connect(ctxt);
3158 llog_ctxt_put(ctxt);
3163 static int mdc_llog_finish(struct obd_device *obd, int count)
3165 struct llog_ctxt *ctxt;
3169 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
3171 llog_cleanup(NULL, ctxt);
3176 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
3178 struct lustre_cfg *lcfg = buf;
3179 int rc = class_process_proc_seq_param(PARAM_MDC, obd->obd_vars,
3181 return (rc > 0 ? 0: rc);
3185 /* get remote permission for current user on fid */
3186 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
3187 struct obd_capa *oc, __u32 suppgid,
3188 struct ptlrpc_request **request)
3190 struct ptlrpc_request *req;
3194 LASSERT(client_is_remote(exp));
3197 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
3201 mdc_set_capa_size(req, &RMF_CAPA1, oc);
3203 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
3205 ptlrpc_request_free(req);
3209 mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
3211 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
3212 sizeof(struct mdt_remote_perm));
3214 ptlrpc_request_set_replen(req);
3216 rc = ptlrpc_queue_wait(req);
3218 ptlrpc_req_finished(req);
3224 static int mdc_interpret_renew_capa(const struct lu_env *env,
3225 struct ptlrpc_request *req, void *args,
3228 struct mdc_renew_capa_args *ra = args;
3229 struct mdt_body *body = NULL;
3230 struct lustre_capa *capa;
3234 GOTO(out, capa = ERR_PTR(status));
3236 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3238 GOTO(out, capa = ERR_PTR(-EFAULT));
3240 if ((body->mbo_valid & OBD_MD_FLOSSCAPA) == 0)
3241 GOTO(out, capa = ERR_PTR(-ENOENT));
3243 capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
3245 GOTO(out, capa = ERR_PTR(-EFAULT));
3248 ra->ra_cb(ra->ra_oc, capa);
3252 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3255 struct ptlrpc_request *req;
3256 struct mdc_renew_capa_args *ra;
3259 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
3260 LUSTRE_MDS_VERSION, MDS_GETATTR);
3264 /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
3265 * capa to renew is oss capa.
3267 mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
3268 ptlrpc_request_set_replen(req);
3270 CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
3271 ra = ptlrpc_req_async_args(req);
3274 req->rq_interpret_reply = mdc_interpret_renew_capa;
3275 ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
3279 struct obd_ops mdc_obd_ops = {
3280 .o_owner = THIS_MODULE,
3281 .o_setup = mdc_setup,
3282 .o_precleanup = mdc_precleanup,
3283 .o_cleanup = mdc_cleanup,
3284 .o_add_conn = client_import_add_conn,
3285 .o_del_conn = client_import_del_conn,
3286 .o_connect = client_connect_import,
3287 .o_disconnect = client_disconnect_export,
3288 .o_iocontrol = mdc_iocontrol,
3289 .o_set_info_async = mdc_set_info_async,
3290 .o_statfs = mdc_statfs,
3291 .o_fid_init = client_fid_init,
3292 .o_fid_fini = client_fid_fini,
3293 .o_fid_alloc = mdc_fid_alloc,
3294 .o_import_event = mdc_import_event,
3295 .o_llog_init = mdc_llog_init,
3296 .o_llog_finish = mdc_llog_finish,
3297 .o_get_info = mdc_get_info,
3298 .o_process_config = mdc_process_config,
3299 .o_get_uuid = mdc_get_uuid,
3300 .o_quotactl = mdc_quotactl,
3301 .o_quotacheck = mdc_quotacheck
3304 struct md_ops mdc_md_ops = {
3305 .m_getstatus = mdc_getstatus,
3306 .m_null_inode = mdc_null_inode,
3307 .m_find_cbdata = mdc_find_cbdata,
3308 .m_close = mdc_close,
3309 .m_create = mdc_create,
3310 .m_done_writing = mdc_done_writing,
3311 .m_enqueue = mdc_enqueue,
3312 .m_getattr = mdc_getattr,
3313 .m_getattr_name = mdc_getattr_name,
3314 .m_intent_lock = mdc_intent_lock,
3316 .m_rename = mdc_rename,
3317 .m_setattr = mdc_setattr,
3318 .m_setxattr = mdc_setxattr,
3319 .m_getxattr = mdc_getxattr,
3320 .m_fsync = mdc_fsync,
3321 .m_read_entry = mdc_read_entry,
3322 .m_unlink = mdc_unlink,
3323 .m_cancel_unused = mdc_cancel_unused,
3324 .m_init_ea_size = mdc_init_ea_size,
3325 .m_set_lock_data = mdc_set_lock_data,
3326 .m_lock_match = mdc_lock_match,
3327 .m_get_lustre_md = mdc_get_lustre_md,
3328 .m_free_lustre_md = mdc_free_lustre_md,
3329 .m_set_open_replay_data = mdc_set_open_replay_data,
3330 .m_clear_open_replay_data = mdc_clear_open_replay_data,
3331 .m_renew_capa = mdc_renew_capa,
3332 .m_unpack_capa = mdc_unpack_capa,
3333 .m_get_remote_perm = mdc_get_remote_perm,
3334 .m_intent_getattr_async = mdc_intent_getattr_async,
3335 .m_revalidate_lock = mdc_revalidate_lock
3338 int __init mdc_init(void)
3340 return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
3341 #ifndef HAVE_ONLY_PROCFS_SEQ
3344 LUSTRE_MDC_NAME, NULL);
3348 static void /*__exit*/ mdc_exit(void)
3350 class_unregister_type(LUSTRE_MDC_NAME);
3353 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3354 MODULE_DESCRIPTION("Lustre Metadata Client");
3355 MODULE_LICENSE("GPL");
3357 module_init(mdc_init);
3358 module_exit(mdc_exit);