projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
staging: lustre: remove ENTRY macro
[linux-2.6-block.git] / drivers / staging / lustre / lustre / mdc / mdc_request.c
CommitLineData
d7e09d03
PT		 1/*
2 * GPL HEADER START
3 *
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
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.
9 *
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).
15 *
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
19 *
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
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26/*
27 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2011, 2012, Intel Corporation.
31 */
32/*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 */
36
37#define DEBUG_SUBSYSTEM S_MDC
38
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>
44
45#include <lustre_acl.h>
46#include <obd_class.h>
47#include <lustre_fid.h>
48#include <lprocfs_status.h>
49#include <lustre_param.h>
50#include <lustre_log.h>
51
52#include "mdc_internal.h"
53
54#define REQUEST_MINOR 244
55
56struct mdc_renew_capa_args {
57 struct obd_capa *ra_oc;
58 renew_capa_cb_t ra_cb;
59};
60
61static int mdc_cleanup(struct obd_device *obd);
62
63int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
64 const struct req_msg_field *field, struct obd_capa **oc)
65{
66 struct lustre_capa *capa;
67 struct obd_capa *c;
d7e09d03
PT		 68
69 /* swabbed already in mdc_enqueue */
70 capa = req_capsule_server_get(&req->rq_pill, field);
71 if (capa == NULL)
72 RETURN(-EPROTO);
73
74 c = alloc_capa(CAPA_SITE_CLIENT);
75 if (IS_ERR(c)) {
76 CDEBUG(D_INFO, "alloc capa failed!\n");
77 RETURN(PTR_ERR(c));
78 } else {
79 c->c_capa = *capa;
80 *oc = c;
81 RETURN(0);
82 }
83}
84
85static inline int mdc_queue_wait(struct ptlrpc_request *req)
86{
87 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
88 int rc;
89
90 /* mdc_enter_request() ensures that this client has no more
91 * than cl_max_rpcs_in_flight RPCs simultaneously inf light
92 * against an MDT. */
93 rc = mdc_enter_request(cli);
94 if (rc != 0)
95 return rc;
96
97 rc = ptlrpc_queue_wait(req);
98 mdc_exit_request(cli);
99
100 return rc;
101}
102
103/* Helper that implements most of mdc_getstatus and signal_completed_replay. */
104/* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
105static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
106 struct obd_capa **pc, int level, int msg_flags)
107{
108 struct ptlrpc_request *req;
109 struct mdt_body *body;
110 int rc;
d7e09d03
PT		 111
112 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
113 LUSTRE_MDS_VERSION, MDS_GETSTATUS);
114 if (req == NULL)
115 RETURN(-ENOMEM);
116
117 mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
118 lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
119 req->rq_send_state = level;
120
121 ptlrpc_request_set_replen(req);
122
123 rc = ptlrpc_queue_wait(req);
124 if (rc)
125 GOTO(out, rc);
126
127 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
128 if (body == NULL)
129 GOTO(out, rc = -EPROTO);
130
131 if (body->valid & OBD_MD_FLMDSCAPA) {
132 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
133 if (rc)
134 GOTO(out, rc);
135 }
136
137 *rootfid = body->fid1;
138 CDEBUG(D_NET,
139 "root fid="DFID", last_committed="LPU64"\n",
140 PFID(rootfid),
141 lustre_msg_get_last_committed(req->rq_repmsg));
142 EXIT;
143out:
144 ptlrpc_req_finished(req);
145 return rc;
146}
147
148/* This should be mdc_get_info("rootfid") */
149int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
150 struct obd_capa **pc)
151{
152 return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
153 LUSTRE_IMP_FULL, 0);
154}
155
156/*
157 * This function now is known to always saying that it will receive 4 buffers
158 * from server. Even for cases when acl_size and md_size is zero, RPC header
159 * will contain 4 fields and RPC itself will contain zero size fields. This is
160 * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
161 * and thus zero, it shrinks it, making zero size. The same story about
162 * md_size. And this is course of problem when client waits for smaller number
163 * of fields. This issue will be fixed later when client gets aware of RPC
164 * layouts. --umka
165 */
166static int mdc_getattr_common(struct obd_export *exp,
167 struct ptlrpc_request *req)
168{
169 struct req_capsule *pill = &req->rq_pill;
170 struct mdt_body *body;
171 void *eadata;
172 int rc;
d7e09d03
PT		 173
174 /* Request message already built. */
175 rc = ptlrpc_queue_wait(req);
176 if (rc != 0)
177 RETURN(rc);
178
179 /* sanity check for the reply */
180 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
181 if (body == NULL)
182 RETURN(-EPROTO);
183
184 CDEBUG(D_NET, "mode: %o\n", body->mode);
185
186 if (body->eadatasize != 0) {
187 mdc_update_max_ea_from_body(exp, body);
188
189 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
190 body->eadatasize);
191 if (eadata == NULL)
192 RETURN(-EPROTO);
193 }
194
195 if (body->valid & OBD_MD_FLRMTPERM) {
196 struct mdt_remote_perm *perm;
197
198 LASSERT(client_is_remote(exp));
199 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
200 lustre_swab_mdt_remote_perm);
201 if (perm == NULL)
202 RETURN(-EPROTO);
203 }
204
205 if (body->valid & OBD_MD_FLMDSCAPA) {
206 struct lustre_capa *capa;
207 capa = req_capsule_server_get(pill, &RMF_CAPA1);
208 if (capa == NULL)
209 RETURN(-EPROTO);
210 }
211
212 RETURN(0);
213}
214
215int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
216 struct ptlrpc_request **request)
217{
218 struct ptlrpc_request *req;
219 int rc;
d7e09d03
PT		 220
221 /* Single MDS without an LMV case */
222 if (op_data->op_flags & MF_GET_MDT_IDX) {
223 op_data->op_mds = 0;
224 RETURN(0);
225 }
226 *request = NULL;
227 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
228 if (req == NULL)
229 RETURN(-ENOMEM);
230
231 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
232
233 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
234 if (rc) {
235 ptlrpc_request_free(req);
236 RETURN(rc);
237 }
238
239 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
240 op_data->op_valid, op_data->op_mode, -1, 0);
241
242 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
243 op_data->op_mode);
244 if (op_data->op_valid & OBD_MD_FLRMTPERM) {
245 LASSERT(client_is_remote(exp));
246 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
247 sizeof(struct mdt_remote_perm));
248 }
249 ptlrpc_request_set_replen(req);
250
251 rc = mdc_getattr_common(exp, req);
252 if (rc)
253 ptlrpc_req_finished(req);
254 else
255 *request = req;
256 RETURN(rc);
257}
258
259int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
260 struct ptlrpc_request **request)
261{
262 struct ptlrpc_request *req;
263 int rc;
d7e09d03
PT		 264
265 *request = NULL;
266 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
267 &RQF_MDS_GETATTR_NAME);
268 if (req == NULL)
269 RETURN(-ENOMEM);
270
271 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
272 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
273 op_data->op_namelen + 1);
274
275 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
276 if (rc) {
277 ptlrpc_request_free(req);
278 RETURN(rc);
279 }
280
281 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
282 op_data->op_valid, op_data->op_mode,
283 op_data->op_suppgids[0], 0);
284
285 if (op_data->op_name) {
286 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
287 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
288 op_data->op_namelen);
289 memcpy(name, op_data->op_name, op_data->op_namelen);
290 }
291
292 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
293 op_data->op_mode);
294 ptlrpc_request_set_replen(req);
295
296 rc = mdc_getattr_common(exp, req);
297 if (rc)
298 ptlrpc_req_finished(req);
299 else
300 *request = req;
301 RETURN(rc);
302}
303
304static int mdc_is_subdir(struct obd_export *exp,
305 const struct lu_fid *pfid,
306 const struct lu_fid *cfid,
307 struct ptlrpc_request **request)
308{
309 struct ptlrpc_request *req;
310 int rc;
311
d7e09d03
PT		 312 *request = NULL;
313 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
314 &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
315 MDS_IS_SUBDIR);
316 if (req == NULL)
317 RETURN(-ENOMEM);
318
319 mdc_is_subdir_pack(req, pfid, cfid, 0);
320 ptlrpc_request_set_replen(req);
321
322 rc = ptlrpc_queue_wait(req);
323 if (rc && rc != -EREMOTE)
324 ptlrpc_req_finished(req);
325 else
326 *request = req;
327 RETURN(rc);
328}
329
330static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
331 const struct lu_fid *fid,
332 struct obd_capa *oc, int opcode, obd_valid valid,
333 const char *xattr_name, const char *input,
334 int input_size, int output_size, int flags,
335 __u32 suppgid, struct ptlrpc_request **request)
336{
337 struct ptlrpc_request *req;
338 int xattr_namelen = 0;
339 char *tmp;
340 int rc;
d7e09d03
PT		 341
342 *request = NULL;
343 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
344 if (req == NULL)
345 RETURN(-ENOMEM);
346
347 mdc_set_capa_size(req, &RMF_CAPA1, oc);
348 if (xattr_name) {
349 xattr_namelen = strlen(xattr_name) + 1;
350 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
351 xattr_namelen);
352 }
353 if (input_size) {
354 LASSERT(input);
355 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
356 input_size);
357 }
358
359 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
360 if (rc) {
361 ptlrpc_request_free(req);
362 RETURN(rc);
363 }
364
365 if (opcode == MDS_REINT) {
366 struct mdt_rec_setxattr *rec;
367
368 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
369 sizeof(struct mdt_rec_reint));
370 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
371 rec->sx_opcode = REINT_SETXATTR;
4b1a25f0
PT		 372 rec->sx_fsuid = from_kuid(&init_user_ns, current_fsuid());
373 rec->sx_fsgid = from_kgid(&init_user_ns, current_fsgid());
d7e09d03
PT		 374 rec->sx_cap = cfs_curproc_cap_pack();
375 rec->sx_suppgid1 = suppgid;
376 rec->sx_suppgid2 = -1;
377 rec->sx_fid = *fid;
378 rec->sx_valid = valid | OBD_MD_FLCTIME;
379 rec->sx_time = cfs_time_current_sec();
380 rec->sx_size = output_size;
381 rec->sx_flags = flags;
382
383 mdc_pack_capa(req, &RMF_CAPA1, oc);
384 } else {
385 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
386 }
387
388 if (xattr_name) {
389 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
390 memcpy(tmp, xattr_name, xattr_namelen);
391 }
392 if (input_size) {
393 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
394 memcpy(tmp, input, input_size);
395 }
396
397 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
398 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
399 RCL_SERVER, output_size);
400 ptlrpc_request_set_replen(req);
401
402 /* make rpc */
403 if (opcode == MDS_REINT)
404 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
405
406 rc = ptlrpc_queue_wait(req);
407
408 if (opcode == MDS_REINT)
409 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
410
411 if (rc)
412 ptlrpc_req_finished(req);
413 else
414 *request = req;
415 RETURN(rc);
416}
417
418int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
419 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
420 const char *input, int input_size, int output_size,
421 int flags, __u32 suppgid, struct ptlrpc_request **request)
422{
423 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
424 fid, oc, MDS_REINT, valid, xattr_name,
425 input, input_size, output_size, flags,
426 suppgid, request);
427}
428
429int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
430 struct obd_capa *oc, obd_valid valid, const char *xattr_name,
431 const char *input, int input_size, int output_size,
432 int flags, struct ptlrpc_request **request)
433{
434 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
435 fid, oc, MDS_GETXATTR, valid, xattr_name,
436 input, input_size, output_size, flags,
437 -1, request);
438}
439
440#ifdef CONFIG_FS_POSIX_ACL
441static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
442{
443 struct req_capsule *pill = &req->rq_pill;
444 struct mdt_body *body = md->body;
445 struct posix_acl *acl;
446 void *buf;
447 int rc;
d7e09d03
PT		 448
449 if (!body->aclsize)
450 RETURN(0);
451
452 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
453
454 if (!buf)
455 RETURN(-EPROTO);
456
457 acl = posix_acl_from_xattr(&init_user_ns, buf, body->aclsize);
458 if (IS_ERR(acl)) {
459 rc = PTR_ERR(acl);
460 CERROR("convert xattr to acl: %d\n", rc);
461 RETURN(rc);
462 }
463
464 rc = posix_acl_valid(acl);
465 if (rc) {
466 CERROR("validate acl: %d\n", rc);
467 posix_acl_release(acl);
468 RETURN(rc);
469 }
470
471 md->posix_acl = acl;
472 RETURN(0);
473}
474#else
475#define mdc_unpack_acl(req, md) 0
476#endif
477
478int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
479 struct obd_export *dt_exp, struct obd_export *md_exp,
480 struct lustre_md *md)
481{
482 struct req_capsule *pill = &req->rq_pill;
483 int rc;
d7e09d03
PT		 484
485 LASSERT(md);
486 memset(md, 0, sizeof(*md));
487
488 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
489 LASSERT(md->body != NULL);
490
491 if (md->body->valid & OBD_MD_FLEASIZE) {
492 int lmmsize;
493 struct lov_mds_md *lmm;
494
495 if (!S_ISREG(md->body->mode)) {
496 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
497 "regular file, but is not\n");
498 GOTO(out, rc = -EPROTO);
499 }
500
501 if (md->body->eadatasize == 0) {
502 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
503 "but eadatasize 0\n");
504 GOTO(out, rc = -EPROTO);
505 }
506 lmmsize = md->body->eadatasize;
507 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
508 if (!lmm)
509 GOTO(out, rc = -EPROTO);
510
511 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
512 if (rc < 0)
513 GOTO(out, rc);
514
515 if (rc < sizeof(*md->lsm)) {
516 CDEBUG(D_INFO, "lsm size too small: "
517 "rc < sizeof (*md->lsm) (%d < %d)\n",
518 rc, (int)sizeof(*md->lsm));
519 GOTO(out, rc = -EPROTO);
520 }
521
522 } else if (md->body->valid & OBD_MD_FLDIREA) {
523 int lmvsize;
524 struct lov_mds_md *lmv;
525
526 if(!S_ISDIR(md->body->mode)) {
527 CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
528 "directory, but is not\n");
529 GOTO(out, rc = -EPROTO);
530 }
531
532 if (md->body->eadatasize == 0) {
533 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
534 "but eadatasize 0\n");
535 RETURN(-EPROTO);
536 }
537 if (md->body->valid & OBD_MD_MEA) {
538 lmvsize = md->body->eadatasize;
539 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
540 lmvsize);
541 if (!lmv)
542 GOTO(out, rc = -EPROTO);
543
544 rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
545 lmvsize);
546 if (rc < 0)
547 GOTO(out, rc);
548
549 if (rc < sizeof(*md->mea)) {
550 CDEBUG(D_INFO, "size too small: "
551 "rc < sizeof(*md->mea) (%d < %d)\n",
552 rc, (int)sizeof(*md->mea));
553 GOTO(out, rc = -EPROTO);
554 }
555 }
556 }
557 rc = 0;
558
559 if (md->body->valid & OBD_MD_FLRMTPERM) {
560 /* remote permission */
561 LASSERT(client_is_remote(exp));
562 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
563 lustre_swab_mdt_remote_perm);
564 if (!md->remote_perm)
565 GOTO(out, rc = -EPROTO);
566 }
567 else if (md->body->valid & OBD_MD_FLACL) {
568 /* for ACL, it's possible that FLACL is set but aclsize is zero.
569 * only when aclsize != 0 there's an actual segment for ACL
570 * in reply buffer.
571 */
572 if (md->body->aclsize) {
573 rc = mdc_unpack_acl(req, md);
574 if (rc)
575 GOTO(out, rc);
576#ifdef CONFIG_FS_POSIX_ACL
577 } else {
578 md->posix_acl = NULL;
579#endif
580 }
581 }
582 if (md->body->valid & OBD_MD_FLMDSCAPA) {
583 struct obd_capa *oc = NULL;
584
585 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
586 if (rc)
587 GOTO(out, rc);
588 md->mds_capa = oc;
589 }
590
591 if (md->body->valid & OBD_MD_FLOSSCAPA) {
592 struct obd_capa *oc = NULL;
593
594 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
595 if (rc)
596 GOTO(out, rc);
597 md->oss_capa = oc;
598 }
599
600 EXIT;
601out:
602 if (rc) {
603 if (md->oss_capa) {
604 capa_put(md->oss_capa);
605 md->oss_capa = NULL;
606 }
607 if (md->mds_capa) {
608 capa_put(md->mds_capa);
609 md->mds_capa = NULL;
610 }
611#ifdef CONFIG_FS_POSIX_ACL
612 posix_acl_release(md->posix_acl);
613#endif
614 if (md->lsm)
615 obd_free_memmd(dt_exp, &md->lsm);
616 }
617 return rc;
618}
619
620int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
621{
d7e09d03
PT		 622 RETURN(0);
623}
624
625/**
626 * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
627 * RPC chains.
628 */
629void mdc_replay_open(struct ptlrpc_request *req)
630{
631 struct md_open_data *mod = req->rq_cb_data;
632 struct ptlrpc_request *close_req;
633 struct obd_client_handle *och;
634 struct lustre_handle old;
635 struct mdt_body *body;
d7e09d03
PT		 636
637 if (mod == NULL) {
638 DEBUG_REQ(D_ERROR, req,
639 "Can't properly replay without open data.");
640 EXIT;
641 return;
642 }
643
644 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
645 LASSERT(body != NULL);
646
647 och = mod->mod_och;
648 if (och != NULL) {
649 struct lustre_handle *file_fh;
650
651 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
652
653 file_fh = &och->och_fh;
654 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
655 file_fh->cookie, body->handle.cookie);
656 old = *file_fh;
657 *file_fh = body->handle;
658 }
659 close_req = mod->mod_close_req;
660 if (close_req != NULL) {
661 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
662 struct mdt_ioepoch *epoch;
663
664 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
665 epoch = req_capsule_client_get(&close_req->rq_pill,
666 &RMF_MDT_EPOCH);
667 LASSERT(epoch);
668
669 if (och != NULL)
670 LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
671 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
672 epoch->handle = body->handle;
673 }
674 EXIT;
675}
676
677void mdc_commit_open(struct ptlrpc_request *req)
678{
679 struct md_open_data *mod = req->rq_cb_data;
680 if (mod == NULL)
681 return;
682
683 /**
684 * No need to touch md_open_data::mod_och, it holds a reference on
685 * \var mod and will zero references to each other, \var mod will be
686 * freed after that when md_open_data::mod_och will put the reference.
687 */
688
689 /**
690 * Do not let open request to disappear as it still may be needed
691 * for close rpc to happen (it may happen on evict only, otherwise
692 * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
693 * called), just mark this rpc as committed to distinguish these 2
694 * cases, see mdc_close() for details. The open request reference will
695 * be put along with freeing \var mod.
696 */
697 ptlrpc_request_addref(req);
698 spin_lock(&req->rq_lock);
699 req->rq_committed = 1;
700 spin_unlock(&req->rq_lock);
701 req->rq_cb_data = NULL;
702 obd_mod_put(mod);
703}
704
705int mdc_set_open_replay_data(struct obd_export *exp,
706 struct obd_client_handle *och,
707 struct ptlrpc_request *open_req)
708{
709 struct md_open_data *mod;
710 struct mdt_rec_create *rec;
711 struct mdt_body *body;
712 struct obd_import *imp = open_req->rq_import;
d7e09d03
PT		 713
714 if (!open_req->rq_replay)
715 RETURN(0);
716
717 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
718 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
719 LASSERT(rec != NULL);
720 /* Incoming message in my byte order (it's been swabbed). */
721 /* Outgoing messages always in my byte order. */
722 LASSERT(body != NULL);
723
724 /* Only if the import is replayable, we set replay_open data */
725 if (och && imp->imp_replayable) {
726 mod = obd_mod_alloc();
727 if (mod == NULL) {
728 DEBUG_REQ(D_ERROR, open_req,
729 "Can't allocate md_open_data");
730 RETURN(0);
731 }
732
733 /**
734 * Take a reference on \var mod, to be freed on mdc_close().
735 * It protects \var mod from being freed on eviction (commit
736 * callback is called despite rq_replay flag).
737 * Another reference for \var och.
738 */
739 obd_mod_get(mod);
740 obd_mod_get(mod);
741
742 spin_lock(&open_req->rq_lock);
743 och->och_mod = mod;
744 mod->mod_och = och;
745 mod->mod_open_req = open_req;
746 open_req->rq_cb_data = mod;
747 open_req->rq_commit_cb = mdc_commit_open;
748 spin_unlock(&open_req->rq_lock);
749 }
750
751 rec->cr_fid2 = body->fid1;
752 rec->cr_ioepoch = body->ioepoch;
753 rec->cr_old_handle.cookie = body->handle.cookie;
754 open_req->rq_replay_cb = mdc_replay_open;
755 if (!fid_is_sane(&body->fid1)) {
756 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
757 "insane fid");
758 LBUG();
759 }
760
761 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
762 RETURN(0);
763}
764
765int mdc_clear_open_replay_data(struct obd_export *exp,
766 struct obd_client_handle *och)
767{
768 struct md_open_data *mod = och->och_mod;
d7e09d03
PT		 769
770 /**
771 * It is possible to not have \var mod in a case of eviction between
772 * lookup and ll_file_open().
773 **/
774 if (mod == NULL)
775 RETURN(0);
776
777 LASSERT(mod != LP_POISON);
778
779 mod->mod_och = NULL;
780 och->och_mod = NULL;
781 obd_mod_put(mod);
782
783 RETURN(0);
784}
785
786/* Prepares the request for the replay by the given reply */
787static void mdc_close_handle_reply(struct ptlrpc_request *req,
788 struct md_op_data *op_data, int rc) {
789 struct mdt_body *repbody;
790 struct mdt_ioepoch *epoch;
791
792 if (req && rc == -EAGAIN) {
793 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
794 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
795
796 epoch->flags |= MF_SOM_AU;
797 if (repbody->valid & OBD_MD_FLGETATTRLOCK)
798 op_data->op_flags |= MF_GETATTR_LOCK;
799 }
800}
801
802int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
803 struct md_open_data *mod, struct ptlrpc_request **request)
804{
805 struct obd_device *obd = class_exp2obd(exp);
806 struct ptlrpc_request *req;
807 int rc;
d7e09d03
PT		 808
809 *request = NULL;
810 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_CLOSE);
811 if (req == NULL)
812 RETURN(-ENOMEM);
813
814 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
815
816 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
817 if (rc) {
818 ptlrpc_request_free(req);
819 RETURN(rc);
820 }
821
822 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
823 * portal whose threads are not taking any DLM locks and are therefore
824 * always progressing */
825 req->rq_request_portal = MDS_READPAGE_PORTAL;
826 ptlrpc_at_set_req_timeout(req);
827
828 /* Ensure that this close's handle is fixed up during replay. */
829 if (likely(mod != NULL)) {
830 LASSERTF(mod->mod_open_req != NULL &&
831 mod->mod_open_req->rq_type != LI_POISON,
832 "POISONED open %p!\n", mod->mod_open_req);
833
834 mod->mod_close_req = req;
835
836 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
837 /* We no longer want to preserve this open for replay even
838 * though the open was committed. b=3632, b=3633 */
839 spin_lock(&mod->mod_open_req->rq_lock);
840 mod->mod_open_req->rq_replay = 0;
841 spin_unlock(&mod->mod_open_req->rq_lock);
842 } else {
843 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
844 }
845
846 mdc_close_pack(req, op_data);
847
848 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
849 obd->u.cli.cl_max_mds_easize);
850 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
851 obd->u.cli.cl_max_mds_cookiesize);
852
853 ptlrpc_request_set_replen(req);
854
855 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
856 rc = ptlrpc_queue_wait(req);
857 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
858
859 if (req->rq_repmsg == NULL) {
860 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
861 req->rq_status);
862 if (rc == 0)
863 rc = req->rq_status ?: -EIO;
864 } else if (rc == 0 || rc == -EAGAIN) {
865 struct mdt_body *body;
866
867 rc = lustre_msg_get_status(req->rq_repmsg);
868 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
869 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
870 "= %d", rc);
871 if (rc > 0)
872 rc = -rc;
873 }
874 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
875 if (body == NULL)
876 rc = -EPROTO;
877 } else if (rc == -ESTALE) {
878 /**
879 * it can be allowed error after 3633 if open was committed and
880 * server failed before close was sent. Let's check if mod
881 * exists and return no error in that case
882 */
883 if (mod) {
884 DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
885 LASSERT(mod->mod_open_req != NULL);
886 if (mod->mod_open_req->rq_committed)
887 rc = 0;
888 }
889 }
890
891 if (mod) {
892 if (rc != 0)
893 mod->mod_close_req = NULL;
894 /* Since now, mod is accessed through open_req only,
895 * thus close req does not keep a reference on mod anymore. */
896 obd_mod_put(mod);
897 }
898 *request = req;
899 mdc_close_handle_reply(req, op_data, rc);
900 RETURN(rc);
901}
902
903int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
904 struct md_open_data *mod)
905{
906 struct obd_device *obd = class_exp2obd(exp);
907 struct ptlrpc_request *req;
908 int rc;
d7e09d03
PT		 909
910 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
911 &RQF_MDS_DONE_WRITING);
912 if (req == NULL)
913 RETURN(-ENOMEM);
914
915 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
916 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
917 if (rc) {
918 ptlrpc_request_free(req);
919 RETURN(rc);
920 }
921
922 if (mod != NULL) {
923 LASSERTF(mod->mod_open_req != NULL &&
924 mod->mod_open_req->rq_type != LI_POISON,
925 "POISONED setattr %p!\n", mod->mod_open_req);
926
927 mod->mod_close_req = req;
928 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
929 /* We no longer want to preserve this setattr for replay even
930 * though the open was committed. b=3632, b=3633 */
931 spin_lock(&mod->mod_open_req->rq_lock);
932 mod->mod_open_req->rq_replay = 0;
933 spin_unlock(&mod->mod_open_req->rq_lock);
934 }
935
936 mdc_close_pack(req, op_data);
937 ptlrpc_request_set_replen(req);
938
939 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
940 rc = ptlrpc_queue_wait(req);
941 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
942
943 if (rc == -ESTALE) {
944 /**
945 * it can be allowed error after 3633 if open or setattr were
946 * committed and server failed before close was sent.
947 * Let's check if mod exists and return no error in that case
948 */
949 if (mod) {
950 LASSERT(mod->mod_open_req != NULL);
951 if (mod->mod_open_req->rq_committed)
952 rc = 0;
953 }
954 }
955
956 if (mod) {
957 if (rc != 0)
958 mod->mod_close_req = NULL;
959 /* Since now, mod is accessed through setattr req only,
960 * thus DW req does not keep a reference on mod anymore. */
961 obd_mod_put(mod);
962 }
963
964 mdc_close_handle_reply(req, op_data, rc);
965 ptlrpc_req_finished(req);
966 RETURN(rc);
967}
968
969
970int mdc_readpage(struct obd_export *exp, struct md_op_data *op_data,
971 struct page **pages, struct ptlrpc_request **request)
972{
973 struct ptlrpc_request *req;
974 struct ptlrpc_bulk_desc *desc;
975 int i;
976 wait_queue_head_t waitq;
977 int resends = 0;
978 struct l_wait_info lwi;
979 int rc;
d7e09d03
PT		 980
981 *request = NULL;
982 init_waitqueue_head(&waitq);
983
984restart_bulk:
985 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
986 if (req == NULL)
987 RETURN(-ENOMEM);
988
989 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
990
991 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
992 if (rc) {
993 ptlrpc_request_free(req);
994 RETURN(rc);
995 }
996
997 req->rq_request_portal = MDS_READPAGE_PORTAL;
998 ptlrpc_at_set_req_timeout(req);
999
1000 desc = ptlrpc_prep_bulk_imp(req, op_data->op_npages, 1, BULK_PUT_SINK,
1001 MDS_BULK_PORTAL);
1002 if (desc == NULL) {
1003 ptlrpc_request_free(req);
1004 RETURN(-ENOMEM);
1005 }
1006
1007 /* NB req now owns desc and will free it when it gets freed */
1008 for (i = 0; i < op_data->op_npages; i++)
1009 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
1010
1011 mdc_readdir_pack(req, op_data->op_offset,
1012 PAGE_CACHE_SIZE * op_data->op_npages,
1013 &op_data->op_fid1, op_data->op_capa1);
1014
1015 ptlrpc_request_set_replen(req);
1016 rc = ptlrpc_queue_wait(req);
1017 if (rc) {
1018 ptlrpc_req_finished(req);
1019 if (rc != -ETIMEDOUT)
1020 RETURN(rc);
1021
1022 resends++;
1023 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1024 CERROR("too many resend retries, returning error\n");
1025 RETURN(-EIO);
1026 }
1027 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL, NULL);
1028 l_wait_event(waitq, 0, &lwi);
1029
1030 goto restart_bulk;
1031 }
1032
1033 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1034 req->rq_bulk->bd_nob_transferred);
1035 if (rc < 0) {
1036 ptlrpc_req_finished(req);
1037 RETURN(rc);
1038 }
1039
1040 if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1041 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1042 req->rq_bulk->bd_nob_transferred,
1043 PAGE_CACHE_SIZE * op_data->op_npages);
1044 ptlrpc_req_finished(req);
1045 RETURN(-EPROTO);
1046 }
1047
1048 *request = req;
1049 RETURN(0);
1050}
1051
1052static int mdc_statfs(const struct lu_env *env,
1053 struct obd_export *exp, struct obd_statfs *osfs,
1054 __u64 max_age, __u32 flags)
1055{
1056 struct obd_device *obd = class_exp2obd(exp);
1057 struct ptlrpc_request *req;
1058 struct obd_statfs *msfs;
1059 struct obd_import *imp = NULL;
1060 int rc;
d7e09d03
PT		 1061
1062 /*
1063 * Since the request might also come from lprocfs, so we need
1064 * sync this with client_disconnect_export Bug15684
1065 */
1066 down_read(&obd->u.cli.cl_sem);
1067 if (obd->u.cli.cl_import)
1068 imp = class_import_get(obd->u.cli.cl_import);
1069 up_read(&obd->u.cli.cl_sem);
1070 if (!imp)
1071 RETURN(-ENODEV);
1072
1073 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1074 LUSTRE_MDS_VERSION, MDS_STATFS);
1075 if (req == NULL)
1076 GOTO(output, rc = -ENOMEM);
1077
1078 ptlrpc_request_set_replen(req);
1079
1080 if (flags & OBD_STATFS_NODELAY) {
1081 /* procfs requests not want stay in wait for avoid deadlock */
1082 req->rq_no_resend = 1;
1083 req->rq_no_delay = 1;
1084 }
1085
1086 rc = ptlrpc_queue_wait(req);
1087 if (rc) {
1088 /* check connection error first */
1089 if (imp->imp_connect_error)
1090 rc = imp->imp_connect_error;
1091 GOTO(out, rc);
1092 }
1093
1094 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1095 if (msfs == NULL)
1096 GOTO(out, rc = -EPROTO);
1097
1098 *osfs = *msfs;
1099 EXIT;
1100out:
1101 ptlrpc_req_finished(req);
1102output:
1103 class_import_put(imp);
1104 return rc;
1105}
1106
1107static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1108{
1109 __u32 keylen, vallen;
1110 void *key;
1111 int rc;
1112
1113 if (gf->gf_pathlen > PATH_MAX)
1114 RETURN(-ENAMETOOLONG);
1115 if (gf->gf_pathlen < 2)
1116 RETURN(-EOVERFLOW);
1117
1118 /* Key is KEY_FID2PATH + getinfo_fid2path description */
1119 keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1120 OBD_ALLOC(key, keylen);
1121 if (key == NULL)
1122 RETURN(-ENOMEM);
1123 memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1124 memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1125
1126 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1127 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1128
1129 if (!fid_is_sane(&gf->gf_fid))
1130 GOTO(out, rc = -EINVAL);
1131
1132 /* Val is struct getinfo_fid2path result plus path */
1133 vallen = sizeof(*gf) + gf->gf_pathlen;
1134
1135 rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1136 if (rc != 0 && rc != -EREMOTE)
1137 GOTO(out, rc);
1138
1139 if (vallen <= sizeof(*gf))
1140 GOTO(out, rc = -EPROTO);
1141 else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1142 GOTO(out, rc = -EOVERFLOW);
1143
1144 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1145 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1146
1147out:
1148 OBD_FREE(key, keylen);
1149 return rc;
1150}
1151
1152static int mdc_ioc_hsm_progress(struct obd_export *exp,
1153 struct hsm_progress_kernel *hpk)
1154{
1155 struct obd_import *imp = class_exp2cliimp(exp);
1156 struct hsm_progress_kernel *req_hpk;
1157 struct ptlrpc_request *req;
1158 int rc;
d7e09d03
PT		 1159
1160 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1161 LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1162 if (req == NULL)
1163 GOTO(out, rc = -ENOMEM);
1164
1165 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1166
1167 /* Copy hsm_progress struct */
1168 req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1169 if (req_hpk == NULL)
1170 GOTO(out, rc = -EPROTO);
1171
1172 *req_hpk = *hpk;
2d58de78 1173 req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
d7e09d03
PT		 1174
1175 ptlrpc_request_set_replen(req);
1176
1177 rc = mdc_queue_wait(req);
1178 GOTO(out, rc);
1179out:
1180 ptlrpc_req_finished(req);
1181 return rc;
1182}
1183
1184static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1185{
1186 __u32 *archive_mask;
1187 struct ptlrpc_request *req;
1188 int rc;
d7e09d03
PT		 1189
1190 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1191 LUSTRE_MDS_VERSION,
1192 MDS_HSM_CT_REGISTER);
1193 if (req == NULL)
1194 GOTO(out, rc = -ENOMEM);
1195
1196 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1197
1198 /* Copy hsm_progress struct */
1199 archive_mask = req_capsule_client_get(&req->rq_pill,
1200 &RMF_MDS_HSM_ARCHIVE);
1201 if (archive_mask == NULL)
1202 GOTO(out, rc = -EPROTO);
1203
1204 *archive_mask = archives;
1205
1206 ptlrpc_request_set_replen(req);
1207
1208 rc = mdc_queue_wait(req);
1209 GOTO(out, rc);
1210out:
1211 ptlrpc_req_finished(req);
1212 return rc;
1213}
1214
1215static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1216 struct md_op_data *op_data)
1217{
1218 struct hsm_current_action *hca = op_data->op_data;
1219 struct hsm_current_action *req_hca;
1220 struct ptlrpc_request *req;
1221 int rc;
d7e09d03
PT		 1222
1223 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1224 &RQF_MDS_HSM_ACTION);
1225 if (req == NULL)
1226 RETURN(-ENOMEM);
1227
1228 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1229
1230 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1231 if (rc) {
1232 ptlrpc_request_free(req);
1233 RETURN(rc);
1234 }
1235
1236 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1237 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1238
1239 ptlrpc_request_set_replen(req);
1240
1241 rc = mdc_queue_wait(req);
1242 if (rc)
1243 GOTO(out, rc);
1244
1245 req_hca = req_capsule_server_get(&req->rq_pill,
1246 &RMF_MDS_HSM_CURRENT_ACTION);
1247 if (req_hca == NULL)
1248 GOTO(out, rc = -EPROTO);
1249
1250 *hca = *req_hca;
1251
1252 EXIT;
1253out:
1254 ptlrpc_req_finished(req);
1255 return rc;
1256}
1257
1258static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1259{
1260 struct ptlrpc_request *req;
1261 int rc;
d7e09d03
PT		 1262
1263 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1264 LUSTRE_MDS_VERSION,
1265 MDS_HSM_CT_UNREGISTER);
1266 if (req == NULL)
1267 GOTO(out, rc = -ENOMEM);
1268
1269 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1270
1271 ptlrpc_request_set_replen(req);
1272
1273 rc = mdc_queue_wait(req);
1274 GOTO(out, rc);
1275out:
1276 ptlrpc_req_finished(req);
1277 return rc;
1278}
1279
1280static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1281 struct md_op_data *op_data)
1282{
1283 struct hsm_user_state *hus = op_data->op_data;
1284 struct hsm_user_state *req_hus;
1285 struct ptlrpc_request *req;
1286 int rc;
d7e09d03
PT		 1287
1288 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1289 &RQF_MDS_HSM_STATE_GET);
1290 if (req == NULL)
1291 RETURN(-ENOMEM);
1292
1293 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1294
1295 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1296 if (rc != 0) {
1297 ptlrpc_request_free(req);
1298 RETURN(rc);
1299 }
1300
1301 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1302 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1303
1304 ptlrpc_request_set_replen(req);
1305
1306 rc = mdc_queue_wait(req);
1307 if (rc)
1308 GOTO(out, rc);
1309
1310 req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1311 if (req_hus == NULL)
1312 GOTO(out, rc = -EPROTO);
1313
1314 *hus = *req_hus;
1315
1316 EXIT;
1317out:
1318 ptlrpc_req_finished(req);
1319 return rc;
1320}
1321
1322static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1323 struct md_op_data *op_data)
1324{
1325 struct hsm_state_set *hss = op_data->op_data;
1326 struct hsm_state_set *req_hss;
1327 struct ptlrpc_request *req;
1328 int rc;
d7e09d03
PT		 1329
1330 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1331 &RQF_MDS_HSM_STATE_SET);
1332 if (req == NULL)
1333 RETURN(-ENOMEM);
1334
1335 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1336
1337 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1338 if (rc) {
1339 ptlrpc_request_free(req);
1340 RETURN(rc);
1341 }
1342
1343 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1344 OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1345
1346 /* Copy states */
1347 req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1348 if (req_hss == NULL)
1349 GOTO(out, rc = -EPROTO);
1350 *req_hss = *hss;
1351
1352 ptlrpc_request_set_replen(req);
1353
1354 rc = mdc_queue_wait(req);
1355 GOTO(out, rc);
1356
1357 EXIT;
1358out:
1359 ptlrpc_req_finished(req);
1360 return rc;
1361}
1362
1363static int mdc_ioc_hsm_request(struct obd_export *exp,
1364 struct hsm_user_request *hur)
1365{
1366 struct obd_import *imp = class_exp2cliimp(exp);
1367 struct ptlrpc_request *req;
1368 struct hsm_request *req_hr;
1369 struct hsm_user_item *req_hui;
1370 char *req_opaque;
1371 int rc;
d7e09d03
PT		 1372
1373 req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1374 if (req == NULL)
1375 GOTO(out, rc = -ENOMEM);
1376
1377 req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1378 hur->hur_request.hr_itemcount
1379 * sizeof(struct hsm_user_item));
1380 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1381 hur->hur_request.hr_data_len);
1382
1383 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1384 if (rc) {
1385 ptlrpc_request_free(req);
1386 RETURN(rc);
1387 }
1388
1389 mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1390
1391 /* Copy hsm_request struct */
1392 req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1393 if (req_hr == NULL)
1394 GOTO(out, rc = -EPROTO);
1395 *req_hr = hur->hur_request;
1396
1397 /* Copy hsm_user_item structs */
1398 req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1399 if (req_hui == NULL)
1400 GOTO(out, rc = -EPROTO);
1401 memcpy(req_hui, hur->hur_user_item,
1402 hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1403
1404 /* Copy opaque field */
1405 req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1406 if (req_opaque == NULL)
1407 GOTO(out, rc = -EPROTO);
1408 memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1409
1410 ptlrpc_request_set_replen(req);
1411
1412 rc = mdc_queue_wait(req);
1413 GOTO(out, rc);
1414
1415out:
1416 ptlrpc_req_finished(req);
1417 return rc;
1418}
1419
1420static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
1421{
1422 struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1423
1424 LASSERT(len <= CR_MAXSIZE);
1425
1426 lh->kuc_magic = KUC_MAGIC;
1427 lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1428 lh->kuc_flags = flags;
1429 lh->kuc_msgtype = CL_RECORD;
1430 lh->kuc_msglen = len;
1431 return lh;
1432}
1433
1434#define D_CHANGELOG 0
1435
1436struct changelog_show {
1437 __u64 cs_startrec;
1438 __u32 cs_flags;
1439 struct file *cs_fp;
1440 char *cs_buf;
1441 struct obd_device *cs_obd;
1442};
1443
e377988e 1444static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
d7e09d03
PT		 1445 struct llog_rec_hdr *hdr, void *data)
1446{
1447 struct changelog_show *cs = data;
1448 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1449 struct kuc_hdr *lh;
1450 int len, rc;
d7e09d03 1451
e377988e
AD		 1452 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1453 rc = -EINVAL;
1454 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1455 cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
1456 rec->cr.cr_type, rc);
1457 RETURN(rc);
d7e09d03
PT		 1458 }
1459
1460 if (rec->cr.cr_index < cs->cs_startrec) {
1461 /* Skip entries earlier than what we are interested in */
1462 CDEBUG(D_CHANGELOG, "rec="LPU64" start="LPU64"\n",
1463 rec->cr.cr_index, cs->cs_startrec);
1464 RETURN(0);
1465 }
1466
1467 CDEBUG(D_CHANGELOG, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID
1468 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
1469 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1470 rec->cr.cr_flags & CLF_FLAGMASK,
1471 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1472 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1473
1474 len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1475
1476 /* Set up the message */
1477 lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1478 memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1479
1480 rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1481 CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len,rc);
1482
1483 RETURN(rc);
1484}
1485
1486static int mdc_changelog_send_thread(void *csdata)
1487{
1488 struct changelog_show *cs = csdata;
1489 struct llog_ctxt *ctxt = NULL;
1490 struct llog_handle *llh = NULL;
1491 struct kuc_hdr *kuch;
1492 int rc;
1493
1494 CDEBUG(D_CHANGELOG, "changelog to fp=%p start "LPU64"\n",
1495 cs->cs_fp, cs->cs_startrec);
1496
1497 OBD_ALLOC(cs->cs_buf, CR_MAXSIZE);
1498 if (cs->cs_buf == NULL)
1499 GOTO(out, rc = -ENOMEM);
1500
1501 /* Set up the remote catalog handle */
1502 ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1503 if (ctxt == NULL)
1504 GOTO(out, rc = -ENOENT);
1505 rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1506 LLOG_OPEN_EXISTS);
1507 if (rc) {
1508 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1509 cs->cs_obd->obd_name, rc);
1510 GOTO(out, rc);
1511 }
1512 rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
1513 if (rc) {
1514 CERROR("llog_init_handle failed %d\n", rc);
1515 GOTO(out, rc);
1516 }
1517
e377988e 1518 rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
d7e09d03
PT		 1519
1520 /* Send EOF no matter what our result */
1521 if ((kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch),
1522 cs->cs_flags))) {
1523 kuch->kuc_msgtype = CL_EOF;
1524 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1525 }
1526
1527out:
1528 fput(cs->cs_fp);
1529 if (llh)
1530 llog_cat_close(NULL, llh);
1531 if (ctxt)
1532 llog_ctxt_put(ctxt);
1533 if (cs->cs_buf)
1534 OBD_FREE(cs->cs_buf, CR_MAXSIZE);
1535 OBD_FREE_PTR(cs);
1536 return rc;
1537}
1538
1539static int mdc_ioc_changelog_send(struct obd_device *obd,
1540 struct ioc_changelog *icc)
1541{
1542 struct changelog_show *cs;
1543 int rc;
1544
1545 /* Freed in mdc_changelog_send_thread */
1546 OBD_ALLOC_PTR(cs);
1547 if (!cs)
1548 return -ENOMEM;
1549
1550 cs->cs_obd = obd;
1551 cs->cs_startrec = icc->icc_recno;
1552 /* matching fput in mdc_changelog_send_thread */
1553 cs->cs_fp = fget(icc->icc_id);
1554 cs->cs_flags = icc->icc_flags;
1555
1556 /*
1557 * New thread because we should return to user app before
1558 * writing into our pipe
1559 */
1560 rc = PTR_ERR(kthread_run(mdc_changelog_send_thread, cs,
1561 "mdc_clg_send_thread"));
1562 if (!IS_ERR_VALUE(rc)) {
1563 CDEBUG(D_CHANGELOG, "start changelog thread\n");
1564 return 0;
1565 }
1566
1567 CERROR("Failed to start changelog thread: %d\n", rc);
1568 OBD_FREE_PTR(cs);
1569 return rc;
1570}
1571
1572static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1573 struct lustre_kernelcomm *lk);
1574
1575static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
1576 struct obd_quotactl *oqctl)
1577{
1578 struct client_obd *cli = &exp->exp_obd->u.cli;
1579 struct ptlrpc_request *req;
1580 struct obd_quotactl *body;
1581 int rc;
d7e09d03
PT		 1582
1583 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1584 &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
1585 MDS_QUOTACHECK);
1586 if (req == NULL)
1587 RETURN(-ENOMEM);
1588
1589 body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1590 *body = *oqctl;
1591
1592 ptlrpc_request_set_replen(req);
1593
1594 /* the next poll will find -ENODATA, that means quotacheck is
1595 * going on */
1596 cli->cl_qchk_stat = -ENODATA;
1597 rc = ptlrpc_queue_wait(req);
1598 if (rc)
1599 cli->cl_qchk_stat = rc;
1600 ptlrpc_req_finished(req);
1601 RETURN(rc);
1602}
1603
1604static int mdc_quota_poll_check(struct obd_export *exp,
1605 struct if_quotacheck *qchk)
1606{
1607 struct client_obd *cli = &exp->exp_obd->u.cli;
1608 int rc;
d7e09d03
PT		 1609
1610 qchk->obd_uuid = cli->cl_target_uuid;
1611 memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
1612
1613 rc = cli->cl_qchk_stat;
1614 /* the client is not the previous one */
1615 if (rc == CL_NOT_QUOTACHECKED)
1616 rc = -EINTR;
1617 RETURN(rc);
1618}
1619
1620static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1621 struct obd_quotactl *oqctl)
1622{
1623 struct ptlrpc_request *req;
1624 struct obd_quotactl *oqc;
1625 int rc;
d7e09d03
PT		 1626
1627 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1628 &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1629 MDS_QUOTACTL);
1630 if (req == NULL)
1631 RETURN(-ENOMEM);
1632
1633 oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1634 *oqc = *oqctl;
1635
1636 ptlrpc_request_set_replen(req);
1637 ptlrpc_at_set_req_timeout(req);
1638 req->rq_no_resend = 1;
1639
1640 rc = ptlrpc_queue_wait(req);
1641 if (rc)
1642 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1643
1644 if (req->rq_repmsg &&
1645 (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
1646 *oqctl = *oqc;
1647 } else if (!rc) {
1648 CERROR ("Can't unpack obd_quotactl\n");
1649 rc = -EPROTO;
1650 }
1651 ptlrpc_req_finished(req);
1652
1653 RETURN(rc);
1654}
1655
1656static int mdc_ioc_swap_layouts(struct obd_export *exp,
1657 struct md_op_data *op_data)
1658{
1659 LIST_HEAD(cancels);
1660 struct ptlrpc_request *req;
1661 int rc, count;
1662 struct mdc_swap_layouts *msl, *payload;
d7e09d03
PT		 1663
1664 msl = op_data->op_data;
1665
1666 /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1667 * first thing it will do is to cancel the 2 layout
1668 * locks hold by this client.
1669 * So the client must cancel its layout locks on the 2 fids
1670 * with the request RPC to avoid extra RPC round trips
1671 */
1672 count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1673 LCK_CR, MDS_INODELOCK_LAYOUT);
1674 count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1675 LCK_CR, MDS_INODELOCK_LAYOUT);
1676
1677 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1678 &RQF_MDS_SWAP_LAYOUTS);
1679 if (req == NULL) {
1680 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1681 RETURN(-ENOMEM);
1682 }
1683
1684 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1685 mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
1686
1687 rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1688 if (rc) {
1689 ptlrpc_request_free(req);
1690 RETURN(rc);
1691 }
1692
1693 mdc_swap_layouts_pack(req, op_data);
1694
1695 payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1696 LASSERT(payload);
1697
1698 *payload = *msl;
1699
1700 ptlrpc_request_set_replen(req);
1701
1702 rc = ptlrpc_queue_wait(req);
1703 if (rc)
1704 GOTO(out, rc);
1705 EXIT;
1706
1707out:
1708 ptlrpc_req_finished(req);
1709 return rc;
1710}
1711
1712static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1713 void *karg, void *uarg)
1714{
1715 struct obd_device *obd = exp->exp_obd;
1716 struct obd_ioctl_data *data = karg;
1717 struct obd_import *imp = obd->u.cli.cl_import;
1718 struct llog_ctxt *ctxt;
1719 int rc;
d7e09d03
PT		 1720
1721 if (!try_module_get(THIS_MODULE)) {
1722 CERROR("Can't get module. Is it alive?");
1723 return -EINVAL;
1724 }
1725 switch (cmd) {
1726 case OBD_IOC_CHANGELOG_SEND:
1727 rc = mdc_ioc_changelog_send(obd, karg);
1728 GOTO(out, rc);
1729 case OBD_IOC_CHANGELOG_CLEAR: {
1730 struct ioc_changelog *icc = karg;
1731 struct changelog_setinfo cs =
1732 {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
1733 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
1734 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1735 NULL);
1736 GOTO(out, rc);
1737 }
1738 case OBD_IOC_FID2PATH:
1739 rc = mdc_ioc_fid2path(exp, karg);
1740 GOTO(out, rc);
1741 case LL_IOC_HSM_CT_START:
1742 rc = mdc_ioc_hsm_ct_start(exp, karg);
78eb9092
TL		 1743 /* ignore if it was already registered on this MDS. */
1744 if (rc == -EEXIST)
1745 rc = 0;
d7e09d03
PT		 1746 GOTO(out, rc);
1747 case LL_IOC_HSM_PROGRESS:
1748 rc = mdc_ioc_hsm_progress(exp, karg);
1749 GOTO(out, rc);
1750 case LL_IOC_HSM_STATE_GET:
1751 rc = mdc_ioc_hsm_state_get(exp, karg);
1752 GOTO(out, rc);
1753 case LL_IOC_HSM_STATE_SET:
1754 rc = mdc_ioc_hsm_state_set(exp, karg);
1755 case LL_IOC_HSM_ACTION:
1756 rc = mdc_ioc_hsm_current_action(exp, karg);
1757 GOTO(out, rc);
1758 case LL_IOC_HSM_REQUEST:
1759 rc = mdc_ioc_hsm_request(exp, karg);
1760 GOTO(out, rc);
1761 case OBD_IOC_CLIENT_RECOVER:
1762 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1763 if (rc < 0)
1764 GOTO(out, rc);
1765 GOTO(out, rc = 0);
1766 case IOC_OSC_SET_ACTIVE:
1767 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1768 GOTO(out, rc);
1769 case OBD_IOC_PARSE: {
1770 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1771 rc = class_config_parse_llog(NULL, ctxt, data->ioc_inlbuf1,
1772 NULL);
1773 llog_ctxt_put(ctxt);
1774 GOTO(out, rc);
1775 }
1776 case OBD_IOC_LLOG_INFO:
1777 case OBD_IOC_LLOG_PRINT: {
1778 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1779 rc = llog_ioctl(NULL, ctxt, cmd, data);
1780 llog_ctxt_put(ctxt);
1781 GOTO(out, rc);
1782 }
1783 case OBD_IOC_POLL_QUOTACHECK:
1784 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
1785 GOTO(out, rc);
1786 case OBD_IOC_PING_TARGET:
1787 rc = ptlrpc_obd_ping(obd);
1788 GOTO(out, rc);
1789 /*
1790 * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1791 * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1792 * there'd be no LMV layer thus we might be called here. Eventually
1793 * this code should be removed.
1794 * bz20731, LU-592.
1795 */
1796 case IOC_OBD_STATFS: {
1797 struct obd_statfs stat_buf = {0};
1798
1799 if (*((__u32 *) data->ioc_inlbuf2) != 0)
1800 GOTO(out, rc = -ENODEV);
1801
1802 /* copy UUID */
1803 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1804 min((int) data->ioc_plen2,
1805 (int) sizeof(struct obd_uuid))))
1806 GOTO(out, rc = -EFAULT);
1807
1808 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1809 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1810 0);
1811 if (rc != 0)
1812 GOTO(out, rc);
1813
1814 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1815 min((int) data->ioc_plen1,
1816 (int) sizeof(stat_buf))))
1817 GOTO(out, rc = -EFAULT);
1818
1819 GOTO(out, rc = 0);
1820 }
1821 case OBD_IOC_QUOTACTL: {
1822 struct if_quotactl *qctl = karg;
1823 struct obd_quotactl *oqctl;
1824
1825 OBD_ALLOC_PTR(oqctl);
1826 if (!oqctl)
1827 RETURN(-ENOMEM);
1828
1829 QCTL_COPY(oqctl, qctl);
1830 rc = obd_quotactl(exp, oqctl);
1831 if (rc == 0) {
1832 QCTL_COPY(qctl, oqctl);
1833 qctl->qc_valid = QC_MDTIDX;
1834 qctl->obd_uuid = obd->u.cli.cl_target_uuid;
1835 }
1836 OBD_FREE_PTR(oqctl);
1837 break;
1838 }
1839 case LL_IOC_GET_CONNECT_FLAGS: {
1840 if (copy_to_user(uarg,
1841 exp_connect_flags_ptr(exp),
1842 sizeof(__u64)))
1843 GOTO(out, rc = -EFAULT);
1844 else
1845 GOTO(out, rc = 0);
1846 }
1847 case LL_IOC_LOV_SWAP_LAYOUTS: {
1848 rc = mdc_ioc_swap_layouts(exp, karg);
1849 break;
1850 }
1851 default:
1852 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1853 GOTO(out, rc = -ENOTTY);
1854 }
1855out:
1856 module_put(THIS_MODULE);
1857
1858 return rc;
1859}
1860
1861int mdc_get_info_rpc(struct obd_export *exp,
1862 obd_count keylen, void *key,
1863 int vallen, void *val)
1864{
1865 struct obd_import *imp = class_exp2cliimp(exp);
1866 struct ptlrpc_request *req;
1867 char *tmp;
1868 int rc = -EINVAL;
d7e09d03
PT		 1869
1870 req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1871 if (req == NULL)
1872 RETURN(-ENOMEM);
1873
1874 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1875 RCL_CLIENT, keylen);
1876 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1877 RCL_CLIENT, sizeof(__u32));
1878
1879 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1880 if (rc) {
1881 ptlrpc_request_free(req);
1882 RETURN(rc);
1883 }
1884
1885 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
1886 memcpy(tmp, key, keylen);
1887 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
1888 memcpy(tmp, &vallen, sizeof(__u32));
1889
1890 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1891 RCL_SERVER, vallen);
1892 ptlrpc_request_set_replen(req);
1893
1894 rc = ptlrpc_queue_wait(req);
1895 /* -EREMOTE means the get_info result is partial, and it needs to
1896 * continue on another MDT, see fid2path part in lmv_iocontrol */
1897 if (rc == 0 || rc == -EREMOTE) {
1898 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
1899 memcpy(val, tmp, vallen);
1900 if (ptlrpc_rep_need_swab(req)) {
1901 if (KEY_IS(KEY_FID2PATH))
1902 lustre_swab_fid2path(val);
1903 }
1904 }
1905 ptlrpc_req_finished(req);
1906
1907 RETURN(rc);
1908}
1909
1910static void lustre_swab_hai(struct hsm_action_item *h)
1911{
1912 __swab32s(&h->hai_len);
1913 __swab32s(&h->hai_action);
1914 lustre_swab_lu_fid(&h->hai_fid);
1915 lustre_swab_lu_fid(&h->hai_dfid);
1916 __swab64s(&h->hai_cookie);
1917 __swab64s(&h->hai_extent.offset);
1918 __swab64s(&h->hai_extent.length);
1919 __swab64s(&h->hai_gid);
1920}
1921
1922static void lustre_swab_hal(struct hsm_action_list *h)
1923{
1924 struct hsm_action_item *hai;
1925 int i;
1926
1927 __swab32s(&h->hal_version);
1928 __swab32s(&h->hal_count);
1929 __swab32s(&h->hal_archive_id);
1930 __swab64s(&h->hal_flags);
1931 hai = hai_zero(h);
1932 for (i = 0; i < h->hal_count; i++) {
1933 lustre_swab_hai(hai);
1934 hai = hai_next(hai);
1935 }
1936}
1937
1938static void lustre_swab_kuch(struct kuc_hdr *l)
1939{
1940 __swab16s(&l->kuc_magic);
1941 /* __u8 l->kuc_transport */
1942 __swab16s(&l->kuc_msgtype);
1943 __swab16s(&l->kuc_msglen);
1944}
1945
1946static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1947 struct lustre_kernelcomm *lk)
1948{
1949 struct obd_import *imp = class_exp2cliimp(exp);
1950 __u32 archive = lk->lk_data;
1951 int rc = 0;
1952
1953 if (lk->lk_group != KUC_GRP_HSM) {
1954 CERROR("Bad copytool group %d\n", lk->lk_group);
1955 return -EINVAL;
1956 }
1957
1958 CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
1959 lk->lk_uid, lk->lk_group, lk->lk_flags);
1960
1961 if (lk->lk_flags & LK_FLG_STOP) {
d7e09d03 1962 /* Unregister with the coordinator */
78eb9092 1963 rc = mdc_ioc_hsm_ct_unregister(imp);
d7e09d03 1964 } else {
78eb9092 1965 rc = mdc_ioc_hsm_ct_register(imp, archive);
d7e09d03
PT		 1966 }
1967
1968 return rc;
1969}
1970
1971/**
1972 * Send a message to any listening copytools
1973 * @param val KUC message (kuc_hdr + hsm_action_list)
1974 * @param len total length of message
1975 */
1976static int mdc_hsm_copytool_send(int len, void *val)
1977{
1978 struct kuc_hdr *lh = (struct kuc_hdr *)val;
1979 struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
1980 int rc;
d7e09d03
PT		 1981
1982 if (len < sizeof(*lh) + sizeof(*hal)) {
1983 CERROR("Short HSM message %d < %d\n", len,
1984 (int) (sizeof(*lh) + sizeof(*hal)));
1985 RETURN(-EPROTO);
1986 }
1987 if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
1988 lustre_swab_kuch(lh);
1989 lustre_swab_hal(hal);
1990 } else if (lh->kuc_magic != KUC_MAGIC) {
1991 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
1992 RETURN(-EPROTO);
1993 }
1994
1995 CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
1996 "on %s\n",
1997 lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
1998 lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
1999
2000 /* Broadcast to HSM listeners */
2001 rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2002
2003 RETURN(rc);
2004}
2005
2006/**
2007 * callback function passed to kuc for re-registering each HSM copytool
2008 * running on MDC, after MDT shutdown/recovery.
2009 * @param data archive id served by the copytool
2010 * @param cb_arg callback argument (obd_import)
2011 */
2012static int mdc_hsm_ct_reregister(__u32 data, void *cb_arg)
2013{
2014 struct obd_import *imp = (struct obd_import *)cb_arg;
2015 __u32 archive = data;
2016 int rc;
2017
2018 CDEBUG(D_HA, "recover copytool registration to MDT (archive=%#x)\n",
2019 archive);
2020 rc = mdc_ioc_hsm_ct_register(imp, archive);
2021
2022 /* ignore error if the copytool is already registered */
2023 return ((rc != 0) && (rc != -EEXIST)) ? rc : 0;
2024}
2025
2026/**
2027 * Re-establish all kuc contexts with MDT
2028 * after MDT shutdown/recovery.
2029 */
2030static int mdc_kuc_reregister(struct obd_import *imp)
2031{
2032 /* re-register HSM agents */
2033 return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2034 (void *)imp);
2035}
2036
2037int mdc_set_info_async(const struct lu_env *env,
2038 struct obd_export *exp,
2039 obd_count keylen, void *key,
2040 obd_count vallen, void *val,
2041 struct ptlrpc_request_set *set)
2042{
2043 struct obd_import *imp = class_exp2cliimp(exp);
2044 int rc;
d7e09d03
PT		 2045
2046 if (KEY_IS(KEY_READ_ONLY)) {
2047 if (vallen != sizeof(int))
2048 RETURN(-EINVAL);
2049
2050 spin_lock(&imp->imp_lock);
2051 if (*((int *)val)) {
2052 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2053 imp->imp_connect_data.ocd_connect_flags |=
2054 OBD_CONNECT_RDONLY;
2055 } else {
2056 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2057 imp->imp_connect_data.ocd_connect_flags &=
2058 ~OBD_CONNECT_RDONLY;
2059 }
2060 spin_unlock(&imp->imp_lock);
2061
2062 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2063 keylen, key, vallen, val, set);
2064 RETURN(rc);
2065 }
2066 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2067 sptlrpc_conf_client_adapt(exp->exp_obd);
2068 RETURN(0);
2069 }
2070 if (KEY_IS(KEY_FLUSH_CTX)) {
2071 sptlrpc_import_flush_my_ctx(imp);
2072 RETURN(0);
2073 }
2074 if (KEY_IS(KEY_MDS_CONN)) {
2075 /* mds-mds import */
2076 spin_lock(&imp->imp_lock);
2077 imp->imp_server_timeout = 1;
2078 spin_unlock(&imp->imp_lock);
2079 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
2080 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
2081 RETURN(0);
2082 }
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);
2086 RETURN(rc);
2087 }
2088 if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2089 rc = mdc_hsm_copytool_send(vallen, val);
2090 RETURN(rc);
2091 }
2092
2093 CERROR("Unknown key %s\n", (char *)key);
2094 RETURN(-EINVAL);
2095}
2096
2097int 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)
2100{
2101 int rc = -EINVAL;
2102
2103 if (KEY_IS(KEY_MAX_EASIZE)) {
2104 int mdsize, *max_easize;
2105
2106 if (*vallen != sizeof(int))
2107 RETURN(-EINVAL);
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;
2111 max_easize = val;
2112 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2113 RETURN(0);
2114 } else if (KEY_IS(KEY_CONN_DATA)) {
2115 struct obd_import *imp = class_exp2cliimp(exp);
2116 struct obd_connect_data *data = val;
2117
2118 if (*vallen != sizeof(*data))
2119 RETURN(-EINVAL);
2120
2121 *data = imp->imp_connect_data;
2122 RETURN(0);
2123 } else if (KEY_IS(KEY_TGT_COUNT)) {
2124 *((int *)val) = 1;
2125 RETURN(0);
2126 }
2127
2128 rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2129
2130 RETURN(rc);
2131}
2132
2133static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
2134 struct obd_capa *oc, struct obd_client_handle *handle,
2135 int flags)
2136{
2137 struct ptlrpc_request *req;
2138 struct mdt_body *body;
2139 int rc;
d7e09d03
PT		 2140
2141 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
2142 if (req == NULL)
2143 RETURN(-ENOMEM);
2144
2145 mdc_set_capa_size(req, &RMF_CAPA1, oc);
2146
2147 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
2148 if (rc) {
2149 ptlrpc_request_free(req);
2150 RETURN(rc);
2151 }
2152
2153 mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
2154
2155 ptlrpc_request_set_replen(req);
2156
2157 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2158 rc = ptlrpc_queue_wait(req);
2159 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2160 if (rc) {
2161 CERROR("Pin failed: %d\n", rc);
2162 GOTO(err_out, rc);
2163 }
2164
2165 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2166 if (body == NULL)
2167 GOTO(err_out, rc = -EPROTO);
2168
2169 handle->och_fh = body->handle;
2170 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
2171
2172 handle->och_mod = obd_mod_alloc();
2173 if (handle->och_mod == NULL) {
2174 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
2175 GOTO(err_out, rc = -ENOMEM);
2176 }
2177 handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
2178
2179 RETURN(0);
2180
2181err_out:
2182 ptlrpc_req_finished(req);
2183 RETURN(rc);
2184}
2185
2186static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
2187 int flag)
2188{
2189 struct ptlrpc_request *req;
2190 struct mdt_body *body;
2191 int rc;
d7e09d03
PT		 2192
2193 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
2194 LUSTRE_MDS_VERSION, MDS_UNPIN);
2195 if (req == NULL)
2196 RETURN(-ENOMEM);
2197
2198 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
2199 body->handle = handle->och_fh;
2200 body->flags = flag;
2201
2202 ptlrpc_request_set_replen(req);
2203
2204 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2205 rc = ptlrpc_queue_wait(req);
2206 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2207
2208 if (rc != 0)
2209 CERROR("Unpin failed: %d\n", rc);
2210
2211 ptlrpc_req_finished(req);
2212 ptlrpc_req_finished(handle->och_mod->mod_open_req);
2213
2214 obd_mod_put(handle->och_mod);
2215 RETURN(rc);
2216}
2217
2218int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
2219 struct obd_capa *oc, struct ptlrpc_request **request)
2220{
2221 struct ptlrpc_request *req;
2222 int rc;
d7e09d03
PT		 2223
2224 *request = NULL;
2225 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2226 if (req == NULL)
2227 RETURN(-ENOMEM);
2228
2229 mdc_set_capa_size(req, &RMF_CAPA1, oc);
2230
2231 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2232 if (rc) {
2233 ptlrpc_request_free(req);
2234 RETURN(rc);
2235 }
2236
2237 mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
2238
2239 ptlrpc_request_set_replen(req);
2240
2241 rc = ptlrpc_queue_wait(req);
2242 if (rc)
2243 ptlrpc_req_finished(req);
2244 else
2245 *request = req;
2246 RETURN(rc);
2247}
2248
2249static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2250 enum obd_import_event event)
2251{
2252 int rc = 0;
2253
2254 LASSERT(imp->imp_obd == obd);
2255
2256 switch (event) {
2257 case IMP_EVENT_DISCON: {
2258#if 0
2259 /* XXX Pass event up to OBDs stack. used only for FLD now */
2260 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2261#endif
2262 break;
2263 }
2264 case IMP_EVENT_INACTIVE: {
2265 struct client_obd *cli = &obd->u.cli;
2266 /*
2267 * Flush current sequence to make client obtain new one
2268 * from server in case of disconnect/reconnect.
2269 */
2270 if (cli->cl_seq != NULL)
2271 seq_client_flush(cli->cl_seq);
2272
2273 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2274 break;
2275 }
2276 case IMP_EVENT_INVALIDATE: {
2277 struct ldlm_namespace *ns = obd->obd_namespace;
2278
2279 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2280
2281 break;
2282 }
2283 case IMP_EVENT_ACTIVE:
2284 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
78eb9092 2285 /* redo the kuc registration after reconnecting */
d7e09d03
PT		 2286 if (rc == 0)
2287 rc = mdc_kuc_reregister(imp);
2288 break;
2289 case IMP_EVENT_OCD:
2290 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2291 break;
2292 case IMP_EVENT_DEACTIVATE:
2293 case IMP_EVENT_ACTIVATE:
2294 break;
2295 default:
2296 CERROR("Unknown import event %x\n", event);
2297 LBUG();
2298 }
2299 RETURN(rc);
2300}
2301
2302int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
2303 struct md_op_data *op_data)
2304{
2305 struct client_obd *cli = &exp->exp_obd->u.cli;
2306 struct lu_client_seq *seq = cli->cl_seq;
29aaf496 2307
d7e09d03
PT		 2308 RETURN(seq_client_alloc_fid(NULL, seq, fid));
2309}
2310
2311struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
2312 struct client_obd *cli = &exp->exp_obd->u.cli;
2313 return &cli->cl_target_uuid;
2314}
2315
2316/**
2317 * Determine whether the lock can be canceled before replaying it during
2318 * recovery, non zero value will be return if the lock can be canceled,
2319 * or zero returned for not
2320 */
2321static int mdc_cancel_for_recovery(struct ldlm_lock *lock)
2322{
2323 if (lock->l_resource->lr_type != LDLM_IBITS)
2324 RETURN(0);
2325
2326 /* FIXME: if we ever get into a situation where there are too many
2327 * opened files with open locks on a single node, then we really
2328 * should replay these open locks to reget it */
2329 if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2330 RETURN(0);
2331
2332 RETURN(1);
2333}
2334
2335static int mdc_resource_inode_free(struct ldlm_resource *res)
2336{
2337 if (res->lr_lvb_inode)
2338 res->lr_lvb_inode = NULL;
2339
2340 return 0;
2341}
2342
2343struct ldlm_valblock_ops inode_lvbo = {
805e517a 2344 .lvbo_free = mdc_resource_inode_free,
d7e09d03
PT		 2345};
2346
2347static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2348{
2349 struct client_obd *cli = &obd->u.cli;
2350 struct lprocfs_static_vars lvars = { 0 };
2351 int rc;
d7e09d03
PT		 2352
2353 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2354 if (!cli->cl_rpc_lock)
2355 RETURN(-ENOMEM);
2356 mdc_init_rpc_lock(cli->cl_rpc_lock);
2357
2358 ptlrpcd_addref();
2359
2360 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2361 if (!cli->cl_close_lock)
2362 GOTO(err_rpc_lock, rc = -ENOMEM);
2363 mdc_init_rpc_lock(cli->cl_close_lock);
2364
2365 rc = client_obd_setup(obd, cfg);
2366 if (rc)
2367 GOTO(err_close_lock, rc);
2368 lprocfs_mdc_init_vars(&lvars);
2369 lprocfs_obd_setup(obd, lvars.obd_vars);
2370 sptlrpc_lprocfs_cliobd_attach(obd);
2371 ptlrpc_lprocfs_register_obd(obd);
2372
2373 ns_register_cancel(obd->obd_namespace, mdc_cancel_for_recovery);
2374
2375 obd->obd_namespace->ns_lvbo = &inode_lvbo;
2376
2377 rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
2378 if (rc) {
2379 mdc_cleanup(obd);
2380 CERROR("failed to setup llogging subsystems\n");
2381 }
2382
2383 RETURN(rc);
2384
2385err_close_lock:
2386 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2387err_rpc_lock:
2388 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2389 ptlrpcd_decref();
2390 RETURN(rc);
2391}
2392
2393/* Initialize the default and maximum LOV EA and cookie sizes. This allows
2394 * us to make MDS RPCs with large enough reply buffers to hold the
2395 * maximum-sized (= maximum striped) EA and cookie without having to
2396 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
2397static int mdc_init_ea_size(struct obd_export *exp, int easize,
2398 int def_easize, int cookiesize)
2399{
2400 struct obd_device *obd = exp->exp_obd;
2401 struct client_obd *cli = &obd->u.cli;
d7e09d03
PT		 2402
2403 if (cli->cl_max_mds_easize < easize)
2404 cli->cl_max_mds_easize = easize;
2405
2406 if (cli->cl_default_mds_easize < def_easize)
2407 cli->cl_default_mds_easize = def_easize;
2408
2409 if (cli->cl_max_mds_cookiesize < cookiesize)
2410 cli->cl_max_mds_cookiesize = cookiesize;
2411
2412 RETURN(0);
2413}
2414
2415static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2416{
2417 int rc = 0;
d7e09d03
PT		 2418
2419 switch (stage) {
2420 case OBD_CLEANUP_EARLY:
2421 break;
2422 case OBD_CLEANUP_EXPORTS:
2423 /* Failsafe, ok if racy */
2424 if (obd->obd_type->typ_refcnt <= 1)
2425 libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2426
2427 obd_cleanup_client_import(obd);
2428 ptlrpc_lprocfs_unregister_obd(obd);
2429 lprocfs_obd_cleanup(obd);
2430
2431 rc = obd_llog_finish(obd, 0);
2432 if (rc != 0)
2433 CERROR("failed to cleanup llogging subsystems\n");
2434 break;
2435 }
2436 RETURN(rc);
2437}
2438
2439static int mdc_cleanup(struct obd_device *obd)
2440{
2441 struct client_obd *cli = &obd->u.cli;
2442
2443 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2444 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2445
2446 ptlrpcd_decref();
2447
2448 return client_obd_cleanup(obd);
2449}
2450
2451
2452static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
2453 struct obd_device *tgt, int *index)
2454{
2455 struct llog_ctxt *ctxt;
2456 int rc;
2457
d7e09d03
PT		 2458 LASSERT(olg == &obd->obd_olg);
2459
2460 rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt,
2461 &llog_client_ops);
2462 if (rc)
2463 RETURN(rc);
2464
2465 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2466 llog_initiator_connect(ctxt);
2467 llog_ctxt_put(ctxt);
2468
2469 RETURN(0);
2470}
2471
2472static int mdc_llog_finish(struct obd_device *obd, int count)
2473{
2474 struct llog_ctxt *ctxt;
2475
d7e09d03
PT		 2476 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2477 if (ctxt)
2478 llog_cleanup(NULL, ctxt);
2479
2480 RETURN(0);
2481}
2482
2483static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
2484{
2485 struct lustre_cfg *lcfg = buf;
2486 struct lprocfs_static_vars lvars = { 0 };
2487 int rc = 0;
2488
2489 lprocfs_mdc_init_vars(&lvars);
2490 switch (lcfg->lcfg_command) {
2491 default:
2492 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
2493 lcfg, obd);
2494 if (rc > 0)
2495 rc = 0;
2496 break;
2497 }
2498 return(rc);
2499}
2500
2501
2502/* get remote permission for current user on fid */
2503int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2504 struct obd_capa *oc, __u32 suppgid,
2505 struct ptlrpc_request **request)
2506{
2507 struct ptlrpc_request *req;
2508 int rc;
d7e09d03
PT		 2509
2510 LASSERT(client_is_remote(exp));
2511
2512 *request = NULL;
2513 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2514 if (req == NULL)
2515 RETURN(-ENOMEM);
2516
2517 mdc_set_capa_size(req, &RMF_CAPA1, oc);
2518
2519 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2520 if (rc) {
2521 ptlrpc_request_free(req);
2522 RETURN(rc);
2523 }
2524
2525 mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2526
2527 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2528 sizeof(struct mdt_remote_perm));
2529
2530 ptlrpc_request_set_replen(req);
2531
2532 rc = ptlrpc_queue_wait(req);
2533 if (rc)
2534 ptlrpc_req_finished(req);
2535 else
2536 *request = req;
2537 RETURN(rc);
2538}
2539
2540static int mdc_interpret_renew_capa(const struct lu_env *env,
2541 struct ptlrpc_request *req, void *args,
2542 int status)
2543{
2544 struct mdc_renew_capa_args *ra = args;
2545 struct mdt_body *body = NULL;
2546 struct lustre_capa *capa;
d7e09d03
PT		 2547
2548 if (status)
2549 GOTO(out, capa = ERR_PTR(status));
2550
2551 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2552 if (body == NULL)
2553 GOTO(out, capa = ERR_PTR(-EFAULT));
2554
2555 if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
2556 GOTO(out, capa = ERR_PTR(-ENOENT));
2557
2558 capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
2559 if (!capa)
2560 GOTO(out, capa = ERR_PTR(-EFAULT));
2561 EXIT;
2562out:
2563 ra->ra_cb(ra->ra_oc, capa);
2564 return 0;
2565}
2566
2567static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2568 renew_capa_cb_t cb)
2569{
2570 struct ptlrpc_request *req;
2571 struct mdc_renew_capa_args *ra;
d7e09d03
PT		 2572
2573 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
2574 LUSTRE_MDS_VERSION, MDS_GETATTR);
2575 if (req == NULL)
2576 RETURN(-ENOMEM);
2577
2578 /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
2579 * capa to renew is oss capa.
2580 */
2581 mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
2582 ptlrpc_request_set_replen(req);
2583
2584 CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
2585 ra = ptlrpc_req_async_args(req);
2586 ra->ra_oc = oc;
2587 ra->ra_cb = cb;
2588 req->rq_interpret_reply = mdc_interpret_renew_capa;
2589 ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
2590 RETURN(0);
2591}
2592
2593static int mdc_connect(const struct lu_env *env,
2594 struct obd_export **exp,
2595 struct obd_device *obd, struct obd_uuid *cluuid,
2596 struct obd_connect_data *data,
2597 void *localdata)
2598{
2599 struct obd_import *imp = obd->u.cli.cl_import;
2600
2601 /* mds-mds import features */
2602 if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
2603 spin_lock(&imp->imp_lock);
2604 imp->imp_server_timeout = 1;
2605 spin_unlock(&imp->imp_lock);
2606 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
2607 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
2608 obd->obd_name);
2609 }
2610
2611 return client_connect_import(env, exp, obd, cluuid, data, NULL);
2612}
2613
2614struct obd_ops mdc_obd_ops = {
2615 .o_owner = THIS_MODULE,
2616 .o_setup = mdc_setup,
2617 .o_precleanup = mdc_precleanup,
2618 .o_cleanup = mdc_cleanup,
2619 .o_add_conn = client_import_add_conn,
2620 .o_del_conn = client_import_del_conn,
2621 .o_connect = mdc_connect,
2622 .o_disconnect = client_disconnect_export,
2623 .o_iocontrol = mdc_iocontrol,
2624 .o_set_info_async = mdc_set_info_async,
2625 .o_statfs = mdc_statfs,
2626 .o_pin = mdc_pin,
2627 .o_unpin = mdc_unpin,
2628 .o_fid_init = client_fid_init,
2629 .o_fid_fini = client_fid_fini,
2630 .o_fid_alloc = mdc_fid_alloc,
2631 .o_import_event = mdc_import_event,
2632 .o_llog_init = mdc_llog_init,
2633 .o_llog_finish = mdc_llog_finish,
2634 .o_get_info = mdc_get_info,
2635 .o_process_config = mdc_process_config,
2636 .o_get_uuid = mdc_get_uuid,
2637 .o_quotactl = mdc_quotactl,
2638 .o_quotacheck = mdc_quotacheck
2639};
2640
2641struct md_ops mdc_md_ops = {
2642 .m_getstatus = mdc_getstatus,
2643 .m_null_inode = mdc_null_inode,
2644 .m_find_cbdata = mdc_find_cbdata,
2645 .m_close = mdc_close,
2646 .m_create = mdc_create,
2647 .m_done_writing = mdc_done_writing,
2648 .m_enqueue = mdc_enqueue,
2649 .m_getattr = mdc_getattr,
2650 .m_getattr_name = mdc_getattr_name,
2651 .m_intent_lock = mdc_intent_lock,
2652 .m_link = mdc_link,
2653 .m_is_subdir = mdc_is_subdir,
2654 .m_rename = mdc_rename,
2655 .m_setattr = mdc_setattr,
2656 .m_setxattr = mdc_setxattr,
2657 .m_getxattr = mdc_getxattr,
2658 .m_sync = mdc_sync,
2659 .m_readpage = mdc_readpage,
2660 .m_unlink = mdc_unlink,
2661 .m_cancel_unused = mdc_cancel_unused,
2662 .m_init_ea_size = mdc_init_ea_size,
2663 .m_set_lock_data = mdc_set_lock_data,
2664 .m_lock_match = mdc_lock_match,
2665 .m_get_lustre_md = mdc_get_lustre_md,
2666 .m_free_lustre_md = mdc_free_lustre_md,
2667 .m_set_open_replay_data = mdc_set_open_replay_data,
2668 .m_clear_open_replay_data = mdc_clear_open_replay_data,
2669 .m_renew_capa = mdc_renew_capa,
2670 .m_unpack_capa = mdc_unpack_capa,
2671 .m_get_remote_perm = mdc_get_remote_perm,
2672 .m_intent_getattr_async = mdc_intent_getattr_async,
2673 .m_revalidate_lock = mdc_revalidate_lock
2674};
2675
2676int __init mdc_init(void)
2677{
2678 int rc;
2679 struct lprocfs_static_vars lvars = { 0 };
2680 lprocfs_mdc_init_vars(&lvars);
2681
2682 rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
2683 LUSTRE_MDC_NAME, NULL);
2684 RETURN(rc);
2685}
2686
2687static void /*__exit*/ mdc_exit(void)
2688{
2689 class_unregister_type(LUSTRE_MDC_NAME);
2690}
2691
2692MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2693MODULE_DESCRIPTION("Lustre Metadata Client");
2694MODULE_LICENSE("GPL");
2695
2696module_init(mdc_init);
2697module_exit(mdc_exit);
Linux 6.x block layer and io_uring tree(s)