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 <obd_class.h>
51 #include <lustre_lmv.h>
52 #include <lustre_fid.h>
53 #include <lprocfs_status.h>
54 #include <lustre_param.h>
55 #include <lustre_log.h>
56 #include <cl_object.h>
59 #include "mdc_internal.h"
61 #define REQUEST_MINOR 244
63 struct mdc_renew_capa_args {
64 struct obd_capa *ra_oc;
65 renew_capa_cb_t ra_cb;
68 static int mdc_cleanup(struct obd_device *obd);
70 int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
71 const struct req_msg_field *field, struct obd_capa **oc)
73 struct lustre_capa *capa;
77 /* swabbed already in mdc_enqueue */
78 capa = req_capsule_server_get(&req->rq_pill, field);
82 c = alloc_capa(CAPA_SITE_CLIENT);
84 CDEBUG(D_INFO, "alloc capa failed!\n");
93 static inline int mdc_queue_wait(struct ptlrpc_request *req)
95 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
98 /* mdc_enter_request() ensures that this client has no more
99 * than cl_max_rpcs_in_flight RPCs simultaneously inf light
101 rc = mdc_enter_request(cli);
105 rc = ptlrpc_queue_wait(req);
106 mdc_exit_request(cli);
111 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
112 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
113 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
114 struct obd_capa **pc, int level, int msg_flags)
116 struct ptlrpc_request *req;
117 struct mdt_body *body;
121 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
122 LUSTRE_MDS_VERSION, MDS_GETSTATUS);
126 mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
127 lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
128 req->rq_send_state = level;
130 ptlrpc_request_set_replen(req);
132 rc = ptlrpc_queue_wait(req);
136 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
138 GOTO(out, rc = -EPROTO);
140 if (body->valid & OBD_MD_FLMDSCAPA) {
141 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
146 *rootfid = body->fid1;
148 "root fid="DFID", last_committed="LPU64"\n",
150 lustre_msg_get_last_committed(req->rq_repmsg));
153 ptlrpc_req_finished(req);
157 /* This should be mdc_get_info("rootfid") */
158 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
159 struct obd_capa **pc)
161 return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
166 * This function now is known to always saying that it will receive 4 buffers
167 * from server. Even for cases when acl_size and md_size is zero, RPC header
168 * will contain 4 fields and RPC itself will contain zero size fields. This is
169 * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
170 * and thus zero, it shrinks it, making zero size. The same story about
171 * md_size. And this is course of problem when client waits for smaller number
172 * of fields. This issue will be fixed later when client gets aware of RPC
175 static int mdc_getattr_common(struct obd_export *exp,
176 struct ptlrpc_request *req)
178 struct req_capsule *pill = &req->rq_pill;
179 struct mdt_body *body;
184 /* Request message already built. */
185 rc = ptlrpc_queue_wait(req);
189 /* sanity check for the reply */
190 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
194 CDEBUG(D_NET, "mode: %o\n", body->mode);
196 mdc_update_max_ea_from_body(exp, body);
197 if (body->eadatasize != 0) {
198 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
204 if (body->valid & OBD_MD_FLRMTPERM) {
205 struct mdt_remote_perm *perm;
207 LASSERT(client_is_remote(exp));
208 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
209 lustre_swab_mdt_remote_perm);
214 if (body->valid & OBD_MD_FLMDSCAPA) {
215 struct lustre_capa *capa;
216 capa = req_capsule_server_get(pill, &RMF_CAPA1);
224 int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
225 struct ptlrpc_request **request)
227 struct ptlrpc_request *req;
231 /* Single MDS without an LMV case */
232 if (op_data->op_flags & MF_GET_MDT_IDX) {
237 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
241 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
243 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
245 ptlrpc_request_free(req);
249 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
250 op_data->op_valid, op_data->op_mode, -1, 0);
252 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
254 if (op_data->op_valid & OBD_MD_FLRMTPERM) {
255 LASSERT(client_is_remote(exp));
256 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
257 sizeof(struct mdt_remote_perm));
259 ptlrpc_request_set_replen(req);
261 rc = mdc_getattr_common(exp, req);
263 ptlrpc_req_finished(req);
269 int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
270 struct ptlrpc_request **request)
272 struct ptlrpc_request *req;
277 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
278 &RQF_MDS_GETATTR_NAME);
282 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
283 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
284 op_data->op_namelen + 1);
286 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
288 ptlrpc_request_free(req);
292 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
293 op_data->op_valid, op_data->op_mode,
294 op_data->op_suppgids[0], 0);
296 if (op_data->op_name) {
297 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
298 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
299 op_data->op_namelen);
300 memcpy(name, op_data->op_name, op_data->op_namelen);
303 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
305 ptlrpc_request_set_replen(req);
307 rc = mdc_getattr_common(exp, req);
309 ptlrpc_req_finished(req);
315 static int mdc_is_subdir(struct obd_export *exp,
316 const struct lu_fid *pfid,
317 const struct lu_fid *cfid,
318 struct ptlrpc_request **request)
320 struct ptlrpc_request *req;
326 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
327 &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
332 mdc_is_subdir_pack(req, pfid, cfid, 0);
333 ptlrpc_request_set_replen(req);
335 rc = ptlrpc_queue_wait(req);
336 if (rc && rc != -EREMOTE)
337 ptlrpc_req_finished(req);
343 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
344 const struct lu_fid *fid,
345 struct obd_capa *oc, int opcode, obd_valid valid,
346 const char *xattr_name, const char *input,
347 int input_size, int output_size, int flags,
348 __u32 suppgid, struct ptlrpc_request **request)
350 struct ptlrpc_request *req;
351 int xattr_namelen = 0;
357 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
361 mdc_set_capa_size(req, &RMF_CAPA1, oc);
363 xattr_namelen = strlen(xattr_name) + 1;
364 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
369 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
373 /* Flush local XATTR locks to get rid of a possible cancel RPC */
374 if (opcode == MDS_REINT && fid_is_sane(fid) &&
375 exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
376 CFS_LIST_HEAD(cancels);
379 /* Without that packing would fail */
381 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
384 count = mdc_resource_get_unused(exp, fid,
386 MDS_INODELOCK_XATTR);
388 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
390 ptlrpc_request_free(req);
394 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
396 ptlrpc_request_free(req);
401 if (opcode == MDS_REINT) {
402 struct mdt_rec_setxattr *rec;
404 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
405 sizeof(struct mdt_rec_reint));
406 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
407 rec->sx_opcode = REINT_SETXATTR;
408 rec->sx_fsuid = from_kuid(&init_user_ns, current_fsuid());
409 rec->sx_fsgid = from_kgid(&init_user_ns, current_fsgid());
410 rec->sx_cap = cfs_curproc_cap_pack();
411 rec->sx_suppgid1 = suppgid;
412 rec->sx_suppgid2 = -1;
414 rec->sx_valid = valid | OBD_MD_FLCTIME;
415 rec->sx_time = cfs_time_current_sec();
416 rec->sx_size = output_size;
417 rec->sx_flags = flags;
419 mdc_pack_capa(req, &RMF_CAPA1, oc);
421 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
425 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
426 memcpy(tmp, xattr_name, xattr_namelen);
429 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
430 memcpy(tmp, input, input_size);
433 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
434 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
435 RCL_SERVER, output_size);
436 ptlrpc_request_set_replen(req);
439 if (opcode == MDS_REINT)
440 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
442 rc = ptlrpc_queue_wait(req);
444 if (opcode == MDS_REINT)
445 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
448 ptlrpc_req_finished(req);
454 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
455 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
456 const char *input, int input_size, int output_size,
457 int flags, __u32 suppgid, struct ptlrpc_request **request)
459 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
460 fid, oc, MDS_REINT, valid, xattr_name,
461 input, input_size, output_size, flags,
465 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
466 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
467 const char *input, int input_size, int output_size,
468 int flags, struct ptlrpc_request **request)
470 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
471 fid, oc, MDS_GETXATTR, valid, xattr_name,
472 input, input_size, output_size, flags,
476 #ifdef CONFIG_FS_POSIX_ACL
477 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
479 struct req_capsule *pill = &req->rq_pill;
480 struct mdt_body *body = md->body;
481 struct posix_acl *acl;
489 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
494 acl = posix_acl_from_xattr(&init_user_ns, buf, body->aclsize);
497 CERROR("convert xattr to acl: %d\n", rc);
501 rc = posix_acl_valid(acl);
503 CERROR("validate acl: %d\n", rc);
504 posix_acl_release(acl);
512 #define mdc_unpack_acl(req, md) 0
515 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
516 struct obd_export *dt_exp, struct obd_export *md_exp,
517 struct lustre_md *md)
519 struct req_capsule *pill = &req->rq_pill;
524 memset(md, 0, sizeof(*md));
526 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
527 LASSERT(md->body != NULL);
529 if (md->body->valid & OBD_MD_FLEASIZE) {
531 struct lov_mds_md *lmm;
533 if (!S_ISREG(md->body->mode)) {
534 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
535 "regular file, but is not\n");
536 GOTO(out, rc = -EPROTO);
539 if (md->body->eadatasize == 0) {
540 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
541 "but eadatasize 0\n");
542 GOTO(out, rc = -EPROTO);
544 lmmsize = md->body->eadatasize;
545 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
547 GOTO(out, rc = -EPROTO);
549 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
553 if (rc < sizeof(*md->lsm)) {
554 CDEBUG(D_INFO, "lsm size too small: "
555 "rc < sizeof (*md->lsm) (%d < %d)\n",
556 rc, (int)sizeof(*md->lsm));
557 GOTO(out, rc = -EPROTO);
560 } else if (md->body->valid & OBD_MD_FLDIREA) {
562 struct lov_mds_md *lmv;
564 if(!S_ISDIR(md->body->mode)) {
565 CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
566 "directory, but is not\n");
567 GOTO(out, rc = -EPROTO);
570 if (md->body->eadatasize == 0) {
571 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
572 "but eadatasize 0\n");
575 if (md->body->valid & OBD_MD_MEA) {
576 lmvsize = md->body->eadatasize;
577 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
580 GOTO(out, rc = -EPROTO);
582 rc = obd_unpackmd(md_exp, (void *)&md->lmv, lmv,
587 if (rc < sizeof(*md->lmv)) {
588 CDEBUG(D_INFO, "size too small: "
589 "rc < sizeof(*md->lmv) (%d < %d)\n",
590 rc, (int)sizeof(*md->lmv));
591 GOTO(out, rc = -EPROTO);
597 if (md->body->valid & OBD_MD_FLRMTPERM) {
598 /* remote permission */
599 LASSERT(client_is_remote(exp));
600 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
601 lustre_swab_mdt_remote_perm);
602 if (!md->remote_perm)
603 GOTO(out, rc = -EPROTO);
605 else if (md->body->valid & OBD_MD_FLACL) {
606 /* for ACL, it's possible that FLACL is set but aclsize is zero.
607 * only when aclsize != 0 there's an actual segment for ACL
610 if (md->body->aclsize) {
611 rc = mdc_unpack_acl(req, md);
614 #ifdef CONFIG_FS_POSIX_ACL
616 md->posix_acl = NULL;
620 if (md->body->valid & OBD_MD_FLMDSCAPA) {
621 struct obd_capa *oc = NULL;
623 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
629 if (md->body->valid & OBD_MD_FLOSSCAPA) {
630 struct obd_capa *oc = NULL;
632 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
642 capa_put(md->oss_capa);
646 capa_put(md->mds_capa);
649 #ifdef CONFIG_FS_POSIX_ACL
650 posix_acl_release(md->posix_acl);
653 obd_free_memmd(dt_exp, &md->lsm);
658 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
665 * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
668 void mdc_replay_open(struct ptlrpc_request *req)
670 struct md_open_data *mod = req->rq_cb_data;
671 struct ptlrpc_request *close_req;
672 struct obd_client_handle *och;
673 struct lustre_handle old;
674 struct mdt_body *body;
678 DEBUG_REQ(D_ERROR, req,
679 "Can't properly replay without open data.");
684 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
685 LASSERT(body != NULL);
689 struct lustre_handle *file_fh;
691 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
693 file_fh = &och->och_fh;
694 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
695 file_fh->cookie, body->handle.cookie);
697 *file_fh = body->handle;
699 close_req = mod->mod_close_req;
700 if (close_req != NULL) {
701 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
702 struct mdt_ioepoch *epoch;
704 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
705 epoch = req_capsule_client_get(&close_req->rq_pill,
710 LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
711 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
712 epoch->handle = body->handle;
717 void mdc_commit_open(struct ptlrpc_request *req)
719 struct md_open_data *mod = req->rq_cb_data;
724 * No need to touch md_open_data::mod_och, it holds a reference on
725 * \var mod and will zero references to each other, \var mod will be
726 * freed after that when md_open_data::mod_och will put the reference.
730 * Do not let open request to disappear as it still may be needed
731 * for close rpc to happen (it may happen on evict only, otherwise
732 * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
733 * called), just mark this rpc as committed to distinguish these 2
734 * cases, see mdc_close() for details. The open request reference will
735 * be put along with freeing \var mod.
737 ptlrpc_request_addref(req);
738 spin_lock(&req->rq_lock);
739 req->rq_committed = 1;
740 spin_unlock(&req->rq_lock);
741 req->rq_cb_data = NULL;
745 int mdc_set_open_replay_data(struct obd_export *exp,
746 struct obd_client_handle *och,
747 struct lookup_intent *it)
749 struct md_open_data *mod;
750 struct mdt_rec_create *rec;
751 struct mdt_body *body;
752 struct ptlrpc_request *open_req = it->d.lustre.it_data;
753 struct obd_import *imp = open_req->rq_import;
756 if (!open_req->rq_replay)
759 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
760 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
761 LASSERT(rec != NULL);
762 /* Incoming message in my byte order (it's been swabbed). */
763 /* Outgoing messages always in my byte order. */
764 LASSERT(body != NULL);
766 /* Only if the import is replayable, we set replay_open data */
767 if (och && imp->imp_replayable) {
768 mod = obd_mod_alloc();
770 DEBUG_REQ(D_ERROR, open_req,
771 "Can't allocate md_open_data");
776 * Take a reference on \var mod, to be freed on mdc_close().
777 * It protects \var mod from being freed on eviction (commit
778 * callback is called despite rq_replay flag).
779 * Another reference for \var och.
784 spin_lock(&open_req->rq_lock);
787 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
788 it_disposition(it, DISP_OPEN_STRIPE);
789 mod->mod_open_req = open_req;
790 open_req->rq_cb_data = mod;
791 open_req->rq_commit_cb = mdc_commit_open;
792 spin_unlock(&open_req->rq_lock);
795 rec->cr_fid2 = body->fid1;
796 rec->cr_ioepoch = body->ioepoch;
797 rec->cr_old_handle.cookie = body->handle.cookie;
798 open_req->rq_replay_cb = mdc_replay_open;
799 if (!fid_is_sane(&body->fid1)) {
800 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
805 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
809 static void mdc_free_open(struct md_open_data *mod)
813 if (mod->mod_is_create == 0 &&
814 imp_connect_disp_stripe(mod->mod_open_req->rq_import))
817 LASSERT(mod->mod_open_req->rq_replay == 0);
819 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
821 ptlrpc_request_committed(mod->mod_open_req, committed);
822 if (mod->mod_close_req)
823 ptlrpc_request_committed(mod->mod_close_req, committed);
826 int mdc_clear_open_replay_data(struct obd_export *exp,
827 struct obd_client_handle *och)
829 struct md_open_data *mod = och->och_mod;
833 * It is possible to not have \var mod in a case of eviction between
834 * lookup and ll_file_open().
839 LASSERT(mod != LP_POISON);
840 LASSERT(mod->mod_open_req != NULL);
850 /* Prepares the request for the replay by the given reply */
851 static void mdc_close_handle_reply(struct ptlrpc_request *req,
852 struct md_op_data *op_data, int rc) {
853 struct mdt_body *repbody;
854 struct mdt_ioepoch *epoch;
856 if (req && rc == -EAGAIN) {
857 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
858 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
860 epoch->flags |= MF_SOM_AU;
861 if (repbody->valid & OBD_MD_FLGETATTRLOCK)
862 op_data->op_flags |= MF_GETATTR_LOCK;
866 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
867 struct md_open_data *mod, struct ptlrpc_request **request)
869 struct obd_device *obd = class_exp2obd(exp);
870 struct ptlrpc_request *req;
871 struct req_format *req_fmt;
876 req_fmt = &RQF_MDS_CLOSE;
877 if (op_data->op_bias & MDS_HSM_RELEASE) {
878 req_fmt = &RQF_MDS_RELEASE_CLOSE;
880 /* allocate a FID for volatile file */
881 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
883 CERROR("%s: "DFID" failed to allocate FID: %d\n",
884 obd->obd_name, PFID(&op_data->op_fid1), rc);
885 /* save the errcode and proceed to close */
891 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
895 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
897 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
899 ptlrpc_request_free(req);
903 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
904 * portal whose threads are not taking any DLM locks and are therefore
905 * always progressing */
906 req->rq_request_portal = MDS_READPAGE_PORTAL;
907 ptlrpc_at_set_req_timeout(req);
909 /* Ensure that this close's handle is fixed up during replay. */
910 if (likely(mod != NULL)) {
911 LASSERTF(mod->mod_open_req != NULL &&
912 mod->mod_open_req->rq_type != LI_POISON,
913 "POISONED open %p!\n", mod->mod_open_req);
915 mod->mod_close_req = req;
917 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
918 /* We no longer want to preserve this open for replay even
919 * though the open was committed. b=3632, b=3633 */
920 spin_lock(&mod->mod_open_req->rq_lock);
921 mod->mod_open_req->rq_replay = 0;
922 spin_unlock(&mod->mod_open_req->rq_lock);
924 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
927 mdc_close_pack(req, op_data);
929 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
930 obd->u.cli.cl_default_mds_easize);
931 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
932 obd->u.cli.cl_default_mds_cookiesize);
934 ptlrpc_request_set_replen(req);
936 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
937 rc = ptlrpc_queue_wait(req);
938 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
940 if (req->rq_repmsg == NULL) {
941 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
944 rc = req->rq_status ?: -EIO;
945 } else if (rc == 0 || rc == -EAGAIN) {
946 struct mdt_body *body;
948 rc = lustre_msg_get_status(req->rq_repmsg);
949 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
950 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
955 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
958 } else if (rc == -ESTALE) {
960 * it can be allowed error after 3633 if open was committed and
961 * server failed before close was sent. Let's check if mod
962 * exists and return no error in that case
965 DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
966 LASSERT(mod->mod_open_req != NULL);
967 if (mod->mod_open_req->rq_committed)
974 mod->mod_close_req = NULL;
975 /* Since now, mod is accessed through open_req only,
976 * thus close req does not keep a reference on mod anymore. */
980 mdc_close_handle_reply(req, op_data, rc);
981 RETURN(rc < 0 ? rc : saved_rc);
984 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
985 struct md_open_data *mod)
987 struct obd_device *obd = class_exp2obd(exp);
988 struct ptlrpc_request *req;
992 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
993 &RQF_MDS_DONE_WRITING);
997 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
998 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
1000 ptlrpc_request_free(req);
1005 LASSERTF(mod->mod_open_req != NULL &&
1006 mod->mod_open_req->rq_type != LI_POISON,
1007 "POISONED setattr %p!\n", mod->mod_open_req);
1009 mod->mod_close_req = req;
1010 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
1011 /* We no longer want to preserve this setattr for replay even
1012 * though the open was committed. b=3632, b=3633 */
1013 spin_lock(&mod->mod_open_req->rq_lock);
1014 mod->mod_open_req->rq_replay = 0;
1015 spin_unlock(&mod->mod_open_req->rq_lock);
1018 mdc_close_pack(req, op_data);
1019 ptlrpc_request_set_replen(req);
1021 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
1022 rc = ptlrpc_queue_wait(req);
1023 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
1025 if (rc == -ESTALE) {
1027 * it can be allowed error after 3633 if open or setattr were
1028 * committed and server failed before close was sent.
1029 * Let's check if mod exists and return no error in that case
1032 LASSERT(mod->mod_open_req != NULL);
1033 if (mod->mod_open_req->rq_committed)
1040 mod->mod_close_req = NULL;
1041 LASSERT(mod->mod_open_req != NULL);
1044 /* Since now, mod is accessed through setattr req only,
1045 * thus DW req does not keep a reference on mod anymore. */
1049 mdc_close_handle_reply(req, op_data, rc);
1050 ptlrpc_req_finished(req);
1054 #ifdef HAVE_SPLIT_SUPPORT
1055 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
1056 const struct page *page, int offset)
1058 struct ptlrpc_request *req;
1059 struct ptlrpc_bulk_desc *desc;
1063 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
1067 /* FIXME: capa doesn't support split yet */
1068 mdc_set_capa_size(req, &RMF_CAPA1, NULL);
1070 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
1072 ptlrpc_request_free(req);
1076 req->rq_request_portal = MDS_READPAGE_PORTAL;
1077 ptlrpc_at_set_req_timeout(req);
1079 desc = ptlrpc_prep_bulk_imp(req, 1, 1,BULK_GET_SOURCE, MDS_BULK_PORTAL);
1081 GOTO(out, rc = -ENOMEM);
1083 /* NB req now owns desc and will free it when it gets freed. */
1084 ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
1085 mdc_readdir_pack(req, 0, offset, fid, NULL);
1087 ptlrpc_request_set_replen(req);
1088 rc = ptlrpc_queue_wait(req);
1092 rc = sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk);
1094 ptlrpc_req_finished(req);
1097 EXPORT_SYMBOL(mdc_sendpage);
1100 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
1101 __u64 offset, struct obd_capa *oc,
1102 struct page **pages, int npages,
1103 struct ptlrpc_request **request)
1105 struct ptlrpc_request *req;
1106 struct ptlrpc_bulk_desc *desc;
1108 wait_queue_head_t waitq;
1110 struct l_wait_info lwi;
1115 init_waitqueue_head(&waitq);
1118 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1122 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1124 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1126 ptlrpc_request_free(req);
1130 req->rq_request_portal = MDS_READPAGE_PORTAL;
1131 ptlrpc_at_set_req_timeout(req);
1133 desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK,
1136 ptlrpc_request_free(req);
1140 /* NB req now owns desc and will free it when it gets freed */
1141 for (i = 0; i < npages; i++)
1142 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
1144 mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE * npages, fid, oc);
1146 ptlrpc_request_set_replen(req);
1147 rc = ptlrpc_queue_wait(req);
1149 ptlrpc_req_finished(req);
1150 if (rc != -ETIMEDOUT)
1154 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1155 CERROR("%s: too many resend retries: rc = %d\n",
1156 exp->exp_obd->obd_name, -EIO);
1159 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
1161 l_wait_event(waitq, 0, &lwi);
1166 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1167 req->rq_bulk->bd_nob_transferred);
1169 ptlrpc_req_finished(req);
1173 if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1174 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1175 exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1176 PAGE_CACHE_SIZE * npages);
1177 ptlrpc_req_finished(req);
1186 static void mdc_release_page(struct page *page, int remove)
1190 if (likely(page->mapping != NULL))
1191 truncate_complete_page(page->mapping, page);
1194 page_cache_release(page);
1197 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1198 __u64 *start, __u64 *end, int hash64)
1201 * Complement of hash is used as an index so that
1202 * radix_tree_gang_lookup() can be used to find a page with starting
1203 * hash _smaller_ than one we are looking for.
1205 unsigned long offset = hash_x_index(*hash, hash64);
1209 spin_lock_irq(&mapping->tree_lock);
1210 found = radix_tree_gang_lookup(&mapping->page_tree,
1211 (void **)&page, offset, 1);
1213 struct lu_dirpage *dp;
1215 page_cache_get(page);
1216 spin_unlock_irq(&mapping->tree_lock);
1218 * In contrast to find_lock_page() we are sure that directory
1219 * page cannot be truncated (while DLM lock is held) and,
1220 * hence, can avoid restart.
1222 * In fact, page cannot be locked here at all, because
1223 * mdc_read_page_remote does synchronous io.
1225 wait_on_page_locked(page);
1226 if (PageUptodate(page)) {
1228 if (BITS_PER_LONG == 32 && hash64) {
1229 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1230 *end = le64_to_cpu(dp->ldp_hash_end) >> 32;
1231 *hash = *hash >> 32;
1233 *start = le64_to_cpu(dp->ldp_hash_start);
1234 *end = le64_to_cpu(dp->ldp_hash_end);
1236 if (unlikely(*start == 1 && *hash == 0))
1239 LASSERTF(*start <= *hash, "start = "LPX64
1240 ",end = "LPX64",hash = "LPX64"\n",
1241 *start, *end, *hash);
1242 CDEBUG(D_VFSTRACE, "page%lu [%llu %llu], hash"LPU64"\n",
1243 offset, *start, *end, *hash);
1246 mdc_release_page(page, 0);
1248 } else if (*end != *start && *hash == *end) {
1250 * upon hash collision, remove this page,
1251 * otherwise put page reference, and
1252 * ll_get_dir_page() will issue RPC to fetch
1256 mdc_release_page(page,
1257 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1261 page_cache_release(page);
1262 page = ERR_PTR(-EIO);
1265 spin_unlock_irq(&mapping->tree_lock);
1272 * Adjust a set of pages, each page containing an array of lu_dirpages,
1273 * so that each page can be used as a single logical lu_dirpage.
1275 * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1276 * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1277 * struct lu_dirent. It has size up to LU_PAGE_SIZE. The ldp_hash_end
1278 * value is used as a cookie to request the next lu_dirpage in a
1279 * directory listing that spans multiple pages (two in this example):
1282 * .|--------v------- -----.
1283 * |s|e|f|p|ent|ent| ... |ent|
1284 * '--|-------------- -----' Each CFS_PAGE contains a single
1285 * '------. lu_dirpage.
1286 * .---------v------- -----.
1287 * |s|e|f|p|ent| 0 | ... | 0 |
1288 * '----------------- -----'
1290 * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
1291 * larger than LU_PAGE_SIZE, a single host page may contain multiple
1292 * lu_dirpages. After reading the lu_dirpages from the MDS, the
1293 * ldp_hash_end of the first lu_dirpage refers to the one immediately
1294 * after it in the same CFS_PAGE (arrows simplified for brevity, but
1295 * in general e0==s1, e1==s2, etc.):
1297 * .-------------------- -----.
1298 * |s0|e0|f0|p|ent|ent| ... |ent|
1299 * |---v---------------- -----|
1300 * |s1|e1|f1|p|ent|ent| ... |ent|
1301 * |---v---------------- -----| Here, each CFS_PAGE contains
1302 * ... multiple lu_dirpages.
1303 * |---v---------------- -----|
1304 * |s'|e'|f'|p|ent|ent| ... |ent|
1305 * '---|---------------- -----'
1307 * .----------------------------.
1310 * This structure is transformed into a single logical lu_dirpage as follows:
1312 * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1313 * labeled 'next CFS_PAGE'.
1315 * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1316 * a hash collision with the next page exists.
1318 * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1319 * to the first entry of the next lu_dirpage.
1321 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
1322 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1326 for (i = 0; i < cfs_pgs; i++) {
1327 struct lu_dirpage *dp = kmap(pages[i]);
1328 struct lu_dirpage *first = dp;
1329 struct lu_dirent *end_dirent = NULL;
1330 struct lu_dirent *ent;
1331 __u64 hash_end = le64_to_cpu(dp->ldp_hash_end);
1332 __u32 flags = le32_to_cpu(dp->ldp_flags);
1334 while (--lu_pgs > 0) {
1335 ent = lu_dirent_start(dp);
1336 for (end_dirent = ent; ent != NULL;
1337 end_dirent = ent, ent = lu_dirent_next(ent));
1339 /* Advance dp to next lu_dirpage. */
1340 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1342 /* Check if we've reached the end of the CFS_PAGE. */
1343 if (!((unsigned long)dp & ~CFS_PAGE_MASK))
1346 /* Save the hash and flags of this lu_dirpage. */
1347 hash_end = le64_to_cpu(dp->ldp_hash_end);
1348 flags = le32_to_cpu(dp->ldp_flags);
1350 /* Check if lu_dirpage contains no entries. */
1351 if (end_dirent == NULL)
1354 /* Enlarge the end entry lde_reclen from 0 to
1355 * first entry of next lu_dirpage. */
1356 LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1357 end_dirent->lde_reclen =
1358 cpu_to_le16((char *)(dp->ldp_entries) -
1359 (char *)end_dirent);
1362 first->ldp_hash_end = hash_end;
1363 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1364 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1368 LASSERTF(lu_pgs == 0, "left = %d", lu_pgs);
1371 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1372 #endif /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
1374 /* parameters for readdir page */
1375 struct readpage_param {
1376 struct md_op_data *rp_mod;
1379 struct obd_export *rp_exp;
1380 struct md_callback *rp_cb;
1384 * Read pages from server.
1386 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1387 * a header lu_dirpage which describes the start/end hash, and whether this
1388 * page is empty (contains no dir entry) or hash collide with next page.
1389 * After client receives reply, several pages will be integrated into dir page
1390 * in CFS_PAGE_SIZE (if CFS_PAGE_SIZE greater than LU_PAGE_SIZE), and the
1391 * lu_dirpage for this integrated page will be adjusted.
1393 static int mdc_read_page_remote(void *data, struct page *page0)
1395 struct readpage_param *rp = data;
1396 struct page **page_pool;
1398 struct lu_dirpage *dp;
1399 int rd_pgs = 0; /* number of pages read actually */
1401 struct md_op_data *op_data = rp->rp_mod;
1402 struct ptlrpc_request *req;
1403 int max_pages = op_data->op_max_pages;
1404 struct inode *inode;
1410 LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
1411 if (op_data->op_mea1 != NULL) {
1412 __u32 index = op_data->op_stripe_offset;
1414 inode = op_data->op_mea1->lsm_md_oinfo[index].lmo_root;
1415 fid = &op_data->op_mea1->lsm_md_oinfo[index].lmo_fid;
1417 inode = op_data->op_data;
1418 fid = &op_data->op_fid1;
1420 LASSERT(inode != NULL);
1422 OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1423 if (page_pool != NULL) {
1424 page_pool[0] = page0;
1430 for (npages = 1; npages < max_pages; npages++) {
1431 page = page_cache_alloc_cold(inode->i_mapping);
1434 page_pool[npages] = page;
1437 rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, op_data->op_capa1,
1438 page_pool, npages, &req);
1442 rd_pgs = (req->rq_bulk->bd_nob_transferred +
1443 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1444 lu_pgs = req->rq_bulk->bd_nob_transferred >>
1446 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1448 CDEBUG(D_INODE, "read %d(%d)/%d pages\n", rd_pgs, lu_pgs,
1449 op_data->op_npages);
1451 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1453 SetPageUptodate(page0);
1457 ptlrpc_req_finished(req);
1458 CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1459 for (i = 1; i < npages; i++) {
1460 unsigned long offset;
1464 page = page_pool[i];
1466 if (rc < 0 || i >= rd_pgs) {
1467 page_cache_release(page);
1471 SetPageUptodate(page);
1474 hash = le64_to_cpu(dp->ldp_hash_start);
1477 offset = hash_x_index(hash, rp->rp_hash64);
1479 prefetchw(&page->flags);
1480 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1485 CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1486 " rc = %d\n", offset, ret);
1487 page_cache_release(page);
1490 if (page_pool != &page0)
1491 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1497 * Read dir page from cache first, if it can not find it, read it from
1498 * server and add into the cache.
1500 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1501 struct md_callback *cb_op, struct page **ppage)
1503 struct lookup_intent it = { .it_op = IT_READDIR };
1505 struct inode *dir = NULL;
1506 struct address_space *mapping;
1507 struct lu_dirpage *dp;
1510 struct lustre_handle lockh;
1511 struct ptlrpc_request *enq_req = NULL;
1512 struct readpage_param rp_param;
1519 if (op_data->op_mea1 != NULL) {
1520 __u32 index = op_data->op_stripe_offset;
1522 dir = op_data->op_mea1->lsm_md_oinfo[index].lmo_root;
1524 dir = op_data->op_data;
1526 LASSERT(dir != NULL);
1528 mapping = dir->i_mapping;
1530 rc = mdc_intent_lock(exp, op_data, NULL, 0, &it, 0, &enq_req,
1531 cb_op->md_blocking_ast, 0);
1532 if (enq_req != NULL)
1533 ptlrpc_req_finished(enq_req);
1536 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1537 exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1542 mdc_set_lock_data(exp, &it.d.lustre.it_lock_handle, dir, NULL);
1544 rp_param.rp_off = op_data->op_hash_offset;
1545 rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1546 page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1547 rp_param.rp_hash64);
1549 CERROR("%s: dir page locate: "DFID" at "LPU64": rc %ld\n",
1550 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1551 rp_param.rp_off, PTR_ERR(page));
1552 GOTO(out_unlock, rc = PTR_ERR(page));
1553 } else if (page != NULL) {
1555 * XXX nikita: not entirely correct handling of a corner case:
1556 * suppose hash chain of entries with hash value HASH crosses
1557 * border between pages P0 and P1. First both P0 and P1 are
1558 * cached, seekdir() is called for some entry from the P0 part
1559 * of the chain. Later P0 goes out of cache. telldir(HASH)
1560 * happens and finds P1, as it starts with matching hash
1561 * value. Remaining entries from P0 part of the chain are
1562 * skipped. (Is that really a bug?)
1564 * Possible solutions: 0. don't cache P1 is such case, handle
1565 * it as an "overflow" page. 1. invalidate all pages at
1566 * once. 2. use HASH|1 as an index for P1.
1568 GOTO(hash_collision, page);
1571 rp_param.rp_exp = exp;
1572 rp_param.rp_mod = op_data;
1573 page = read_cache_page(mapping,
1574 hash_x_index(rp_param.rp_off,
1575 rp_param.rp_hash64),
1576 mdc_read_page_remote, &rp_param);
1578 CERROR("%s: read cache page: "DFID" at "LPU64": rc %ld\n",
1579 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1580 rp_param.rp_off, PTR_ERR(page));
1581 GOTO(out_unlock, rc = PTR_ERR(page));
1584 wait_on_page_locked(page);
1586 if (!PageUptodate(page)) {
1587 CERROR("%s: page not updated: "DFID" at "LPU64": rc %d\n",
1588 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1589 rp_param.rp_off, -5);
1592 if (!PageChecked(page))
1593 SetPageChecked(page);
1594 if (PageError(page)) {
1595 CERROR("%s: page error: "DFID" at "LPU64": rc %d\n",
1596 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1597 rp_param.rp_off, -5);
1602 dp = page_address(page);
1603 if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1604 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1605 end = le64_to_cpu(dp->ldp_hash_end) >> 32;
1606 rp_param.rp_off = op_data->op_hash_offset >> 32;
1608 start = le64_to_cpu(dp->ldp_hash_start);
1609 end = le64_to_cpu(dp->ldp_hash_end);
1610 rp_param.rp_off = op_data->op_hash_offset;
1613 LASSERT(start == rp_param.rp_off);
1614 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1615 #if BITS_PER_LONG == 32
1616 CWARN("Real page-wide hash collision at ["LPU64" "LPU64"] with "
1617 "hash "LPU64"\n", le64_to_cpu(dp->ldp_hash_start),
1618 le64_to_cpu(dp->ldp_hash_end), op_data->op_hash_offset);
1622 * Fetch whole overflow chain...
1630 lockh.cookie = it.d.lustre.it_lock_handle;
1631 ldlm_lock_decref(&lockh, it.d.lustre.it_lock_mode);
1632 it.d.lustre.it_lock_handle = 0;
1636 mdc_release_page(page, 1);
1642 * Read one directory entry from the cache.
1644 int mdc_read_entry(struct obd_export *exp, struct md_op_data *op_data,
1645 struct md_callback *cb_op, struct lu_dirent **entp,
1646 struct page **ppage)
1648 struct page *page = NULL;
1649 struct lu_dirpage *dp;
1650 struct lu_dirent *ent;
1654 CDEBUG(D_INFO, DFID "offset = "LPU64"\n", PFID(&op_data->op_fid1),
1655 op_data->op_hash_offset);
1660 if (op_data->op_hash_offset == MDS_DIR_END_OFF)
1663 rc = mdc_read_page(exp, op_data, cb_op, &page);
1667 dp = page_address(page);
1668 for (ent = lu_dirent_start(dp); ent != NULL;
1669 ent = lu_dirent_next(ent)) {
1670 /* Skip dummy entry */
1671 if (le16_to_cpu(ent->lde_namelen) == 0)
1674 if (le64_to_cpu(ent->lde_hash) > op_data->op_hash_offset)
1678 /* If it can not find entry in current page, try next page. */
1680 __u64 orig_offset = op_data->op_hash_offset;
1682 if (le64_to_cpu(dp->ldp_hash_end) == MDS_DIR_END_OFF) {
1683 mdc_release_page(page, 0);
1687 op_data->op_hash_offset = le64_to_cpu(dp->ldp_hash_end);
1688 mdc_release_page(page,
1689 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1690 rc = mdc_read_page(exp, op_data, cb_op, &page);
1695 dp = page_address(page);
1696 ent = lu_dirent_start(dp);
1699 op_data->op_hash_offset = orig_offset;
1708 #else /* __KERNEL__ */
1711 *mdc_read_page_remote(struct obd_export *exp, const struct lmv_oinfo *lmo,
1712 const __u64 hash, struct obd_capa *oc)
1714 struct ptlrpc_request *req = NULL;
1718 OBD_PAGE_ALLOC(page, 0);
1720 return ERR_PTR(-ENOMEM);
1722 rc = mdc_getpage(exp, &lmo->lmo_fid, hash, oc, &page, 1, &req);
1724 ptlrpc_req_finished(req);
1727 OBD_PAGE_FREE(page);
1734 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1735 struct md_callback *cb_op,
1736 struct page **ppage)
1739 struct lmv_oinfo *lmo;
1742 /* No local cache for liblustre, always read entry remotely */
1743 lmo = &op_data->op_mea1->lsm_md_oinfo[op_data->op_stripe_offset];
1744 page = mdc_read_page_remote(exp, lmo, op_data->op_hash_offset,
1747 return PTR_ERR(page);
1754 int mdc_read_entry(struct obd_export *exp, struct md_op_data *op_data,
1755 struct md_callback *cb_op, struct lu_dirent **entp,
1756 struct page **ppage)
1758 struct page *page = NULL;
1759 struct lu_dirpage *dp;
1760 struct lu_dirent *ent;
1764 rc = mdc_read_page(exp, op_data, cb_op, &page);
1768 dp = page_address(page);
1769 if (le64_to_cpu(dp->ldp_hash_end) < op_data->op_hash_offset)
1770 GOTO(out, *entp = NULL);
1772 for (ent = lu_dirent_start(dp); ent != NULL;
1773 ent = lu_dirent_next(ent))
1774 if (le64_to_cpu(ent->lde_hash) >= op_data->op_hash_offset)
1779 OBD_PAGE_FREE(page);
1785 static int mdc_statfs(const struct lu_env *env,
1786 struct obd_export *exp, struct obd_statfs *osfs,
1787 __u64 max_age, __u32 flags)
1789 struct obd_device *obd = class_exp2obd(exp);
1790 struct ptlrpc_request *req;
1791 struct obd_statfs *msfs;
1792 struct obd_import *imp = NULL;
1797 * Since the request might also come from lprocfs, so we need
1798 * sync this with client_disconnect_export Bug15684
1800 down_read(&obd->u.cli.cl_sem);
1801 if (obd->u.cli.cl_import)
1802 imp = class_import_get(obd->u.cli.cl_import);
1803 up_read(&obd->u.cli.cl_sem);
1807 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1808 LUSTRE_MDS_VERSION, MDS_STATFS);
1810 GOTO(output, rc = -ENOMEM);
1812 ptlrpc_request_set_replen(req);
1814 if (flags & OBD_STATFS_NODELAY) {
1815 /* procfs requests not want stay in wait for avoid deadlock */
1816 req->rq_no_resend = 1;
1817 req->rq_no_delay = 1;
1820 rc = ptlrpc_queue_wait(req);
1822 /* check connection error first */
1823 if (imp->imp_connect_error)
1824 rc = imp->imp_connect_error;
1828 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1830 GOTO(out, rc = -EPROTO);
1835 ptlrpc_req_finished(req);
1837 class_import_put(imp);
1841 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1843 __u32 keylen, vallen;
1847 if (gf->gf_pathlen > PATH_MAX)
1848 RETURN(-ENAMETOOLONG);
1849 if (gf->gf_pathlen < 2)
1852 /* Key is KEY_FID2PATH + getinfo_fid2path description */
1853 keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1854 OBD_ALLOC(key, keylen);
1857 memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1858 memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1860 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1861 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1863 if (!fid_is_sane(&gf->gf_fid))
1864 GOTO(out, rc = -EINVAL);
1866 /* Val is struct getinfo_fid2path result plus path */
1867 vallen = sizeof(*gf) + gf->gf_pathlen;
1869 rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1870 if (rc != 0 && rc != -EREMOTE)
1873 if (vallen <= sizeof(*gf))
1874 GOTO(out, rc = -EPROTO);
1875 else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1876 GOTO(out, rc = -EOVERFLOW);
1878 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1879 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1882 OBD_FREE(key, keylen);
1886 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1887 struct hsm_progress_kernel *hpk)
1889 struct obd_import *imp = class_exp2cliimp(exp);
1890 struct hsm_progress_kernel *req_hpk;
1891 struct ptlrpc_request *req;
1895 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1896 LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1898 GOTO(out, rc = -ENOMEM);
1900 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1902 /* Copy hsm_progress struct */
1903 req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1904 if (req_hpk == NULL)
1905 GOTO(out, rc = -EPROTO);
1908 req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1910 ptlrpc_request_set_replen(req);
1912 rc = mdc_queue_wait(req);
1915 ptlrpc_req_finished(req);
1919 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1921 __u32 *archive_mask;
1922 struct ptlrpc_request *req;
1926 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1928 MDS_HSM_CT_REGISTER);
1930 GOTO(out, rc = -ENOMEM);
1932 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1934 /* Copy hsm_progress struct */
1935 archive_mask = req_capsule_client_get(&req->rq_pill,
1936 &RMF_MDS_HSM_ARCHIVE);
1937 if (archive_mask == NULL)
1938 GOTO(out, rc = -EPROTO);
1940 *archive_mask = archives;
1942 ptlrpc_request_set_replen(req);
1944 rc = mdc_queue_wait(req);
1947 ptlrpc_req_finished(req);
1951 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1952 struct md_op_data *op_data)
1954 struct hsm_current_action *hca = op_data->op_data;
1955 struct hsm_current_action *req_hca;
1956 struct ptlrpc_request *req;
1960 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1961 &RQF_MDS_HSM_ACTION);
1965 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1967 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1969 ptlrpc_request_free(req);
1973 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1974 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1976 ptlrpc_request_set_replen(req);
1978 rc = mdc_queue_wait(req);
1982 req_hca = req_capsule_server_get(&req->rq_pill,
1983 &RMF_MDS_HSM_CURRENT_ACTION);
1984 if (req_hca == NULL)
1985 GOTO(out, rc = -EPROTO);
1991 ptlrpc_req_finished(req);
1995 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1997 struct ptlrpc_request *req;
2001 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
2003 MDS_HSM_CT_UNREGISTER);
2005 GOTO(out, rc = -ENOMEM);
2007 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
2009 ptlrpc_request_set_replen(req);
2011 rc = mdc_queue_wait(req);
2014 ptlrpc_req_finished(req);
2018 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
2019 struct md_op_data *op_data)
2021 struct hsm_user_state *hus = op_data->op_data;
2022 struct hsm_user_state *req_hus;
2023 struct ptlrpc_request *req;
2027 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2028 &RQF_MDS_HSM_STATE_GET);
2032 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2034 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
2036 ptlrpc_request_free(req);
2040 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
2041 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
2043 ptlrpc_request_set_replen(req);
2045 rc = mdc_queue_wait(req);
2049 req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
2050 if (req_hus == NULL)
2051 GOTO(out, rc = -EPROTO);
2057 ptlrpc_req_finished(req);
2061 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
2062 struct md_op_data *op_data)
2064 struct hsm_state_set *hss = op_data->op_data;
2065 struct hsm_state_set *req_hss;
2066 struct ptlrpc_request *req;
2070 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2071 &RQF_MDS_HSM_STATE_SET);
2075 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2077 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
2079 ptlrpc_request_free(req);
2083 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
2084 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
2087 req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
2088 if (req_hss == NULL)
2089 GOTO(out, rc = -EPROTO);
2092 ptlrpc_request_set_replen(req);
2094 rc = mdc_queue_wait(req);
2099 ptlrpc_req_finished(req);
2103 static int mdc_ioc_hsm_request(struct obd_export *exp,
2104 struct hsm_user_request *hur)
2106 struct obd_import *imp = class_exp2cliimp(exp);
2107 struct ptlrpc_request *req;
2108 struct hsm_request *req_hr;
2109 struct hsm_user_item *req_hui;
2114 req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
2116 GOTO(out, rc = -ENOMEM);
2118 req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
2119 hur->hur_request.hr_itemcount
2120 * sizeof(struct hsm_user_item));
2121 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
2122 hur->hur_request.hr_data_len);
2124 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
2126 ptlrpc_request_free(req);
2130 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
2132 /* Copy hsm_request struct */
2133 req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
2135 GOTO(out, rc = -EPROTO);
2136 *req_hr = hur->hur_request;
2138 /* Copy hsm_user_item structs */
2139 req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
2140 if (req_hui == NULL)
2141 GOTO(out, rc = -EPROTO);
2142 memcpy(req_hui, hur->hur_user_item,
2143 hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
2145 /* Copy opaque field */
2146 req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
2147 if (req_opaque == NULL)
2148 GOTO(out, rc = -EPROTO);
2149 memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
2151 ptlrpc_request_set_replen(req);
2153 rc = mdc_queue_wait(req);
2157 ptlrpc_req_finished(req);
2161 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
2163 struct kuc_hdr *lh = (struct kuc_hdr *)buf;
2165 LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
2167 lh->kuc_magic = KUC_MAGIC;
2168 lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
2169 lh->kuc_flags = flags;
2170 lh->kuc_msgtype = CL_RECORD;
2171 lh->kuc_msglen = len;
2175 #define D_CHANGELOG 0
2177 struct changelog_show {
2182 struct obd_device *cs_obd;
2185 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
2186 struct llog_rec_hdr *hdr, void *data)
2188 struct changelog_show *cs = data;
2189 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
2194 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
2196 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
2197 cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
2198 rec->cr.cr_type, rc);
2202 if (rec->cr.cr_index < cs->cs_startrec) {
2203 /* Skip entries earlier than what we are interested in */
2204 CDEBUG(D_CHANGELOG, "rec="LPU64" start="LPU64"\n",
2205 rec->cr.cr_index, cs->cs_startrec);
2209 CDEBUG(D_CHANGELOG, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID
2210 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
2211 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
2212 rec->cr.cr_flags & CLF_FLAGMASK,
2213 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
2214 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
2216 len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
2218 /* Set up the message */
2219 lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
2220 memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
2222 rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
2223 CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len,rc);
2228 static int mdc_changelog_send_thread(void *csdata)
2230 struct changelog_show *cs = csdata;
2231 struct llog_ctxt *ctxt = NULL;
2232 struct llog_handle *llh = NULL;
2233 struct kuc_hdr *kuch;
2236 CDEBUG(D_CHANGELOG, "changelog to fp=%p start "LPU64"\n",
2237 cs->cs_fp, cs->cs_startrec);
2239 OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
2240 if (cs->cs_buf == NULL)
2241 GOTO(out, rc = -ENOMEM);
2243 /* Set up the remote catalog handle */
2244 ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
2246 GOTO(out, rc = -ENOENT);
2247 rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
2250 CERROR("%s: fail to open changelog catalog: rc = %d\n",
2251 cs->cs_obd->obd_name, rc);
2254 rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
2256 CERROR("llog_init_handle failed %d\n", rc);
2260 rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
2262 /* Send EOF no matter what our result */
2263 if ((kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch),
2265 kuch->kuc_msgtype = CL_EOF;
2266 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
2272 llog_cat_close(NULL, llh);
2274 llog_ctxt_put(ctxt);
2276 OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
2281 static int mdc_ioc_changelog_send(struct obd_device *obd,
2282 struct ioc_changelog *icc)
2284 struct changelog_show *cs;
2287 /* Freed in mdc_changelog_send_thread */
2293 cs->cs_startrec = icc->icc_recno;
2294 /* matching fput in mdc_changelog_send_thread */
2295 cs->cs_fp = fget(icc->icc_id);
2296 cs->cs_flags = icc->icc_flags;
2299 * New thread because we should return to user app before
2300 * writing into our pipe
2302 rc = PTR_ERR(kthread_run(mdc_changelog_send_thread, cs,
2303 "mdc_clg_send_thread"));
2304 if (!IS_ERR_VALUE(rc)) {
2305 CDEBUG(D_CHANGELOG, "start changelog thread\n");
2309 CERROR("Failed to start changelog thread: %d\n", rc);
2314 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2315 struct lustre_kernelcomm *lk);
2317 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
2318 struct obd_quotactl *oqctl)
2320 struct client_obd *cli = &exp->exp_obd->u.cli;
2321 struct ptlrpc_request *req;
2322 struct obd_quotactl *body;
2326 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2327 &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
2332 body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2335 ptlrpc_request_set_replen(req);
2337 /* the next poll will find -ENODATA, that means quotacheck is
2339 cli->cl_qchk_stat = -ENODATA;
2340 rc = ptlrpc_queue_wait(req);
2342 cli->cl_qchk_stat = rc;
2343 ptlrpc_req_finished(req);
2347 static int mdc_quota_poll_check(struct obd_export *exp,
2348 struct if_quotacheck *qchk)
2350 struct client_obd *cli = &exp->exp_obd->u.cli;
2354 qchk->obd_uuid = cli->cl_target_uuid;
2355 memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
2357 rc = cli->cl_qchk_stat;
2358 /* the client is not the previous one */
2359 if (rc == CL_NOT_QUOTACHECKED)
2364 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
2365 struct obd_quotactl *oqctl)
2367 struct ptlrpc_request *req;
2368 struct obd_quotactl *oqc;
2372 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2373 &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
2378 oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2381 ptlrpc_request_set_replen(req);
2382 ptlrpc_at_set_req_timeout(req);
2383 req->rq_no_resend = 1;
2385 rc = ptlrpc_queue_wait(req);
2387 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
2389 if (req->rq_repmsg &&
2390 (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2393 CERROR ("Can't unpack obd_quotactl\n");
2396 ptlrpc_req_finished(req);
2401 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2402 struct md_op_data *op_data)
2404 CFS_LIST_HEAD(cancels);
2405 struct ptlrpc_request *req;
2407 struct mdc_swap_layouts *msl, *payload;
2410 msl = op_data->op_data;
2412 /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2413 * first thing it will do is to cancel the 2 layout
2414 * locks hold by this client.
2415 * So the client must cancel its layout locks on the 2 fids
2416 * with the request RPC to avoid extra RPC round trips
2418 count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2419 LCK_EX, MDS_INODELOCK_LAYOUT |
2420 MDS_INODELOCK_XATTR);
2421 count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2422 LCK_EX, MDS_INODELOCK_LAYOUT |
2423 MDS_INODELOCK_XATTR);
2425 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2426 &RQF_MDS_SWAP_LAYOUTS);
2428 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2432 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2433 mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
2435 rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2437 ptlrpc_request_free(req);
2441 mdc_swap_layouts_pack(req, op_data);
2443 payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2448 ptlrpc_request_set_replen(req);
2450 rc = ptlrpc_queue_wait(req);
2456 ptlrpc_req_finished(req);
2460 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2461 void *karg, void *uarg)
2463 struct obd_device *obd = exp->exp_obd;
2464 struct obd_ioctl_data *data = karg;
2465 struct obd_import *imp = obd->u.cli.cl_import;
2469 if (!try_module_get(THIS_MODULE)) {
2470 CERROR("Can't get module. Is it alive?");
2474 case OBD_IOC_CHANGELOG_SEND:
2475 rc = mdc_ioc_changelog_send(obd, karg);
2477 case OBD_IOC_CHANGELOG_CLEAR: {
2478 struct ioc_changelog *icc = karg;
2479 struct changelog_setinfo cs =
2480 {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
2481 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
2482 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
2486 case OBD_IOC_FID2PATH:
2487 rc = mdc_ioc_fid2path(exp, karg);
2489 case LL_IOC_HSM_CT_START:
2490 rc = mdc_ioc_hsm_ct_start(exp, karg);
2491 /* ignore if it was already registered on this MDS. */
2495 case LL_IOC_HSM_PROGRESS:
2496 rc = mdc_ioc_hsm_progress(exp, karg);
2498 case LL_IOC_HSM_STATE_GET:
2499 rc = mdc_ioc_hsm_state_get(exp, karg);
2501 case LL_IOC_HSM_STATE_SET:
2502 rc = mdc_ioc_hsm_state_set(exp, karg);
2504 case LL_IOC_HSM_ACTION:
2505 rc = mdc_ioc_hsm_current_action(exp, karg);
2507 case LL_IOC_HSM_REQUEST:
2508 rc = mdc_ioc_hsm_request(exp, karg);
2510 case OBD_IOC_CLIENT_RECOVER:
2511 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2515 case IOC_OSC_SET_ACTIVE:
2516 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2518 case OBD_IOC_POLL_QUOTACHECK:
2519 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
2521 case OBD_IOC_PING_TARGET:
2522 rc = ptlrpc_obd_ping(obd);
2525 * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2526 * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2527 * there'd be no LMV layer thus we might be called here. Eventually
2528 * this code should be removed.
2531 case IOC_OBD_STATFS: {
2532 struct obd_statfs stat_buf = {0};
2534 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2535 GOTO(out, rc = -ENODEV);
2538 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2539 min((int)data->ioc_plen2,
2540 (int)sizeof(struct obd_uuid))))
2541 GOTO(out, rc = -EFAULT);
2543 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2544 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2549 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2550 min((int) data->ioc_plen1,
2551 (int) sizeof(stat_buf))))
2552 GOTO(out, rc = -EFAULT);
2556 case OBD_IOC_QUOTACTL: {
2557 struct if_quotactl *qctl = karg;
2558 struct obd_quotactl *oqctl;
2560 OBD_ALLOC_PTR(oqctl);
2562 GOTO(out, rc = -ENOMEM);
2564 QCTL_COPY(oqctl, qctl);
2565 rc = obd_quotactl(exp, oqctl);
2567 QCTL_COPY(qctl, oqctl);
2568 qctl->qc_valid = QC_MDTIDX;
2569 qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2572 OBD_FREE_PTR(oqctl);
2575 case LL_IOC_GET_CONNECT_FLAGS:
2576 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2577 sizeof(*exp_connect_flags_ptr(exp))))
2578 GOTO(out, rc = -EFAULT);
2581 case LL_IOC_LOV_SWAP_LAYOUTS:
2582 rc = mdc_ioc_swap_layouts(exp, karg);
2585 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2586 GOTO(out, rc = -ENOTTY);
2589 module_put(THIS_MODULE);
2594 int mdc_get_info_rpc(struct obd_export *exp,
2595 obd_count keylen, void *key,
2596 int vallen, void *val)
2598 struct obd_import *imp = class_exp2cliimp(exp);
2599 struct ptlrpc_request *req;
2604 req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2608 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2609 RCL_CLIENT, keylen);
2610 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2611 RCL_CLIENT, sizeof(__u32));
2613 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2615 ptlrpc_request_free(req);
2619 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2620 memcpy(tmp, key, keylen);
2621 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2622 memcpy(tmp, &vallen, sizeof(__u32));
2624 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2625 RCL_SERVER, vallen);
2626 ptlrpc_request_set_replen(req);
2628 rc = ptlrpc_queue_wait(req);
2629 /* -EREMOTE means the get_info result is partial, and it needs to
2630 * continue on another MDT, see fid2path part in lmv_iocontrol */
2631 if (rc == 0 || rc == -EREMOTE) {
2632 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2633 memcpy(val, tmp, vallen);
2634 if (ptlrpc_rep_need_swab(req)) {
2635 if (KEY_IS(KEY_FID2PATH))
2636 lustre_swab_fid2path(val);
2639 ptlrpc_req_finished(req);
2644 static void lustre_swab_hai(struct hsm_action_item *h)
2646 __swab32s(&h->hai_len);
2647 __swab32s(&h->hai_action);
2648 lustre_swab_lu_fid(&h->hai_fid);
2649 lustre_swab_lu_fid(&h->hai_dfid);
2650 __swab64s(&h->hai_cookie);
2651 __swab64s(&h->hai_extent.offset);
2652 __swab64s(&h->hai_extent.length);
2653 __swab64s(&h->hai_gid);
2656 static void lustre_swab_hal(struct hsm_action_list *h)
2658 struct hsm_action_item *hai;
2661 __swab32s(&h->hal_version);
2662 __swab32s(&h->hal_count);
2663 __swab32s(&h->hal_archive_id);
2664 __swab64s(&h->hal_flags);
2666 for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2667 lustre_swab_hai(hai);
2670 static void lustre_swab_kuch(struct kuc_hdr *l)
2672 __swab16s(&l->kuc_magic);
2673 /* __u8 l->kuc_transport */
2674 __swab16s(&l->kuc_msgtype);
2675 __swab16s(&l->kuc_msglen);
2678 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2679 struct lustre_kernelcomm *lk)
2681 struct obd_import *imp = class_exp2cliimp(exp);
2682 __u32 archive = lk->lk_data;
2685 if (lk->lk_group != KUC_GRP_HSM) {
2686 CERROR("Bad copytool group %d\n", lk->lk_group);
2690 CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2691 lk->lk_uid, lk->lk_group, lk->lk_flags);
2693 if (lk->lk_flags & LK_FLG_STOP) {
2694 /* Unregister with the coordinator */
2695 rc = mdc_ioc_hsm_ct_unregister(imp);
2697 rc = mdc_ioc_hsm_ct_register(imp, archive);
2704 * Send a message to any listening copytools
2705 * @param val KUC message (kuc_hdr + hsm_action_list)
2706 * @param len total length of message
2708 static int mdc_hsm_copytool_send(int len, void *val)
2710 struct kuc_hdr *lh = (struct kuc_hdr *)val;
2711 struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
2715 if (len < sizeof(*lh) + sizeof(*hal)) {
2716 CERROR("Short HSM message %d < %d\n", len,
2717 (int) (sizeof(*lh) + sizeof(*hal)));
2720 if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2721 lustre_swab_kuch(lh);
2722 lustre_swab_hal(hal);
2723 } else if (lh->kuc_magic != KUC_MAGIC) {
2724 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2728 CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2730 lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2731 lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2733 /* Broadcast to HSM listeners */
2734 rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2740 * callback function passed to kuc for re-registering each HSM copytool
2741 * running on MDC, after MDT shutdown/recovery.
2742 * @param data copytool registration data
2743 * @param cb_arg callback argument (obd_import)
2745 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2747 struct kkuc_ct_data *kcd = data;
2748 struct obd_import *imp = (struct obd_import *)cb_arg;
2751 if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2754 if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2757 CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2758 imp->imp_obd->obd_name, kcd->kcd_archive);
2759 rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2761 /* ignore error if the copytool is already registered */
2762 return (rc == -EEXIST) ? 0 : rc;
2766 * Re-establish all kuc contexts with MDT
2767 * after MDT shutdown/recovery.
2769 static int mdc_kuc_reregister(struct obd_import *imp)
2771 /* re-register HSM agents */
2772 return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2776 int mdc_set_info_async(const struct lu_env *env,
2777 struct obd_export *exp,
2778 obd_count keylen, void *key,
2779 obd_count vallen, void *val,
2780 struct ptlrpc_request_set *set)
2782 struct obd_import *imp = class_exp2cliimp(exp);
2786 if (KEY_IS(KEY_READ_ONLY)) {
2787 if (vallen != sizeof(int))
2790 spin_lock(&imp->imp_lock);
2791 if (*((int *)val)) {
2792 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2793 imp->imp_connect_data.ocd_connect_flags |=
2796 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2797 imp->imp_connect_data.ocd_connect_flags &=
2798 ~OBD_CONNECT_RDONLY;
2800 spin_unlock(&imp->imp_lock);
2802 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2803 keylen, key, vallen, val, set);
2806 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2807 sptlrpc_conf_client_adapt(exp->exp_obd);
2810 if (KEY_IS(KEY_FLUSH_CTX)) {
2811 sptlrpc_import_flush_my_ctx(imp);
2814 if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2815 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2816 keylen, key, vallen, val, set);
2819 if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2820 rc = mdc_hsm_copytool_send(vallen, val);
2824 CERROR("Unknown key %s\n", (char *)key);
2828 int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2829 __u32 keylen, void *key, __u32 *vallen, void *val,
2830 struct lov_stripe_md *lsm)
2834 if (KEY_IS(KEY_MAX_EASIZE)) {
2835 int mdsize, *max_easize;
2837 if (*vallen != sizeof(int))
2839 mdsize = *(int *)val;
2840 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2841 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2843 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2845 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2846 int *default_easize;
2848 if (*vallen != sizeof(int))
2850 default_easize = val;
2851 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2853 } else if (KEY_IS(KEY_MAX_COOKIESIZE)) {
2854 int mdsize, *max_cookiesize;
2856 if (*vallen != sizeof(int))
2858 mdsize = *(int *)val;
2859 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_cookiesize)
2860 exp->exp_obd->u.cli.cl_max_mds_cookiesize = mdsize;
2861 max_cookiesize = val;
2862 *max_cookiesize = exp->exp_obd->u.cli.cl_max_mds_cookiesize;
2864 } else if (KEY_IS(KEY_DEFAULT_COOKIESIZE)) {
2865 int *default_cookiesize;
2867 if (*vallen != sizeof(int))
2869 default_cookiesize = val;
2870 *default_cookiesize =
2871 exp->exp_obd->u.cli.cl_default_mds_cookiesize;
2873 } else if (KEY_IS(KEY_CONN_DATA)) {
2874 struct obd_import *imp = class_exp2cliimp(exp);
2875 struct obd_connect_data *data = val;
2877 if (*vallen != sizeof(*data))
2880 *data = imp->imp_connect_data;
2882 } else if (KEY_IS(KEY_TGT_COUNT)) {
2887 rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2892 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
2893 struct obd_capa *oc, struct obd_client_handle *handle,
2896 struct ptlrpc_request *req;
2897 struct mdt_body *body;
2901 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
2905 mdc_set_capa_size(req, &RMF_CAPA1, oc);
2907 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
2909 ptlrpc_request_free(req);
2913 mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
2915 ptlrpc_request_set_replen(req);
2917 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2918 rc = ptlrpc_queue_wait(req);
2919 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2921 CERROR("Pin failed: %d\n", rc);
2925 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2927 GOTO(err_out, rc = -EPROTO);
2929 handle->och_fh = body->handle;
2930 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
2932 handle->och_mod = obd_mod_alloc();
2933 if (handle->och_mod == NULL) {
2934 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
2935 GOTO(err_out, rc = -ENOMEM);
2937 handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
2942 ptlrpc_req_finished(req);
2946 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
2949 struct ptlrpc_request *req;
2950 struct mdt_body *body;
2954 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
2955 LUSTRE_MDS_VERSION, MDS_UNPIN);
2959 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
2960 body->handle = handle->och_fh;
2963 ptlrpc_request_set_replen(req);
2965 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2966 rc = ptlrpc_queue_wait(req);
2967 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2970 CERROR("Unpin failed: %d\n", rc);
2972 ptlrpc_req_finished(req);
2973 ptlrpc_req_finished(handle->och_mod->mod_open_req);
2975 obd_mod_put(handle->och_mod);
2979 int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2980 struct obd_capa *oc, struct ptlrpc_request **request)
2982 struct ptlrpc_request *req;
2987 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2991 mdc_set_capa_size(req, &RMF_CAPA1, oc);
2993 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2995 ptlrpc_request_free(req);
2999 mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
3001 ptlrpc_request_set_replen(req);
3003 rc = ptlrpc_queue_wait(req);
3005 ptlrpc_req_finished(req);
3011 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
3012 enum obd_import_event event)
3016 LASSERT(imp->imp_obd == obd);
3019 case IMP_EVENT_DISCON: {
3021 /* XXX Pass event up to OBDs stack. used only for FLD now */
3022 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
3026 case IMP_EVENT_INACTIVE: {
3027 struct client_obd *cli = &obd->u.cli;
3029 * Flush current sequence to make client obtain new one
3030 * from server in case of disconnect/reconnect.
3032 if (cli->cl_seq != NULL)
3033 seq_client_flush(cli->cl_seq);
3035 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
3038 case IMP_EVENT_INVALIDATE: {
3039 struct ldlm_namespace *ns = obd->obd_namespace;
3041 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
3045 case IMP_EVENT_ACTIVE:
3046 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
3047 /* redo the kuc registration after reconnecting */
3049 rc = mdc_kuc_reregister(imp);
3052 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
3054 case IMP_EVENT_DEACTIVATE:
3055 case IMP_EVENT_ACTIVATE:
3058 CERROR("Unknown import event %x\n", event);
3064 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
3065 struct md_op_data *op_data)
3067 struct client_obd *cli = &exp->exp_obd->u.cli;
3068 struct lu_client_seq *seq = cli->cl_seq;
3070 RETURN(seq_client_alloc_fid(NULL, seq, fid));
3073 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
3074 struct client_obd *cli = &exp->exp_obd->u.cli;
3075 return &cli->cl_target_uuid;
3079 * Determine whether the lock can be canceled before replaying it during
3080 * recovery, non zero value will be return if the lock can be canceled,
3081 * or zero returned for not
3083 static int mdc_cancel_weight(struct ldlm_lock *lock)
3085 if (lock->l_resource->lr_type != LDLM_IBITS)
3088 /* FIXME: if we ever get into a situation where there are too many
3089 * opened files with open locks on a single node, then we really
3090 * should replay these open locks to reget it */
3091 if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
3097 static int mdc_resource_inode_free(struct ldlm_resource *res)
3099 if (res->lr_lvb_inode)
3100 res->lr_lvb_inode = NULL;
3105 struct ldlm_valblock_ops inode_lvbo = {
3106 .lvbo_free = mdc_resource_inode_free
3109 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
3111 struct client_obd *cli = &obd->u.cli;
3115 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3116 if (!cli->cl_rpc_lock)
3118 mdc_init_rpc_lock(cli->cl_rpc_lock);
3120 rc = ptlrpcd_addref();
3122 GOTO(err_rpc_lock, rc);
3124 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3125 if (!cli->cl_close_lock)
3126 GOTO(err_ptlrpcd_decref, rc = -ENOMEM);
3127 mdc_init_rpc_lock(cli->cl_close_lock);
3129 rc = client_obd_setup(obd, cfg);
3131 GOTO(err_close_lock, rc);
3133 obd->obd_vars = lprocfs_mdc_obd_vars;
3134 lprocfs_seq_obd_setup(obd);
3135 lprocfs_alloc_md_stats(obd, 0);
3137 sptlrpc_lprocfs_cliobd_attach(obd);
3138 ptlrpc_lprocfs_register_obd(obd);
3140 ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
3142 obd->obd_namespace->ns_lvbo = &inode_lvbo;
3144 rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
3147 CERROR("failed to setup llogging subsystems\n");
3153 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3157 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3161 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
3162 * us to make MDS RPCs with large enough reply buffers to hold a default
3163 * sized EA and cookie without having to calculate this (via a call into the
3164 * LOV + OSCs) each time we make an RPC. The maximum size is also tracked
3165 * but not used to avoid wastefully vmalloc()'ing large reply buffers when
3166 * a large number of stripes is possible. If a larger reply buffer is
3167 * required it will be reallocated in the ptlrpc layer due to overflow.
3169 static int mdc_init_ea_size(struct obd_export *exp, int easize,
3170 int def_easize, int cookiesize, int def_cookiesize)
3172 struct obd_device *obd = exp->exp_obd;
3173 struct client_obd *cli = &obd->u.cli;
3176 if (cli->cl_max_mds_easize < easize)
3177 cli->cl_max_mds_easize = easize;
3179 if (cli->cl_default_mds_easize < def_easize)
3180 cli->cl_default_mds_easize = def_easize;
3182 if (cli->cl_max_mds_cookiesize < cookiesize)
3183 cli->cl_max_mds_cookiesize = cookiesize;
3185 if (cli->cl_default_mds_cookiesize < def_cookiesize)
3186 cli->cl_default_mds_cookiesize = def_cookiesize;
3191 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
3197 case OBD_CLEANUP_EARLY:
3199 case OBD_CLEANUP_EXPORTS:
3200 /* Failsafe, ok if racy */
3201 if (obd->obd_type->typ_refcnt <= 1)
3202 libcfs_kkuc_group_rem(0, KUC_GRP_HSM, NULL);
3204 obd_cleanup_client_import(obd);
3205 ptlrpc_lprocfs_unregister_obd(obd);
3206 lprocfs_obd_cleanup(obd);
3207 lprocfs_free_md_stats(obd);
3209 rc = obd_llog_finish(obd, 0);
3211 CERROR("failed to cleanup llogging subsystems\n");
3217 static int mdc_cleanup(struct obd_device *obd)
3219 struct client_obd *cli = &obd->u.cli;
3221 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3222 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3226 return client_obd_cleanup(obd);
3230 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
3231 struct obd_device *tgt, int *index)
3233 struct llog_ctxt *ctxt;
3238 LASSERT(olg == &obd->obd_olg);
3240 rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt,
3245 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
3246 llog_initiator_connect(ctxt);
3247 llog_ctxt_put(ctxt);
3252 static int mdc_llog_finish(struct obd_device *obd, int count)
3254 struct llog_ctxt *ctxt;
3258 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
3260 llog_cleanup(NULL, ctxt);
3265 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
3267 struct lustre_cfg *lcfg = buf;
3268 int rc = class_process_proc_seq_param(PARAM_MDC, obd->obd_vars,
3270 return (rc > 0 ? 0: rc);
3274 /* get remote permission for current user on fid */
3275 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
3276 struct obd_capa *oc, __u32 suppgid,
3277 struct ptlrpc_request **request)
3279 struct ptlrpc_request *req;
3283 LASSERT(client_is_remote(exp));
3286 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
3290 mdc_set_capa_size(req, &RMF_CAPA1, oc);
3292 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
3294 ptlrpc_request_free(req);
3298 mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
3300 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
3301 sizeof(struct mdt_remote_perm));
3303 ptlrpc_request_set_replen(req);
3305 rc = ptlrpc_queue_wait(req);
3307 ptlrpc_req_finished(req);
3313 static int mdc_interpret_renew_capa(const struct lu_env *env,
3314 struct ptlrpc_request *req, void *args,
3317 struct mdc_renew_capa_args *ra = args;
3318 struct mdt_body *body = NULL;
3319 struct lustre_capa *capa;
3323 GOTO(out, capa = ERR_PTR(status));
3325 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3327 GOTO(out, capa = ERR_PTR(-EFAULT));
3329 if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
3330 GOTO(out, capa = ERR_PTR(-ENOENT));
3332 capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
3334 GOTO(out, capa = ERR_PTR(-EFAULT));
3337 ra->ra_cb(ra->ra_oc, capa);
3341 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3344 struct ptlrpc_request *req;
3345 struct mdc_renew_capa_args *ra;
3348 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
3349 LUSTRE_MDS_VERSION, MDS_GETATTR);
3353 /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
3354 * capa to renew is oss capa.
3356 mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
3357 ptlrpc_request_set_replen(req);
3359 CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
3360 ra = ptlrpc_req_async_args(req);
3363 req->rq_interpret_reply = mdc_interpret_renew_capa;
3364 ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
3368 struct obd_ops mdc_obd_ops = {
3369 .o_owner = THIS_MODULE,
3370 .o_setup = mdc_setup,
3371 .o_precleanup = mdc_precleanup,
3372 .o_cleanup = mdc_cleanup,
3373 .o_add_conn = client_import_add_conn,
3374 .o_del_conn = client_import_del_conn,
3375 .o_connect = client_connect_import,
3376 .o_disconnect = client_disconnect_export,
3377 .o_iocontrol = mdc_iocontrol,
3378 .o_set_info_async = mdc_set_info_async,
3379 .o_statfs = mdc_statfs,
3381 .o_unpin = mdc_unpin,
3382 .o_fid_init = client_fid_init,
3383 .o_fid_fini = client_fid_fini,
3384 .o_fid_alloc = mdc_fid_alloc,
3385 .o_import_event = mdc_import_event,
3386 .o_llog_init = mdc_llog_init,
3387 .o_llog_finish = mdc_llog_finish,
3388 .o_get_info = mdc_get_info,
3389 .o_process_config = mdc_process_config,
3390 .o_get_uuid = mdc_get_uuid,
3391 .o_quotactl = mdc_quotactl,
3392 .o_quotacheck = mdc_quotacheck
3395 struct md_ops mdc_md_ops = {
3396 .m_getstatus = mdc_getstatus,
3397 .m_null_inode = mdc_null_inode,
3398 .m_find_cbdata = mdc_find_cbdata,
3399 .m_close = mdc_close,
3400 .m_create = mdc_create,
3401 .m_done_writing = mdc_done_writing,
3402 .m_enqueue = mdc_enqueue,
3403 .m_getattr = mdc_getattr,
3404 .m_getattr_name = mdc_getattr_name,
3405 .m_intent_lock = mdc_intent_lock,
3407 .m_is_subdir = mdc_is_subdir,
3408 .m_rename = mdc_rename,
3409 .m_setattr = mdc_setattr,
3410 .m_setxattr = mdc_setxattr,
3411 .m_getxattr = mdc_getxattr,
3412 .m_fsync = mdc_fsync,
3413 .m_read_entry = mdc_read_entry,
3414 .m_unlink = mdc_unlink,
3415 .m_cancel_unused = mdc_cancel_unused,
3416 .m_init_ea_size = mdc_init_ea_size,
3417 .m_set_lock_data = mdc_set_lock_data,
3418 .m_lock_match = mdc_lock_match,
3419 .m_get_lustre_md = mdc_get_lustre_md,
3420 .m_free_lustre_md = mdc_free_lustre_md,
3421 .m_set_open_replay_data = mdc_set_open_replay_data,
3422 .m_clear_open_replay_data = mdc_clear_open_replay_data,
3423 .m_renew_capa = mdc_renew_capa,
3424 .m_unpack_capa = mdc_unpack_capa,
3425 .m_get_remote_perm = mdc_get_remote_perm,
3426 .m_intent_getattr_async = mdc_intent_getattr_async,
3427 .m_revalidate_lock = mdc_revalidate_lock
3430 int __init mdc_init(void)
3432 return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
3433 #ifndef HAVE_ONLY_PROCFS_SEQ
3436 LUSTRE_MDC_NAME, NULL);
3440 static void /*__exit*/ mdc_exit(void)
3442 class_unregister_type(LUSTRE_MDC_NAME);
3445 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3446 MODULE_DESCRIPTION("Lustre Metadata Client");
3447 MODULE_LICENSE("GPL");
3449 module_init(mdc_init);
3450 module_exit(mdc_exit);