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->valid & OBD_MD_FLMDSCAPA) {
142 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
147 *rootfid = body->fid1;
149 "root fid="DFID", last_committed="LPU64"\n",
151 lustre_msg_get_last_committed(req->rq_repmsg));
154 ptlrpc_req_finished(req);
158 /* This should be mdc_get_info("rootfid") */
159 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
160 struct obd_capa **pc)
162 return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
167 * This function now is known to always saying that it will receive 4 buffers
168 * from server. Even for cases when acl_size and md_size is zero, RPC header
169 * will contain 4 fields and RPC itself will contain zero size fields. This is
170 * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
171 * and thus zero, it shrinks it, making zero size. The same story about
172 * md_size. And this is course of problem when client waits for smaller number
173 * of fields. This issue will be fixed later when client gets aware of RPC
176 static int mdc_getattr_common(struct obd_export *exp,
177 struct ptlrpc_request *req)
179 struct req_capsule *pill = &req->rq_pill;
180 struct mdt_body *body;
185 /* Request message already built. */
186 rc = ptlrpc_queue_wait(req);
190 /* sanity check for the reply */
191 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
195 CDEBUG(D_NET, "mode: %o\n", body->mode);
197 mdc_update_max_ea_from_body(exp, body);
198 if (body->eadatasize != 0) {
199 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
205 if (body->valid & OBD_MD_FLRMTPERM) {
206 struct mdt_remote_perm *perm;
208 LASSERT(client_is_remote(exp));
209 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
210 lustre_swab_mdt_remote_perm);
215 if (body->valid & OBD_MD_FLMDSCAPA) {
216 struct lustre_capa *capa;
217 capa = req_capsule_server_get(pill, &RMF_CAPA1);
225 int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
226 struct ptlrpc_request **request)
228 struct ptlrpc_request *req;
232 /* Single MDS without an LMV case */
233 if (op_data->op_flags & MF_GET_MDT_IDX) {
238 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
242 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
244 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
246 ptlrpc_request_free(req);
250 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
251 op_data->op_valid, op_data->op_mode, -1, 0);
253 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
255 if (op_data->op_valid & OBD_MD_FLRMTPERM) {
256 LASSERT(client_is_remote(exp));
257 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
258 sizeof(struct mdt_remote_perm));
260 ptlrpc_request_set_replen(req);
262 rc = mdc_getattr_common(exp, req);
264 ptlrpc_req_finished(req);
270 int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
271 struct ptlrpc_request **request)
273 struct ptlrpc_request *req;
278 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
279 &RQF_MDS_GETATTR_NAME);
283 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
284 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
285 op_data->op_namelen + 1);
287 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
289 ptlrpc_request_free(req);
293 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
294 op_data->op_valid, op_data->op_mode,
295 op_data->op_suppgids[0], 0);
297 if (op_data->op_name) {
298 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
299 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
300 op_data->op_namelen);
301 memcpy(name, op_data->op_name, op_data->op_namelen);
304 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
306 ptlrpc_request_set_replen(req);
308 rc = mdc_getattr_common(exp, req);
310 ptlrpc_req_finished(req);
316 static int mdc_is_subdir(struct obd_export *exp,
317 const struct lu_fid *pfid,
318 const struct lu_fid *cfid,
319 struct ptlrpc_request **request)
321 struct ptlrpc_request *req;
327 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
328 &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
333 mdc_is_subdir_pack(req, pfid, cfid, 0);
334 ptlrpc_request_set_replen(req);
336 rc = ptlrpc_queue_wait(req);
337 if (rc && rc != -EREMOTE)
338 ptlrpc_req_finished(req);
344 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
345 const struct lu_fid *fid,
346 struct obd_capa *oc, int opcode, obd_valid valid,
347 const char *xattr_name, const char *input,
348 int input_size, int output_size, int flags,
349 __u32 suppgid, struct ptlrpc_request **request)
351 struct ptlrpc_request *req;
352 int xattr_namelen = 0;
358 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
362 mdc_set_capa_size(req, &RMF_CAPA1, oc);
364 xattr_namelen = strlen(xattr_name) + 1;
365 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
370 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
374 /* Flush local XATTR locks to get rid of a possible cancel RPC */
375 if (opcode == MDS_REINT && fid_is_sane(fid) &&
376 exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
377 struct list_head cancels = LIST_HEAD_INIT(cancels);
380 /* Without that packing would fail */
382 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
385 count = mdc_resource_get_unused(exp, fid,
387 MDS_INODELOCK_XATTR);
389 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
391 ptlrpc_request_free(req);
395 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
397 ptlrpc_request_free(req);
402 if (opcode == MDS_REINT) {
403 struct mdt_rec_setxattr *rec;
405 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
406 sizeof(struct mdt_rec_reint));
407 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
408 rec->sx_opcode = REINT_SETXATTR;
409 rec->sx_fsuid = from_kuid(&init_user_ns, current_fsuid());
410 rec->sx_fsgid = from_kgid(&init_user_ns, current_fsgid());
411 rec->sx_cap = cfs_curproc_cap_pack();
412 rec->sx_suppgid1 = suppgid;
413 rec->sx_suppgid2 = -1;
415 rec->sx_valid = valid | OBD_MD_FLCTIME;
416 rec->sx_time = cfs_time_current_sec();
417 rec->sx_size = output_size;
418 rec->sx_flags = flags;
420 mdc_pack_capa(req, &RMF_CAPA1, oc);
422 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
426 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
427 memcpy(tmp, xattr_name, xattr_namelen);
430 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
431 memcpy(tmp, input, input_size);
434 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
435 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
436 RCL_SERVER, output_size);
437 ptlrpc_request_set_replen(req);
440 if (opcode == MDS_REINT)
441 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
443 rc = ptlrpc_queue_wait(req);
445 if (opcode == MDS_REINT)
446 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
449 ptlrpc_req_finished(req);
455 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
456 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
457 const char *input, int input_size, int output_size,
458 int flags, __u32 suppgid, struct ptlrpc_request **request)
460 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
461 fid, oc, MDS_REINT, valid, xattr_name,
462 input, input_size, output_size, flags,
466 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
467 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
468 const char *input, int input_size, int output_size,
469 int flags, struct ptlrpc_request **request)
471 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
472 fid, oc, MDS_GETXATTR, valid, xattr_name,
473 input, input_size, output_size, flags,
477 #ifdef CONFIG_FS_POSIX_ACL
478 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
480 struct req_capsule *pill = &req->rq_pill;
481 struct mdt_body *body = md->body;
482 struct posix_acl *acl;
490 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
495 acl = posix_acl_from_xattr(&init_user_ns, buf, body->aclsize);
498 CERROR("convert xattr to acl: %d\n", rc);
502 rc = posix_acl_valid(acl);
504 CERROR("validate acl: %d\n", rc);
505 posix_acl_release(acl);
513 #define mdc_unpack_acl(req, md) 0
516 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
517 struct obd_export *dt_exp, struct obd_export *md_exp,
518 struct lustre_md *md)
520 struct req_capsule *pill = &req->rq_pill;
525 memset(md, 0, sizeof(*md));
527 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
528 LASSERT(md->body != NULL);
530 if (md->body->valid & OBD_MD_FLEASIZE) {
532 struct lov_mds_md *lmm;
534 if (!S_ISREG(md->body->mode)) {
535 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
536 "regular file, but is not\n");
537 GOTO(out, rc = -EPROTO);
540 if (md->body->eadatasize == 0) {
541 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
542 "but eadatasize 0\n");
543 GOTO(out, rc = -EPROTO);
545 lmmsize = md->body->eadatasize;
546 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
548 GOTO(out, rc = -EPROTO);
550 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
554 if (rc < sizeof(*md->lsm)) {
555 CDEBUG(D_INFO, "lsm size too small: "
556 "rc < sizeof (*md->lsm) (%d < %d)\n",
557 rc, (int)sizeof(*md->lsm));
558 GOTO(out, rc = -EPROTO);
561 } else if (md->body->valid & OBD_MD_FLDIREA) {
563 struct lov_mds_md *lmv;
565 if(!S_ISDIR(md->body->mode)) {
566 CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
567 "directory, but is not\n");
568 GOTO(out, rc = -EPROTO);
571 if (md->body->eadatasize == 0) {
572 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
573 "but eadatasize 0\n");
576 if (md->body->valid & OBD_MD_MEA) {
577 lmvsize = md->body->eadatasize;
578 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
581 GOTO(out, rc = -EPROTO);
583 rc = obd_unpackmd(md_exp, (void *)&md->lmv, lmv,
588 if (rc < sizeof(*md->lmv)) {
589 CDEBUG(D_INFO, "size too small: "
590 "rc < sizeof(*md->lmv) (%d < %d)\n",
591 rc, (int)sizeof(*md->lmv));
592 GOTO(out, rc = -EPROTO);
598 if (md->body->valid & OBD_MD_FLRMTPERM) {
599 /* remote permission */
600 LASSERT(client_is_remote(exp));
601 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
602 lustre_swab_mdt_remote_perm);
603 if (!md->remote_perm)
604 GOTO(out, rc = -EPROTO);
606 else if (md->body->valid & OBD_MD_FLACL) {
607 /* for ACL, it's possible that FLACL is set but aclsize is zero.
608 * only when aclsize != 0 there's an actual segment for ACL
611 if (md->body->aclsize) {
612 rc = mdc_unpack_acl(req, md);
615 #ifdef CONFIG_FS_POSIX_ACL
617 md->posix_acl = NULL;
621 if (md->body->valid & OBD_MD_FLMDSCAPA) {
622 struct obd_capa *oc = NULL;
624 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
630 if (md->body->valid & OBD_MD_FLOSSCAPA) {
631 struct obd_capa *oc = NULL;
633 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
643 capa_put(md->oss_capa);
647 capa_put(md->mds_capa);
650 #ifdef CONFIG_FS_POSIX_ACL
651 posix_acl_release(md->posix_acl);
654 obd_free_memmd(dt_exp, &md->lsm);
659 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
666 * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
669 void mdc_replay_open(struct ptlrpc_request *req)
671 struct md_open_data *mod = req->rq_cb_data;
672 struct ptlrpc_request *close_req;
673 struct obd_client_handle *och;
674 struct lustre_handle old;
675 struct mdt_body *body;
679 DEBUG_REQ(D_ERROR, req,
680 "Can't properly replay without open data.");
685 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
686 LASSERT(body != NULL);
690 struct lustre_handle *file_fh;
692 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
694 file_fh = &och->och_fh;
695 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
696 file_fh->cookie, body->handle.cookie);
698 *file_fh = body->handle;
700 close_req = mod->mod_close_req;
701 if (close_req != NULL) {
702 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
703 struct mdt_ioepoch *epoch;
705 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
706 epoch = req_capsule_client_get(&close_req->rq_pill,
711 LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
712 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
713 epoch->handle = body->handle;
718 void mdc_commit_open(struct ptlrpc_request *req)
720 struct md_open_data *mod = req->rq_cb_data;
725 * No need to touch md_open_data::mod_och, it holds a reference on
726 * \var mod and will zero references to each other, \var mod will be
727 * freed after that when md_open_data::mod_och will put the reference.
731 * Do not let open request to disappear as it still may be needed
732 * for close rpc to happen (it may happen on evict only, otherwise
733 * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
734 * called), just mark this rpc as committed to distinguish these 2
735 * cases, see mdc_close() for details. The open request reference will
736 * be put along with freeing \var mod.
738 ptlrpc_request_addref(req);
739 spin_lock(&req->rq_lock);
740 req->rq_committed = 1;
741 spin_unlock(&req->rq_lock);
742 req->rq_cb_data = NULL;
746 int mdc_set_open_replay_data(struct obd_export *exp,
747 struct obd_client_handle *och,
748 struct lookup_intent *it)
750 struct md_open_data *mod;
751 struct mdt_rec_create *rec;
752 struct mdt_body *body;
753 struct ptlrpc_request *open_req = it->d.lustre.it_data;
754 struct obd_import *imp = open_req->rq_import;
757 if (!open_req->rq_replay)
760 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
761 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
762 LASSERT(rec != NULL);
763 /* Incoming message in my byte order (it's been swabbed). */
764 /* Outgoing messages always in my byte order. */
765 LASSERT(body != NULL);
767 /* Only if the import is replayable, we set replay_open data */
768 if (och && imp->imp_replayable) {
769 mod = obd_mod_alloc();
771 DEBUG_REQ(D_ERROR, open_req,
772 "Can't allocate md_open_data");
777 * Take a reference on \var mod, to be freed on mdc_close().
778 * It protects \var mod from being freed on eviction (commit
779 * callback is called despite rq_replay flag).
780 * Another reference for \var och.
785 spin_lock(&open_req->rq_lock);
788 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
789 it_disposition(it, DISP_OPEN_STRIPE);
790 mod->mod_open_req = open_req;
791 open_req->rq_cb_data = mod;
792 open_req->rq_commit_cb = mdc_commit_open;
793 spin_unlock(&open_req->rq_lock);
796 rec->cr_fid2 = body->fid1;
797 rec->cr_ioepoch = body->ioepoch;
798 rec->cr_old_handle.cookie = body->handle.cookie;
799 open_req->rq_replay_cb = mdc_replay_open;
800 if (!fid_is_sane(&body->fid1)) {
801 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
806 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
810 static void mdc_free_open(struct md_open_data *mod)
814 if (mod->mod_is_create == 0 &&
815 imp_connect_disp_stripe(mod->mod_open_req->rq_import))
818 LASSERT(mod->mod_open_req->rq_replay == 0);
820 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
822 ptlrpc_request_committed(mod->mod_open_req, committed);
823 if (mod->mod_close_req)
824 ptlrpc_request_committed(mod->mod_close_req, committed);
827 int mdc_clear_open_replay_data(struct obd_export *exp,
828 struct obd_client_handle *och)
830 struct md_open_data *mod = och->och_mod;
834 * It is possible to not have \var mod in a case of eviction between
835 * lookup and ll_file_open().
840 LASSERT(mod != LP_POISON);
841 LASSERT(mod->mod_open_req != NULL);
851 /* Prepares the request for the replay by the given reply */
852 static void mdc_close_handle_reply(struct ptlrpc_request *req,
853 struct md_op_data *op_data, int rc) {
854 struct mdt_body *repbody;
855 struct mdt_ioepoch *epoch;
857 if (req && rc == -EAGAIN) {
858 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
859 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
861 epoch->flags |= MF_SOM_AU;
862 if (repbody->valid & OBD_MD_FLGETATTRLOCK)
863 op_data->op_flags |= MF_GETATTR_LOCK;
867 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
868 struct md_open_data *mod, struct ptlrpc_request **request)
870 struct obd_device *obd = class_exp2obd(exp);
871 struct ptlrpc_request *req;
872 struct req_format *req_fmt;
877 req_fmt = &RQF_MDS_CLOSE;
878 if (op_data->op_bias & MDS_HSM_RELEASE) {
879 req_fmt = &RQF_MDS_RELEASE_CLOSE;
881 /* allocate a FID for volatile file */
882 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
884 CERROR("%s: "DFID" failed to allocate FID: %d\n",
885 obd->obd_name, PFID(&op_data->op_fid1), rc);
886 /* save the errcode and proceed to close */
892 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
896 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
898 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
900 ptlrpc_request_free(req);
904 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
905 * portal whose threads are not taking any DLM locks and are therefore
906 * always progressing */
907 req->rq_request_portal = MDS_READPAGE_PORTAL;
908 ptlrpc_at_set_req_timeout(req);
910 /* Ensure that this close's handle is fixed up during replay. */
911 if (likely(mod != NULL)) {
912 LASSERTF(mod->mod_open_req != NULL &&
913 mod->mod_open_req->rq_type != LI_POISON,
914 "POISONED open %p!\n", mod->mod_open_req);
916 mod->mod_close_req = req;
918 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
919 /* We no longer want to preserve this open for replay even
920 * though the open was committed. b=3632, b=3633 */
921 spin_lock(&mod->mod_open_req->rq_lock);
922 mod->mod_open_req->rq_replay = 0;
923 spin_unlock(&mod->mod_open_req->rq_lock);
925 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
928 mdc_close_pack(req, op_data);
930 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
931 obd->u.cli.cl_default_mds_easize);
932 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
933 obd->u.cli.cl_default_mds_cookiesize);
935 ptlrpc_request_set_replen(req);
937 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
938 rc = ptlrpc_queue_wait(req);
939 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
941 if (req->rq_repmsg == NULL) {
942 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
945 rc = req->rq_status ?: -EIO;
946 } else if (rc == 0 || rc == -EAGAIN) {
947 struct mdt_body *body;
949 rc = lustre_msg_get_status(req->rq_repmsg);
950 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
951 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
956 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
959 } else if (rc == -ESTALE) {
961 * it can be allowed error after 3633 if open was committed and
962 * server failed before close was sent. Let's check if mod
963 * exists and return no error in that case
966 DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
967 LASSERT(mod->mod_open_req != NULL);
968 if (mod->mod_open_req->rq_committed)
975 mod->mod_close_req = NULL;
976 /* Since now, mod is accessed through open_req only,
977 * thus close req does not keep a reference on mod anymore. */
981 mdc_close_handle_reply(req, op_data, rc);
982 RETURN(rc < 0 ? rc : saved_rc);
985 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
986 struct md_open_data *mod)
988 struct obd_device *obd = class_exp2obd(exp);
989 struct ptlrpc_request *req;
993 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
994 &RQF_MDS_DONE_WRITING);
998 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
999 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
1001 ptlrpc_request_free(req);
1006 LASSERTF(mod->mod_open_req != NULL &&
1007 mod->mod_open_req->rq_type != LI_POISON,
1008 "POISONED setattr %p!\n", mod->mod_open_req);
1010 mod->mod_close_req = req;
1011 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
1012 /* We no longer want to preserve this setattr for replay even
1013 * though the open was committed. b=3632, b=3633 */
1014 spin_lock(&mod->mod_open_req->rq_lock);
1015 mod->mod_open_req->rq_replay = 0;
1016 spin_unlock(&mod->mod_open_req->rq_lock);
1019 mdc_close_pack(req, op_data);
1020 ptlrpc_request_set_replen(req);
1022 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
1023 rc = ptlrpc_queue_wait(req);
1024 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
1026 if (rc == -ESTALE) {
1028 * it can be allowed error after 3633 if open or setattr were
1029 * committed and server failed before close was sent.
1030 * Let's check if mod exists and return no error in that case
1033 LASSERT(mod->mod_open_req != NULL);
1034 if (mod->mod_open_req->rq_committed)
1041 mod->mod_close_req = NULL;
1042 LASSERT(mod->mod_open_req != NULL);
1045 /* Since now, mod is accessed through setattr req only,
1046 * thus DW req does not keep a reference on mod anymore. */
1050 mdc_close_handle_reply(req, op_data, rc);
1051 ptlrpc_req_finished(req);
1055 #ifdef HAVE_SPLIT_SUPPORT
1056 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
1057 const struct page *page, int offset)
1059 struct ptlrpc_request *req;
1060 struct ptlrpc_bulk_desc *desc;
1064 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
1068 /* FIXME: capa doesn't support split yet */
1069 mdc_set_capa_size(req, &RMF_CAPA1, NULL);
1071 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
1073 ptlrpc_request_free(req);
1077 req->rq_request_portal = MDS_READPAGE_PORTAL;
1078 ptlrpc_at_set_req_timeout(req);
1080 desc = ptlrpc_prep_bulk_imp(req, 1, 1,BULK_GET_SOURCE, MDS_BULK_PORTAL);
1082 GOTO(out, rc = -ENOMEM);
1084 /* NB req now owns desc and will free it when it gets freed. */
1085 ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
1086 mdc_readdir_pack(req, 0, offset, fid, NULL);
1088 ptlrpc_request_set_replen(req);
1089 rc = ptlrpc_queue_wait(req);
1093 rc = sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk);
1095 ptlrpc_req_finished(req);
1098 EXPORT_SYMBOL(mdc_sendpage);
1101 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
1102 __u64 offset, struct obd_capa *oc,
1103 struct page **pages, int npages,
1104 struct ptlrpc_request **request)
1106 struct ptlrpc_request *req;
1107 struct ptlrpc_bulk_desc *desc;
1109 wait_queue_head_t waitq;
1111 struct l_wait_info lwi;
1116 init_waitqueue_head(&waitq);
1119 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1123 mdc_set_capa_size(req, &RMF_CAPA1, oc);
1125 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1127 ptlrpc_request_free(req);
1131 req->rq_request_portal = MDS_READPAGE_PORTAL;
1132 ptlrpc_at_set_req_timeout(req);
1134 desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK,
1137 ptlrpc_request_free(req);
1141 /* NB req now owns desc and will free it when it gets freed */
1142 for (i = 0; i < npages; i++)
1143 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
1145 mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE * npages, fid, oc);
1147 ptlrpc_request_set_replen(req);
1148 rc = ptlrpc_queue_wait(req);
1150 ptlrpc_req_finished(req);
1151 if (rc != -ETIMEDOUT)
1155 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1156 CERROR("%s: too many resend retries: rc = %d\n",
1157 exp->exp_obd->obd_name, -EIO);
1160 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
1162 l_wait_event(waitq, 0, &lwi);
1167 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1168 req->rq_bulk->bd_nob_transferred);
1170 ptlrpc_req_finished(req);
1174 if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1175 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1176 exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1177 PAGE_CACHE_SIZE * npages);
1178 ptlrpc_req_finished(req);
1187 static void mdc_release_page(struct page *page, int remove)
1191 if (likely(page->mapping != NULL))
1192 truncate_complete_page(page->mapping, page);
1195 page_cache_release(page);
1198 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1199 __u64 *start, __u64 *end, int hash64)
1202 * Complement of hash is used as an index so that
1203 * radix_tree_gang_lookup() can be used to find a page with starting
1204 * hash _smaller_ than one we are looking for.
1206 unsigned long offset = hash_x_index(*hash, hash64);
1210 spin_lock_irq(&mapping->tree_lock);
1211 found = radix_tree_gang_lookup(&mapping->page_tree,
1212 (void **)&page, offset, 1);
1214 struct lu_dirpage *dp;
1216 page_cache_get(page);
1217 spin_unlock_irq(&mapping->tree_lock);
1219 * In contrast to find_lock_page() we are sure that directory
1220 * page cannot be truncated (while DLM lock is held) and,
1221 * hence, can avoid restart.
1223 * In fact, page cannot be locked here at all, because
1224 * mdc_read_page_remote does synchronous io.
1226 wait_on_page_locked(page);
1227 if (PageUptodate(page)) {
1229 if (BITS_PER_LONG == 32 && hash64) {
1230 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1231 *end = le64_to_cpu(dp->ldp_hash_end) >> 32;
1232 *hash = *hash >> 32;
1234 *start = le64_to_cpu(dp->ldp_hash_start);
1235 *end = le64_to_cpu(dp->ldp_hash_end);
1237 if (unlikely(*start == 1 && *hash == 0))
1240 LASSERTF(*start <= *hash, "start = "LPX64
1241 ",end = "LPX64",hash = "LPX64"\n",
1242 *start, *end, *hash);
1243 CDEBUG(D_VFSTRACE, "offset %lx ["LPX64" "LPX64"],"
1244 " hash "LPX64"\n", offset, *start, *end, *hash);
1247 mdc_release_page(page, 0);
1249 } else if (*end != *start && *hash == *end) {
1251 * upon hash collision, remove this page,
1252 * otherwise put page reference, and
1253 * ll_get_dir_page() will issue RPC to fetch
1257 mdc_release_page(page,
1258 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1262 page_cache_release(page);
1263 page = ERR_PTR(-EIO);
1266 spin_unlock_irq(&mapping->tree_lock);
1273 * Adjust a set of pages, each page containing an array of lu_dirpages,
1274 * so that each page can be used as a single logical lu_dirpage.
1276 * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1277 * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1278 * struct lu_dirent. It has size up to LU_PAGE_SIZE. The ldp_hash_end
1279 * value is used as a cookie to request the next lu_dirpage in a
1280 * directory listing that spans multiple pages (two in this example):
1283 * .|--------v------- -----.
1284 * |s|e|f|p|ent|ent| ... |ent|
1285 * '--|-------------- -----' Each CFS_PAGE contains a single
1286 * '------. lu_dirpage.
1287 * .---------v------- -----.
1288 * |s|e|f|p|ent| 0 | ... | 0 |
1289 * '----------------- -----'
1291 * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
1292 * larger than LU_PAGE_SIZE, a single host page may contain multiple
1293 * lu_dirpages. After reading the lu_dirpages from the MDS, the
1294 * ldp_hash_end of the first lu_dirpage refers to the one immediately
1295 * after it in the same CFS_PAGE (arrows simplified for brevity, but
1296 * in general e0==s1, e1==s2, etc.):
1298 * .-------------------- -----.
1299 * |s0|e0|f0|p|ent|ent| ... |ent|
1300 * |---v---------------- -----|
1301 * |s1|e1|f1|p|ent|ent| ... |ent|
1302 * |---v---------------- -----| Here, each CFS_PAGE contains
1303 * ... multiple lu_dirpages.
1304 * |---v---------------- -----|
1305 * |s'|e'|f'|p|ent|ent| ... |ent|
1306 * '---|---------------- -----'
1308 * .----------------------------.
1311 * This structure is transformed into a single logical lu_dirpage as follows:
1313 * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1314 * labeled 'next CFS_PAGE'.
1316 * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1317 * a hash collision with the next page exists.
1319 * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1320 * to the first entry of the next lu_dirpage.
1322 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
1323 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1327 for (i = 0; i < cfs_pgs; i++) {
1328 struct lu_dirpage *dp = kmap(pages[i]);
1329 struct lu_dirpage *first = dp;
1330 struct lu_dirent *end_dirent = NULL;
1331 struct lu_dirent *ent;
1332 __u64 hash_end = le64_to_cpu(dp->ldp_hash_end);
1333 __u32 flags = le32_to_cpu(dp->ldp_flags);
1335 while (--lu_pgs > 0) {
1336 ent = lu_dirent_start(dp);
1337 for (end_dirent = ent; ent != NULL;
1338 end_dirent = ent, ent = lu_dirent_next(ent));
1340 /* Advance dp to next lu_dirpage. */
1341 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1343 /* Check if we've reached the end of the CFS_PAGE. */
1344 if (!((unsigned long)dp & ~CFS_PAGE_MASK))
1347 /* Save the hash and flags of this lu_dirpage. */
1348 hash_end = le64_to_cpu(dp->ldp_hash_end);
1349 flags = le32_to_cpu(dp->ldp_flags);
1351 /* Check if lu_dirpage contains no entries. */
1352 if (end_dirent == NULL)
1355 /* Enlarge the end entry lde_reclen from 0 to
1356 * first entry of next lu_dirpage. */
1357 LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1358 end_dirent->lde_reclen =
1359 cpu_to_le16((char *)(dp->ldp_entries) -
1360 (char *)end_dirent);
1363 first->ldp_hash_end = hash_end;
1364 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1365 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1369 LASSERTF(lu_pgs == 0, "left = %d", lu_pgs);
1372 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1373 #endif /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
1375 /* parameters for readdir page */
1376 struct readpage_param {
1377 struct md_op_data *rp_mod;
1380 struct obd_export *rp_exp;
1381 struct md_callback *rp_cb;
1385 * Read pages from server.
1387 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1388 * a header lu_dirpage which describes the start/end hash, and whether this
1389 * page is empty (contains no dir entry) or hash collide with next page.
1390 * After client receives reply, several pages will be integrated into dir page
1391 * in CFS_PAGE_SIZE (if CFS_PAGE_SIZE greater than LU_PAGE_SIZE), and the
1392 * lu_dirpage for this integrated page will be adjusted.
1394 static int mdc_read_page_remote(void *data, struct page *page0)
1396 struct readpage_param *rp = data;
1397 struct page **page_pool;
1399 struct lu_dirpage *dp;
1400 int rd_pgs = 0; /* number of pages read actually */
1402 struct md_op_data *op_data = rp->rp_mod;
1403 struct ptlrpc_request *req;
1404 int max_pages = op_data->op_max_pages;
1405 struct inode *inode;
1411 LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
1412 if (op_data->op_mea1 != NULL) {
1413 __u32 index = op_data->op_stripe_offset;
1415 inode = op_data->op_mea1->lsm_md_oinfo[index].lmo_root;
1416 fid = &op_data->op_mea1->lsm_md_oinfo[index].lmo_fid;
1418 inode = op_data->op_data;
1419 fid = &op_data->op_fid1;
1421 LASSERT(inode != NULL);
1423 OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1424 if (page_pool != NULL) {
1425 page_pool[0] = page0;
1431 for (npages = 1; npages < max_pages; npages++) {
1432 page = page_cache_alloc_cold(inode->i_mapping);
1435 page_pool[npages] = page;
1438 rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, op_data->op_capa1,
1439 page_pool, npages, &req);
1443 rd_pgs = (req->rq_bulk->bd_nob_transferred +
1444 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1445 lu_pgs = req->rq_bulk->bd_nob_transferred >>
1447 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1449 CDEBUG(D_INODE, "read %d(%d)/%d pages\n", rd_pgs, lu_pgs,
1450 op_data->op_npages);
1452 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1454 SetPageUptodate(page0);
1458 ptlrpc_req_finished(req);
1459 CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1460 for (i = 1; i < npages; i++) {
1461 unsigned long offset;
1465 page = page_pool[i];
1467 if (rc < 0 || i >= rd_pgs) {
1468 page_cache_release(page);
1472 SetPageUptodate(page);
1475 hash = le64_to_cpu(dp->ldp_hash_start);
1478 offset = hash_x_index(hash, rp->rp_hash64);
1480 prefetchw(&page->flags);
1481 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1486 CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1487 " rc = %d\n", offset, ret);
1488 page_cache_release(page);
1491 if (page_pool != &page0)
1492 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1498 * Read dir page from cache first, if it can not find it, read it from
1499 * server and add into the cache.
1501 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1502 struct md_callback *cb_op, struct page **ppage)
1504 struct lookup_intent it = { .it_op = IT_READDIR };
1506 struct inode *dir = op_data->op_data;
1507 struct address_space *mapping;
1508 struct lu_dirpage *dp;
1511 struct lustre_handle lockh;
1512 struct ptlrpc_request *enq_req = NULL;
1513 struct readpage_param rp_param;
1520 LASSERT(dir != NULL);
1521 mapping = dir->i_mapping;
1523 rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1524 cb_op->md_blocking_ast, 0);
1525 if (enq_req != NULL)
1526 ptlrpc_req_finished(enq_req);
1529 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1530 exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1535 mdc_set_lock_data(exp, &it.d.lustre.it_lock_handle, dir, NULL);
1537 rp_param.rp_off = op_data->op_hash_offset;
1538 rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1539 page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1540 rp_param.rp_hash64);
1542 CERROR("%s: dir page locate: "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));
1546 } else if (page != NULL) {
1548 * XXX nikita: not entirely correct handling of a corner case:
1549 * suppose hash chain of entries with hash value HASH crosses
1550 * border between pages P0 and P1. First both P0 and P1 are
1551 * cached, seekdir() is called for some entry from the P0 part
1552 * of the chain. Later P0 goes out of cache. telldir(HASH)
1553 * happens and finds P1, as it starts with matching hash
1554 * value. Remaining entries from P0 part of the chain are
1555 * skipped. (Is that really a bug?)
1557 * Possible solutions: 0. don't cache P1 is such case, handle
1558 * it as an "overflow" page. 1. invalidate all pages at
1559 * once. 2. use HASH|1 as an index for P1.
1561 GOTO(hash_collision, page);
1564 rp_param.rp_exp = exp;
1565 rp_param.rp_mod = op_data;
1566 page = read_cache_page(mapping,
1567 hash_x_index(rp_param.rp_off,
1568 rp_param.rp_hash64),
1569 mdc_read_page_remote, &rp_param);
1571 CERROR("%s: read cache page: "DFID" at "LPU64": rc %ld\n",
1572 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1573 rp_param.rp_off, PTR_ERR(page));
1574 GOTO(out_unlock, rc = PTR_ERR(page));
1577 wait_on_page_locked(page);
1579 if (!PageUptodate(page)) {
1580 CERROR("%s: page not updated: "DFID" at "LPU64": rc %d\n",
1581 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1582 rp_param.rp_off, -5);
1585 if (!PageChecked(page))
1586 SetPageChecked(page);
1587 if (PageError(page)) {
1588 CERROR("%s: page error: "DFID" at "LPU64": rc %d\n",
1589 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1590 rp_param.rp_off, -5);
1595 dp = page_address(page);
1596 if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1597 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1598 end = le64_to_cpu(dp->ldp_hash_end) >> 32;
1599 rp_param.rp_off = op_data->op_hash_offset >> 32;
1601 start = le64_to_cpu(dp->ldp_hash_start);
1602 end = le64_to_cpu(dp->ldp_hash_end);
1603 rp_param.rp_off = op_data->op_hash_offset;
1606 LASSERT(start == rp_param.rp_off);
1607 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1608 #if BITS_PER_LONG == 32
1609 CWARN("Real page-wide hash collision at ["LPU64" "LPU64"] with "
1610 "hash "LPU64"\n", le64_to_cpu(dp->ldp_hash_start),
1611 le64_to_cpu(dp->ldp_hash_end), op_data->op_hash_offset);
1615 * Fetch whole overflow chain...
1623 lockh.cookie = it.d.lustre.it_lock_handle;
1624 ldlm_lock_decref(&lockh, it.d.lustre.it_lock_mode);
1625 it.d.lustre.it_lock_handle = 0;
1629 mdc_release_page(page, 1);
1635 * Read one directory entry from the cache.
1637 int mdc_read_entry(struct obd_export *exp, struct md_op_data *op_data,
1638 struct md_callback *cb_op, struct lu_dirent **entp,
1639 struct page **ppage)
1641 struct page *page = NULL;
1642 struct lu_dirpage *dp;
1643 struct lu_dirent *ent;
1645 __u32 same_hash_count;
1646 __u64 hash_offset = op_data->op_hash_offset;
1649 CDEBUG(D_INFO, DFID " offset = "LPU64", flags %#x\n",
1650 PFID(&op_data->op_fid1), op_data->op_hash_offset,
1651 op_data->op_cli_flags);
1656 if (op_data->op_hash_offset == MDS_DIR_END_OFF)
1659 rc = mdc_read_page(exp, op_data, cb_op, &page);
1663 /* same_hash_count means how many entries with this
1664 * hash value has been read */
1665 same_hash_count = op_data->op_same_hash_offset + 1;
1666 dp = page_address(page);
1667 for (ent = lu_dirent_start(dp); ent != NULL;
1668 ent = lu_dirent_next(ent)) {
1669 /* Skip dummy entry */
1670 if (le16_to_cpu(ent->lde_namelen) == 0)
1673 if (le64_to_cpu(ent->lde_hash) <
1674 op_data->op_hash_offset)
1677 if (unlikely(le64_to_cpu(ent->lde_hash) ==
1678 op_data->op_hash_offset)) {
1679 /* If it is not for next entry, which usually from
1680 * ll_dir_entry_start, return this entry. */
1681 if (!(op_data->op_cli_flags & CLI_NEXT_ENTRY))
1684 /* Keep reading until all of entries being read are
1686 if (same_hash_count > 0) {
1694 /* If it can not find entry in current page, try next page. */
1696 if (le64_to_cpu(dp->ldp_hash_end) == MDS_DIR_END_OFF) {
1697 op_data->op_same_hash_offset = 0;
1698 mdc_release_page(page,
1699 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1703 op_data->op_hash_offset = le64_to_cpu(dp->ldp_hash_end);
1704 mdc_release_page(page,
1705 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1706 rc = mdc_read_page(exp, op_data, cb_op, &page);
1711 dp = page_address(page);
1712 ent = lu_dirent_start(dp);
1716 /* If the next hash is the same as the current hash, increase
1717 * the op_same_hash_offset to resolve the same hash conflict */
1718 if (ent != NULL && op_data->op_cli_flags & CLI_NEXT_ENTRY) {
1719 if (unlikely(le64_to_cpu(ent->lde_hash) == hash_offset))
1720 op_data->op_same_hash_offset++;
1722 op_data->op_same_hash_offset = 0;
1730 #else /* __KERNEL__ */
1733 *mdc_read_page_remote(struct obd_export *exp, const struct lmv_oinfo *lmo,
1734 const __u64 hash, struct obd_capa *oc)
1736 struct ptlrpc_request *req = NULL;
1740 OBD_PAGE_ALLOC(page, 0);
1742 return ERR_PTR(-ENOMEM);
1744 rc = mdc_getpage(exp, &lmo->lmo_fid, hash, oc, &page, 1, &req);
1746 ptlrpc_req_finished(req);
1749 OBD_PAGE_FREE(page);
1756 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1757 struct md_callback *cb_op,
1758 struct page **ppage)
1761 struct lmv_oinfo *lmo;
1764 /* No local cache for liblustre, always read entry remotely */
1765 lmo = &op_data->op_mea1->lsm_md_oinfo[op_data->op_stripe_offset];
1766 page = mdc_read_page_remote(exp, lmo, op_data->op_hash_offset,
1769 return PTR_ERR(page);
1776 int mdc_read_entry(struct obd_export *exp, struct md_op_data *op_data,
1777 struct md_callback *cb_op, struct lu_dirent **entp,
1778 struct page **ppage)
1780 struct page *page = NULL;
1781 struct lu_dirpage *dp;
1782 struct lu_dirent *ent;
1786 rc = mdc_read_page(exp, op_data, cb_op, &page);
1790 dp = page_address(page);
1791 if (le64_to_cpu(dp->ldp_hash_end) < op_data->op_hash_offset)
1792 GOTO(out, *entp = NULL);
1794 for (ent = lu_dirent_start(dp); ent != NULL;
1795 ent = lu_dirent_next(ent))
1796 if (le64_to_cpu(ent->lde_hash) >= op_data->op_hash_offset)
1801 OBD_PAGE_FREE(page);
1807 static int mdc_statfs(const struct lu_env *env,
1808 struct obd_export *exp, struct obd_statfs *osfs,
1809 __u64 max_age, __u32 flags)
1811 struct obd_device *obd = class_exp2obd(exp);
1812 struct ptlrpc_request *req;
1813 struct obd_statfs *msfs;
1814 struct obd_import *imp = NULL;
1819 * Since the request might also come from lprocfs, so we need
1820 * sync this with client_disconnect_export Bug15684
1822 down_read(&obd->u.cli.cl_sem);
1823 if (obd->u.cli.cl_import)
1824 imp = class_import_get(obd->u.cli.cl_import);
1825 up_read(&obd->u.cli.cl_sem);
1829 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1830 LUSTRE_MDS_VERSION, MDS_STATFS);
1832 GOTO(output, rc = -ENOMEM);
1834 ptlrpc_request_set_replen(req);
1836 if (flags & OBD_STATFS_NODELAY) {
1837 /* procfs requests not want stay in wait for avoid deadlock */
1838 req->rq_no_resend = 1;
1839 req->rq_no_delay = 1;
1842 rc = ptlrpc_queue_wait(req);
1844 /* check connection error first */
1845 if (imp->imp_connect_error)
1846 rc = imp->imp_connect_error;
1850 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1852 GOTO(out, rc = -EPROTO);
1857 ptlrpc_req_finished(req);
1859 class_import_put(imp);
1863 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1865 __u32 keylen, vallen;
1869 if (gf->gf_pathlen > PATH_MAX)
1870 RETURN(-ENAMETOOLONG);
1871 if (gf->gf_pathlen < 2)
1874 /* Key is KEY_FID2PATH + getinfo_fid2path description */
1875 keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1876 OBD_ALLOC(key, keylen);
1879 memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1880 memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1882 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1883 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1885 if (!fid_is_sane(&gf->gf_fid))
1886 GOTO(out, rc = -EINVAL);
1888 /* Val is struct getinfo_fid2path result plus path */
1889 vallen = sizeof(*gf) + gf->gf_pathlen;
1891 rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1892 if (rc != 0 && rc != -EREMOTE)
1895 if (vallen <= sizeof(*gf))
1896 GOTO(out, rc = -EPROTO);
1897 else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1898 GOTO(out, rc = -EOVERFLOW);
1900 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1901 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1904 OBD_FREE(key, keylen);
1908 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1909 struct hsm_progress_kernel *hpk)
1911 struct obd_import *imp = class_exp2cliimp(exp);
1912 struct hsm_progress_kernel *req_hpk;
1913 struct ptlrpc_request *req;
1917 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1918 LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1920 GOTO(out, rc = -ENOMEM);
1922 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1924 /* Copy hsm_progress struct */
1925 req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1926 if (req_hpk == NULL)
1927 GOTO(out, rc = -EPROTO);
1930 req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1932 ptlrpc_request_set_replen(req);
1934 rc = mdc_queue_wait(req);
1937 ptlrpc_req_finished(req);
1941 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1943 __u32 *archive_mask;
1944 struct ptlrpc_request *req;
1948 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1950 MDS_HSM_CT_REGISTER);
1952 GOTO(out, rc = -ENOMEM);
1954 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1956 /* Copy hsm_progress struct */
1957 archive_mask = req_capsule_client_get(&req->rq_pill,
1958 &RMF_MDS_HSM_ARCHIVE);
1959 if (archive_mask == NULL)
1960 GOTO(out, rc = -EPROTO);
1962 *archive_mask = archives;
1964 ptlrpc_request_set_replen(req);
1966 rc = mdc_queue_wait(req);
1969 ptlrpc_req_finished(req);
1973 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1974 struct md_op_data *op_data)
1976 struct hsm_current_action *hca = op_data->op_data;
1977 struct hsm_current_action *req_hca;
1978 struct ptlrpc_request *req;
1982 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1983 &RQF_MDS_HSM_ACTION);
1987 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1989 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1991 ptlrpc_request_free(req);
1995 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1996 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1998 ptlrpc_request_set_replen(req);
2000 rc = mdc_queue_wait(req);
2004 req_hca = req_capsule_server_get(&req->rq_pill,
2005 &RMF_MDS_HSM_CURRENT_ACTION);
2006 if (req_hca == NULL)
2007 GOTO(out, rc = -EPROTO);
2013 ptlrpc_req_finished(req);
2017 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
2019 struct ptlrpc_request *req;
2023 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
2025 MDS_HSM_CT_UNREGISTER);
2027 GOTO(out, rc = -ENOMEM);
2029 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
2031 ptlrpc_request_set_replen(req);
2033 rc = mdc_queue_wait(req);
2036 ptlrpc_req_finished(req);
2040 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
2041 struct md_op_data *op_data)
2043 struct hsm_user_state *hus = op_data->op_data;
2044 struct hsm_user_state *req_hus;
2045 struct ptlrpc_request *req;
2049 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2050 &RQF_MDS_HSM_STATE_GET);
2054 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2056 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
2058 ptlrpc_request_free(req);
2062 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
2063 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
2065 ptlrpc_request_set_replen(req);
2067 rc = mdc_queue_wait(req);
2071 req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
2072 if (req_hus == NULL)
2073 GOTO(out, rc = -EPROTO);
2079 ptlrpc_req_finished(req);
2083 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
2084 struct md_op_data *op_data)
2086 struct hsm_state_set *hss = op_data->op_data;
2087 struct hsm_state_set *req_hss;
2088 struct ptlrpc_request *req;
2092 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2093 &RQF_MDS_HSM_STATE_SET);
2097 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2099 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
2101 ptlrpc_request_free(req);
2105 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
2106 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
2109 req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
2110 if (req_hss == NULL)
2111 GOTO(out, rc = -EPROTO);
2114 ptlrpc_request_set_replen(req);
2116 rc = mdc_queue_wait(req);
2121 ptlrpc_req_finished(req);
2125 static int mdc_ioc_hsm_request(struct obd_export *exp,
2126 struct hsm_user_request *hur)
2128 struct obd_import *imp = class_exp2cliimp(exp);
2129 struct ptlrpc_request *req;
2130 struct hsm_request *req_hr;
2131 struct hsm_user_item *req_hui;
2136 req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
2138 GOTO(out, rc = -ENOMEM);
2140 req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
2141 hur->hur_request.hr_itemcount
2142 * sizeof(struct hsm_user_item));
2143 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
2144 hur->hur_request.hr_data_len);
2146 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
2148 ptlrpc_request_free(req);
2152 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
2154 /* Copy hsm_request struct */
2155 req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
2157 GOTO(out, rc = -EPROTO);
2158 *req_hr = hur->hur_request;
2160 /* Copy hsm_user_item structs */
2161 req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
2162 if (req_hui == NULL)
2163 GOTO(out, rc = -EPROTO);
2164 memcpy(req_hui, hur->hur_user_item,
2165 hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
2167 /* Copy opaque field */
2168 req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
2169 if (req_opaque == NULL)
2170 GOTO(out, rc = -EPROTO);
2171 memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
2173 ptlrpc_request_set_replen(req);
2175 rc = mdc_queue_wait(req);
2179 ptlrpc_req_finished(req);
2183 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
2185 struct kuc_hdr *lh = (struct kuc_hdr *)buf;
2187 LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
2189 lh->kuc_magic = KUC_MAGIC;
2190 lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
2191 lh->kuc_flags = flags;
2192 lh->kuc_msgtype = CL_RECORD;
2193 lh->kuc_msglen = len;
2197 #define D_CHANGELOG 0
2199 struct changelog_show {
2204 struct obd_device *cs_obd;
2207 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
2208 struct llog_rec_hdr *hdr, void *data)
2210 struct changelog_show *cs = data;
2211 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
2216 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
2218 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
2219 cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
2220 rec->cr.cr_type, rc);
2224 if (rec->cr.cr_index < cs->cs_startrec) {
2225 /* Skip entries earlier than what we are interested in */
2226 CDEBUG(D_CHANGELOG, "rec="LPU64" start="LPU64"\n",
2227 rec->cr.cr_index, cs->cs_startrec);
2231 CDEBUG(D_CHANGELOG, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID
2232 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
2233 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
2234 rec->cr.cr_flags & CLF_FLAGMASK,
2235 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
2236 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
2238 len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
2240 /* Set up the message */
2241 lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
2242 memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
2244 rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
2245 CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len,rc);
2250 static int mdc_changelog_send_thread(void *csdata)
2252 struct changelog_show *cs = csdata;
2253 struct llog_ctxt *ctxt = NULL;
2254 struct llog_handle *llh = NULL;
2255 struct kuc_hdr *kuch;
2258 CDEBUG(D_CHANGELOG, "changelog to fp=%p start "LPU64"\n",
2259 cs->cs_fp, cs->cs_startrec);
2261 OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
2262 if (cs->cs_buf == NULL)
2263 GOTO(out, rc = -ENOMEM);
2265 /* Set up the remote catalog handle */
2266 ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
2268 GOTO(out, rc = -ENOENT);
2269 rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
2272 CERROR("%s: fail to open changelog catalog: rc = %d\n",
2273 cs->cs_obd->obd_name, rc);
2276 rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
2278 CERROR("llog_init_handle failed %d\n", rc);
2282 rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
2284 /* Send EOF no matter what our result */
2285 if ((kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch),
2287 kuch->kuc_msgtype = CL_EOF;
2288 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
2294 llog_cat_close(NULL, llh);
2296 llog_ctxt_put(ctxt);
2298 OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
2303 static int mdc_ioc_changelog_send(struct obd_device *obd,
2304 struct ioc_changelog *icc)
2306 struct changelog_show *cs;
2307 struct task_struct *task;
2310 /* Freed in mdc_changelog_send_thread */
2316 cs->cs_startrec = icc->icc_recno;
2317 /* matching fput in mdc_changelog_send_thread */
2318 cs->cs_fp = fget(icc->icc_id);
2319 cs->cs_flags = icc->icc_flags;
2322 * New thread because we should return to user app before
2323 * writing into our pipe
2325 task = kthread_run(mdc_changelog_send_thread, cs,
2326 "mdc_clg_send_thread");
2329 CERROR("%s: cannot start changelog thread: rc = %d\n",
2334 CDEBUG(D_CHANGELOG, "%s: started changelog thread\n",
2341 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2342 struct lustre_kernelcomm *lk);
2344 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
2345 struct obd_quotactl *oqctl)
2347 struct client_obd *cli = &exp->exp_obd->u.cli;
2348 struct ptlrpc_request *req;
2349 struct obd_quotactl *body;
2353 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2354 &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
2359 body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2362 ptlrpc_request_set_replen(req);
2364 /* the next poll will find -ENODATA, that means quotacheck is
2366 cli->cl_qchk_stat = -ENODATA;
2367 rc = ptlrpc_queue_wait(req);
2369 cli->cl_qchk_stat = rc;
2370 ptlrpc_req_finished(req);
2374 static int mdc_quota_poll_check(struct obd_export *exp,
2375 struct if_quotacheck *qchk)
2377 struct client_obd *cli = &exp->exp_obd->u.cli;
2381 qchk->obd_uuid = cli->cl_target_uuid;
2382 memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
2384 rc = cli->cl_qchk_stat;
2385 /* the client is not the previous one */
2386 if (rc == CL_NOT_QUOTACHECKED)
2391 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
2392 struct obd_quotactl *oqctl)
2394 struct ptlrpc_request *req;
2395 struct obd_quotactl *oqc;
2399 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2400 &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
2405 oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2408 ptlrpc_request_set_replen(req);
2409 ptlrpc_at_set_req_timeout(req);
2410 req->rq_no_resend = 1;
2412 rc = ptlrpc_queue_wait(req);
2414 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
2416 if (req->rq_repmsg &&
2417 (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2420 CERROR ("Can't unpack obd_quotactl\n");
2423 ptlrpc_req_finished(req);
2428 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2429 struct md_op_data *op_data)
2431 struct list_head cancels = LIST_HEAD_INIT(cancels);
2432 struct ptlrpc_request *req;
2434 struct mdc_swap_layouts *msl, *payload;
2437 msl = op_data->op_data;
2439 /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2440 * first thing it will do is to cancel the 2 layout
2441 * locks hold by this client.
2442 * So the client must cancel its layout locks on the 2 fids
2443 * with the request RPC to avoid extra RPC round trips
2445 count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2446 LCK_EX, MDS_INODELOCK_LAYOUT |
2447 MDS_INODELOCK_XATTR);
2448 count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2449 LCK_EX, MDS_INODELOCK_LAYOUT |
2450 MDS_INODELOCK_XATTR);
2452 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2453 &RQF_MDS_SWAP_LAYOUTS);
2455 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2459 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2460 mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
2462 rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2464 ptlrpc_request_free(req);
2468 mdc_swap_layouts_pack(req, op_data);
2470 payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2475 ptlrpc_request_set_replen(req);
2477 rc = ptlrpc_queue_wait(req);
2483 ptlrpc_req_finished(req);
2487 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2488 void *karg, void *uarg)
2490 struct obd_device *obd = exp->exp_obd;
2491 struct obd_ioctl_data *data = karg;
2492 struct obd_import *imp = obd->u.cli.cl_import;
2496 if (!try_module_get(THIS_MODULE)) {
2497 CERROR("Can't get module. Is it alive?");
2501 case OBD_IOC_CHANGELOG_SEND:
2502 rc = mdc_ioc_changelog_send(obd, karg);
2504 case OBD_IOC_CHANGELOG_CLEAR: {
2505 struct ioc_changelog *icc = karg;
2506 struct changelog_setinfo cs =
2507 {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
2508 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
2509 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
2513 case OBD_IOC_FID2PATH:
2514 rc = mdc_ioc_fid2path(exp, karg);
2516 case LL_IOC_HSM_CT_START:
2517 rc = mdc_ioc_hsm_ct_start(exp, karg);
2518 /* ignore if it was already registered on this MDS. */
2522 case LL_IOC_HSM_PROGRESS:
2523 rc = mdc_ioc_hsm_progress(exp, karg);
2525 case LL_IOC_HSM_STATE_GET:
2526 rc = mdc_ioc_hsm_state_get(exp, karg);
2528 case LL_IOC_HSM_STATE_SET:
2529 rc = mdc_ioc_hsm_state_set(exp, karg);
2531 case LL_IOC_HSM_ACTION:
2532 rc = mdc_ioc_hsm_current_action(exp, karg);
2534 case LL_IOC_HSM_REQUEST:
2535 rc = mdc_ioc_hsm_request(exp, karg);
2537 case OBD_IOC_CLIENT_RECOVER:
2538 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2542 case IOC_OSC_SET_ACTIVE:
2543 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2545 case OBD_IOC_POLL_QUOTACHECK:
2546 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
2548 case OBD_IOC_PING_TARGET:
2549 rc = ptlrpc_obd_ping(obd);
2552 * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2553 * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2554 * there'd be no LMV layer thus we might be called here. Eventually
2555 * this code should be removed.
2558 case IOC_OBD_STATFS: {
2559 struct obd_statfs stat_buf = {0};
2561 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2562 GOTO(out, rc = -ENODEV);
2565 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2566 min((int)data->ioc_plen2,
2567 (int)sizeof(struct obd_uuid))))
2568 GOTO(out, rc = -EFAULT);
2570 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2571 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2576 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2577 min((int) data->ioc_plen1,
2578 (int) sizeof(stat_buf))))
2579 GOTO(out, rc = -EFAULT);
2583 case OBD_IOC_QUOTACTL: {
2584 struct if_quotactl *qctl = karg;
2585 struct obd_quotactl *oqctl;
2587 OBD_ALLOC_PTR(oqctl);
2589 GOTO(out, rc = -ENOMEM);
2591 QCTL_COPY(oqctl, qctl);
2592 rc = obd_quotactl(exp, oqctl);
2594 QCTL_COPY(qctl, oqctl);
2595 qctl->qc_valid = QC_MDTIDX;
2596 qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2599 OBD_FREE_PTR(oqctl);
2602 case LL_IOC_GET_CONNECT_FLAGS:
2603 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2604 sizeof(*exp_connect_flags_ptr(exp))))
2605 GOTO(out, rc = -EFAULT);
2608 case LL_IOC_LOV_SWAP_LAYOUTS:
2609 rc = mdc_ioc_swap_layouts(exp, karg);
2612 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2613 GOTO(out, rc = -ENOTTY);
2616 module_put(THIS_MODULE);
2621 int mdc_get_info_rpc(struct obd_export *exp,
2622 obd_count keylen, void *key,
2623 int vallen, void *val)
2625 struct obd_import *imp = class_exp2cliimp(exp);
2626 struct ptlrpc_request *req;
2631 req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2635 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2636 RCL_CLIENT, keylen);
2637 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2638 RCL_CLIENT, sizeof(__u32));
2640 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2642 ptlrpc_request_free(req);
2646 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2647 memcpy(tmp, key, keylen);
2648 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2649 memcpy(tmp, &vallen, sizeof(__u32));
2651 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2652 RCL_SERVER, vallen);
2653 ptlrpc_request_set_replen(req);
2655 rc = ptlrpc_queue_wait(req);
2656 /* -EREMOTE means the get_info result is partial, and it needs to
2657 * continue on another MDT, see fid2path part in lmv_iocontrol */
2658 if (rc == 0 || rc == -EREMOTE) {
2659 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2660 memcpy(val, tmp, vallen);
2661 if (ptlrpc_rep_need_swab(req)) {
2662 if (KEY_IS(KEY_FID2PATH))
2663 lustre_swab_fid2path(val);
2666 ptlrpc_req_finished(req);
2671 static void lustre_swab_hai(struct hsm_action_item *h)
2673 __swab32s(&h->hai_len);
2674 __swab32s(&h->hai_action);
2675 lustre_swab_lu_fid(&h->hai_fid);
2676 lustre_swab_lu_fid(&h->hai_dfid);
2677 __swab64s(&h->hai_cookie);
2678 __swab64s(&h->hai_extent.offset);
2679 __swab64s(&h->hai_extent.length);
2680 __swab64s(&h->hai_gid);
2683 static void lustre_swab_hal(struct hsm_action_list *h)
2685 struct hsm_action_item *hai;
2688 __swab32s(&h->hal_version);
2689 __swab32s(&h->hal_count);
2690 __swab32s(&h->hal_archive_id);
2691 __swab64s(&h->hal_flags);
2693 for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2694 lustre_swab_hai(hai);
2697 static void lustre_swab_kuch(struct kuc_hdr *l)
2699 __swab16s(&l->kuc_magic);
2700 /* __u8 l->kuc_transport */
2701 __swab16s(&l->kuc_msgtype);
2702 __swab16s(&l->kuc_msglen);
2705 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2706 struct lustre_kernelcomm *lk)
2708 struct obd_import *imp = class_exp2cliimp(exp);
2709 __u32 archive = lk->lk_data;
2712 if (lk->lk_group != KUC_GRP_HSM) {
2713 CERROR("Bad copytool group %d\n", lk->lk_group);
2717 CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2718 lk->lk_uid, lk->lk_group, lk->lk_flags);
2720 if (lk->lk_flags & LK_FLG_STOP) {
2721 /* Unregister with the coordinator */
2722 rc = mdc_ioc_hsm_ct_unregister(imp);
2724 rc = mdc_ioc_hsm_ct_register(imp, archive);
2731 * Send a message to any listening copytools
2732 * @param val KUC message (kuc_hdr + hsm_action_list)
2733 * @param len total length of message
2735 static int mdc_hsm_copytool_send(int len, void *val)
2737 struct kuc_hdr *lh = (struct kuc_hdr *)val;
2738 struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
2742 if (len < sizeof(*lh) + sizeof(*hal)) {
2743 CERROR("Short HSM message %d < %d\n", len,
2744 (int) (sizeof(*lh) + sizeof(*hal)));
2747 if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2748 lustre_swab_kuch(lh);
2749 lustre_swab_hal(hal);
2750 } else if (lh->kuc_magic != KUC_MAGIC) {
2751 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2755 CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2757 lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2758 lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2760 /* Broadcast to HSM listeners */
2761 rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2767 * callback function passed to kuc for re-registering each HSM copytool
2768 * running on MDC, after MDT shutdown/recovery.
2769 * @param data copytool registration data
2770 * @param cb_arg callback argument (obd_import)
2772 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2774 struct kkuc_ct_data *kcd = data;
2775 struct obd_import *imp = (struct obd_import *)cb_arg;
2778 if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2781 if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2784 CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2785 imp->imp_obd->obd_name, kcd->kcd_archive);
2786 rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2788 /* ignore error if the copytool is already registered */
2789 return (rc == -EEXIST) ? 0 : rc;
2793 * Re-establish all kuc contexts with MDT
2794 * after MDT shutdown/recovery.
2796 static int mdc_kuc_reregister(struct obd_import *imp)
2798 /* re-register HSM agents */
2799 return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2803 int mdc_set_info_async(const struct lu_env *env,
2804 struct obd_export *exp,
2805 obd_count keylen, void *key,
2806 obd_count vallen, void *val,
2807 struct ptlrpc_request_set *set)
2809 struct obd_import *imp = class_exp2cliimp(exp);
2813 if (KEY_IS(KEY_READ_ONLY)) {
2814 if (vallen != sizeof(int))
2817 spin_lock(&imp->imp_lock);
2818 if (*((int *)val)) {
2819 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2820 imp->imp_connect_data.ocd_connect_flags |=
2823 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2824 imp->imp_connect_data.ocd_connect_flags &=
2825 ~OBD_CONNECT_RDONLY;
2827 spin_unlock(&imp->imp_lock);
2829 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2830 keylen, key, vallen, val, set);
2833 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2834 sptlrpc_conf_client_adapt(exp->exp_obd);
2837 if (KEY_IS(KEY_FLUSH_CTX)) {
2838 sptlrpc_import_flush_my_ctx(imp);
2841 if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2842 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2843 keylen, key, vallen, val, set);
2846 if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2847 rc = mdc_hsm_copytool_send(vallen, val);
2851 CERROR("Unknown key %s\n", (char *)key);
2855 int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2856 __u32 keylen, void *key, __u32 *vallen, void *val,
2857 struct lov_stripe_md *lsm)
2861 if (KEY_IS(KEY_MAX_EASIZE)) {
2862 int mdsize, *max_easize;
2864 if (*vallen != sizeof(int))
2866 mdsize = *(int *)val;
2867 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2868 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2870 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2872 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2873 int *default_easize;
2875 if (*vallen != sizeof(int))
2877 default_easize = val;
2878 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2880 } else if (KEY_IS(KEY_MAX_COOKIESIZE)) {
2881 int mdsize, *max_cookiesize;
2883 if (*vallen != sizeof(int))
2885 mdsize = *(int *)val;
2886 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_cookiesize)
2887 exp->exp_obd->u.cli.cl_max_mds_cookiesize = mdsize;
2888 max_cookiesize = val;
2889 *max_cookiesize = exp->exp_obd->u.cli.cl_max_mds_cookiesize;
2891 } else if (KEY_IS(KEY_DEFAULT_COOKIESIZE)) {
2892 int *default_cookiesize;
2894 if (*vallen != sizeof(int))
2896 default_cookiesize = val;
2897 *default_cookiesize =
2898 exp->exp_obd->u.cli.cl_default_mds_cookiesize;
2900 } else if (KEY_IS(KEY_CONN_DATA)) {
2901 struct obd_import *imp = class_exp2cliimp(exp);
2902 struct obd_connect_data *data = val;
2904 if (*vallen != sizeof(*data))
2907 *data = imp->imp_connect_data;
2909 } else if (KEY_IS(KEY_TGT_COUNT)) {
2914 rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2919 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
2920 struct obd_capa *oc, struct obd_client_handle *handle,
2923 struct ptlrpc_request *req;
2924 struct mdt_body *body;
2928 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
2932 mdc_set_capa_size(req, &RMF_CAPA1, oc);
2934 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
2936 ptlrpc_request_free(req);
2940 mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
2942 ptlrpc_request_set_replen(req);
2944 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2945 rc = ptlrpc_queue_wait(req);
2946 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2948 CERROR("Pin failed: %d\n", rc);
2952 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2954 GOTO(err_out, rc = -EPROTO);
2956 handle->och_fh = body->handle;
2957 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
2959 handle->och_mod = obd_mod_alloc();
2960 if (handle->och_mod == NULL) {
2961 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
2962 GOTO(err_out, rc = -ENOMEM);
2964 handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
2969 ptlrpc_req_finished(req);
2973 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
2976 struct ptlrpc_request *req;
2977 struct mdt_body *body;
2981 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
2982 LUSTRE_MDS_VERSION, MDS_UNPIN);
2986 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
2987 body->handle = handle->och_fh;
2990 ptlrpc_request_set_replen(req);
2992 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2993 rc = ptlrpc_queue_wait(req);
2994 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2997 CERROR("Unpin failed: %d\n", rc);
2999 ptlrpc_req_finished(req);
3000 ptlrpc_req_finished(handle->och_mod->mod_open_req);
3002 obd_mod_put(handle->och_mod);
3006 int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
3007 struct obd_capa *oc, struct ptlrpc_request **request)
3009 struct ptlrpc_request *req;
3014 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
3018 mdc_set_capa_size(req, &RMF_CAPA1, oc);
3020 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
3022 ptlrpc_request_free(req);
3026 mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
3028 ptlrpc_request_set_replen(req);
3030 rc = ptlrpc_queue_wait(req);
3032 ptlrpc_req_finished(req);
3038 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
3039 enum obd_import_event event)
3043 LASSERT(imp->imp_obd == obd);
3046 case IMP_EVENT_DISCON: {
3048 /* XXX Pass event up to OBDs stack. used only for FLD now */
3049 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
3053 case IMP_EVENT_INACTIVE: {
3054 struct client_obd *cli = &obd->u.cli;
3056 * Flush current sequence to make client obtain new one
3057 * from server in case of disconnect/reconnect.
3059 if (cli->cl_seq != NULL)
3060 seq_client_flush(cli->cl_seq);
3062 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
3065 case IMP_EVENT_INVALIDATE: {
3066 struct ldlm_namespace *ns = obd->obd_namespace;
3068 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
3072 case IMP_EVENT_ACTIVE:
3073 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
3074 /* redo the kuc registration after reconnecting */
3076 rc = mdc_kuc_reregister(imp);
3079 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
3081 case IMP_EVENT_DEACTIVATE:
3082 case IMP_EVENT_ACTIVATE:
3085 CERROR("Unknown import event %x\n", event);
3091 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
3092 struct lu_fid *fid, struct md_op_data *op_data)
3094 struct client_obd *cli = &exp->exp_obd->u.cli;
3095 struct lu_client_seq *seq = cli->cl_seq;
3097 RETURN(seq_client_alloc_fid(env, seq, fid));
3100 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
3101 struct client_obd *cli = &exp->exp_obd->u.cli;
3102 return &cli->cl_target_uuid;
3106 * Determine whether the lock can be canceled before replaying it during
3107 * recovery, non zero value will be return if the lock can be canceled,
3108 * or zero returned for not
3110 static int mdc_cancel_weight(struct ldlm_lock *lock)
3112 if (lock->l_resource->lr_type != LDLM_IBITS)
3115 /* FIXME: if we ever get into a situation where there are too many
3116 * opened files with open locks on a single node, then we really
3117 * should replay these open locks to reget it */
3118 if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
3124 static int mdc_resource_inode_free(struct ldlm_resource *res)
3126 if (res->lr_lvb_inode)
3127 res->lr_lvb_inode = NULL;
3132 struct ldlm_valblock_ops inode_lvbo = {
3133 .lvbo_free = mdc_resource_inode_free
3136 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
3138 struct client_obd *cli = &obd->u.cli;
3142 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3143 if (!cli->cl_rpc_lock)
3145 mdc_init_rpc_lock(cli->cl_rpc_lock);
3147 rc = ptlrpcd_addref();
3149 GOTO(err_rpc_lock, rc);
3151 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3152 if (!cli->cl_close_lock)
3153 GOTO(err_ptlrpcd_decref, rc = -ENOMEM);
3154 mdc_init_rpc_lock(cli->cl_close_lock);
3156 rc = client_obd_setup(obd, cfg);
3158 GOTO(err_close_lock, rc);
3160 obd->obd_vars = lprocfs_mdc_obd_vars;
3161 lprocfs_seq_obd_setup(obd);
3162 lprocfs_alloc_md_stats(obd, 0);
3164 sptlrpc_lprocfs_cliobd_attach(obd);
3165 ptlrpc_lprocfs_register_obd(obd);
3167 ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
3169 obd->obd_namespace->ns_lvbo = &inode_lvbo;
3171 rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
3174 CERROR("failed to setup llogging subsystems\n");
3180 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3184 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3188 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
3189 * us to make MDS RPCs with large enough reply buffers to hold a default
3190 * sized EA and cookie without having to calculate this (via a call into the
3191 * LOV + OSCs) each time we make an RPC. The maximum size is also tracked
3192 * but not used to avoid wastefully vmalloc()'ing large reply buffers when
3193 * a large number of stripes is possible. If a larger reply buffer is
3194 * required it will be reallocated in the ptlrpc layer due to overflow.
3196 static int mdc_init_ea_size(struct obd_export *exp, int easize,
3197 int def_easize, int cookiesize, int def_cookiesize)
3199 struct obd_device *obd = exp->exp_obd;
3200 struct client_obd *cli = &obd->u.cli;
3203 if (cli->cl_max_mds_easize < easize)
3204 cli->cl_max_mds_easize = easize;
3206 if (cli->cl_default_mds_easize < def_easize)
3207 cli->cl_default_mds_easize = def_easize;
3209 if (cli->cl_max_mds_cookiesize < cookiesize)
3210 cli->cl_max_mds_cookiesize = cookiesize;
3212 if (cli->cl_default_mds_cookiesize < def_cookiesize)
3213 cli->cl_default_mds_cookiesize = def_cookiesize;
3218 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
3224 case OBD_CLEANUP_EARLY:
3226 case OBD_CLEANUP_EXPORTS:
3227 /* Failsafe, ok if racy */
3228 if (obd->obd_type->typ_refcnt <= 1)
3229 libcfs_kkuc_group_rem(0, KUC_GRP_HSM, NULL);
3231 obd_cleanup_client_import(obd);
3232 ptlrpc_lprocfs_unregister_obd(obd);
3233 lprocfs_obd_cleanup(obd);
3234 lprocfs_free_md_stats(obd);
3236 rc = obd_llog_finish(obd, 0);
3238 CERROR("failed to cleanup llogging subsystems\n");
3244 static int mdc_cleanup(struct obd_device *obd)
3246 struct client_obd *cli = &obd->u.cli;
3248 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3249 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3253 return client_obd_cleanup(obd);
3257 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
3258 struct obd_device *tgt, int *index)
3260 struct llog_ctxt *ctxt;
3265 LASSERT(olg == &obd->obd_olg);
3267 rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt,
3272 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
3273 llog_initiator_connect(ctxt);
3274 llog_ctxt_put(ctxt);
3279 static int mdc_llog_finish(struct obd_device *obd, int count)
3281 struct llog_ctxt *ctxt;
3285 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
3287 llog_cleanup(NULL, ctxt);
3292 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
3294 struct lustre_cfg *lcfg = buf;
3295 int rc = class_process_proc_seq_param(PARAM_MDC, obd->obd_vars,
3297 return (rc > 0 ? 0: rc);
3301 /* get remote permission for current user on fid */
3302 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
3303 struct obd_capa *oc, __u32 suppgid,
3304 struct ptlrpc_request **request)
3306 struct ptlrpc_request *req;
3310 LASSERT(client_is_remote(exp));
3313 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
3317 mdc_set_capa_size(req, &RMF_CAPA1, oc);
3319 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
3321 ptlrpc_request_free(req);
3325 mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
3327 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
3328 sizeof(struct mdt_remote_perm));
3330 ptlrpc_request_set_replen(req);
3332 rc = ptlrpc_queue_wait(req);
3334 ptlrpc_req_finished(req);
3340 static int mdc_interpret_renew_capa(const struct lu_env *env,
3341 struct ptlrpc_request *req, void *args,
3344 struct mdc_renew_capa_args *ra = args;
3345 struct mdt_body *body = NULL;
3346 struct lustre_capa *capa;
3350 GOTO(out, capa = ERR_PTR(status));
3352 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3354 GOTO(out, capa = ERR_PTR(-EFAULT));
3356 if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
3357 GOTO(out, capa = ERR_PTR(-ENOENT));
3359 capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
3361 GOTO(out, capa = ERR_PTR(-EFAULT));
3364 ra->ra_cb(ra->ra_oc, capa);
3368 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3371 struct ptlrpc_request *req;
3372 struct mdc_renew_capa_args *ra;
3375 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
3376 LUSTRE_MDS_VERSION, MDS_GETATTR);
3380 /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
3381 * capa to renew is oss capa.
3383 mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
3384 ptlrpc_request_set_replen(req);
3386 CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
3387 ra = ptlrpc_req_async_args(req);
3390 req->rq_interpret_reply = mdc_interpret_renew_capa;
3391 ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
3395 struct obd_ops mdc_obd_ops = {
3396 .o_owner = THIS_MODULE,
3397 .o_setup = mdc_setup,
3398 .o_precleanup = mdc_precleanup,
3399 .o_cleanup = mdc_cleanup,
3400 .o_add_conn = client_import_add_conn,
3401 .o_del_conn = client_import_del_conn,
3402 .o_connect = client_connect_import,
3403 .o_disconnect = client_disconnect_export,
3404 .o_iocontrol = mdc_iocontrol,
3405 .o_set_info_async = mdc_set_info_async,
3406 .o_statfs = mdc_statfs,
3408 .o_unpin = mdc_unpin,
3409 .o_fid_init = client_fid_init,
3410 .o_fid_fini = client_fid_fini,
3411 .o_fid_alloc = mdc_fid_alloc,
3412 .o_import_event = mdc_import_event,
3413 .o_llog_init = mdc_llog_init,
3414 .o_llog_finish = mdc_llog_finish,
3415 .o_get_info = mdc_get_info,
3416 .o_process_config = mdc_process_config,
3417 .o_get_uuid = mdc_get_uuid,
3418 .o_quotactl = mdc_quotactl,
3419 .o_quotacheck = mdc_quotacheck
3422 struct md_ops mdc_md_ops = {
3423 .m_getstatus = mdc_getstatus,
3424 .m_null_inode = mdc_null_inode,
3425 .m_find_cbdata = mdc_find_cbdata,
3426 .m_close = mdc_close,
3427 .m_create = mdc_create,
3428 .m_done_writing = mdc_done_writing,
3429 .m_enqueue = mdc_enqueue,
3430 .m_getattr = mdc_getattr,
3431 .m_getattr_name = mdc_getattr_name,
3432 .m_intent_lock = mdc_intent_lock,
3434 .m_is_subdir = mdc_is_subdir,
3435 .m_rename = mdc_rename,
3436 .m_setattr = mdc_setattr,
3437 .m_setxattr = mdc_setxattr,
3438 .m_getxattr = mdc_getxattr,
3439 .m_fsync = mdc_fsync,
3440 .m_read_entry = mdc_read_entry,
3441 .m_unlink = mdc_unlink,
3442 .m_cancel_unused = mdc_cancel_unused,
3443 .m_init_ea_size = mdc_init_ea_size,
3444 .m_set_lock_data = mdc_set_lock_data,
3445 .m_lock_match = mdc_lock_match,
3446 .m_get_lustre_md = mdc_get_lustre_md,
3447 .m_free_lustre_md = mdc_free_lustre_md,
3448 .m_set_open_replay_data = mdc_set_open_replay_data,
3449 .m_clear_open_replay_data = mdc_clear_open_replay_data,
3450 .m_renew_capa = mdc_renew_capa,
3451 .m_unpack_capa = mdc_unpack_capa,
3452 .m_get_remote_perm = mdc_get_remote_perm,
3453 .m_intent_getattr_async = mdc_intent_getattr_async,
3454 .m_revalidate_lock = mdc_revalidate_lock
3457 int __init mdc_init(void)
3459 return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
3460 #ifndef HAVE_ONLY_PROCFS_SEQ
3463 LUSTRE_MDC_NAME, NULL);
3467 static void /*__exit*/ mdc_exit(void)
3469 class_unregister_type(LUSTRE_MDC_NAME);
3472 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3473 MODULE_DESCRIPTION("Lustre Metadata Client");
3474 MODULE_LICENSE("GPL");
3476 module_init(mdc_init);
3477 module_exit(mdc_exit);