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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #define DEBUG_SUBSYSTEM S_MDC
35 # include <linux/module.h>
36 # include <linux/pagemap.h>
37 # include <linux/miscdevice.h>
38 # include <linux/init.h>
39 # include <linux/utsname.h>
41 #include "../include/lustre_acl.h"
42 #include "../include/obd_class.h"
43 #include "../include/lustre_fid.h"
44 #include "../include/lprocfs_status.h"
45 #include "../include/lustre_param.h"
46 #include "../include/lustre_log.h"
47 #include "../include/lustre_kernelcomm.h"
49 #include "mdc_internal.h"
51 #define REQUEST_MINOR 244
53 static int mdc_cleanup(struct obd_device
*obd
);
55 static inline int mdc_queue_wait(struct ptlrpc_request
*req
)
57 struct client_obd
*cli
= &req
->rq_import
->imp_obd
->u
.cli
;
60 /* mdc_enter_request() ensures that this client has no more
61 * than cl_max_rpcs_in_flight RPCs simultaneously inf light
64 rc
= mdc_enter_request(cli
);
68 rc
= ptlrpc_queue_wait(req
);
69 mdc_exit_request(cli
);
74 static int mdc_getstatus(struct obd_export
*exp
, struct lu_fid
*rootfid
)
76 struct ptlrpc_request
*req
;
77 struct mdt_body
*body
;
80 req
= ptlrpc_request_alloc_pack(class_exp2cliimp(exp
),
82 LUSTRE_MDS_VERSION
, MDS_GETSTATUS
);
86 mdc_pack_body(req
, NULL
, 0, 0, -1, 0);
87 req
->rq_send_state
= LUSTRE_IMP_FULL
;
89 ptlrpc_request_set_replen(req
);
91 rc
= ptlrpc_queue_wait(req
);
95 body
= req_capsule_server_get(&req
->rq_pill
, &RMF_MDT_BODY
);
101 *rootfid
= body
->fid1
;
103 "root fid="DFID
", last_committed=%llu\n",
105 lustre_msg_get_last_committed(req
->rq_repmsg
));
107 ptlrpc_req_finished(req
);
112 * This function now is known to always saying that it will receive 4 buffers
113 * from server. Even for cases when acl_size and md_size is zero, RPC header
114 * will contain 4 fields and RPC itself will contain zero size fields. This is
115 * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
116 * and thus zero, it shrinks it, making zero size. The same story about
117 * md_size. And this is course of problem when client waits for smaller number
118 * of fields. This issue will be fixed later when client gets aware of RPC
121 static int mdc_getattr_common(struct obd_export
*exp
,
122 struct ptlrpc_request
*req
)
124 struct req_capsule
*pill
= &req
->rq_pill
;
125 struct mdt_body
*body
;
129 /* Request message already built. */
130 rc
= ptlrpc_queue_wait(req
);
134 /* sanity check for the reply */
135 body
= req_capsule_server_get(pill
, &RMF_MDT_BODY
);
139 CDEBUG(D_NET
, "mode: %o\n", body
->mode
);
141 mdc_update_max_ea_from_body(exp
, body
);
142 if (body
->eadatasize
!= 0) {
143 eadata
= req_capsule_server_sized_get(pill
, &RMF_MDT_MD
,
149 if (body
->valid
& OBD_MD_FLRMTPERM
) {
150 struct mdt_remote_perm
*perm
;
152 LASSERT(client_is_remote(exp
));
153 perm
= req_capsule_server_swab_get(pill
, &RMF_ACL
,
154 lustre_swab_mdt_remote_perm
);
162 static int mdc_getattr(struct obd_export
*exp
, struct md_op_data
*op_data
,
163 struct ptlrpc_request
**request
)
165 struct ptlrpc_request
*req
;
168 /* Single MDS without an LMV case */
169 if (op_data
->op_flags
& MF_GET_MDT_IDX
) {
174 req
= ptlrpc_request_alloc(class_exp2cliimp(exp
), &RQF_MDS_GETATTR
);
178 rc
= ptlrpc_request_pack(req
, LUSTRE_MDS_VERSION
, MDS_GETATTR
);
180 ptlrpc_request_free(req
);
184 mdc_pack_body(req
, &op_data
->op_fid1
, op_data
->op_valid
,
185 op_data
->op_mode
, -1, 0);
187 req_capsule_set_size(&req
->rq_pill
, &RMF_MDT_MD
, RCL_SERVER
,
189 if (op_data
->op_valid
& OBD_MD_FLRMTPERM
) {
190 LASSERT(client_is_remote(exp
));
191 req_capsule_set_size(&req
->rq_pill
, &RMF_ACL
, RCL_SERVER
,
192 sizeof(struct mdt_remote_perm
));
194 ptlrpc_request_set_replen(req
);
196 rc
= mdc_getattr_common(exp
, req
);
198 ptlrpc_req_finished(req
);
204 static int mdc_getattr_name(struct obd_export
*exp
, struct md_op_data
*op_data
,
205 struct ptlrpc_request
**request
)
207 struct ptlrpc_request
*req
;
211 req
= ptlrpc_request_alloc(class_exp2cliimp(exp
),
212 &RQF_MDS_GETATTR_NAME
);
216 req_capsule_set_size(&req
->rq_pill
, &RMF_NAME
, RCL_CLIENT
,
217 op_data
->op_namelen
+ 1);
219 rc
= ptlrpc_request_pack(req
, LUSTRE_MDS_VERSION
, MDS_GETATTR_NAME
);
221 ptlrpc_request_free(req
);
225 mdc_pack_body(req
, &op_data
->op_fid1
, op_data
->op_valid
,
226 op_data
->op_mode
, op_data
->op_suppgids
[0], 0);
228 if (op_data
->op_name
) {
229 char *name
= req_capsule_client_get(&req
->rq_pill
, &RMF_NAME
);
231 LASSERT(strnlen(op_data
->op_name
, op_data
->op_namelen
) ==
232 op_data
->op_namelen
);
233 memcpy(name
, op_data
->op_name
, op_data
->op_namelen
);
236 req_capsule_set_size(&req
->rq_pill
, &RMF_MDT_MD
, RCL_SERVER
,
238 ptlrpc_request_set_replen(req
);
240 rc
= mdc_getattr_common(exp
, req
);
242 ptlrpc_req_finished(req
);
248 static int mdc_is_subdir(struct obd_export
*exp
,
249 const struct lu_fid
*pfid
,
250 const struct lu_fid
*cfid
,
251 struct ptlrpc_request
**request
)
253 struct ptlrpc_request
*req
;
257 req
= ptlrpc_request_alloc_pack(class_exp2cliimp(exp
),
258 &RQF_MDS_IS_SUBDIR
, LUSTRE_MDS_VERSION
,
263 mdc_is_subdir_pack(req
, pfid
, cfid
, 0);
264 ptlrpc_request_set_replen(req
);
266 rc
= ptlrpc_queue_wait(req
);
267 if (rc
&& rc
!= -EREMOTE
)
268 ptlrpc_req_finished(req
);
274 static int mdc_xattr_common(struct obd_export
*exp
,
275 const struct req_format
*fmt
,
276 const struct lu_fid
*fid
,
277 int opcode
, u64 valid
,
278 const char *xattr_name
, const char *input
,
279 int input_size
, int output_size
, int flags
,
280 __u32 suppgid
, struct ptlrpc_request
**request
)
282 struct ptlrpc_request
*req
;
283 int xattr_namelen
= 0;
288 req
= ptlrpc_request_alloc(class_exp2cliimp(exp
), fmt
);
293 xattr_namelen
= strlen(xattr_name
) + 1;
294 req_capsule_set_size(&req
->rq_pill
, &RMF_NAME
, RCL_CLIENT
,
299 req_capsule_set_size(&req
->rq_pill
, &RMF_EADATA
, RCL_CLIENT
,
303 /* Flush local XATTR locks to get rid of a possible cancel RPC */
304 if (opcode
== MDS_REINT
&& fid_is_sane(fid
) &&
305 exp
->exp_connect_data
.ocd_ibits_known
& MDS_INODELOCK_XATTR
) {
309 /* Without that packing would fail */
311 req_capsule_set_size(&req
->rq_pill
, &RMF_EADATA
,
314 count
= mdc_resource_get_unused(exp
, fid
,
316 MDS_INODELOCK_XATTR
);
318 rc
= mdc_prep_elc_req(exp
, req
, MDS_REINT
, &cancels
, count
);
320 ptlrpc_request_free(req
);
324 rc
= ptlrpc_request_pack(req
, LUSTRE_MDS_VERSION
, opcode
);
326 ptlrpc_request_free(req
);
331 if (opcode
== MDS_REINT
) {
332 struct mdt_rec_setxattr
*rec
;
334 CLASSERT(sizeof(struct mdt_rec_setxattr
) ==
335 sizeof(struct mdt_rec_reint
));
336 rec
= req_capsule_client_get(&req
->rq_pill
, &RMF_REC_REINT
);
337 rec
->sx_opcode
= REINT_SETXATTR
;
338 rec
->sx_fsuid
= from_kuid(&init_user_ns
, current_fsuid());
339 rec
->sx_fsgid
= from_kgid(&init_user_ns
, current_fsgid());
340 rec
->sx_cap
= cfs_curproc_cap_pack();
341 rec
->sx_suppgid1
= suppgid
;
342 rec
->sx_suppgid2
= -1;
344 rec
->sx_valid
= valid
| OBD_MD_FLCTIME
;
345 rec
->sx_time
= ktime_get_real_seconds();
346 rec
->sx_size
= output_size
;
347 rec
->sx_flags
= flags
;
350 mdc_pack_body(req
, fid
, valid
, output_size
, suppgid
, flags
);
354 tmp
= req_capsule_client_get(&req
->rq_pill
, &RMF_NAME
);
355 memcpy(tmp
, xattr_name
, xattr_namelen
);
358 tmp
= req_capsule_client_get(&req
->rq_pill
, &RMF_EADATA
);
359 memcpy(tmp
, input
, input_size
);
362 if (req_capsule_has_field(&req
->rq_pill
, &RMF_EADATA
, RCL_SERVER
))
363 req_capsule_set_size(&req
->rq_pill
, &RMF_EADATA
,
364 RCL_SERVER
, output_size
);
365 ptlrpc_request_set_replen(req
);
368 if (opcode
== MDS_REINT
)
369 mdc_get_rpc_lock(exp
->exp_obd
->u
.cli
.cl_rpc_lock
, NULL
);
371 rc
= ptlrpc_queue_wait(req
);
373 if (opcode
== MDS_REINT
)
374 mdc_put_rpc_lock(exp
->exp_obd
->u
.cli
.cl_rpc_lock
, NULL
);
377 ptlrpc_req_finished(req
);
383 static int mdc_setxattr(struct obd_export
*exp
, const struct lu_fid
*fid
,
384 u64 valid
, const char *xattr_name
,
385 const char *input
, int input_size
, int output_size
,
386 int flags
, __u32 suppgid
,
387 struct ptlrpc_request
**request
)
389 return mdc_xattr_common(exp
, &RQF_MDS_REINT_SETXATTR
,
390 fid
, MDS_REINT
, valid
, xattr_name
,
391 input
, input_size
, output_size
, flags
,
395 static int mdc_getxattr(struct obd_export
*exp
, const struct lu_fid
*fid
,
396 u64 valid
, const char *xattr_name
,
397 const char *input
, int input_size
, int output_size
,
398 int flags
, struct ptlrpc_request
**request
)
400 return mdc_xattr_common(exp
, &RQF_MDS_GETXATTR
,
401 fid
, MDS_GETXATTR
, valid
, xattr_name
,
402 input
, input_size
, output_size
, flags
,
406 #ifdef CONFIG_FS_POSIX_ACL
407 static int mdc_unpack_acl(struct ptlrpc_request
*req
, struct lustre_md
*md
)
409 struct req_capsule
*pill
= &req
->rq_pill
;
410 struct mdt_body
*body
= md
->body
;
411 struct posix_acl
*acl
;
418 buf
= req_capsule_server_sized_get(pill
, &RMF_ACL
, body
->aclsize
);
423 acl
= posix_acl_from_xattr(&init_user_ns
, buf
, body
->aclsize
);
429 CERROR("convert xattr to acl: %d\n", rc
);
433 rc
= posix_acl_valid(acl
);
435 CERROR("validate acl: %d\n", rc
);
436 posix_acl_release(acl
);
444 #define mdc_unpack_acl(req, md) 0
447 static int mdc_get_lustre_md(struct obd_export
*exp
,
448 struct ptlrpc_request
*req
,
449 struct obd_export
*dt_exp
,
450 struct obd_export
*md_exp
,
451 struct lustre_md
*md
)
453 struct req_capsule
*pill
= &req
->rq_pill
;
457 memset(md
, 0, sizeof(*md
));
459 md
->body
= req_capsule_server_get(pill
, &RMF_MDT_BODY
);
461 if (md
->body
->valid
& OBD_MD_FLEASIZE
) {
463 struct lov_mds_md
*lmm
;
465 if (!S_ISREG(md
->body
->mode
)) {
467 "OBD_MD_FLEASIZE set, should be a regular file, but is not\n");
472 if (md
->body
->eadatasize
== 0) {
474 "OBD_MD_FLEASIZE set, but eadatasize 0\n");
478 lmmsize
= md
->body
->eadatasize
;
479 lmm
= req_capsule_server_sized_get(pill
, &RMF_MDT_MD
, lmmsize
);
485 rc
= obd_unpackmd(dt_exp
, &md
->lsm
, lmm
, lmmsize
);
489 if (rc
< sizeof(*md
->lsm
)) {
491 "lsm size too small: rc < sizeof (*md->lsm) (%d < %d)\n",
492 rc
, (int)sizeof(*md
->lsm
));
497 } else if (md
->body
->valid
& OBD_MD_FLDIREA
) {
499 struct lov_mds_md
*lmv
;
501 if (!S_ISDIR(md
->body
->mode
)) {
503 "OBD_MD_FLDIREA set, should be a directory, but is not\n");
508 if (md
->body
->eadatasize
== 0) {
510 "OBD_MD_FLDIREA is set, but eadatasize 0\n");
513 if (md
->body
->valid
& OBD_MD_MEA
) {
514 lmvsize
= md
->body
->eadatasize
;
515 lmv
= req_capsule_server_sized_get(pill
, &RMF_MDT_MD
,
522 rc
= obd_unpackmd(md_exp
, (void *)&md
->mea
, lmv
,
527 if (rc
< sizeof(*md
->mea
)) {
529 "size too small: rc < sizeof(*md->mea) (%d < %d)\n",
530 rc
, (int)sizeof(*md
->mea
));
538 if (md
->body
->valid
& OBD_MD_FLRMTPERM
) {
539 /* remote permission */
540 LASSERT(client_is_remote(exp
));
541 md
->remote_perm
= req_capsule_server_swab_get(pill
, &RMF_ACL
,
542 lustre_swab_mdt_remote_perm
);
543 if (!md
->remote_perm
) {
547 } else if (md
->body
->valid
& OBD_MD_FLACL
) {
548 /* for ACL, it's possible that FLACL is set but aclsize is zero.
549 * only when aclsize != 0 there's an actual segment for ACL
552 if (md
->body
->aclsize
) {
553 rc
= mdc_unpack_acl(req
, md
);
556 #ifdef CONFIG_FS_POSIX_ACL
558 md
->posix_acl
= NULL
;
565 #ifdef CONFIG_FS_POSIX_ACL
566 posix_acl_release(md
->posix_acl
);
569 obd_free_memmd(dt_exp
, &md
->lsm
);
574 static int mdc_free_lustre_md(struct obd_export
*exp
, struct lustre_md
*md
)
580 * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
583 void mdc_replay_open(struct ptlrpc_request
*req
)
585 struct md_open_data
*mod
= req
->rq_cb_data
;
586 struct ptlrpc_request
*close_req
;
587 struct obd_client_handle
*och
;
588 struct lustre_handle old
;
589 struct mdt_body
*body
;
592 DEBUG_REQ(D_ERROR
, req
,
593 "Can't properly replay without open data.");
597 body
= req_capsule_server_get(&req
->rq_pill
, &RMF_MDT_BODY
);
601 struct lustre_handle
*file_fh
;
603 LASSERT(och
->och_magic
== OBD_CLIENT_HANDLE_MAGIC
);
605 file_fh
= &och
->och_fh
;
606 CDEBUG(D_HA
, "updating handle from %#llx to %#llx\n",
607 file_fh
->cookie
, body
->handle
.cookie
);
609 *file_fh
= body
->handle
;
611 close_req
= mod
->mod_close_req
;
613 __u32 opc
= lustre_msg_get_opc(close_req
->rq_reqmsg
);
614 struct mdt_ioepoch
*epoch
;
616 LASSERT(opc
== MDS_CLOSE
|| opc
== MDS_DONE_WRITING
);
617 epoch
= req_capsule_client_get(&close_req
->rq_pill
,
622 LASSERT(!memcmp(&old
, &epoch
->handle
, sizeof(old
)));
623 DEBUG_REQ(D_HA
, close_req
, "updating close body with new fh");
624 epoch
->handle
= body
->handle
;
628 void mdc_commit_open(struct ptlrpc_request
*req
)
630 struct md_open_data
*mod
= req
->rq_cb_data
;
636 * No need to touch md_open_data::mod_och, it holds a reference on
637 * \var mod and will zero references to each other, \var mod will be
638 * freed after that when md_open_data::mod_och will put the reference.
642 * Do not let open request to disappear as it still may be needed
643 * for close rpc to happen (it may happen on evict only, otherwise
644 * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
645 * called), just mark this rpc as committed to distinguish these 2
646 * cases, see mdc_close() for details. The open request reference will
647 * be put along with freeing \var mod.
649 ptlrpc_request_addref(req
);
650 spin_lock(&req
->rq_lock
);
651 req
->rq_committed
= 1;
652 spin_unlock(&req
->rq_lock
);
653 req
->rq_cb_data
= NULL
;
657 int mdc_set_open_replay_data(struct obd_export
*exp
,
658 struct obd_client_handle
*och
,
659 struct lookup_intent
*it
)
661 struct md_open_data
*mod
;
662 struct mdt_rec_create
*rec
;
663 struct mdt_body
*body
;
664 struct ptlrpc_request
*open_req
= it
->d
.lustre
.it_data
;
665 struct obd_import
*imp
= open_req
->rq_import
;
667 if (!open_req
->rq_replay
)
670 rec
= req_capsule_client_get(&open_req
->rq_pill
, &RMF_REC_REINT
);
671 body
= req_capsule_server_get(&open_req
->rq_pill
, &RMF_MDT_BODY
);
673 /* Incoming message in my byte order (it's been swabbed). */
674 /* Outgoing messages always in my byte order. */
677 /* Only if the import is replayable, we set replay_open data */
678 if (och
&& imp
->imp_replayable
) {
679 mod
= obd_mod_alloc();
681 DEBUG_REQ(D_ERROR
, open_req
,
682 "Can't allocate md_open_data");
687 * Take a reference on \var mod, to be freed on mdc_close().
688 * It protects \var mod from being freed on eviction (commit
689 * callback is called despite rq_replay flag).
690 * Another reference for \var och.
695 spin_lock(&open_req
->rq_lock
);
698 mod
->mod_is_create
= it_disposition(it
, DISP_OPEN_CREATE
) ||
699 it_disposition(it
, DISP_OPEN_STRIPE
);
700 mod
->mod_open_req
= open_req
;
701 open_req
->rq_cb_data
= mod
;
702 open_req
->rq_commit_cb
= mdc_commit_open
;
703 spin_unlock(&open_req
->rq_lock
);
706 rec
->cr_fid2
= body
->fid1
;
707 rec
->cr_ioepoch
= body
->ioepoch
;
708 rec
->cr_old_handle
.cookie
= body
->handle
.cookie
;
709 open_req
->rq_replay_cb
= mdc_replay_open
;
710 if (!fid_is_sane(&body
->fid1
)) {
711 DEBUG_REQ(D_ERROR
, open_req
,
712 "Saving replay request with insane fid");
716 DEBUG_REQ(D_RPCTRACE
, open_req
, "Set up open replay data");
720 static void mdc_free_open(struct md_open_data
*mod
)
724 if (mod
->mod_is_create
== 0 &&
725 imp_connect_disp_stripe(mod
->mod_open_req
->rq_import
))
728 LASSERT(mod
->mod_open_req
->rq_replay
== 0);
730 DEBUG_REQ(D_RPCTRACE
, mod
->mod_open_req
, "free open request\n");
732 ptlrpc_request_committed(mod
->mod_open_req
, committed
);
733 if (mod
->mod_close_req
)
734 ptlrpc_request_committed(mod
->mod_close_req
, committed
);
737 static int mdc_clear_open_replay_data(struct obd_export
*exp
,
738 struct obd_client_handle
*och
)
740 struct md_open_data
*mod
= och
->och_mod
;
743 * It is possible to not have \var mod in a case of eviction between
744 * lookup and ll_file_open().
749 LASSERT(mod
!= LP_POISON
);
750 LASSERT(mod
->mod_open_req
);
760 /* Prepares the request for the replay by the given reply */
761 static void mdc_close_handle_reply(struct ptlrpc_request
*req
,
762 struct md_op_data
*op_data
, int rc
) {
763 struct mdt_body
*repbody
;
764 struct mdt_ioepoch
*epoch
;
766 if (req
&& rc
== -EAGAIN
) {
767 repbody
= req_capsule_server_get(&req
->rq_pill
, &RMF_MDT_BODY
);
768 epoch
= req_capsule_client_get(&req
->rq_pill
, &RMF_MDT_EPOCH
);
770 epoch
->flags
|= MF_SOM_AU
;
771 if (repbody
->valid
& OBD_MD_FLGETATTRLOCK
)
772 op_data
->op_flags
|= MF_GETATTR_LOCK
;
776 static int mdc_close(struct obd_export
*exp
, struct md_op_data
*op_data
,
777 struct md_open_data
*mod
, struct ptlrpc_request
**request
)
779 struct obd_device
*obd
= class_exp2obd(exp
);
780 struct ptlrpc_request
*req
;
781 struct req_format
*req_fmt
;
785 req_fmt
= &RQF_MDS_CLOSE
;
786 if (op_data
->op_bias
& MDS_HSM_RELEASE
) {
787 req_fmt
= &RQF_MDS_RELEASE_CLOSE
;
789 /* allocate a FID for volatile file */
790 rc
= mdc_fid_alloc(exp
, &op_data
->op_fid2
, op_data
);
792 CERROR("%s: "DFID
" failed to allocate FID: %d\n",
793 obd
->obd_name
, PFID(&op_data
->op_fid1
), rc
);
794 /* save the errcode and proceed to close */
800 req
= ptlrpc_request_alloc(class_exp2cliimp(exp
), req_fmt
);
804 rc
= ptlrpc_request_pack(req
, LUSTRE_MDS_VERSION
, MDS_CLOSE
);
806 ptlrpc_request_free(req
);
810 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
811 * portal whose threads are not taking any DLM locks and are therefore
814 req
->rq_request_portal
= MDS_READPAGE_PORTAL
;
815 ptlrpc_at_set_req_timeout(req
);
817 /* Ensure that this close's handle is fixed up during replay. */
819 LASSERTF(mod
->mod_open_req
&&
820 mod
->mod_open_req
->rq_type
!= LI_POISON
,
821 "POISONED open %p!\n", mod
->mod_open_req
);
823 mod
->mod_close_req
= req
;
825 DEBUG_REQ(D_HA
, mod
->mod_open_req
, "matched open");
826 /* We no longer want to preserve this open for replay even
827 * though the open was committed. b=3632, b=3633
829 spin_lock(&mod
->mod_open_req
->rq_lock
);
830 mod
->mod_open_req
->rq_replay
= 0;
831 spin_unlock(&mod
->mod_open_req
->rq_lock
);
834 "couldn't find open req; expecting close error\n");
837 mdc_close_pack(req
, op_data
);
839 req_capsule_set_size(&req
->rq_pill
, &RMF_MDT_MD
, RCL_SERVER
,
840 obd
->u
.cli
.cl_default_mds_easize
);
841 req_capsule_set_size(&req
->rq_pill
, &RMF_LOGCOOKIES
, RCL_SERVER
,
842 obd
->u
.cli
.cl_default_mds_cookiesize
);
844 ptlrpc_request_set_replen(req
);
846 mdc_get_rpc_lock(obd
->u
.cli
.cl_close_lock
, NULL
);
847 rc
= ptlrpc_queue_wait(req
);
848 mdc_put_rpc_lock(obd
->u
.cli
.cl_close_lock
, NULL
);
850 if (!req
->rq_repmsg
) {
851 CDEBUG(D_RPCTRACE
, "request failed to send: %p, %d\n", req
,
854 rc
= req
->rq_status
?: -EIO
;
855 } else if (rc
== 0 || rc
== -EAGAIN
) {
856 struct mdt_body
*body
;
858 rc
= lustre_msg_get_status(req
->rq_repmsg
);
859 if (lustre_msg_get_type(req
->rq_repmsg
) == PTL_RPC_MSG_ERR
) {
860 DEBUG_REQ(D_ERROR
, req
,
861 "type == PTL_RPC_MSG_ERR, err = %d", rc
);
865 body
= req_capsule_server_get(&req
->rq_pill
, &RMF_MDT_BODY
);
868 } else if (rc
== -ESTALE
) {
870 * it can be allowed error after 3633 if open was committed and
871 * server failed before close was sent. Let's check if mod
872 * exists and return no error in that case
875 DEBUG_REQ(D_HA
, req
, "Reset ESTALE = %d", rc
);
876 if (mod
->mod_open_req
->rq_committed
)
883 mod
->mod_close_req
= NULL
;
884 /* Since now, mod is accessed through open_req only,
885 * thus close req does not keep a reference on mod anymore.
890 mdc_close_handle_reply(req
, op_data
, rc
);
891 return rc
< 0 ? rc
: saved_rc
;
894 static int mdc_done_writing(struct obd_export
*exp
, struct md_op_data
*op_data
,
895 struct md_open_data
*mod
)
897 struct obd_device
*obd
= class_exp2obd(exp
);
898 struct ptlrpc_request
*req
;
901 req
= ptlrpc_request_alloc(class_exp2cliimp(exp
),
902 &RQF_MDS_DONE_WRITING
);
906 rc
= ptlrpc_request_pack(req
, LUSTRE_MDS_VERSION
, MDS_DONE_WRITING
);
908 ptlrpc_request_free(req
);
913 LASSERTF(mod
->mod_open_req
&&
914 mod
->mod_open_req
->rq_type
!= LI_POISON
,
915 "POISONED setattr %p!\n", mod
->mod_open_req
);
917 mod
->mod_close_req
= req
;
918 DEBUG_REQ(D_HA
, mod
->mod_open_req
, "matched setattr");
919 /* We no longer want to preserve this setattr for replay even
920 * though the open was committed. b=3632, b=3633
922 spin_lock(&mod
->mod_open_req
->rq_lock
);
923 mod
->mod_open_req
->rq_replay
= 0;
924 spin_unlock(&mod
->mod_open_req
->rq_lock
);
927 mdc_close_pack(req
, op_data
);
928 ptlrpc_request_set_replen(req
);
930 mdc_get_rpc_lock(obd
->u
.cli
.cl_close_lock
, NULL
);
931 rc
= ptlrpc_queue_wait(req
);
932 mdc_put_rpc_lock(obd
->u
.cli
.cl_close_lock
, NULL
);
936 * it can be allowed error after 3633 if open or setattr were
937 * committed and server failed before close was sent.
938 * Let's check if mod exists and return no error in that case
941 if (mod
->mod_open_req
->rq_committed
)
948 mod
->mod_close_req
= NULL
;
949 LASSERT(mod
->mod_open_req
);
952 /* Since now, mod is accessed through setattr req only,
953 * thus DW req does not keep a reference on mod anymore.
958 mdc_close_handle_reply(req
, op_data
, rc
);
959 ptlrpc_req_finished(req
);
963 static int mdc_readpage(struct obd_export
*exp
, struct md_op_data
*op_data
,
964 struct page
**pages
, struct ptlrpc_request
**request
)
966 struct ptlrpc_request
*req
;
967 struct ptlrpc_bulk_desc
*desc
;
969 wait_queue_head_t waitq
;
971 struct l_wait_info lwi
;
975 init_waitqueue_head(&waitq
);
978 req
= ptlrpc_request_alloc(class_exp2cliimp(exp
), &RQF_MDS_READPAGE
);
982 rc
= ptlrpc_request_pack(req
, LUSTRE_MDS_VERSION
, MDS_READPAGE
);
984 ptlrpc_request_free(req
);
988 req
->rq_request_portal
= MDS_READPAGE_PORTAL
;
989 ptlrpc_at_set_req_timeout(req
);
991 desc
= ptlrpc_prep_bulk_imp(req
, op_data
->op_npages
, 1, BULK_PUT_SINK
,
994 ptlrpc_request_free(req
);
998 /* NB req now owns desc and will free it when it gets freed */
999 for (i
= 0; i
< op_data
->op_npages
; i
++)
1000 ptlrpc_prep_bulk_page_pin(desc
, pages
[i
], 0, PAGE_SIZE
);
1002 mdc_readdir_pack(req
, op_data
->op_offset
,
1003 PAGE_SIZE
* op_data
->op_npages
,
1006 ptlrpc_request_set_replen(req
);
1007 rc
= ptlrpc_queue_wait(req
);
1009 ptlrpc_req_finished(req
);
1010 if (rc
!= -ETIMEDOUT
)
1014 if (!client_should_resend(resends
, &exp
->exp_obd
->u
.cli
)) {
1015 CERROR("too many resend retries, returning error\n");
1018 lwi
= LWI_TIMEOUT_INTR(cfs_time_seconds(resends
),
1020 l_wait_event(waitq
, 0, &lwi
);
1025 rc
= sptlrpc_cli_unwrap_bulk_read(req
, req
->rq_bulk
,
1026 req
->rq_bulk
->bd_nob_transferred
);
1028 ptlrpc_req_finished(req
);
1032 if (req
->rq_bulk
->bd_nob_transferred
& ~LU_PAGE_MASK
) {
1033 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1034 req
->rq_bulk
->bd_nob_transferred
,
1035 PAGE_SIZE
* op_data
->op_npages
);
1036 ptlrpc_req_finished(req
);
1044 static int mdc_statfs(const struct lu_env
*env
,
1045 struct obd_export
*exp
, struct obd_statfs
*osfs
,
1046 __u64 max_age
, __u32 flags
)
1048 struct obd_device
*obd
= class_exp2obd(exp
);
1049 struct ptlrpc_request
*req
;
1050 struct obd_statfs
*msfs
;
1051 struct obd_import
*imp
= NULL
;
1055 * Since the request might also come from lprocfs, so we need
1056 * sync this with client_disconnect_export Bug15684
1058 down_read(&obd
->u
.cli
.cl_sem
);
1059 if (obd
->u
.cli
.cl_import
)
1060 imp
= class_import_get(obd
->u
.cli
.cl_import
);
1061 up_read(&obd
->u
.cli
.cl_sem
);
1065 req
= ptlrpc_request_alloc_pack(imp
, &RQF_MDS_STATFS
,
1066 LUSTRE_MDS_VERSION
, MDS_STATFS
);
1072 ptlrpc_request_set_replen(req
);
1074 if (flags
& OBD_STATFS_NODELAY
) {
1075 /* procfs requests not want stay in wait for avoid deadlock */
1076 req
->rq_no_resend
= 1;
1077 req
->rq_no_delay
= 1;
1080 rc
= ptlrpc_queue_wait(req
);
1082 /* check connection error first */
1083 if (imp
->imp_connect_error
)
1084 rc
= imp
->imp_connect_error
;
1088 msfs
= req_capsule_server_get(&req
->rq_pill
, &RMF_OBD_STATFS
);
1096 ptlrpc_req_finished(req
);
1098 class_import_put(imp
);
1102 static int mdc_ioc_fid2path(struct obd_export
*exp
, struct getinfo_fid2path
*gf
)
1104 __u32 keylen
, vallen
;
1108 if (gf
->gf_pathlen
> PATH_MAX
)
1109 return -ENAMETOOLONG
;
1110 if (gf
->gf_pathlen
< 2)
1113 /* Key is KEY_FID2PATH + getinfo_fid2path description */
1114 keylen
= cfs_size_round(sizeof(KEY_FID2PATH
)) + sizeof(*gf
);
1115 key
= kzalloc(keylen
, GFP_NOFS
);
1118 memcpy(key
, KEY_FID2PATH
, sizeof(KEY_FID2PATH
));
1119 memcpy(key
+ cfs_size_round(sizeof(KEY_FID2PATH
)), gf
, sizeof(*gf
));
1121 CDEBUG(D_IOCTL
, "path get "DFID
" from %llu #%d\n",
1122 PFID(&gf
->gf_fid
), gf
->gf_recno
, gf
->gf_linkno
);
1124 if (!fid_is_sane(&gf
->gf_fid
)) {
1129 /* Val is struct getinfo_fid2path result plus path */
1130 vallen
= sizeof(*gf
) + gf
->gf_pathlen
;
1132 rc
= obd_get_info(NULL
, exp
, keylen
, key
, &vallen
, gf
, NULL
);
1133 if (rc
!= 0 && rc
!= -EREMOTE
)
1136 if (vallen
<= sizeof(*gf
)) {
1139 } else if (vallen
> sizeof(*gf
) + gf
->gf_pathlen
) {
1144 CDEBUG(D_IOCTL
, "path get "DFID
" from %llu #%d\n%s\n",
1145 PFID(&gf
->gf_fid
), gf
->gf_recno
, gf
->gf_linkno
, gf
->gf_path
);
1152 static int mdc_ioc_hsm_progress(struct obd_export
*exp
,
1153 struct hsm_progress_kernel
*hpk
)
1155 struct obd_import
*imp
= class_exp2cliimp(exp
);
1156 struct hsm_progress_kernel
*req_hpk
;
1157 struct ptlrpc_request
*req
;
1160 req
= ptlrpc_request_alloc_pack(imp
, &RQF_MDS_HSM_PROGRESS
,
1161 LUSTRE_MDS_VERSION
, MDS_HSM_PROGRESS
);
1167 mdc_pack_body(req
, NULL
, OBD_MD_FLRMTPERM
, 0, -1, 0);
1169 /* Copy hsm_progress struct */
1170 req_hpk
= req_capsule_client_get(&req
->rq_pill
, &RMF_MDS_HSM_PROGRESS
);
1177 req_hpk
->hpk_errval
= lustre_errno_hton(hpk
->hpk_errval
);
1179 ptlrpc_request_set_replen(req
);
1181 rc
= mdc_queue_wait(req
);
1183 ptlrpc_req_finished(req
);
1187 static int mdc_ioc_hsm_ct_register(struct obd_import
*imp
, __u32 archives
)
1189 __u32
*archive_mask
;
1190 struct ptlrpc_request
*req
;
1193 req
= ptlrpc_request_alloc_pack(imp
, &RQF_MDS_HSM_CT_REGISTER
,
1195 MDS_HSM_CT_REGISTER
);
1201 mdc_pack_body(req
, NULL
, OBD_MD_FLRMTPERM
, 0, -1, 0);
1203 /* Copy hsm_progress struct */
1204 archive_mask
= req_capsule_client_get(&req
->rq_pill
,
1205 &RMF_MDS_HSM_ARCHIVE
);
1206 if (!archive_mask
) {
1211 *archive_mask
= archives
;
1213 ptlrpc_request_set_replen(req
);
1215 rc
= mdc_queue_wait(req
);
1217 ptlrpc_req_finished(req
);
1221 static int mdc_ioc_hsm_current_action(struct obd_export
*exp
,
1222 struct md_op_data
*op_data
)
1224 struct hsm_current_action
*hca
= op_data
->op_data
;
1225 struct hsm_current_action
*req_hca
;
1226 struct ptlrpc_request
*req
;
1229 req
= ptlrpc_request_alloc(class_exp2cliimp(exp
),
1230 &RQF_MDS_HSM_ACTION
);
1234 rc
= ptlrpc_request_pack(req
, LUSTRE_MDS_VERSION
, MDS_HSM_ACTION
);
1236 ptlrpc_request_free(req
);
1240 mdc_pack_body(req
, &op_data
->op_fid1
, OBD_MD_FLRMTPERM
, 0,
1241 op_data
->op_suppgids
[0], 0);
1243 ptlrpc_request_set_replen(req
);
1245 rc
= mdc_queue_wait(req
);
1249 req_hca
= req_capsule_server_get(&req
->rq_pill
,
1250 &RMF_MDS_HSM_CURRENT_ACTION
);
1259 ptlrpc_req_finished(req
);
1263 static int mdc_ioc_hsm_ct_unregister(struct obd_import
*imp
)
1265 struct ptlrpc_request
*req
;
1268 req
= ptlrpc_request_alloc_pack(imp
, &RQF_MDS_HSM_CT_UNREGISTER
,
1270 MDS_HSM_CT_UNREGISTER
);
1276 mdc_pack_body(req
, NULL
, OBD_MD_FLRMTPERM
, 0, -1, 0);
1278 ptlrpc_request_set_replen(req
);
1280 rc
= mdc_queue_wait(req
);
1282 ptlrpc_req_finished(req
);
1286 static int mdc_ioc_hsm_state_get(struct obd_export
*exp
,
1287 struct md_op_data
*op_data
)
1289 struct hsm_user_state
*hus
= op_data
->op_data
;
1290 struct hsm_user_state
*req_hus
;
1291 struct ptlrpc_request
*req
;
1294 req
= ptlrpc_request_alloc(class_exp2cliimp(exp
),
1295 &RQF_MDS_HSM_STATE_GET
);
1299 rc
= ptlrpc_request_pack(req
, LUSTRE_MDS_VERSION
, MDS_HSM_STATE_GET
);
1301 ptlrpc_request_free(req
);
1305 mdc_pack_body(req
, &op_data
->op_fid1
, OBD_MD_FLRMTPERM
, 0,
1306 op_data
->op_suppgids
[0], 0);
1308 ptlrpc_request_set_replen(req
);
1310 rc
= mdc_queue_wait(req
);
1314 req_hus
= req_capsule_server_get(&req
->rq_pill
, &RMF_HSM_USER_STATE
);
1323 ptlrpc_req_finished(req
);
1327 static int mdc_ioc_hsm_state_set(struct obd_export
*exp
,
1328 struct md_op_data
*op_data
)
1330 struct hsm_state_set
*hss
= op_data
->op_data
;
1331 struct hsm_state_set
*req_hss
;
1332 struct ptlrpc_request
*req
;
1335 req
= ptlrpc_request_alloc(class_exp2cliimp(exp
),
1336 &RQF_MDS_HSM_STATE_SET
);
1340 rc
= ptlrpc_request_pack(req
, LUSTRE_MDS_VERSION
, MDS_HSM_STATE_SET
);
1342 ptlrpc_request_free(req
);
1346 mdc_pack_body(req
, &op_data
->op_fid1
, OBD_MD_FLRMTPERM
, 0,
1347 op_data
->op_suppgids
[0], 0);
1350 req_hss
= req_capsule_client_get(&req
->rq_pill
, &RMF_HSM_STATE_SET
);
1357 ptlrpc_request_set_replen(req
);
1359 rc
= mdc_queue_wait(req
);
1361 ptlrpc_req_finished(req
);
1365 static int mdc_ioc_hsm_request(struct obd_export
*exp
,
1366 struct hsm_user_request
*hur
)
1368 struct obd_import
*imp
= class_exp2cliimp(exp
);
1369 struct ptlrpc_request
*req
;
1370 struct hsm_request
*req_hr
;
1371 struct hsm_user_item
*req_hui
;
1375 req
= ptlrpc_request_alloc(imp
, &RQF_MDS_HSM_REQUEST
);
1381 req_capsule_set_size(&req
->rq_pill
, &RMF_MDS_HSM_USER_ITEM
, RCL_CLIENT
,
1382 hur
->hur_request
.hr_itemcount
1383 * sizeof(struct hsm_user_item
));
1384 req_capsule_set_size(&req
->rq_pill
, &RMF_GENERIC_DATA
, RCL_CLIENT
,
1385 hur
->hur_request
.hr_data_len
);
1387 rc
= ptlrpc_request_pack(req
, LUSTRE_MDS_VERSION
, MDS_HSM_REQUEST
);
1389 ptlrpc_request_free(req
);
1393 mdc_pack_body(req
, NULL
, OBD_MD_FLRMTPERM
, 0, -1, 0);
1395 /* Copy hsm_request struct */
1396 req_hr
= req_capsule_client_get(&req
->rq_pill
, &RMF_MDS_HSM_REQUEST
);
1401 *req_hr
= hur
->hur_request
;
1403 /* Copy hsm_user_item structs */
1404 req_hui
= req_capsule_client_get(&req
->rq_pill
, &RMF_MDS_HSM_USER_ITEM
);
1409 memcpy(req_hui
, hur
->hur_user_item
,
1410 hur
->hur_request
.hr_itemcount
* sizeof(struct hsm_user_item
));
1412 /* Copy opaque field */
1413 req_opaque
= req_capsule_client_get(&req
->rq_pill
, &RMF_GENERIC_DATA
);
1418 memcpy(req_opaque
, hur_data(hur
), hur
->hur_request
.hr_data_len
);
1420 ptlrpc_request_set_replen(req
);
1422 rc
= mdc_queue_wait(req
);
1424 ptlrpc_req_finished(req
);
1428 static struct kuc_hdr
*changelog_kuc_hdr(char *buf
, int len
, int flags
)
1430 struct kuc_hdr
*lh
= (struct kuc_hdr
*)buf
;
1432 LASSERT(len
<= KUC_CHANGELOG_MSG_MAXSIZE
);
1434 lh
->kuc_magic
= KUC_MAGIC
;
1435 lh
->kuc_transport
= KUC_TRANSPORT_CHANGELOG
;
1436 lh
->kuc_flags
= flags
;
1437 lh
->kuc_msgtype
= CL_RECORD
;
1438 lh
->kuc_msglen
= len
;
1442 #define D_CHANGELOG 0
1444 struct changelog_show
{
1449 struct obd_device
*cs_obd
;
1452 static int changelog_kkuc_cb(const struct lu_env
*env
, struct llog_handle
*llh
,
1453 struct llog_rec_hdr
*hdr
, void *data
)
1455 struct changelog_show
*cs
= data
;
1456 struct llog_changelog_rec
*rec
= (struct llog_changelog_rec
*)hdr
;
1460 if (rec
->cr_hdr
.lrh_type
!= CHANGELOG_REC
) {
1462 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1463 cs
->cs_obd
->obd_name
, rec
->cr_hdr
.lrh_type
,
1464 rec
->cr
.cr_type
, rc
);
1468 if (rec
->cr
.cr_index
< cs
->cs_startrec
) {
1469 /* Skip entries earlier than what we are interested in */
1470 CDEBUG(D_CHANGELOG
, "rec=%llu start=%llu\n",
1471 rec
->cr
.cr_index
, cs
->cs_startrec
);
1475 CDEBUG(D_CHANGELOG
, "%llu %02d%-5s %llu 0x%x t="DFID
" p="DFID
1476 " %.*s\n", rec
->cr
.cr_index
, rec
->cr
.cr_type
,
1477 changelog_type2str(rec
->cr
.cr_type
), rec
->cr
.cr_time
,
1478 rec
->cr
.cr_flags
& CLF_FLAGMASK
,
1479 PFID(&rec
->cr
.cr_tfid
), PFID(&rec
->cr
.cr_pfid
),
1480 rec
->cr
.cr_namelen
, changelog_rec_name(&rec
->cr
));
1482 len
= sizeof(*lh
) + changelog_rec_size(&rec
->cr
) + rec
->cr
.cr_namelen
;
1484 /* Set up the message */
1485 lh
= changelog_kuc_hdr(cs
->cs_buf
, len
, cs
->cs_flags
);
1486 memcpy(lh
+ 1, &rec
->cr
, len
- sizeof(*lh
));
1488 rc
= libcfs_kkuc_msg_put(cs
->cs_fp
, lh
);
1489 CDEBUG(D_CHANGELOG
, "kucmsg fp %p len %d rc %d\n", cs
->cs_fp
, len
, rc
);
1494 static int mdc_changelog_send_thread(void *csdata
)
1496 struct changelog_show
*cs
= csdata
;
1497 struct llog_ctxt
*ctxt
= NULL
;
1498 struct llog_handle
*llh
= NULL
;
1499 struct kuc_hdr
*kuch
;
1502 CDEBUG(D_CHANGELOG
, "changelog to fp=%p start %llu\n",
1503 cs
->cs_fp
, cs
->cs_startrec
);
1505 cs
->cs_buf
= kzalloc(KUC_CHANGELOG_MSG_MAXSIZE
, GFP_NOFS
);
1511 /* Set up the remote catalog handle */
1512 ctxt
= llog_get_context(cs
->cs_obd
, LLOG_CHANGELOG_REPL_CTXT
);
1517 rc
= llog_open(NULL
, ctxt
, &llh
, NULL
, CHANGELOG_CATALOG
,
1520 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1521 cs
->cs_obd
->obd_name
, rc
);
1524 rc
= llog_init_handle(NULL
, llh
, LLOG_F_IS_CAT
, NULL
);
1526 CERROR("llog_init_handle failed %d\n", rc
);
1530 rc
= llog_cat_process(NULL
, llh
, changelog_kkuc_cb
, cs
, 0, 0);
1532 /* Send EOF no matter what our result */
1533 kuch
= changelog_kuc_hdr(cs
->cs_buf
, sizeof(*kuch
), cs
->cs_flags
);
1535 kuch
->kuc_msgtype
= CL_EOF
;
1536 libcfs_kkuc_msg_put(cs
->cs_fp
, kuch
);
1542 llog_cat_close(NULL
, llh
);
1544 llog_ctxt_put(ctxt
);
1550 static int mdc_ioc_changelog_send(struct obd_device
*obd
,
1551 struct ioc_changelog
*icc
)
1553 struct changelog_show
*cs
;
1554 struct task_struct
*task
;
1557 /* Freed in mdc_changelog_send_thread */
1558 cs
= kzalloc(sizeof(*cs
), GFP_NOFS
);
1563 cs
->cs_startrec
= icc
->icc_recno
;
1564 /* matching fput in mdc_changelog_send_thread */
1565 cs
->cs_fp
= fget(icc
->icc_id
);
1566 cs
->cs_flags
= icc
->icc_flags
;
1569 * New thread because we should return to user app before
1570 * writing into our pipe
1572 task
= kthread_run(mdc_changelog_send_thread
, cs
,
1573 "mdc_clg_send_thread");
1576 CERROR("%s: can't start changelog thread: rc = %d\n",
1581 CDEBUG(D_CHANGELOG
, "%s: started changelog thread\n",
1585 CERROR("Failed to start changelog thread: %d\n", rc
);
1589 static int mdc_ioc_hsm_ct_start(struct obd_export
*exp
,
1590 struct lustre_kernelcomm
*lk
);
1592 static int mdc_quotacheck(struct obd_device
*unused
, struct obd_export
*exp
,
1593 struct obd_quotactl
*oqctl
)
1595 struct client_obd
*cli
= &exp
->exp_obd
->u
.cli
;
1596 struct ptlrpc_request
*req
;
1597 struct obd_quotactl
*body
;
1600 req
= ptlrpc_request_alloc_pack(class_exp2cliimp(exp
),
1601 &RQF_MDS_QUOTACHECK
, LUSTRE_MDS_VERSION
,
1606 body
= req_capsule_client_get(&req
->rq_pill
, &RMF_OBD_QUOTACTL
);
1609 ptlrpc_request_set_replen(req
);
1611 /* the next poll will find -ENODATA, that means quotacheck is
1614 cli
->cl_qchk_stat
= -ENODATA
;
1615 rc
= ptlrpc_queue_wait(req
);
1617 cli
->cl_qchk_stat
= rc
;
1618 ptlrpc_req_finished(req
);
1622 static int mdc_quota_poll_check(struct obd_export
*exp
,
1623 struct if_quotacheck
*qchk
)
1625 struct client_obd
*cli
= &exp
->exp_obd
->u
.cli
;
1628 qchk
->obd_uuid
= cli
->cl_target_uuid
;
1629 memcpy(qchk
->obd_type
, LUSTRE_MDS_NAME
, strlen(LUSTRE_MDS_NAME
));
1631 rc
= cli
->cl_qchk_stat
;
1632 /* the client is not the previous one */
1633 if (rc
== CL_NOT_QUOTACHECKED
)
1638 static int mdc_quotactl(struct obd_device
*unused
, struct obd_export
*exp
,
1639 struct obd_quotactl
*oqctl
)
1641 struct ptlrpc_request
*req
;
1642 struct obd_quotactl
*oqc
;
1645 req
= ptlrpc_request_alloc_pack(class_exp2cliimp(exp
),
1646 &RQF_MDS_QUOTACTL
, LUSTRE_MDS_VERSION
,
1651 oqc
= req_capsule_client_get(&req
->rq_pill
, &RMF_OBD_QUOTACTL
);
1654 ptlrpc_request_set_replen(req
);
1655 ptlrpc_at_set_req_timeout(req
);
1656 req
->rq_no_resend
= 1;
1658 rc
= ptlrpc_queue_wait(req
);
1660 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc
);
1662 if (req
->rq_repmsg
) {
1663 oqc
= req_capsule_server_get(&req
->rq_pill
, &RMF_OBD_QUOTACTL
);
1667 CERROR("Can't unpack obd_quotactl\n");
1671 CERROR("Can't unpack obd_quotactl\n");
1674 ptlrpc_req_finished(req
);
1679 static int mdc_ioc_swap_layouts(struct obd_export
*exp
,
1680 struct md_op_data
*op_data
)
1683 struct ptlrpc_request
*req
;
1685 struct mdc_swap_layouts
*msl
, *payload
;
1687 msl
= op_data
->op_data
;
1689 /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1690 * first thing it will do is to cancel the 2 layout
1691 * locks hold by this client.
1692 * So the client must cancel its layout locks on the 2 fids
1693 * with the request RPC to avoid extra RPC round trips
1695 count
= mdc_resource_get_unused(exp
, &op_data
->op_fid1
, &cancels
,
1696 LCK_CR
, MDS_INODELOCK_LAYOUT
);
1697 count
+= mdc_resource_get_unused(exp
, &op_data
->op_fid2
, &cancels
,
1698 LCK_CR
, MDS_INODELOCK_LAYOUT
);
1700 req
= ptlrpc_request_alloc(class_exp2cliimp(exp
),
1701 &RQF_MDS_SWAP_LAYOUTS
);
1703 ldlm_lock_list_put(&cancels
, l_bl_ast
, count
);
1707 rc
= mdc_prep_elc_req(exp
, req
, MDS_SWAP_LAYOUTS
, &cancels
, count
);
1709 ptlrpc_request_free(req
);
1713 mdc_swap_layouts_pack(req
, op_data
);
1715 payload
= req_capsule_client_get(&req
->rq_pill
, &RMF_SWAP_LAYOUTS
);
1720 ptlrpc_request_set_replen(req
);
1722 rc
= ptlrpc_queue_wait(req
);
1724 ptlrpc_req_finished(req
);
1728 static int mdc_iocontrol(unsigned int cmd
, struct obd_export
*exp
, int len
,
1729 void *karg
, void __user
*uarg
)
1731 struct obd_device
*obd
= exp
->exp_obd
;
1732 struct obd_ioctl_data
*data
= karg
;
1733 struct obd_import
*imp
= obd
->u
.cli
.cl_import
;
1736 if (!try_module_get(THIS_MODULE
)) {
1737 CERROR("%s: cannot get module '%s'\n", obd
->obd_name
,
1738 module_name(THIS_MODULE
));
1742 case OBD_IOC_CHANGELOG_SEND
:
1743 rc
= mdc_ioc_changelog_send(obd
, karg
);
1745 case OBD_IOC_CHANGELOG_CLEAR
: {
1746 struct ioc_changelog
*icc
= karg
;
1747 struct changelog_setinfo cs
= {
1748 .cs_recno
= icc
->icc_recno
,
1749 .cs_id
= icc
->icc_id
1752 rc
= obd_set_info_async(NULL
, exp
, strlen(KEY_CHANGELOG_CLEAR
),
1753 KEY_CHANGELOG_CLEAR
, sizeof(cs
), &cs
,
1757 case OBD_IOC_FID2PATH
:
1758 rc
= mdc_ioc_fid2path(exp
, karg
);
1760 case LL_IOC_HSM_CT_START
:
1761 rc
= mdc_ioc_hsm_ct_start(exp
, karg
);
1762 /* ignore if it was already registered on this MDS. */
1766 case LL_IOC_HSM_PROGRESS
:
1767 rc
= mdc_ioc_hsm_progress(exp
, karg
);
1769 case LL_IOC_HSM_STATE_GET
:
1770 rc
= mdc_ioc_hsm_state_get(exp
, karg
);
1772 case LL_IOC_HSM_STATE_SET
:
1773 rc
= mdc_ioc_hsm_state_set(exp
, karg
);
1775 case LL_IOC_HSM_ACTION
:
1776 rc
= mdc_ioc_hsm_current_action(exp
, karg
);
1778 case LL_IOC_HSM_REQUEST
:
1779 rc
= mdc_ioc_hsm_request(exp
, karg
);
1781 case OBD_IOC_CLIENT_RECOVER
:
1782 rc
= ptlrpc_recover_import(imp
, data
->ioc_inlbuf1
, 0);
1787 case IOC_OSC_SET_ACTIVE
:
1788 rc
= ptlrpc_set_import_active(imp
, data
->ioc_offset
);
1790 case OBD_IOC_POLL_QUOTACHECK
:
1791 rc
= mdc_quota_poll_check(exp
, (struct if_quotacheck
*)karg
);
1793 case OBD_IOC_PING_TARGET
:
1794 rc
= ptlrpc_obd_ping(obd
);
1797 * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1798 * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1799 * there'd be no LMV layer thus we might be called here. Eventually
1800 * this code should be removed.
1803 case IOC_OBD_STATFS
: {
1804 struct obd_statfs stat_buf
= {0};
1806 if (*((__u32
*) data
->ioc_inlbuf2
) != 0) {
1812 if (copy_to_user(data
->ioc_pbuf2
, obd2cli_tgt(obd
),
1813 min_t(size_t, data
->ioc_plen2
,
1814 sizeof(struct obd_uuid
)))) {
1819 rc
= mdc_statfs(NULL
, obd
->obd_self_export
, &stat_buf
,
1820 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS
),
1825 if (copy_to_user(data
->ioc_pbuf1
, &stat_buf
,
1826 min_t(size_t, data
->ioc_plen1
,
1827 sizeof(stat_buf
)))) {
1835 case OBD_IOC_QUOTACTL
: {
1836 struct if_quotactl
*qctl
= karg
;
1837 struct obd_quotactl
*oqctl
;
1839 oqctl
= kzalloc(sizeof(*oqctl
), GFP_NOFS
);
1845 QCTL_COPY(oqctl
, qctl
);
1846 rc
= obd_quotactl(exp
, oqctl
);
1848 QCTL_COPY(qctl
, oqctl
);
1849 qctl
->qc_valid
= QC_MDTIDX
;
1850 qctl
->obd_uuid
= obd
->u
.cli
.cl_target_uuid
;
1856 case LL_IOC_GET_CONNECT_FLAGS
:
1857 if (copy_to_user(uarg
, exp_connect_flags_ptr(exp
),
1858 sizeof(*exp_connect_flags_ptr(exp
)))) {
1865 case LL_IOC_LOV_SWAP_LAYOUTS
:
1866 rc
= mdc_ioc_swap_layouts(exp
, karg
);
1869 CERROR("unrecognised ioctl: cmd = %#x\n", cmd
);
1874 module_put(THIS_MODULE
);
1879 static int mdc_get_info_rpc(struct obd_export
*exp
,
1880 u32 keylen
, void *key
,
1881 int vallen
, void *val
)
1883 struct obd_import
*imp
= class_exp2cliimp(exp
);
1884 struct ptlrpc_request
*req
;
1888 req
= ptlrpc_request_alloc(imp
, &RQF_MDS_GET_INFO
);
1892 req_capsule_set_size(&req
->rq_pill
, &RMF_GETINFO_KEY
,
1893 RCL_CLIENT
, keylen
);
1894 req_capsule_set_size(&req
->rq_pill
, &RMF_GETINFO_VALLEN
,
1895 RCL_CLIENT
, sizeof(__u32
));
1897 rc
= ptlrpc_request_pack(req
, LUSTRE_MDS_VERSION
, MDS_GET_INFO
);
1899 ptlrpc_request_free(req
);
1903 tmp
= req_capsule_client_get(&req
->rq_pill
, &RMF_GETINFO_KEY
);
1904 memcpy(tmp
, key
, keylen
);
1905 tmp
= req_capsule_client_get(&req
->rq_pill
, &RMF_GETINFO_VALLEN
);
1906 memcpy(tmp
, &vallen
, sizeof(__u32
));
1908 req_capsule_set_size(&req
->rq_pill
, &RMF_GETINFO_VAL
,
1909 RCL_SERVER
, vallen
);
1910 ptlrpc_request_set_replen(req
);
1912 rc
= ptlrpc_queue_wait(req
);
1913 /* -EREMOTE means the get_info result is partial, and it needs to
1914 * continue on another MDT, see fid2path part in lmv_iocontrol
1916 if (rc
== 0 || rc
== -EREMOTE
) {
1917 tmp
= req_capsule_server_get(&req
->rq_pill
, &RMF_GETINFO_VAL
);
1918 memcpy(val
, tmp
, vallen
);
1919 if (ptlrpc_rep_need_swab(req
)) {
1920 if (KEY_IS(KEY_FID2PATH
))
1921 lustre_swab_fid2path(val
);
1924 ptlrpc_req_finished(req
);
1929 static void lustre_swab_hai(struct hsm_action_item
*h
)
1931 __swab32s(&h
->hai_len
);
1932 __swab32s(&h
->hai_action
);
1933 lustre_swab_lu_fid(&h
->hai_fid
);
1934 lustre_swab_lu_fid(&h
->hai_dfid
);
1935 __swab64s(&h
->hai_cookie
);
1936 __swab64s(&h
->hai_extent
.offset
);
1937 __swab64s(&h
->hai_extent
.length
);
1938 __swab64s(&h
->hai_gid
);
1941 static void lustre_swab_hal(struct hsm_action_list
*h
)
1943 struct hsm_action_item
*hai
;
1946 __swab32s(&h
->hal_version
);
1947 __swab32s(&h
->hal_count
);
1948 __swab32s(&h
->hal_archive_id
);
1949 __swab64s(&h
->hal_flags
);
1951 for (i
= 0; i
< h
->hal_count
; i
++, hai
= hai_next(hai
))
1952 lustre_swab_hai(hai
);
1955 static void lustre_swab_kuch(struct kuc_hdr
*l
)
1957 __swab16s(&l
->kuc_magic
);
1958 /* __u8 l->kuc_transport */
1959 __swab16s(&l
->kuc_msgtype
);
1960 __swab16s(&l
->kuc_msglen
);
1963 static int mdc_ioc_hsm_ct_start(struct obd_export
*exp
,
1964 struct lustre_kernelcomm
*lk
)
1966 struct obd_import
*imp
= class_exp2cliimp(exp
);
1967 __u32 archive
= lk
->lk_data
;
1970 if (lk
->lk_group
!= KUC_GRP_HSM
) {
1971 CERROR("Bad copytool group %d\n", lk
->lk_group
);
1975 CDEBUG(D_HSM
, "CT start r%d w%d u%d g%d f%#x\n", lk
->lk_rfd
, lk
->lk_wfd
,
1976 lk
->lk_uid
, lk
->lk_group
, lk
->lk_flags
);
1978 if (lk
->lk_flags
& LK_FLG_STOP
) {
1979 /* Unregister with the coordinator */
1980 rc
= mdc_ioc_hsm_ct_unregister(imp
);
1982 rc
= mdc_ioc_hsm_ct_register(imp
, archive
);
1989 * Send a message to any listening copytools
1990 * @param val KUC message (kuc_hdr + hsm_action_list)
1991 * @param len total length of message
1993 static int mdc_hsm_copytool_send(int len
, void *val
)
1995 struct kuc_hdr
*lh
= (struct kuc_hdr
*)val
;
1996 struct hsm_action_list
*hal
= (struct hsm_action_list
*)(lh
+ 1);
1998 if (len
< sizeof(*lh
) + sizeof(*hal
)) {
1999 CERROR("Short HSM message %d < %d\n", len
,
2000 (int) (sizeof(*lh
) + sizeof(*hal
)));
2003 if (lh
->kuc_magic
== __swab16(KUC_MAGIC
)) {
2004 lustre_swab_kuch(lh
);
2005 lustre_swab_hal(hal
);
2006 } else if (lh
->kuc_magic
!= KUC_MAGIC
) {
2007 CERROR("Bad magic %x!=%x\n", lh
->kuc_magic
, KUC_MAGIC
);
2012 "Received message mg=%x t=%d m=%d l=%d actions=%d on %s\n",
2013 lh
->kuc_magic
, lh
->kuc_transport
, lh
->kuc_msgtype
,
2014 lh
->kuc_msglen
, hal
->hal_count
, hal
->hal_fsname
);
2016 /* Broadcast to HSM listeners */
2017 return libcfs_kkuc_group_put(KUC_GRP_HSM
, lh
);
2021 * callback function passed to kuc for re-registering each HSM copytool
2022 * running on MDC, after MDT shutdown/recovery.
2023 * @param data copytool registration data
2024 * @param cb_arg callback argument (obd_import)
2026 static int mdc_hsm_ct_reregister(void *data
, void *cb_arg
)
2028 struct kkuc_ct_data
*kcd
= data
;
2029 struct obd_import
*imp
= (struct obd_import
*)cb_arg
;
2032 if (!kcd
|| kcd
->kcd_magic
!= KKUC_CT_DATA_MAGIC
)
2035 if (!obd_uuid_equals(&kcd
->kcd_uuid
, &imp
->imp_obd
->obd_uuid
))
2038 CDEBUG(D_HA
, "%s: recover copytool registration to MDT (archive=%#x)\n",
2039 imp
->imp_obd
->obd_name
, kcd
->kcd_archive
);
2040 rc
= mdc_ioc_hsm_ct_register(imp
, kcd
->kcd_archive
);
2042 /* ignore error if the copytool is already registered */
2043 return (rc
== -EEXIST
) ? 0 : rc
;
2046 static int mdc_set_info_async(const struct lu_env
*env
,
2047 struct obd_export
*exp
,
2048 u32 keylen
, void *key
,
2049 u32 vallen
, void *val
,
2050 struct ptlrpc_request_set
*set
)
2052 struct obd_import
*imp
= class_exp2cliimp(exp
);
2055 if (KEY_IS(KEY_READ_ONLY
)) {
2056 if (vallen
!= sizeof(int))
2059 spin_lock(&imp
->imp_lock
);
2060 if (*((int *)val
)) {
2061 imp
->imp_connect_flags_orig
|= OBD_CONNECT_RDONLY
;
2062 imp
->imp_connect_data
.ocd_connect_flags
|=
2065 imp
->imp_connect_flags_orig
&= ~OBD_CONNECT_RDONLY
;
2066 imp
->imp_connect_data
.ocd_connect_flags
&=
2067 ~OBD_CONNECT_RDONLY
;
2069 spin_unlock(&imp
->imp_lock
);
2071 rc
= do_set_info_async(imp
, MDS_SET_INFO
, LUSTRE_MDS_VERSION
,
2072 keylen
, key
, vallen
, val
, set
);
2075 if (KEY_IS(KEY_SPTLRPC_CONF
)) {
2076 sptlrpc_conf_client_adapt(exp
->exp_obd
);
2079 if (KEY_IS(KEY_FLUSH_CTX
)) {
2080 sptlrpc_import_flush_my_ctx(imp
);
2083 if (KEY_IS(KEY_CHANGELOG_CLEAR
)) {
2084 rc
= do_set_info_async(imp
, MDS_SET_INFO
, LUSTRE_MDS_VERSION
,
2085 keylen
, key
, vallen
, val
, set
);
2088 if (KEY_IS(KEY_HSM_COPYTOOL_SEND
)) {
2089 rc
= mdc_hsm_copytool_send(vallen
, val
);
2093 CERROR("Unknown key %s\n", (char *)key
);
2097 static int mdc_get_info(const struct lu_env
*env
, struct obd_export
*exp
,
2098 __u32 keylen
, void *key
, __u32
*vallen
, void *val
,
2099 struct lov_stripe_md
*lsm
)
2103 if (KEY_IS(KEY_MAX_EASIZE
)) {
2104 int mdsize
, *max_easize
;
2106 if (*vallen
!= sizeof(int))
2108 mdsize
= *(int *)val
;
2109 if (mdsize
> exp
->exp_obd
->u
.cli
.cl_max_mds_easize
)
2110 exp
->exp_obd
->u
.cli
.cl_max_mds_easize
= mdsize
;
2112 *max_easize
= exp
->exp_obd
->u
.cli
.cl_max_mds_easize
;
2114 } else if (KEY_IS(KEY_DEFAULT_EASIZE
)) {
2115 int *default_easize
;
2117 if (*vallen
!= sizeof(int))
2119 default_easize
= val
;
2120 *default_easize
= exp
->exp_obd
->u
.cli
.cl_default_mds_easize
;
2122 } else if (KEY_IS(KEY_CONN_DATA
)) {
2123 struct obd_import
*imp
= class_exp2cliimp(exp
);
2124 struct obd_connect_data
*data
= val
;
2126 if (*vallen
!= sizeof(*data
))
2129 *data
= imp
->imp_connect_data
;
2131 } else if (KEY_IS(KEY_TGT_COUNT
)) {
2136 rc
= mdc_get_info_rpc(exp
, keylen
, key
, *vallen
, val
);
2141 static int mdc_sync(struct obd_export
*exp
, const struct lu_fid
*fid
,
2142 struct ptlrpc_request
**request
)
2144 struct ptlrpc_request
*req
;
2148 req
= ptlrpc_request_alloc(class_exp2cliimp(exp
), &RQF_MDS_SYNC
);
2152 rc
= ptlrpc_request_pack(req
, LUSTRE_MDS_VERSION
, MDS_SYNC
);
2154 ptlrpc_request_free(req
);
2158 mdc_pack_body(req
, fid
, 0, 0, -1, 0);
2160 ptlrpc_request_set_replen(req
);
2162 rc
= ptlrpc_queue_wait(req
);
2164 ptlrpc_req_finished(req
);
2170 static int mdc_import_event(struct obd_device
*obd
, struct obd_import
*imp
,
2171 enum obd_import_event event
)
2175 LASSERT(imp
->imp_obd
== obd
);
2178 case IMP_EVENT_DISCON
: {
2180 /* XXX Pass event up to OBDs stack. used only for FLD now */
2181 rc
= obd_notify_observer(obd
, obd
, OBD_NOTIFY_DISCON
, NULL
);
2185 case IMP_EVENT_INACTIVE
: {
2186 struct client_obd
*cli
= &obd
->u
.cli
;
2188 * Flush current sequence to make client obtain new one
2189 * from server in case of disconnect/reconnect.
2192 seq_client_flush(cli
->cl_seq
);
2194 rc
= obd_notify_observer(obd
, obd
, OBD_NOTIFY_INACTIVE
, NULL
);
2197 case IMP_EVENT_INVALIDATE
: {
2198 struct ldlm_namespace
*ns
= obd
->obd_namespace
;
2200 ldlm_namespace_cleanup(ns
, LDLM_FL_LOCAL_ONLY
);
2204 case IMP_EVENT_ACTIVE
:
2205 rc
= obd_notify_observer(obd
, obd
, OBD_NOTIFY_ACTIVE
, NULL
);
2206 /* redo the kuc registration after reconnecting */
2208 /* re-register HSM agents */
2209 rc
= libcfs_kkuc_group_foreach(KUC_GRP_HSM
,
2210 mdc_hsm_ct_reregister
,
2214 rc
= obd_notify_observer(obd
, obd
, OBD_NOTIFY_OCD
, NULL
);
2216 case IMP_EVENT_DEACTIVATE
:
2217 case IMP_EVENT_ACTIVATE
:
2220 CERROR("Unknown import event %x\n", event
);
2226 int mdc_fid_alloc(struct obd_export
*exp
, struct lu_fid
*fid
,
2227 struct md_op_data
*op_data
)
2229 struct client_obd
*cli
= &exp
->exp_obd
->u
.cli
;
2230 struct lu_client_seq
*seq
= cli
->cl_seq
;
2232 return seq_client_alloc_fid(NULL
, seq
, fid
);
2235 static struct obd_uuid
*mdc_get_uuid(struct obd_export
*exp
)
2237 struct client_obd
*cli
= &exp
->exp_obd
->u
.cli
;
2239 return &cli
->cl_target_uuid
;
2243 * Determine whether the lock can be canceled before replaying it during
2244 * recovery, non zero value will be return if the lock can be canceled,
2245 * or zero returned for not
2247 static int mdc_cancel_weight(struct ldlm_lock
*lock
)
2249 if (lock
->l_resource
->lr_type
!= LDLM_IBITS
)
2252 /* FIXME: if we ever get into a situation where there are too many
2253 * opened files with open locks on a single node, then we really
2254 * should replay these open locks to reget it
2256 if (lock
->l_policy_data
.l_inodebits
.bits
& MDS_INODELOCK_OPEN
)
2262 static int mdc_resource_inode_free(struct ldlm_resource
*res
)
2264 if (res
->lr_lvb_inode
)
2265 res
->lr_lvb_inode
= NULL
;
2270 static struct ldlm_valblock_ops inode_lvbo
= {
2271 .lvbo_free
= mdc_resource_inode_free
,
2274 static int mdc_llog_init(struct obd_device
*obd
)
2276 struct obd_llog_group
*olg
= &obd
->obd_olg
;
2277 struct llog_ctxt
*ctxt
;
2280 rc
= llog_setup(NULL
, obd
, olg
, LLOG_CHANGELOG_REPL_CTXT
, obd
,
2285 ctxt
= llog_group_get_ctxt(olg
, LLOG_CHANGELOG_REPL_CTXT
);
2286 llog_initiator_connect(ctxt
);
2287 llog_ctxt_put(ctxt
);
2292 static void mdc_llog_finish(struct obd_device
*obd
)
2294 struct llog_ctxt
*ctxt
;
2296 ctxt
= llog_get_context(obd
, LLOG_CHANGELOG_REPL_CTXT
);
2298 llog_cleanup(NULL
, ctxt
);
2301 static int mdc_setup(struct obd_device
*obd
, struct lustre_cfg
*cfg
)
2303 struct client_obd
*cli
= &obd
->u
.cli
;
2304 struct lprocfs_static_vars lvars
= { NULL
};
2307 cli
->cl_rpc_lock
= kzalloc(sizeof(*cli
->cl_rpc_lock
), GFP_NOFS
);
2308 if (!cli
->cl_rpc_lock
)
2310 mdc_init_rpc_lock(cli
->cl_rpc_lock
);
2312 rc
= ptlrpcd_addref();
2316 cli
->cl_close_lock
= kzalloc(sizeof(*cli
->cl_close_lock
), GFP_NOFS
);
2317 if (!cli
->cl_close_lock
) {
2319 goto err_ptlrpcd_decref
;
2321 mdc_init_rpc_lock(cli
->cl_close_lock
);
2323 rc
= client_obd_setup(obd
, cfg
);
2325 goto err_close_lock
;
2326 lprocfs_mdc_init_vars(&lvars
);
2327 lprocfs_obd_setup(obd
, lvars
.obd_vars
, lvars
.sysfs_vars
);
2328 sptlrpc_lprocfs_cliobd_attach(obd
);
2329 ptlrpc_lprocfs_register_obd(obd
);
2331 ns_register_cancel(obd
->obd_namespace
, mdc_cancel_weight
);
2333 obd
->obd_namespace
->ns_lvbo
= &inode_lvbo
;
2335 rc
= mdc_llog_init(obd
);
2338 CERROR("failed to setup llogging subsystems\n");
2344 kfree(cli
->cl_close_lock
);
2348 kfree(cli
->cl_rpc_lock
);
2352 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
2353 * us to make MDS RPCs with large enough reply buffers to hold a default
2354 * sized EA and cookie without having to calculate this (via a call into the
2355 * LOV + OSCs) each time we make an RPC. The maximum size is also tracked
2356 * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2357 * a large number of stripes is possible. If a larger reply buffer is
2358 * required it will be reallocated in the ptlrpc layer due to overflow.
2360 static int mdc_init_ea_size(struct obd_export
*exp
, int easize
,
2361 int def_easize
, int cookiesize
, int def_cookiesize
)
2363 struct obd_device
*obd
= exp
->exp_obd
;
2364 struct client_obd
*cli
= &obd
->u
.cli
;
2366 if (cli
->cl_max_mds_easize
< easize
)
2367 cli
->cl_max_mds_easize
= easize
;
2369 if (cli
->cl_default_mds_easize
< def_easize
)
2370 cli
->cl_default_mds_easize
= def_easize
;
2372 if (cli
->cl_max_mds_cookiesize
< cookiesize
)
2373 cli
->cl_max_mds_cookiesize
= cookiesize
;
2375 if (cli
->cl_default_mds_cookiesize
< def_cookiesize
)
2376 cli
->cl_default_mds_cookiesize
= def_cookiesize
;
2381 static int mdc_precleanup(struct obd_device
*obd
, enum obd_cleanup_stage stage
)
2384 case OBD_CLEANUP_EARLY
:
2386 case OBD_CLEANUP_EXPORTS
:
2387 /* Failsafe, ok if racy */
2388 if (obd
->obd_type
->typ_refcnt
<= 1)
2389 libcfs_kkuc_group_rem(0, KUC_GRP_HSM
);
2391 obd_cleanup_client_import(obd
);
2392 ptlrpc_lprocfs_unregister_obd(obd
);
2393 lprocfs_obd_cleanup(obd
);
2395 mdc_llog_finish(obd
);
2401 static int mdc_cleanup(struct obd_device
*obd
)
2403 struct client_obd
*cli
= &obd
->u
.cli
;
2405 kfree(cli
->cl_rpc_lock
);
2406 kfree(cli
->cl_close_lock
);
2410 return client_obd_cleanup(obd
);
2413 static int mdc_process_config(struct obd_device
*obd
, u32 len
, void *buf
)
2415 struct lustre_cfg
*lcfg
= buf
;
2416 struct lprocfs_static_vars lvars
= { NULL
};
2419 lprocfs_mdc_init_vars(&lvars
);
2420 switch (lcfg
->lcfg_command
) {
2422 rc
= class_process_proc_param(PARAM_MDC
, lvars
.obd_vars
,
2431 /* get remote permission for current user on fid */
2432 static int mdc_get_remote_perm(struct obd_export
*exp
, const struct lu_fid
*fid
,
2433 __u32 suppgid
, struct ptlrpc_request
**request
)
2435 struct ptlrpc_request
*req
;
2438 LASSERT(client_is_remote(exp
));
2441 req
= ptlrpc_request_alloc(class_exp2cliimp(exp
), &RQF_MDS_GETATTR
);
2445 rc
= ptlrpc_request_pack(req
, LUSTRE_MDS_VERSION
, MDS_GETATTR
);
2447 ptlrpc_request_free(req
);
2451 mdc_pack_body(req
, fid
, OBD_MD_FLRMTPERM
, 0, suppgid
, 0);
2453 req_capsule_set_size(&req
->rq_pill
, &RMF_ACL
, RCL_SERVER
,
2454 sizeof(struct mdt_remote_perm
));
2456 ptlrpc_request_set_replen(req
);
2458 rc
= ptlrpc_queue_wait(req
);
2460 ptlrpc_req_finished(req
);
2466 static struct obd_ops mdc_obd_ops
= {
2467 .owner
= THIS_MODULE
,
2469 .precleanup
= mdc_precleanup
,
2470 .cleanup
= mdc_cleanup
,
2471 .add_conn
= client_import_add_conn
,
2472 .del_conn
= client_import_del_conn
,
2473 .connect
= client_connect_import
,
2474 .disconnect
= client_disconnect_export
,
2475 .iocontrol
= mdc_iocontrol
,
2476 .set_info_async
= mdc_set_info_async
,
2477 .statfs
= mdc_statfs
,
2478 .fid_init
= client_fid_init
,
2479 .fid_fini
= client_fid_fini
,
2480 .fid_alloc
= mdc_fid_alloc
,
2481 .import_event
= mdc_import_event
,
2482 .get_info
= mdc_get_info
,
2483 .process_config
= mdc_process_config
,
2484 .get_uuid
= mdc_get_uuid
,
2485 .quotactl
= mdc_quotactl
,
2486 .quotacheck
= mdc_quotacheck
2489 static struct md_ops mdc_md_ops
= {
2490 .getstatus
= mdc_getstatus
,
2491 .null_inode
= mdc_null_inode
,
2492 .find_cbdata
= mdc_find_cbdata
,
2494 .create
= mdc_create
,
2495 .done_writing
= mdc_done_writing
,
2496 .enqueue
= mdc_enqueue
,
2497 .getattr
= mdc_getattr
,
2498 .getattr_name
= mdc_getattr_name
,
2499 .intent_lock
= mdc_intent_lock
,
2501 .is_subdir
= mdc_is_subdir
,
2502 .rename
= mdc_rename
,
2503 .setattr
= mdc_setattr
,
2504 .setxattr
= mdc_setxattr
,
2505 .getxattr
= mdc_getxattr
,
2507 .readpage
= mdc_readpage
,
2508 .unlink
= mdc_unlink
,
2509 .cancel_unused
= mdc_cancel_unused
,
2510 .init_ea_size
= mdc_init_ea_size
,
2511 .set_lock_data
= mdc_set_lock_data
,
2512 .lock_match
= mdc_lock_match
,
2513 .get_lustre_md
= mdc_get_lustre_md
,
2514 .free_lustre_md
= mdc_free_lustre_md
,
2515 .set_open_replay_data
= mdc_set_open_replay_data
,
2516 .clear_open_replay_data
= mdc_clear_open_replay_data
,
2517 .get_remote_perm
= mdc_get_remote_perm
,
2518 .intent_getattr_async
= mdc_intent_getattr_async
,
2519 .revalidate_lock
= mdc_revalidate_lock
2522 static int __init
mdc_init(void)
2524 struct lprocfs_static_vars lvars
= { NULL
};
2526 lprocfs_mdc_init_vars(&lvars
);
2528 return class_register_type(&mdc_obd_ops
, &mdc_md_ops
,
2529 LUSTRE_MDC_NAME
, NULL
);
2532 static void /*__exit*/ mdc_exit(void)
2534 class_unregister_type(LUSTRE_MDC_NAME
);
2537 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2538 MODULE_DESCRIPTION("Lustre Metadata Client");
2539 MODULE_VERSION(LUSTRE_VERSION_STRING
);
2540 MODULE_LICENSE("GPL");
2542 module_init(mdc_init
);
2543 module_exit(mdc_exit
);