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, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_MDC
39 #include <linux/module.h>
40 #include <linux/pagemap.h>
41 #include <linux/miscdevice.h>
42 #include <linux/init.h>
43 #include <linux/utsname.h>
45 #include <lustre_acl.h>
46 #include <lustre_ioctl.h>
47 #include <obd_class.h>
48 #include <lustre_lmv.h>
49 #include <lustre_fid.h>
50 #include <lprocfs_status.h>
51 #include <lustre_param.h>
52 #include <lustre_log.h>
53 #include <lustre_kernelcomm.h>
54 #include <cl_object.h>
56 #include "mdc_internal.h"
58 #define REQUEST_MINOR 244
60 struct mdc_renew_capa_args {
61 struct obd_capa *ra_oc;
62 renew_capa_cb_t ra_cb;
65 static int mdc_cleanup(struct obd_device *obd);
67 static int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
68 const struct req_msg_field *field,
71 struct lustre_capa *capa;
75 /* swabbed already in mdc_enqueue */
76 capa = req_capsule_server_get(&req->rq_pill, field);
80 c = alloc_capa(CAPA_SITE_CLIENT);
82 CDEBUG(D_INFO, "alloc capa failed!\n");
91 static inline int mdc_queue_wait(struct ptlrpc_request *req)
93 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
96 /* obd_get_request_slot() ensures that this client has no more
97 * than cl_max_rpcs_in_flight RPCs simultaneously inf light
99 rc = obd_get_request_slot(cli);
103 rc = ptlrpc_queue_wait(req);
104 obd_put_request_slot(cli);
109 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
110 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
111 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
112 struct obd_capa **pc, int level, int msg_flags)
114 struct ptlrpc_request *req;
115 struct mdt_body *body;
119 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
120 LUSTRE_MDS_VERSION, MDS_GETSTATUS);
124 mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
125 lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
126 req->rq_send_state = level;
128 ptlrpc_request_set_replen(req);
130 rc = ptlrpc_queue_wait(req);
134 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
136 GOTO(out, rc = -EPROTO);
138 if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
139 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
144 *rootfid = body->mbo_fid1;
145 CDEBUG(D_NET, "root fid="DFID", last_committed="LPU64"\n",
146 PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
149 ptlrpc_req_finished(req);
153 /* This should be mdc_get_info("rootfid") */
154 static int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
155 struct obd_capa **pc)
157 return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
162 * This function now is known to always saying that it will receive 4 buffers
163 * from server. Even for cases when acl_size and md_size is zero, RPC header
164 * will contain 4 fields and RPC itself will contain zero size fields. This is
165 * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
166 * and thus zero, it shrinks it, making zero size. The same story about
167 * md_size. And this is course of problem when client waits for smaller number
168 * of fields. This issue will be fixed later when client gets aware of RPC
171 static int mdc_getattr_common(struct obd_export *exp,
172 struct ptlrpc_request *req)
174 struct req_capsule *pill = &req->rq_pill;
175 struct mdt_body *body;
180 /* Request message already built. */
181 rc = ptlrpc_queue_wait(req);
185 /* sanity check for the reply */
186 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
190 CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
192 mdc_update_max_ea_from_body(exp, body);
193 if (body->mbo_eadatasize != 0) {
194 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
195 body->mbo_eadatasize);
200 if (body->mbo_valid & OBD_MD_FLRMTPERM) {
201 struct mdt_remote_perm *perm;
203 LASSERT(client_is_remote(exp));
204 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
205 lustre_swab_mdt_remote_perm);
210 if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
211 struct lustre_capa *capa;
212 capa = req_capsule_server_get(pill, &RMF_CAPA1);
220 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
221 struct ptlrpc_request **request)
223 struct ptlrpc_request *req;
227 /* Single MDS without an LMV case */
228 if (op_data->op_flags & MF_GET_MDT_IDX) {
233 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
237 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
239 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
241 ptlrpc_request_free(req);
245 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
246 op_data->op_valid, op_data->op_mode, -1, 0);
248 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
250 if (op_data->op_valid & OBD_MD_FLRMTPERM) {
251 LASSERT(client_is_remote(exp));
252 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
253 sizeof(struct mdt_remote_perm));
255 ptlrpc_request_set_replen(req);
257 rc = mdc_getattr_common(exp, req);
259 ptlrpc_req_finished(req);
265 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
266 struct ptlrpc_request **request)
268 struct ptlrpc_request *req;
273 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
274 &RQF_MDS_GETATTR_NAME);
278 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
279 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
280 op_data->op_namelen + 1);
282 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
284 ptlrpc_request_free(req);
288 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
289 op_data->op_valid, op_data->op_mode,
290 op_data->op_suppgids[0], 0);
292 if (op_data->op_name) {
293 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
294 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
295 op_data->op_namelen);
296 memcpy(name, op_data->op_name, op_data->op_namelen);
299 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
301 ptlrpc_request_set_replen(req);
303 rc = mdc_getattr_common(exp, req);
305 ptlrpc_req_finished(req);
311 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
312 const struct lu_fid *fid,
313 struct obd_capa *oc, int opcode, u64 valid,
314 const char *xattr_name, const char *input,
315 int input_size, int output_size, int flags,
316 __u32 suppgid, struct ptlrpc_request **request)
318 struct ptlrpc_request *req;
319 int xattr_namelen = 0;
325 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
329 mdc_set_capa_size(req, &RMF_CAPA1, oc);
331 xattr_namelen = strlen(xattr_name) + 1;
332 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
337 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
341 /* Flush local XATTR locks to get rid of a possible cancel RPC */
342 if (opcode == MDS_REINT && fid_is_sane(fid) &&
343 exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
344 struct list_head cancels = LIST_HEAD_INIT(cancels);
347 /* Without that packing would fail */
349 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
352 count = mdc_resource_get_unused(exp, fid,
354 MDS_INODELOCK_XATTR);
356 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
358 ptlrpc_request_free(req);
362 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
364 ptlrpc_request_free(req);
369 if (opcode == MDS_REINT) {
370 struct mdt_rec_setxattr *rec;
372 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
373 sizeof(struct mdt_rec_reint));
374 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
375 rec->sx_opcode = REINT_SETXATTR;
376 rec->sx_fsuid = from_kuid(&init_user_ns, current_fsuid());
377 rec->sx_fsgid = from_kgid(&init_user_ns, current_fsgid());
378 rec->sx_cap = cfs_curproc_cap_pack();
379 rec->sx_suppgid1 = suppgid;
380 rec->sx_suppgid2 = -1;
382 rec->sx_valid = valid | OBD_MD_FLCTIME;
383 rec->sx_time = cfs_time_current_sec();
384 rec->sx_size = output_size;
385 rec->sx_flags = flags;
387 mdc_pack_capa(req, &RMF_CAPA1, oc);
389 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
393 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
394 memcpy(tmp, xattr_name, xattr_namelen);
397 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
398 memcpy(tmp, input, input_size);
401 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
402 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
403 RCL_SERVER, output_size);
404 ptlrpc_request_set_replen(req);
407 if (opcode == MDS_REINT)
408 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
410 rc = ptlrpc_queue_wait(req);
412 if (opcode == MDS_REINT)
413 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
416 ptlrpc_req_finished(req);
422 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
423 struct obd_capa *oc, u64 valid,
424 const char *xattr_name,
425 const char *input, int input_size, int output_size,
426 int flags, __u32 suppgid,
427 struct ptlrpc_request **request)
429 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
430 fid, oc, MDS_REINT, valid, xattr_name,
431 input, input_size, output_size, flags,
435 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
436 struct obd_capa *oc, u64 valid,
437 const char *xattr_name,
438 const char *input, int input_size, int output_size,
439 int flags, struct ptlrpc_request **request)
441 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
442 fid, oc, MDS_GETXATTR, valid, xattr_name,
443 input, input_size, output_size, flags,
447 #ifdef CONFIG_FS_POSIX_ACL
448 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
450 struct req_capsule *pill = &req->rq_pill;
451 struct mdt_body *body = md->body;
452 struct posix_acl *acl;
457 if (!body->mbo_aclsize)
460 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
465 acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
470 CERROR("convert xattr to acl: %d\n", rc);
474 rc = posix_acl_valid(acl);
476 CERROR("validate acl: %d\n", rc);
477 posix_acl_release(acl);
485 #define mdc_unpack_acl(req, md) 0
488 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
489 struct obd_export *dt_exp, struct obd_export *md_exp,
490 struct lustre_md *md)
492 struct req_capsule *pill = &req->rq_pill;
497 memset(md, 0, sizeof(*md));
499 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
500 LASSERT(md->body != NULL);
502 if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
504 struct lov_mds_md *lmm;
506 if (!S_ISREG(md->body->mbo_mode)) {
507 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
508 "regular file, but is not\n");
509 GOTO(out, rc = -EPROTO);
512 if (md->body->mbo_eadatasize == 0) {
513 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
514 "but eadatasize 0\n");
515 GOTO(out, rc = -EPROTO);
518 lmmsize = md->body->mbo_eadatasize;
519 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
521 GOTO(out, rc = -EPROTO);
523 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
527 if (rc < (typeof(rc))sizeof(*md->lsm)) {
528 CDEBUG(D_INFO, "lsm size too small: "
529 "rc < sizeof (*md->lsm) (%d < %d)\n",
530 rc, (int)sizeof(*md->lsm));
531 GOTO(out, rc = -EPROTO);
534 } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
536 struct lov_mds_md *lmv;
538 if (!S_ISDIR(md->body->mbo_mode)) {
539 CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
540 "directory, but is not\n");
541 GOTO(out, rc = -EPROTO);
544 if (md->body->mbo_eadatasize == 0) {
545 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
546 "but eadatasize 0\n");
550 if (md->body->mbo_valid & OBD_MD_MEA) {
551 lmvsize = md->body->mbo_eadatasize;
552 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
555 GOTO(out, rc = -EPROTO);
557 rc = obd_unpackmd(md_exp, (void *)&md->lmv, lmv,
562 if (rc < (typeof(rc))sizeof(*md->lmv)) {
563 CDEBUG(D_INFO, "size too small: "
564 "rc < sizeof(*md->lmv) (%d < %d)\n",
565 rc, (int)sizeof(*md->lmv));
566 GOTO(out, rc = -EPROTO);
572 if (md->body->mbo_valid & OBD_MD_FLRMTPERM) {
573 /* remote permission */
574 LASSERT(client_is_remote(exp));
575 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
576 lustre_swab_mdt_remote_perm);
577 if (!md->remote_perm)
578 GOTO(out, rc = -EPROTO);
579 } else if (md->body->mbo_valid & OBD_MD_FLACL) {
580 /* for ACL, it's possible that FLACL is set but aclsize is zero.
581 * only when aclsize != 0 there's an actual segment for ACL
584 if (md->body->mbo_aclsize) {
585 rc = mdc_unpack_acl(req, md);
588 #ifdef CONFIG_FS_POSIX_ACL
590 md->posix_acl = NULL;
594 if (md->body->mbo_valid & OBD_MD_FLMDSCAPA) {
595 struct obd_capa *oc = NULL;
597 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
603 if (md->body->mbo_valid & OBD_MD_FLOSSCAPA) {
604 struct obd_capa *oc = NULL;
606 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
616 capa_put(md->oss_capa);
620 capa_put(md->mds_capa);
623 #ifdef CONFIG_FS_POSIX_ACL
624 posix_acl_release(md->posix_acl);
627 obd_free_memmd(dt_exp, &md->lsm);
632 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
638 void mdc_replay_open(struct ptlrpc_request *req)
640 struct md_open_data *mod = req->rq_cb_data;
641 struct ptlrpc_request *close_req;
642 struct obd_client_handle *och;
643 struct lustre_handle old;
644 struct mdt_body *body;
648 DEBUG_REQ(D_ERROR, req,
649 "Can't properly replay without open data.");
654 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
655 LASSERT(body != NULL);
659 struct lustre_handle *file_fh;
661 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
663 file_fh = &och->och_fh;
664 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
665 file_fh->cookie, body->mbo_handle.cookie);
667 *file_fh = body->mbo_handle;
669 close_req = mod->mod_close_req;
670 if (close_req != NULL) {
671 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
672 struct mdt_ioepoch *epoch;
674 LASSERT(opc == MDS_CLOSE);
675 epoch = req_capsule_client_get(&close_req->rq_pill,
680 LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
681 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
682 epoch->handle = body->mbo_handle;
687 void mdc_commit_open(struct ptlrpc_request *req)
689 struct md_open_data *mod = req->rq_cb_data;
694 * No need to touch md_open_data::mod_och, it holds a reference on
695 * \var mod and will zero references to each other, \var mod will be
696 * freed after that when md_open_data::mod_och will put the reference.
700 * Do not let open request to disappear as it still may be needed
701 * for close rpc to happen (it may happen on evict only, otherwise
702 * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
703 * called), just mark this rpc as committed to distinguish these 2
704 * cases, see mdc_close() for details. The open request reference will
705 * be put along with freeing \var mod.
707 ptlrpc_request_addref(req);
708 spin_lock(&req->rq_lock);
709 req->rq_committed = 1;
710 spin_unlock(&req->rq_lock);
711 req->rq_cb_data = NULL;
715 int mdc_set_open_replay_data(struct obd_export *exp,
716 struct obd_client_handle *och,
717 struct lookup_intent *it)
719 struct md_open_data *mod;
720 struct mdt_rec_create *rec;
721 struct mdt_body *body;
722 struct ptlrpc_request *open_req = it->d.lustre.it_data;
723 struct obd_import *imp = open_req->rq_import;
726 if (!open_req->rq_replay)
729 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
730 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
731 LASSERT(rec != NULL);
732 /* Incoming message in my byte order (it's been swabbed). */
733 /* Outgoing messages always in my byte order. */
734 LASSERT(body != NULL);
736 /* Only if the import is replayable, we set replay_open data */
737 if (och && imp->imp_replayable) {
738 mod = obd_mod_alloc();
740 DEBUG_REQ(D_ERROR, open_req,
741 "Can't allocate md_open_data");
746 * Take a reference on \var mod, to be freed on mdc_close().
747 * It protects \var mod from being freed on eviction (commit
748 * callback is called despite rq_replay flag).
749 * Another reference for \var och.
754 spin_lock(&open_req->rq_lock);
757 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
758 it_disposition(it, DISP_OPEN_STRIPE);
759 mod->mod_open_req = open_req;
760 open_req->rq_cb_data = mod;
761 open_req->rq_commit_cb = mdc_commit_open;
762 spin_unlock(&open_req->rq_lock);
765 rec->cr_fid2 = body->mbo_fid1;
766 rec->cr_ioepoch = body->mbo_ioepoch;
767 rec->cr_old_handle.cookie = body->mbo_handle.cookie;
768 open_req->rq_replay_cb = mdc_replay_open;
769 if (!fid_is_sane(&body->mbo_fid1)) {
770 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
775 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
779 static void mdc_free_open(struct md_open_data *mod)
783 if (mod->mod_is_create == 0 &&
784 imp_connect_disp_stripe(mod->mod_open_req->rq_import))
787 LASSERT(mod->mod_open_req->rq_replay == 0);
789 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
791 ptlrpc_request_committed(mod->mod_open_req, committed);
792 if (mod->mod_close_req)
793 ptlrpc_request_committed(mod->mod_close_req, committed);
796 int mdc_clear_open_replay_data(struct obd_export *exp,
797 struct obd_client_handle *och)
799 struct md_open_data *mod = och->och_mod;
803 * It is possible to not have \var mod in a case of eviction between
804 * lookup and ll_file_open().
809 LASSERT(mod != LP_POISON);
810 LASSERT(mod->mod_open_req != NULL);
820 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
821 struct md_open_data *mod, struct ptlrpc_request **request)
823 struct obd_device *obd = class_exp2obd(exp);
824 struct ptlrpc_request *req;
825 struct req_format *req_fmt;
830 req_fmt = &RQF_MDS_CLOSE;
831 if (op_data->op_bias & MDS_HSM_RELEASE) {
832 req_fmt = &RQF_MDS_RELEASE_CLOSE;
834 /* allocate a FID for volatile file */
835 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
837 CERROR("%s: "DFID" failed to allocate FID: %d\n",
838 obd->obd_name, PFID(&op_data->op_fid1), rc);
839 /* save the errcode and proceed to close */
845 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
849 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
851 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
853 ptlrpc_request_free(req);
857 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
858 * portal whose threads are not taking any DLM locks and are therefore
859 * always progressing */
860 req->rq_request_portal = MDS_READPAGE_PORTAL;
861 ptlrpc_at_set_req_timeout(req);
863 /* Ensure that this close's handle is fixed up during replay. */
864 if (likely(mod != NULL)) {
865 LASSERTF(mod->mod_open_req != NULL &&
866 mod->mod_open_req->rq_type != LI_POISON,
867 "POISONED open %p!\n", mod->mod_open_req);
869 mod->mod_close_req = req;
871 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
872 /* We no longer want to preserve this open for replay even
873 * though the open was committed. b=3632, b=3633 */
874 spin_lock(&mod->mod_open_req->rq_lock);
875 mod->mod_open_req->rq_replay = 0;
876 spin_unlock(&mod->mod_open_req->rq_lock);
878 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
881 mdc_close_pack(req, op_data);
883 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
884 obd->u.cli.cl_default_mds_easize);
885 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
886 obd->u.cli.cl_default_mds_cookiesize);
888 ptlrpc_request_set_replen(req);
890 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
891 rc = ptlrpc_queue_wait(req);
892 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
894 if (req->rq_repmsg == NULL) {
895 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
898 rc = req->rq_status ?: -EIO;
899 } else if (rc == 0 || rc == -EAGAIN) {
900 struct mdt_body *body;
902 rc = lustre_msg_get_status(req->rq_repmsg);
903 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
904 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
909 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
912 } else if (rc == -ESTALE) {
914 * it can be allowed error after 3633 if open was committed and
915 * server failed before close was sent. Let's check if mod
916 * exists and return no error in that case
919 DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
920 LASSERT(mod->mod_open_req != NULL);
921 if (mod->mod_open_req->rq_committed)
928 mod->mod_close_req = NULL;
929 /* Since now, mod is accessed through open_req only,
930 * thus close req does not keep a reference on mod anymore. */
935 RETURN(rc < 0 ? rc : saved_rc);
938 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
939 __u64 offset, struct obd_capa *oc,
940 struct page **pages, int npages,
941 struct ptlrpc_request **request)
943 struct ptlrpc_request *req;
944 struct ptlrpc_bulk_desc *desc;
946 wait_queue_head_t waitq;
948 struct l_wait_info lwi;
953 init_waitqueue_head(&waitq);
956 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
960 mdc_set_capa_size(req, &RMF_CAPA1, oc);
962 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
964 ptlrpc_request_free(req);
968 req->rq_request_portal = MDS_READPAGE_PORTAL;
969 ptlrpc_at_set_req_timeout(req);
971 desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK,
974 ptlrpc_request_free(req);
978 /* NB req now owns desc and will free it when it gets freed */
979 for (i = 0; i < npages; i++)
980 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
982 mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE * npages, fid, oc);
984 ptlrpc_request_set_replen(req);
985 rc = ptlrpc_queue_wait(req);
987 ptlrpc_req_finished(req);
988 if (rc != -ETIMEDOUT)
992 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
993 CERROR("%s: too many resend retries: rc = %d\n",
994 exp->exp_obd->obd_name, -EIO);
997 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
999 l_wait_event(waitq, 0, &lwi);
1004 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1005 req->rq_bulk->bd_nob_transferred);
1007 ptlrpc_req_finished(req);
1011 if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1012 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1013 exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1014 PAGE_CACHE_SIZE * npages);
1015 ptlrpc_req_finished(req);
1023 static void mdc_release_page(struct page *page, int remove)
1027 if (likely(page->mapping != NULL))
1028 truncate_complete_page(page->mapping, page);
1031 page_cache_release(page);
1034 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1035 __u64 *start, __u64 *end, int hash64)
1038 * Complement of hash is used as an index so that
1039 * radix_tree_gang_lookup() can be used to find a page with starting
1040 * hash _smaller_ than one we are looking for.
1042 unsigned long offset = hash_x_index(*hash, hash64);
1046 spin_lock_irq(&mapping->tree_lock);
1047 found = radix_tree_gang_lookup(&mapping->page_tree,
1048 (void **)&page, offset, 1);
1049 if (found > 0 && !radix_tree_exceptional_entry(page)) {
1050 struct lu_dirpage *dp;
1052 page_cache_get(page);
1053 spin_unlock_irq(&mapping->tree_lock);
1055 * In contrast to find_lock_page() we are sure that directory
1056 * page cannot be truncated (while DLM lock is held) and,
1057 * hence, can avoid restart.
1059 * In fact, page cannot be locked here at all, because
1060 * mdc_read_page_remote does synchronous io.
1062 wait_on_page_locked(page);
1063 if (PageUptodate(page)) {
1065 if (BITS_PER_LONG == 32 && hash64) {
1066 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1067 *end = le64_to_cpu(dp->ldp_hash_end) >> 32;
1068 *hash = *hash >> 32;
1070 *start = le64_to_cpu(dp->ldp_hash_start);
1071 *end = le64_to_cpu(dp->ldp_hash_end);
1073 if (unlikely(*start == 1 && *hash == 0))
1076 LASSERTF(*start <= *hash, "start = "LPX64
1077 ",end = "LPX64",hash = "LPX64"\n",
1078 *start, *end, *hash);
1079 CDEBUG(D_VFSTRACE, "offset %lx ["LPX64" "LPX64"],"
1080 " hash "LPX64"\n", offset, *start, *end, *hash);
1083 mdc_release_page(page, 0);
1085 } else if (*end != *start && *hash == *end) {
1087 * upon hash collision, remove this page,
1088 * otherwise put page reference, and
1089 * mdc_read_page_remote() will issue RPC to
1090 * fetch the page we want.
1093 mdc_release_page(page,
1094 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1098 page_cache_release(page);
1099 page = ERR_PTR(-EIO);
1102 spin_unlock_irq(&mapping->tree_lock);
1109 * Adjust a set of pages, each page containing an array of lu_dirpages,
1110 * so that each page can be used as a single logical lu_dirpage.
1112 * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1113 * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1114 * struct lu_dirent. It has size up to LU_PAGE_SIZE. The ldp_hash_end
1115 * value is used as a cookie to request the next lu_dirpage in a
1116 * directory listing that spans multiple pages (two in this example):
1119 * .|--------v------- -----.
1120 * |s|e|f|p|ent|ent| ... |ent|
1121 * '--|-------------- -----' Each PAGE contains a single
1122 * '------. lu_dirpage.
1123 * .---------v------- -----.
1124 * |s|e|f|p|ent| 0 | ... | 0 |
1125 * '----------------- -----'
1127 * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
1128 * larger than LU_PAGE_SIZE, a single host page may contain multiple
1129 * lu_dirpages. After reading the lu_dirpages from the MDS, the
1130 * ldp_hash_end of the first lu_dirpage refers to the one immediately
1131 * after it in the same PAGE (arrows simplified for brevity, but
1132 * in general e0==s1, e1==s2, etc.):
1134 * .-------------------- -----.
1135 * |s0|e0|f0|p|ent|ent| ... |ent|
1136 * |---v---------------- -----|
1137 * |s1|e1|f1|p|ent|ent| ... |ent|
1138 * |---v---------------- -----| Here, each PAGE contains
1139 * ... multiple lu_dirpages.
1140 * |---v---------------- -----|
1141 * |s'|e'|f'|p|ent|ent| ... |ent|
1142 * '---|---------------- -----'
1144 * .----------------------------.
1147 * This structure is transformed into a single logical lu_dirpage as follows:
1149 * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1150 * labeled 'next PAGE'.
1152 * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1153 * a hash collision with the next page exists.
1155 * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1156 * to the first entry of the next lu_dirpage.
1158 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
1159 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1163 for (i = 0; i < cfs_pgs; i++) {
1164 struct lu_dirpage *dp = kmap(pages[i]);
1165 struct lu_dirpage *first = dp;
1166 struct lu_dirent *end_dirent = NULL;
1167 struct lu_dirent *ent;
1168 __u64 hash_end = le64_to_cpu(dp->ldp_hash_end);
1169 __u32 flags = le32_to_cpu(dp->ldp_flags);
1171 while (--lu_pgs > 0) {
1172 ent = lu_dirent_start(dp);
1173 for (end_dirent = ent; ent != NULL;
1174 end_dirent = ent, ent = lu_dirent_next(ent));
1176 /* Advance dp to next lu_dirpage. */
1177 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1179 /* Check if we've reached the end of the PAGE. */
1180 if (!((unsigned long)dp & ~PAGE_MASK))
1183 /* Save the hash and flags of this lu_dirpage. */
1184 hash_end = le64_to_cpu(dp->ldp_hash_end);
1185 flags = le32_to_cpu(dp->ldp_flags);
1187 /* Check if lu_dirpage contains no entries. */
1188 if (end_dirent == NULL)
1191 /* Enlarge the end entry lde_reclen from 0 to
1192 * first entry of next lu_dirpage. */
1193 LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1194 end_dirent->lde_reclen =
1195 cpu_to_le16((char *)(dp->ldp_entries) -
1196 (char *)end_dirent);
1199 first->ldp_hash_end = hash_end;
1200 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1201 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1205 LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1208 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1209 #endif /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
1211 /* parameters for readdir page */
1212 struct readpage_param {
1213 struct md_op_data *rp_mod;
1216 struct obd_export *rp_exp;
1217 struct md_callback *rp_cb;
1220 #ifndef HAVE_DELETE_FROM_PAGE_CACHE
1221 static inline void delete_from_page_cache(struct page *page)
1223 remove_from_page_cache(page);
1224 page_cache_release(page);
1229 * Read pages from server.
1231 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1232 * a header lu_dirpage which describes the start/end hash, and whether this
1233 * page is empty (contains no dir entry) or hash collide with next page.
1234 * After client receives reply, several pages will be integrated into dir page
1235 * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1236 * lu_dirpage for this integrated page will be adjusted.
1238 static int mdc_read_page_remote(void *data, struct page *page0)
1240 struct readpage_param *rp = data;
1241 struct page **page_pool;
1243 struct lu_dirpage *dp;
1244 int rd_pgs = 0; /* number of pages read actually */
1246 struct md_op_data *op_data = rp->rp_mod;
1247 struct ptlrpc_request *req;
1248 int max_pages = op_data->op_max_pages;
1249 struct inode *inode;
1255 LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
1256 inode = op_data->op_data;
1257 fid = &op_data->op_fid1;
1258 LASSERT(inode != NULL);
1260 OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1261 if (page_pool != NULL) {
1262 page_pool[0] = page0;
1268 for (npages = 1; npages < max_pages; npages++) {
1269 page = page_cache_alloc_cold(inode->i_mapping);
1272 page_pool[npages] = page;
1275 rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, op_data->op_capa1,
1276 page_pool, npages, &req);
1278 /* page0 is special, which was added into page cache early */
1279 delete_from_page_cache(page0);
1283 rd_pgs = (req->rq_bulk->bd_nob_transferred +
1284 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1285 lu_pgs = req->rq_bulk->bd_nob_transferred >>
1287 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1289 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1291 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1293 SetPageUptodate(page0);
1297 ptlrpc_req_finished(req);
1298 CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1299 for (i = 1; i < npages; i++) {
1300 unsigned long offset;
1304 page = page_pool[i];
1306 if (rc < 0 || i >= rd_pgs) {
1307 page_cache_release(page);
1311 SetPageUptodate(page);
1314 hash = le64_to_cpu(dp->ldp_hash_start);
1317 offset = hash_x_index(hash, rp->rp_hash64);
1319 prefetchw(&page->flags);
1320 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1325 CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1326 " rc = %d\n", offset, ret);
1327 page_cache_release(page);
1330 if (page_pool != &page0)
1331 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1337 * Read dir page from cache first, if it can not find it, read it from
1338 * server and add into the cache.
1340 * \param[in] exp MDC export
1341 * \param[in] op_data client MD stack parameters, transfering parameters
1342 * between different layers on client MD stack.
1343 * \param[in] cb_op callback required for ldlm lock enqueue during
1345 * \param[in] hash_offset the hash offset of the page to be read
1346 * \param[in] ppage the page to be read
1348 * retval = 0 get the page successfully
1349 * errno(<0) get the page failed
1351 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1352 struct md_callback *cb_op, __u64 hash_offset,
1353 struct page **ppage)
1355 struct lookup_intent it = { .it_op = IT_READDIR };
1357 struct inode *dir = op_data->op_data;
1358 struct address_space *mapping;
1359 struct lu_dirpage *dp;
1362 struct lustre_handle lockh;
1363 struct ptlrpc_request *enq_req = NULL;
1364 struct readpage_param rp_param;
1371 LASSERT(dir != NULL);
1372 mapping = dir->i_mapping;
1374 rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1375 cb_op->md_blocking_ast, 0);
1376 if (enq_req != NULL)
1377 ptlrpc_req_finished(enq_req);
1380 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1381 exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1386 mdc_set_lock_data(exp, &it.d.lustre.it_lock_handle, dir, NULL);
1388 rp_param.rp_off = hash_offset;
1389 rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1390 page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1391 rp_param.rp_hash64);
1393 CERROR("%s: dir page locate: "DFID" at "LPU64": rc %ld\n",
1394 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1395 rp_param.rp_off, PTR_ERR(page));
1396 GOTO(out_unlock, rc = PTR_ERR(page));
1397 } else if (page != NULL) {
1399 * XXX nikita: not entirely correct handling of a corner case:
1400 * suppose hash chain of entries with hash value HASH crosses
1401 * border between pages P0 and P1. First both P0 and P1 are
1402 * cached, seekdir() is called for some entry from the P0 part
1403 * of the chain. Later P0 goes out of cache. telldir(HASH)
1404 * happens and finds P1, as it starts with matching hash
1405 * value. Remaining entries from P0 part of the chain are
1406 * skipped. (Is that really a bug?)
1408 * Possible solutions: 0. don't cache P1 is such case, handle
1409 * it as an "overflow" page. 1. invalidate all pages at
1410 * once. 2. use HASH|1 as an index for P1.
1412 GOTO(hash_collision, page);
1415 rp_param.rp_exp = exp;
1416 rp_param.rp_mod = op_data;
1417 page = read_cache_page(mapping,
1418 hash_x_index(rp_param.rp_off,
1419 rp_param.rp_hash64),
1420 mdc_read_page_remote, &rp_param);
1422 CDEBUG(D_INFO, "%s: read cache page: "DFID" at "LPU64": %ld\n",
1423 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1424 rp_param.rp_off, PTR_ERR(page));
1425 GOTO(out_unlock, rc = PTR_ERR(page));
1428 wait_on_page_locked(page);
1430 if (!PageUptodate(page)) {
1431 CERROR("%s: page not updated: "DFID" at "LPU64": rc %d\n",
1432 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1433 rp_param.rp_off, -5);
1436 if (!PageChecked(page))
1437 SetPageChecked(page);
1438 if (PageError(page)) {
1439 CERROR("%s: page error: "DFID" at "LPU64": rc %d\n",
1440 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1441 rp_param.rp_off, -5);
1446 dp = page_address(page);
1447 if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1448 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1449 end = le64_to_cpu(dp->ldp_hash_end) >> 32;
1450 rp_param.rp_off = hash_offset >> 32;
1452 start = le64_to_cpu(dp->ldp_hash_start);
1453 end = le64_to_cpu(dp->ldp_hash_end);
1454 rp_param.rp_off = hash_offset;
1457 LASSERT(start == rp_param.rp_off);
1458 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1459 #if BITS_PER_LONG == 32
1460 CWARN("Real page-wide hash collision at ["LPU64" "LPU64"] with "
1461 "hash "LPU64"\n", le64_to_cpu(dp->ldp_hash_start),
1462 le64_to_cpu(dp->ldp_hash_end), hash_offset);
1466 * Fetch whole overflow chain...
1474 lockh.cookie = it.d.lustre.it_lock_handle;
1475 ldlm_lock_decref(&lockh, it.d.lustre.it_lock_mode);
1476 it.d.lustre.it_lock_handle = 0;
1480 mdc_release_page(page, 1);
1486 static int mdc_statfs(const struct lu_env *env,
1487 struct obd_export *exp, struct obd_statfs *osfs,
1488 __u64 max_age, __u32 flags)
1490 struct obd_device *obd = class_exp2obd(exp);
1491 struct ptlrpc_request *req;
1492 struct obd_statfs *msfs;
1493 struct obd_import *imp = NULL;
1498 * Since the request might also come from lprocfs, so we need
1499 * sync this with client_disconnect_export Bug15684
1501 down_read(&obd->u.cli.cl_sem);
1502 if (obd->u.cli.cl_import)
1503 imp = class_import_get(obd->u.cli.cl_import);
1504 up_read(&obd->u.cli.cl_sem);
1508 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1509 LUSTRE_MDS_VERSION, MDS_STATFS);
1511 GOTO(output, rc = -ENOMEM);
1513 ptlrpc_request_set_replen(req);
1515 if (flags & OBD_STATFS_NODELAY) {
1516 /* procfs requests not want stay in wait for avoid deadlock */
1517 req->rq_no_resend = 1;
1518 req->rq_no_delay = 1;
1521 rc = ptlrpc_queue_wait(req);
1523 /* check connection error first */
1524 if (imp->imp_connect_error)
1525 rc = imp->imp_connect_error;
1529 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1531 GOTO(out, rc = -EPROTO);
1536 ptlrpc_req_finished(req);
1538 class_import_put(imp);
1542 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1544 __u32 keylen, vallen;
1548 if (gf->gf_pathlen > PATH_MAX)
1549 RETURN(-ENAMETOOLONG);
1550 if (gf->gf_pathlen < 2)
1553 /* Key is KEY_FID2PATH + getinfo_fid2path description */
1554 keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1555 OBD_ALLOC(key, keylen);
1558 memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1559 memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1561 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1562 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1564 if (!fid_is_sane(&gf->gf_fid))
1565 GOTO(out, rc = -EINVAL);
1567 /* Val is struct getinfo_fid2path result plus path */
1568 vallen = sizeof(*gf) + gf->gf_pathlen;
1570 rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1571 if (rc != 0 && rc != -EREMOTE)
1574 if (vallen <= sizeof(*gf))
1575 GOTO(out, rc = -EPROTO);
1576 else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1577 GOTO(out, rc = -EOVERFLOW);
1579 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1580 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1583 OBD_FREE(key, keylen);
1587 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1588 struct hsm_progress_kernel *hpk)
1590 struct obd_import *imp = class_exp2cliimp(exp);
1591 struct hsm_progress_kernel *req_hpk;
1592 struct ptlrpc_request *req;
1596 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1597 LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1599 GOTO(out, rc = -ENOMEM);
1601 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1603 /* Copy hsm_progress struct */
1604 req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1605 if (req_hpk == NULL)
1606 GOTO(out, rc = -EPROTO);
1609 req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1611 ptlrpc_request_set_replen(req);
1613 rc = mdc_queue_wait(req);
1616 ptlrpc_req_finished(req);
1620 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1622 __u32 *archive_mask;
1623 struct ptlrpc_request *req;
1627 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1629 MDS_HSM_CT_REGISTER);
1631 GOTO(out, rc = -ENOMEM);
1633 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1635 /* Copy hsm_progress struct */
1636 archive_mask = req_capsule_client_get(&req->rq_pill,
1637 &RMF_MDS_HSM_ARCHIVE);
1638 if (archive_mask == NULL)
1639 GOTO(out, rc = -EPROTO);
1641 *archive_mask = archives;
1643 ptlrpc_request_set_replen(req);
1645 rc = mdc_queue_wait(req);
1648 ptlrpc_req_finished(req);
1652 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1653 struct md_op_data *op_data)
1655 struct hsm_current_action *hca = op_data->op_data;
1656 struct hsm_current_action *req_hca;
1657 struct ptlrpc_request *req;
1661 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1662 &RQF_MDS_HSM_ACTION);
1666 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1668 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1670 ptlrpc_request_free(req);
1674 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1675 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1677 ptlrpc_request_set_replen(req);
1679 rc = mdc_queue_wait(req);
1683 req_hca = req_capsule_server_get(&req->rq_pill,
1684 &RMF_MDS_HSM_CURRENT_ACTION);
1685 if (req_hca == NULL)
1686 GOTO(out, rc = -EPROTO);
1692 ptlrpc_req_finished(req);
1696 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1698 struct ptlrpc_request *req;
1702 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1704 MDS_HSM_CT_UNREGISTER);
1706 GOTO(out, rc = -ENOMEM);
1708 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1710 ptlrpc_request_set_replen(req);
1712 rc = mdc_queue_wait(req);
1715 ptlrpc_req_finished(req);
1719 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1720 struct md_op_data *op_data)
1722 struct hsm_user_state *hus = op_data->op_data;
1723 struct hsm_user_state *req_hus;
1724 struct ptlrpc_request *req;
1728 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1729 &RQF_MDS_HSM_STATE_GET);
1733 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1735 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1737 ptlrpc_request_free(req);
1741 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1742 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1744 ptlrpc_request_set_replen(req);
1746 rc = mdc_queue_wait(req);
1750 req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1751 if (req_hus == NULL)
1752 GOTO(out, rc = -EPROTO);
1758 ptlrpc_req_finished(req);
1762 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1763 struct md_op_data *op_data)
1765 struct hsm_state_set *hss = op_data->op_data;
1766 struct hsm_state_set *req_hss;
1767 struct ptlrpc_request *req;
1771 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1772 &RQF_MDS_HSM_STATE_SET);
1776 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1778 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1780 ptlrpc_request_free(req);
1784 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1785 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1788 req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1789 if (req_hss == NULL)
1790 GOTO(out, rc = -EPROTO);
1793 ptlrpc_request_set_replen(req);
1795 rc = mdc_queue_wait(req);
1800 ptlrpc_req_finished(req);
1804 static int mdc_ioc_hsm_request(struct obd_export *exp,
1805 struct hsm_user_request *hur)
1807 struct obd_import *imp = class_exp2cliimp(exp);
1808 struct ptlrpc_request *req;
1809 struct hsm_request *req_hr;
1810 struct hsm_user_item *req_hui;
1815 req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1817 GOTO(out, rc = -ENOMEM);
1819 req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1820 hur->hur_request.hr_itemcount
1821 * sizeof(struct hsm_user_item));
1822 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1823 hur->hur_request.hr_data_len);
1825 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1827 ptlrpc_request_free(req);
1831 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1833 /* Copy hsm_request struct */
1834 req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1836 GOTO(out, rc = -EPROTO);
1837 *req_hr = hur->hur_request;
1839 /* Copy hsm_user_item structs */
1840 req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1841 if (req_hui == NULL)
1842 GOTO(out, rc = -EPROTO);
1843 memcpy(req_hui, hur->hur_user_item,
1844 hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1846 /* Copy opaque field */
1847 req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1848 if (req_opaque == NULL)
1849 GOTO(out, rc = -EPROTO);
1850 memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1852 ptlrpc_request_set_replen(req);
1854 rc = mdc_queue_wait(req);
1858 ptlrpc_req_finished(req);
1862 static struct kuc_hdr *changelog_kuc_hdr(char *buf, size_t len, __u32 flags)
1864 struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1866 LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1868 lh->kuc_magic = KUC_MAGIC;
1869 lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1870 lh->kuc_flags = flags;
1871 lh->kuc_msgtype = CL_RECORD;
1872 lh->kuc_msglen = len;
1876 struct changelog_show {
1878 enum changelog_send_flag cs_flags;
1881 struct obd_device *cs_obd;
1884 static inline char *cs_obd_name(struct changelog_show *cs)
1886 return cs->cs_obd->obd_name;
1889 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1890 struct llog_rec_hdr *hdr, void *data)
1892 struct changelog_show *cs = data;
1893 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1899 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1901 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1902 cs_obd_name(cs), rec->cr_hdr.lrh_type,
1903 rec->cr.cr_type, rc);
1907 if (rec->cr.cr_index < cs->cs_startrec) {
1908 /* Skip entries earlier than what we are interested in */
1909 CDEBUG(D_HSM, "rec="LPU64" start="LPU64"\n",
1910 rec->cr.cr_index, cs->cs_startrec);
1914 CDEBUG(D_HSM, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID" %.*s\n",
1915 rec->cr.cr_index, rec->cr.cr_type,
1916 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1917 rec->cr.cr_flags & CLF_FLAGMASK,
1918 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1919 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1921 len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1923 /* Set up the message */
1924 lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1925 memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1927 rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1928 CDEBUG(D_HSM, "kucmsg fp %p len %zu rc %d\n", cs->cs_fp, len, rc);
1933 static int mdc_changelog_send_thread(void *csdata)
1935 struct changelog_show *cs = csdata;
1936 struct llog_ctxt *ctxt = NULL;
1937 struct llog_handle *llh = NULL;
1938 struct kuc_hdr *kuch;
1939 enum llog_flag flags = LLOG_F_IS_CAT;
1942 CDEBUG(D_HSM, "changelog to fp=%p start "LPU64"\n",
1943 cs->cs_fp, cs->cs_startrec);
1945 OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1946 if (cs->cs_buf == NULL)
1947 GOTO(out, rc = -ENOMEM);
1949 /* Set up the remote catalog handle */
1950 ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1952 GOTO(out, rc = -ENOENT);
1953 rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1956 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1957 cs_obd_name(cs), rc);
1961 if (cs->cs_flags & CHANGELOG_FLAG_JOBID)
1962 flags |= LLOG_F_EXT_JOBID;
1964 rc = llog_init_handle(NULL, llh, flags, NULL);
1966 CERROR("llog_init_handle failed %d\n", rc);
1970 rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1972 /* Send EOF no matter what our result */
1973 if ((kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch),
1975 kuch->kuc_msgtype = CL_EOF;
1976 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1982 llog_cat_close(NULL, llh);
1984 llog_ctxt_put(ctxt);
1986 OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1991 static int mdc_ioc_changelog_send(struct obd_device *obd,
1992 struct ioc_changelog *icc)
1994 struct changelog_show *cs;
1995 struct task_struct *task;
1998 /* Freed in mdc_changelog_send_thread */
2004 cs->cs_startrec = icc->icc_recno;
2005 /* matching fput in mdc_changelog_send_thread */
2006 cs->cs_fp = fget(icc->icc_id);
2007 cs->cs_flags = icc->icc_flags;
2010 * New thread because we should return to user app before
2011 * writing into our pipe
2013 task = kthread_run(mdc_changelog_send_thread, cs,
2014 "mdc_clg_send_thread");
2017 CERROR("%s: cannot start changelog thread: rc = %d\n",
2018 cs_obd_name(cs), rc);
2022 CDEBUG(D_HSM, "%s: started changelog thread\n",
2029 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2030 struct lustre_kernelcomm *lk);
2032 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
2033 struct obd_quotactl *oqctl)
2035 struct client_obd *cli = &exp->exp_obd->u.cli;
2036 struct ptlrpc_request *req;
2037 struct obd_quotactl *body;
2041 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2042 &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
2047 body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2050 ptlrpc_request_set_replen(req);
2052 /* the next poll will find -ENODATA, that means quotacheck is
2054 cli->cl_qchk_stat = -ENODATA;
2055 rc = ptlrpc_queue_wait(req);
2057 cli->cl_qchk_stat = rc;
2058 ptlrpc_req_finished(req);
2062 static int mdc_quota_poll_check(struct obd_export *exp,
2063 struct if_quotacheck *qchk)
2065 struct client_obd *cli = &exp->exp_obd->u.cli;
2069 qchk->obd_uuid = cli->cl_target_uuid;
2070 memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
2072 rc = cli->cl_qchk_stat;
2073 /* the client is not the previous one */
2074 if (rc == CL_NOT_QUOTACHECKED)
2079 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
2080 struct obd_quotactl *oqctl)
2082 struct ptlrpc_request *req;
2083 struct obd_quotactl *oqc;
2087 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2088 &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
2093 oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2096 ptlrpc_request_set_replen(req);
2097 ptlrpc_at_set_req_timeout(req);
2098 req->rq_no_resend = 1;
2100 rc = ptlrpc_queue_wait(req);
2102 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
2104 if (req->rq_repmsg &&
2105 (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2108 CERROR ("Can't unpack obd_quotactl\n");
2111 ptlrpc_req_finished(req);
2116 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2117 struct md_op_data *op_data)
2119 struct list_head cancels = LIST_HEAD_INIT(cancels);
2120 struct ptlrpc_request *req;
2122 struct mdc_swap_layouts *msl, *payload;
2125 msl = op_data->op_data;
2127 /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2128 * first thing it will do is to cancel the 2 layout
2129 * locks hold by this client.
2130 * So the client must cancel its layout locks on the 2 fids
2131 * with the request RPC to avoid extra RPC round trips
2133 count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2134 LCK_EX, MDS_INODELOCK_LAYOUT |
2135 MDS_INODELOCK_XATTR);
2136 count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2137 LCK_EX, MDS_INODELOCK_LAYOUT |
2138 MDS_INODELOCK_XATTR);
2140 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2141 &RQF_MDS_SWAP_LAYOUTS);
2143 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2147 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2148 mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
2150 rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2152 ptlrpc_request_free(req);
2156 mdc_swap_layouts_pack(req, op_data);
2158 payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2163 ptlrpc_request_set_replen(req);
2165 rc = ptlrpc_queue_wait(req);
2171 ptlrpc_req_finished(req);
2175 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2176 void *karg, void __user *uarg)
2178 struct obd_device *obd = exp->exp_obd;
2179 struct obd_ioctl_data *data = karg;
2180 struct obd_import *imp = obd->u.cli.cl_import;
2184 if (!try_module_get(THIS_MODULE)) {
2185 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2186 module_name(THIS_MODULE));
2190 case OBD_IOC_CHANGELOG_SEND:
2191 rc = mdc_ioc_changelog_send(obd, karg);
2193 case OBD_IOC_CHANGELOG_CLEAR: {
2194 struct ioc_changelog *icc = karg;
2195 struct changelog_setinfo cs =
2196 {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
2197 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
2198 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
2202 case OBD_IOC_FID2PATH:
2203 rc = mdc_ioc_fid2path(exp, karg);
2205 case LL_IOC_HSM_CT_START:
2206 rc = mdc_ioc_hsm_ct_start(exp, karg);
2207 /* ignore if it was already registered on this MDS. */
2211 case LL_IOC_HSM_PROGRESS:
2212 rc = mdc_ioc_hsm_progress(exp, karg);
2214 case LL_IOC_HSM_STATE_GET:
2215 rc = mdc_ioc_hsm_state_get(exp, karg);
2217 case LL_IOC_HSM_STATE_SET:
2218 rc = mdc_ioc_hsm_state_set(exp, karg);
2220 case LL_IOC_HSM_ACTION:
2221 rc = mdc_ioc_hsm_current_action(exp, karg);
2223 case LL_IOC_HSM_REQUEST:
2224 rc = mdc_ioc_hsm_request(exp, karg);
2226 case OBD_IOC_CLIENT_RECOVER:
2227 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2231 case IOC_OSC_SET_ACTIVE:
2232 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2234 case OBD_IOC_POLL_QUOTACHECK:
2235 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
2237 case OBD_IOC_PING_TARGET:
2238 rc = ptlrpc_obd_ping(obd);
2241 * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2242 * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2243 * there'd be no LMV layer thus we might be called here. Eventually
2244 * this code should be removed.
2247 case IOC_OBD_STATFS: {
2248 struct obd_statfs stat_buf = {0};
2250 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2251 GOTO(out, rc = -ENODEV);
2254 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2255 min((int)data->ioc_plen2,
2256 (int)sizeof(struct obd_uuid))))
2257 GOTO(out, rc = -EFAULT);
2259 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2260 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2265 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2266 min((int) data->ioc_plen1,
2267 (int) sizeof(stat_buf))))
2268 GOTO(out, rc = -EFAULT);
2272 case OBD_IOC_QUOTACTL: {
2273 struct if_quotactl *qctl = karg;
2274 struct obd_quotactl *oqctl;
2276 OBD_ALLOC_PTR(oqctl);
2278 GOTO(out, rc = -ENOMEM);
2280 QCTL_COPY(oqctl, qctl);
2281 rc = obd_quotactl(exp, oqctl);
2283 QCTL_COPY(qctl, oqctl);
2284 qctl->qc_valid = QC_MDTIDX;
2285 qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2288 OBD_FREE_PTR(oqctl);
2291 case LL_IOC_GET_CONNECT_FLAGS:
2292 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2293 sizeof(*exp_connect_flags_ptr(exp))))
2294 GOTO(out, rc = -EFAULT);
2297 case LL_IOC_LOV_SWAP_LAYOUTS:
2298 rc = mdc_ioc_swap_layouts(exp, karg);
2301 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2302 GOTO(out, rc = -ENOTTY);
2305 module_put(THIS_MODULE);
2310 static int mdc_get_info_rpc(struct obd_export *exp,
2311 u32 keylen, void *key,
2312 u32 vallen, void *val)
2314 struct obd_import *imp = class_exp2cliimp(exp);
2315 struct ptlrpc_request *req;
2320 req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2324 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2325 RCL_CLIENT, keylen);
2326 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2327 RCL_CLIENT, sizeof(vallen));
2329 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2331 ptlrpc_request_free(req);
2335 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2336 memcpy(tmp, key, keylen);
2337 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2338 memcpy(tmp, &vallen, sizeof(vallen));
2340 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2341 RCL_SERVER, vallen);
2342 ptlrpc_request_set_replen(req);
2344 rc = ptlrpc_queue_wait(req);
2345 /* -EREMOTE means the get_info result is partial, and it needs to
2346 * continue on another MDT, see fid2path part in lmv_iocontrol */
2347 if (rc == 0 || rc == -EREMOTE) {
2348 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2349 memcpy(val, tmp, vallen);
2350 if (ptlrpc_rep_need_swab(req)) {
2351 if (KEY_IS(KEY_FID2PATH))
2352 lustre_swab_fid2path(val);
2355 ptlrpc_req_finished(req);
2360 static void lustre_swab_hai(struct hsm_action_item *h)
2362 __swab32s(&h->hai_len);
2363 __swab32s(&h->hai_action);
2364 lustre_swab_lu_fid(&h->hai_fid);
2365 lustre_swab_lu_fid(&h->hai_dfid);
2366 __swab64s(&h->hai_cookie);
2367 __swab64s(&h->hai_extent.offset);
2368 __swab64s(&h->hai_extent.length);
2369 __swab64s(&h->hai_gid);
2372 static void lustre_swab_hal(struct hsm_action_list *h)
2374 struct hsm_action_item *hai;
2377 __swab32s(&h->hal_version);
2378 __swab32s(&h->hal_count);
2379 __swab32s(&h->hal_archive_id);
2380 __swab64s(&h->hal_flags);
2382 for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2383 lustre_swab_hai(hai);
2386 static void lustre_swab_kuch(struct kuc_hdr *l)
2388 __swab16s(&l->kuc_magic);
2389 /* __u8 l->kuc_transport */
2390 __swab16s(&l->kuc_msgtype);
2391 __swab16s(&l->kuc_msglen);
2394 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2395 struct lustre_kernelcomm *lk)
2397 struct obd_import *imp = class_exp2cliimp(exp);
2398 __u32 archive = lk->lk_data;
2401 if (lk->lk_group != KUC_GRP_HSM) {
2402 CERROR("Bad copytool group %d\n", lk->lk_group);
2406 CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2407 lk->lk_uid, lk->lk_group, lk->lk_flags);
2409 if (lk->lk_flags & LK_FLG_STOP) {
2410 /* Unregister with the coordinator */
2411 rc = mdc_ioc_hsm_ct_unregister(imp);
2413 rc = mdc_ioc_hsm_ct_register(imp, archive);
2420 * Send a message to any listening copytools
2421 * @param val KUC message (kuc_hdr + hsm_action_list)
2422 * @param len total length of message
2424 static int mdc_hsm_copytool_send(size_t len, void *val)
2426 struct kuc_hdr *lh = (struct kuc_hdr *)val;
2427 struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
2431 if (len < sizeof(*lh) + sizeof(*hal)) {
2432 CERROR("Short HSM message %zu < %zu\n", len,
2433 sizeof(*lh) + sizeof(*hal));
2436 if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2437 lustre_swab_kuch(lh);
2438 lustre_swab_hal(hal);
2439 } else if (lh->kuc_magic != KUC_MAGIC) {
2440 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2444 CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2446 lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2447 lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2449 /* Broadcast to HSM listeners */
2450 rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2456 * callback function passed to kuc for re-registering each HSM copytool
2457 * running on MDC, after MDT shutdown/recovery.
2458 * @param data copytool registration data
2459 * @param cb_arg callback argument (obd_import)
2461 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2463 struct kkuc_ct_data *kcd = data;
2464 struct obd_import *imp = (struct obd_import *)cb_arg;
2467 if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2470 if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2473 CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2474 imp->imp_obd->obd_name, kcd->kcd_archive);
2475 rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2477 /* ignore error if the copytool is already registered */
2478 return (rc == -EEXIST) ? 0 : rc;
2482 * Re-establish all kuc contexts with MDT
2483 * after MDT shutdown/recovery.
2485 static int mdc_kuc_reregister(struct obd_import *imp)
2487 /* re-register HSM agents */
2488 return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2492 static int mdc_set_info_async(const struct lu_env *env,
2493 struct obd_export *exp,
2494 u32 keylen, void *key,
2495 u32 vallen, void *val,
2496 struct ptlrpc_request_set *set)
2498 struct obd_import *imp = class_exp2cliimp(exp);
2502 if (KEY_IS(KEY_READ_ONLY)) {
2503 if (vallen != sizeof(int))
2506 spin_lock(&imp->imp_lock);
2507 if (*((int *)val)) {
2508 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2509 imp->imp_connect_data.ocd_connect_flags |=
2512 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2513 imp->imp_connect_data.ocd_connect_flags &=
2514 ~OBD_CONNECT_RDONLY;
2516 spin_unlock(&imp->imp_lock);
2518 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2519 keylen, key, vallen, val, set);
2522 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2523 sptlrpc_conf_client_adapt(exp->exp_obd);
2526 if (KEY_IS(KEY_FLUSH_CTX)) {
2527 sptlrpc_import_flush_my_ctx(imp);
2530 if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2531 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2532 keylen, key, vallen, val, set);
2535 if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2536 rc = mdc_hsm_copytool_send(vallen, val);
2540 if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2541 __u32 *default_easize = val;
2543 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2547 CERROR("Unknown key %s\n", (char *)key);
2551 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2552 __u32 keylen, void *key,
2553 __u32 *vallen, void *val,
2554 struct lov_stripe_md *lsm)
2558 if (KEY_IS(KEY_MAX_EASIZE)) {
2559 __u32 mdsize, *max_easize;
2561 if (*vallen != sizeof(int))
2563 mdsize = *(__u32 *)val;
2564 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2565 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2567 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2569 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2570 __u32 *default_easize;
2572 if (*vallen != sizeof(int))
2574 default_easize = val;
2575 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2577 } else if (KEY_IS(KEY_MAX_COOKIESIZE)) {
2578 __u32 *max_cookiesize;
2580 if (*vallen != sizeof(int))
2582 max_cookiesize = val;
2583 *max_cookiesize = exp->exp_obd->u.cli.cl_max_mds_cookiesize;
2585 } else if (KEY_IS(KEY_DEFAULT_COOKIESIZE)) {
2586 __u32 *default_cookiesize;
2588 if (*vallen != sizeof(int))
2590 default_cookiesize = val;
2591 *default_cookiesize =
2592 exp->exp_obd->u.cli.cl_default_mds_cookiesize;
2594 } else if (KEY_IS(KEY_CONN_DATA)) {
2595 struct obd_import *imp = class_exp2cliimp(exp);
2596 struct obd_connect_data *data = val;
2598 if (*vallen != sizeof(*data))
2601 *data = imp->imp_connect_data;
2603 } else if (KEY_IS(KEY_TGT_COUNT)) {
2604 *((__u32 *)val) = 1;
2608 rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2613 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2614 struct obd_capa *oc, struct ptlrpc_request **request)
2616 struct ptlrpc_request *req;
2621 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2625 mdc_set_capa_size(req, &RMF_CAPA1, oc);
2627 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2629 ptlrpc_request_free(req);
2633 mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
2635 ptlrpc_request_set_replen(req);
2637 rc = ptlrpc_queue_wait(req);
2639 ptlrpc_req_finished(req);
2645 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2646 enum obd_import_event event)
2650 LASSERT(imp->imp_obd == obd);
2653 case IMP_EVENT_DISCON: {
2655 /* XXX Pass event up to OBDs stack. used only for FLD now */
2656 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2660 case IMP_EVENT_INACTIVE: {
2661 struct client_obd *cli = &obd->u.cli;
2663 * Flush current sequence to make client obtain new one
2664 * from server in case of disconnect/reconnect.
2666 if (cli->cl_seq != NULL)
2667 seq_client_flush(cli->cl_seq);
2669 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2672 case IMP_EVENT_INVALIDATE: {
2673 struct ldlm_namespace *ns = obd->obd_namespace;
2675 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2679 case IMP_EVENT_ACTIVE:
2680 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2681 /* redo the kuc registration after reconnecting */
2683 rc = mdc_kuc_reregister(imp);
2686 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2688 case IMP_EVENT_DEACTIVATE:
2689 case IMP_EVENT_ACTIVATE:
2692 CERROR("Unknown import event %x\n", event);
2698 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2699 struct lu_fid *fid, struct md_op_data *op_data)
2701 struct client_obd *cli = &exp->exp_obd->u.cli;
2702 struct lu_client_seq *seq = cli->cl_seq;
2704 RETURN(seq_client_alloc_fid(env, seq, fid));
2707 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2709 struct client_obd *cli = &exp->exp_obd->u.cli;
2710 return &cli->cl_target_uuid;
2714 * Determine whether the lock can be canceled before replaying it during
2715 * recovery, non zero value will be return if the lock can be canceled,
2716 * or zero returned for not
2718 static int mdc_cancel_weight(struct ldlm_lock *lock)
2720 if (lock->l_resource->lr_type != LDLM_IBITS)
2723 /* FIXME: if we ever get into a situation where there are too many
2724 * opened files with open locks on a single node, then we really
2725 * should replay these open locks to reget it */
2726 if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2732 static int mdc_resource_inode_free(struct ldlm_resource *res)
2734 if (res->lr_lvb_inode)
2735 res->lr_lvb_inode = NULL;
2740 static struct ldlm_valblock_ops inode_lvbo = {
2741 .lvbo_free = mdc_resource_inode_free
2744 static int mdc_llog_init(struct obd_device *obd)
2746 struct obd_llog_group *olg = &obd->obd_olg;
2747 struct llog_ctxt *ctxt;
2752 rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2757 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2758 llog_initiator_connect(ctxt);
2759 llog_ctxt_put(ctxt);
2764 static void mdc_llog_finish(struct obd_device *obd)
2766 struct llog_ctxt *ctxt;
2770 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2772 llog_cleanup(NULL, ctxt);
2777 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2779 struct client_obd *cli = &obd->u.cli;
2783 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2784 if (!cli->cl_rpc_lock)
2786 mdc_init_rpc_lock(cli->cl_rpc_lock);
2788 rc = ptlrpcd_addref();
2790 GOTO(err_rpc_lock, rc);
2792 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2793 if (!cli->cl_close_lock)
2794 GOTO(err_ptlrpcd_decref, rc = -ENOMEM);
2795 mdc_init_rpc_lock(cli->cl_close_lock);
2797 rc = client_obd_setup(obd, cfg);
2799 GOTO(err_close_lock, rc);
2800 #ifdef CONFIG_PROC_FS
2801 obd->obd_vars = lprocfs_mdc_obd_vars;
2802 lprocfs_obd_setup(obd);
2803 lprocfs_alloc_md_stats(obd, 0);
2805 sptlrpc_lprocfs_cliobd_attach(obd);
2806 ptlrpc_lprocfs_register_obd(obd);
2808 ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2810 obd->obd_namespace->ns_lvbo = &inode_lvbo;
2812 rc = mdc_llog_init(obd);
2815 CERROR("failed to setup llogging subsystems\n");
2821 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2825 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2829 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
2830 * us to make MDS RPCs with large enough reply buffers to hold a default
2831 * sized EA and cookie without having to calculate this (via a call into the
2832 * LOV + OSCs) each time we make an RPC. The maximum size is also tracked
2833 * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2834 * a large number of stripes is possible. If a larger reply buffer is
2835 * required it will be reallocated in the ptlrpc layer due to overflow.
2837 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2838 __u32 def_easize, __u32 cookiesize,
2839 __u32 def_cookiesize)
2841 struct obd_device *obd = exp->exp_obd;
2842 struct client_obd *cli = &obd->u.cli;
2845 if (cli->cl_max_mds_easize < easize)
2846 cli->cl_max_mds_easize = easize;
2848 if (cli->cl_default_mds_easize < def_easize)
2849 cli->cl_default_mds_easize = def_easize;
2851 if (cli->cl_max_mds_cookiesize < cookiesize)
2852 cli->cl_max_mds_cookiesize = cookiesize;
2854 if (cli->cl_default_mds_cookiesize < def_cookiesize)
2855 cli->cl_default_mds_cookiesize = def_cookiesize;
2860 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2866 case OBD_CLEANUP_EARLY:
2868 case OBD_CLEANUP_EXPORTS:
2869 /* Failsafe, ok if racy */
2870 if (obd->obd_type->typ_refcnt <= 1)
2871 libcfs_kkuc_group_rem(0, KUC_GRP_HSM, NULL);
2873 obd_cleanup_client_import(obd);
2874 ptlrpc_lprocfs_unregister_obd(obd);
2875 lprocfs_obd_cleanup(obd);
2876 lprocfs_free_md_stats(obd);
2877 mdc_llog_finish(obd);
2883 static int mdc_cleanup(struct obd_device *obd)
2885 struct client_obd *cli = &obd->u.cli;
2887 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2888 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2892 return client_obd_cleanup(obd);
2895 static int mdc_process_config(struct obd_device *obd, size_t len, void *buf)
2897 struct lustre_cfg *lcfg = buf;
2898 int rc = class_process_proc_param(PARAM_MDC, obd->obd_vars, lcfg, obd);
2899 return (rc > 0 ? 0: rc);
2903 /* get remote permission for current user on fid */
2904 static int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2905 struct obd_capa *oc, __u32 suppgid,
2906 struct ptlrpc_request **request)
2908 struct ptlrpc_request *req;
2912 LASSERT(client_is_remote(exp));
2915 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2919 mdc_set_capa_size(req, &RMF_CAPA1, oc);
2921 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2923 ptlrpc_request_free(req);
2927 mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2929 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2930 sizeof(struct mdt_remote_perm));
2932 ptlrpc_request_set_replen(req);
2934 rc = ptlrpc_queue_wait(req);
2936 ptlrpc_req_finished(req);
2942 static int mdc_interpret_renew_capa(const struct lu_env *env,
2943 struct ptlrpc_request *req, void *args,
2946 struct mdc_renew_capa_args *ra = args;
2947 struct mdt_body *body = NULL;
2948 struct lustre_capa *capa;
2952 GOTO(out, capa = ERR_PTR(status));
2954 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2956 GOTO(out, capa = ERR_PTR(-EFAULT));
2958 if ((body->mbo_valid & OBD_MD_FLOSSCAPA) == 0)
2959 GOTO(out, capa = ERR_PTR(-ENOENT));
2961 capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
2963 GOTO(out, capa = ERR_PTR(-EFAULT));
2966 ra->ra_cb(ra->ra_oc, capa);
2970 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2973 struct ptlrpc_request *req;
2974 struct mdc_renew_capa_args *ra;
2977 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
2978 LUSTRE_MDS_VERSION, MDS_GETATTR);
2982 /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
2983 * capa to renew is oss capa.
2985 mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
2986 ptlrpc_request_set_replen(req);
2988 CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
2989 ra = ptlrpc_req_async_args(req);
2992 req->rq_interpret_reply = mdc_interpret_renew_capa;
2993 ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
2997 static struct obd_ops mdc_obd_ops = {
2998 .o_owner = THIS_MODULE,
2999 .o_setup = mdc_setup,
3000 .o_precleanup = mdc_precleanup,
3001 .o_cleanup = mdc_cleanup,
3002 .o_add_conn = client_import_add_conn,
3003 .o_del_conn = client_import_del_conn,
3004 .o_connect = client_connect_import,
3005 .o_disconnect = client_disconnect_export,
3006 .o_iocontrol = mdc_iocontrol,
3007 .o_set_info_async = mdc_set_info_async,
3008 .o_statfs = mdc_statfs,
3009 .o_fid_init = client_fid_init,
3010 .o_fid_fini = client_fid_fini,
3011 .o_fid_alloc = mdc_fid_alloc,
3012 .o_import_event = mdc_import_event,
3013 .o_get_info = mdc_get_info,
3014 .o_process_config = mdc_process_config,
3015 .o_get_uuid = mdc_get_uuid,
3016 .o_quotactl = mdc_quotactl,
3017 .o_quotacheck = mdc_quotacheck
3020 static struct md_ops mdc_md_ops = {
3021 .m_getstatus = mdc_getstatus,
3022 .m_null_inode = mdc_null_inode,
3023 .m_find_cbdata = mdc_find_cbdata,
3024 .m_close = mdc_close,
3025 .m_create = mdc_create,
3026 .m_enqueue = mdc_enqueue,
3027 .m_getattr = mdc_getattr,
3028 .m_getattr_name = mdc_getattr_name,
3029 .m_intent_lock = mdc_intent_lock,
3031 .m_rename = mdc_rename,
3032 .m_setattr = mdc_setattr,
3033 .m_setxattr = mdc_setxattr,
3034 .m_getxattr = mdc_getxattr,
3035 .m_fsync = mdc_fsync,
3036 .m_read_page = mdc_read_page,
3037 .m_unlink = mdc_unlink,
3038 .m_cancel_unused = mdc_cancel_unused,
3039 .m_init_ea_size = mdc_init_ea_size,
3040 .m_set_lock_data = mdc_set_lock_data,
3041 .m_lock_match = mdc_lock_match,
3042 .m_get_lustre_md = mdc_get_lustre_md,
3043 .m_free_lustre_md = mdc_free_lustre_md,
3044 .m_set_open_replay_data = mdc_set_open_replay_data,
3045 .m_clear_open_replay_data = mdc_clear_open_replay_data,
3046 .m_renew_capa = mdc_renew_capa,
3047 .m_unpack_capa = mdc_unpack_capa,
3048 .m_get_remote_perm = mdc_get_remote_perm,
3049 .m_intent_getattr_async = mdc_intent_getattr_async,
3050 .m_revalidate_lock = mdc_revalidate_lock
3053 static int __init mdc_init(void)
3055 return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
3056 LUSTRE_MDC_NAME, NULL);
3059 static void /*__exit*/ mdc_exit(void)
3061 class_unregister_type(LUSTRE_MDC_NAME);
3064 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3065 MODULE_DESCRIPTION("Lustre Metadata Client");
3066 MODULE_LICENSE("GPL");
3068 module_init(mdc_init);
3069 module_exit(mdc_exit);