projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
staging/lustre: Replace sun.com GPLv2 URL with gnu.org one.
[deliverable/linux.git] / drivers / staging / lustre / lustre / mdc / mdc_request.c
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.gnu.org/licenses/gpl-2.0.html
19 *
20 * GPL HEADER END
21 */
22 /*
23 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
25 *
26 * Copyright (c) 2011, 2015, Intel Corporation.
27 */
28 /*
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
31 */
32
33 #define DEBUG_SUBSYSTEM S_MDC
34
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>
40
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"
48
49 #include "mdc_internal.h"
50
51 #define REQUEST_MINOR 244
52
53 static int mdc_cleanup(struct obd_device *obd);
54
55 static inline int mdc_queue_wait(struct ptlrpc_request *req)
56 {
57 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
58 int rc;
59
60 /* mdc_enter_request() ensures that this client has no more
61 * than cl_max_rpcs_in_flight RPCs simultaneously inf light
62 * against an MDT.
63 */
64 rc = mdc_enter_request(cli);
65 if (rc != 0)
66 return rc;
67
68 rc = ptlrpc_queue_wait(req);
69 mdc_exit_request(cli);
70
71 return rc;
72 }
73
74 static int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid)
75 {
76 struct ptlrpc_request *req;
77 struct mdt_body *body;
78 int rc;
79
80 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
81 &RQF_MDS_GETSTATUS,
82 LUSTRE_MDS_VERSION, MDS_GETSTATUS);
83 if (!req)
84 return -ENOMEM;
85
86 mdc_pack_body(req, NULL, 0, 0, -1, 0);
87 req->rq_send_state = LUSTRE_IMP_FULL;
88
89 ptlrpc_request_set_replen(req);
90
91 rc = ptlrpc_queue_wait(req);
92 if (rc)
93 goto out;
94
95 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
96 if (!body) {
97 rc = -EPROTO;
98 goto out;
99 }
100
101 *rootfid = body->fid1;
102 CDEBUG(D_NET,
103 "root fid="DFID", last_committed=%llu\n",
104 PFID(rootfid),
105 lustre_msg_get_last_committed(req->rq_repmsg));
106 out:
107 ptlrpc_req_finished(req);
108 return rc;
109 }
110
111 /*
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
119 * layouts. --umka
120 */
121 static int mdc_getattr_common(struct obd_export *exp,
122 struct ptlrpc_request *req)
123 {
124 struct req_capsule *pill = &req->rq_pill;
125 struct mdt_body *body;
126 void *eadata;
127 int rc;
128
129 /* Request message already built. */
130 rc = ptlrpc_queue_wait(req);
131 if (rc != 0)
132 return rc;
133
134 /* sanity check for the reply */
135 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
136 if (!body)
137 return -EPROTO;
138
139 CDEBUG(D_NET, "mode: %o\n", body->mode);
140
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,
144 body->eadatasize);
145 if (!eadata)
146 return -EPROTO;
147 }
148
149 if (body->valid & OBD_MD_FLRMTPERM) {
150 struct mdt_remote_perm *perm;
151
152 LASSERT(client_is_remote(exp));
153 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
154 lustre_swab_mdt_remote_perm);
155 if (!perm)
156 return -EPROTO;
157 }
158
159 return 0;
160 }
161
162 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
163 struct ptlrpc_request **request)
164 {
165 struct ptlrpc_request *req;
166 int rc;
167
168 /* Single MDS without an LMV case */
169 if (op_data->op_flags & MF_GET_MDT_IDX) {
170 op_data->op_mds = 0;
171 return 0;
172 }
173 *request = NULL;
174 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
175 if (!req)
176 return -ENOMEM;
177
178 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
179 if (rc) {
180 ptlrpc_request_free(req);
181 return rc;
182 }
183
184 mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
185 op_data->op_mode, -1, 0);
186
187 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
188 op_data->op_mode);
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));
193 }
194 ptlrpc_request_set_replen(req);
195
196 rc = mdc_getattr_common(exp, req);
197 if (rc)
198 ptlrpc_req_finished(req);
199 else
200 *request = req;
201 return rc;
202 }
203
204 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
205 struct ptlrpc_request **request)
206 {
207 struct ptlrpc_request *req;
208 int rc;
209
210 *request = NULL;
211 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
212 &RQF_MDS_GETATTR_NAME);
213 if (!req)
214 return -ENOMEM;
215
216 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
217 op_data->op_namelen + 1);
218
219 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
220 if (rc) {
221 ptlrpc_request_free(req);
222 return rc;
223 }
224
225 mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
226 op_data->op_mode, op_data->op_suppgids[0], 0);
227
228 if (op_data->op_name) {
229 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
230
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);
234 }
235
236 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
237 op_data->op_mode);
238 ptlrpc_request_set_replen(req);
239
240 rc = mdc_getattr_common(exp, req);
241 if (rc)
242 ptlrpc_req_finished(req);
243 else
244 *request = req;
245 return rc;
246 }
247
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)
252 {
253 struct ptlrpc_request *req;
254 int rc;
255
256 *request = NULL;
257 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
258 &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
259 MDS_IS_SUBDIR);
260 if (!req)
261 return -ENOMEM;
262
263 mdc_is_subdir_pack(req, pfid, cfid, 0);
264 ptlrpc_request_set_replen(req);
265
266 rc = ptlrpc_queue_wait(req);
267 if (rc && rc != -EREMOTE)
268 ptlrpc_req_finished(req);
269 else
270 *request = req;
271 return rc;
272 }
273
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)
281 {
282 struct ptlrpc_request *req;
283 int xattr_namelen = 0;
284 char *tmp;
285 int rc;
286
287 *request = NULL;
288 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
289 if (!req)
290 return -ENOMEM;
291
292 if (xattr_name) {
293 xattr_namelen = strlen(xattr_name) + 1;
294 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
295 xattr_namelen);
296 }
297 if (input_size) {
298 LASSERT(input);
299 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
300 input_size);
301 }
302
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) {
306 LIST_HEAD(cancels);
307 int count;
308
309 /* Without that packing would fail */
310 if (input_size == 0)
311 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
312 RCL_CLIENT, 0);
313
314 count = mdc_resource_get_unused(exp, fid,
315 &cancels, LCK_EX,
316 MDS_INODELOCK_XATTR);
317
318 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
319 if (rc) {
320 ptlrpc_request_free(req);
321 return rc;
322 }
323 } else {
324 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
325 if (rc) {
326 ptlrpc_request_free(req);
327 return rc;
328 }
329 }
330
331 if (opcode == MDS_REINT) {
332 struct mdt_rec_setxattr *rec;
333
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;
343 rec->sx_fid = *fid;
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;
348
349 } else {
350 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
351 }
352
353 if (xattr_name) {
354 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
355 memcpy(tmp, xattr_name, xattr_namelen);
356 }
357 if (input_size) {
358 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
359 memcpy(tmp, input, input_size);
360 }
361
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);
366
367 /* make rpc */
368 if (opcode == MDS_REINT)
369 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
370
371 rc = ptlrpc_queue_wait(req);
372
373 if (opcode == MDS_REINT)
374 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
375
376 if (rc)
377 ptlrpc_req_finished(req);
378 else
379 *request = req;
380 return rc;
381 }
382
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)
388 {
389 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
390 fid, MDS_REINT, valid, xattr_name,
391 input, input_size, output_size, flags,
392 suppgid, request);
393 }
394
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)
399 {
400 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
401 fid, MDS_GETXATTR, valid, xattr_name,
402 input, input_size, output_size, flags,
403 -1, request);
404 }
405
406 #ifdef CONFIG_FS_POSIX_ACL
407 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
408 {
409 struct req_capsule *pill = &req->rq_pill;
410 struct mdt_body *body = md->body;
411 struct posix_acl *acl;
412 void *buf;
413 int rc;
414
415 if (!body->aclsize)
416 return 0;
417
418 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
419
420 if (!buf)
421 return -EPROTO;
422
423 acl = posix_acl_from_xattr(&init_user_ns, buf, body->aclsize);
424 if (!acl)
425 return 0;
426
427 if (IS_ERR(acl)) {
428 rc = PTR_ERR(acl);
429 CERROR("convert xattr to acl: %d\n", rc);
430 return rc;
431 }
432
433 rc = posix_acl_valid(acl);
434 if (rc) {
435 CERROR("validate acl: %d\n", rc);
436 posix_acl_release(acl);
437 return rc;
438 }
439
440 md->posix_acl = acl;
441 return 0;
442 }
443 #else
444 #define mdc_unpack_acl(req, md) 0
445 #endif
446
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)
452 {
453 struct req_capsule *pill = &req->rq_pill;
454 int rc;
455
456 LASSERT(md);
457 memset(md, 0, sizeof(*md));
458
459 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
460
461 if (md->body->valid & OBD_MD_FLEASIZE) {
462 int lmmsize;
463 struct lov_mds_md *lmm;
464
465 if (!S_ISREG(md->body->mode)) {
466 CDEBUG(D_INFO,
467 "OBD_MD_FLEASIZE set, should be a regular file, but is not\n");
468 rc = -EPROTO;
469 goto out;
470 }
471
472 if (md->body->eadatasize == 0) {
473 CDEBUG(D_INFO,
474 "OBD_MD_FLEASIZE set, but eadatasize 0\n");
475 rc = -EPROTO;
476 goto out;
477 }
478 lmmsize = md->body->eadatasize;
479 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
480 if (!lmm) {
481 rc = -EPROTO;
482 goto out;
483 }
484
485 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
486 if (rc < 0)
487 goto out;
488
489 if (rc < sizeof(*md->lsm)) {
490 CDEBUG(D_INFO,
491 "lsm size too small: rc < sizeof (*md->lsm) (%d < %d)\n",
492 rc, (int)sizeof(*md->lsm));
493 rc = -EPROTO;
494 goto out;
495 }
496
497 } else if (md->body->valid & OBD_MD_FLDIREA) {
498 int lmvsize;
499 struct lov_mds_md *lmv;
500
501 if (!S_ISDIR(md->body->mode)) {
502 CDEBUG(D_INFO,
503 "OBD_MD_FLDIREA set, should be a directory, but is not\n");
504 rc = -EPROTO;
505 goto out;
506 }
507
508 if (md->body->eadatasize == 0) {
509 CDEBUG(D_INFO,
510 "OBD_MD_FLDIREA is set, but eadatasize 0\n");
511 return -EPROTO;
512 }
513 if (md->body->valid & OBD_MD_MEA) {
514 lmvsize = md->body->eadatasize;
515 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
516 lmvsize);
517 if (!lmv) {
518 rc = -EPROTO;
519 goto out;
520 }
521
522 rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
523 lmvsize);
524 if (rc < 0)
525 goto out;
526
527 if (rc < sizeof(*md->mea)) {
528 CDEBUG(D_INFO,
529 "size too small: rc < sizeof(*md->mea) (%d < %d)\n",
530 rc, (int)sizeof(*md->mea));
531 rc = -EPROTO;
532 goto out;
533 }
534 }
535 }
536 rc = 0;
537
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) {
544 rc = -EPROTO;
545 goto out;
546 }
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
550 * in reply buffer.
551 */
552 if (md->body->aclsize) {
553 rc = mdc_unpack_acl(req, md);
554 if (rc)
555 goto out;
556 #ifdef CONFIG_FS_POSIX_ACL
557 } else {
558 md->posix_acl = NULL;
559 #endif
560 }
561 }
562
563 out:
564 if (rc) {
565 #ifdef CONFIG_FS_POSIX_ACL
566 posix_acl_release(md->posix_acl);
567 #endif
568 if (md->lsm)
569 obd_free_memmd(dt_exp, &md->lsm);
570 }
571 return rc;
572 }
573
574 static int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
575 {
576 return 0;
577 }
578
579 /**
580 * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
581 * RPC chains.
582 */
583 void mdc_replay_open(struct ptlrpc_request *req)
584 {
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;
590
591 if (!mod) {
592 DEBUG_REQ(D_ERROR, req,
593 "Can't properly replay without open data.");
594 return;
595 }
596
597 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
598
599 och = mod->mod_och;
600 if (och) {
601 struct lustre_handle *file_fh;
602
603 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
604
605 file_fh = &och->och_fh;
606 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
607 file_fh->cookie, body->handle.cookie);
608 old = *file_fh;
609 *file_fh = body->handle;
610 }
611 close_req = mod->mod_close_req;
612 if (close_req) {
613 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
614 struct mdt_ioepoch *epoch;
615
616 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
617 epoch = req_capsule_client_get(&close_req->rq_pill,
618 &RMF_MDT_EPOCH);
619 LASSERT(epoch);
620
621 if (och)
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;
625 }
626 }
627
628 void mdc_commit_open(struct ptlrpc_request *req)
629 {
630 struct md_open_data *mod = req->rq_cb_data;
631
632 if (!mod)
633 return;
634
635 /**
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.
639 */
640
641 /**
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.
648 */
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;
654 obd_mod_put(mod);
655 }
656
657 int mdc_set_open_replay_data(struct obd_export *exp,
658 struct obd_client_handle *och,
659 struct lookup_intent *it)
660 {
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;
666
667 if (!open_req->rq_replay)
668 return 0;
669
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);
672 LASSERT(rec);
673 /* Incoming message in my byte order (it's been swabbed). */
674 /* Outgoing messages always in my byte order. */
675 LASSERT(body);
676
677 /* Only if the import is replayable, we set replay_open data */
678 if (och && imp->imp_replayable) {
679 mod = obd_mod_alloc();
680 if (!mod) {
681 DEBUG_REQ(D_ERROR, open_req,
682 "Can't allocate md_open_data");
683 return 0;
684 }
685
686 /**
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.
691 */
692 obd_mod_get(mod);
693 obd_mod_get(mod);
694
695 spin_lock(&open_req->rq_lock);
696 och->och_mod = mod;
697 mod->mod_och = och;
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);
704 }
705
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");
713 LBUG();
714 }
715
716 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
717 return 0;
718 }
719
720 static void mdc_free_open(struct md_open_data *mod)
721 {
722 int committed = 0;
723
724 if (mod->mod_is_create == 0 &&
725 imp_connect_disp_stripe(mod->mod_open_req->rq_import))
726 committed = 1;
727
728 LASSERT(mod->mod_open_req->rq_replay == 0);
729
730 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
731
732 ptlrpc_request_committed(mod->mod_open_req, committed);
733 if (mod->mod_close_req)
734 ptlrpc_request_committed(mod->mod_close_req, committed);
735 }
736
737 static int mdc_clear_open_replay_data(struct obd_export *exp,
738 struct obd_client_handle *och)
739 {
740 struct md_open_data *mod = och->och_mod;
741
742 /**
743 * It is possible to not have \var mod in a case of eviction between
744 * lookup and ll_file_open().
745 **/
746 if (!mod)
747 return 0;
748
749 LASSERT(mod != LP_POISON);
750 LASSERT(mod->mod_open_req);
751 mdc_free_open(mod);
752
753 mod->mod_och = NULL;
754 och->och_mod = NULL;
755 obd_mod_put(mod);
756
757 return 0;
758 }
759
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;
765
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);
769
770 epoch->flags |= MF_SOM_AU;
771 if (repbody->valid & OBD_MD_FLGETATTRLOCK)
772 op_data->op_flags |= MF_GETATTR_LOCK;
773 }
774 }
775
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)
778 {
779 struct obd_device *obd = class_exp2obd(exp);
780 struct ptlrpc_request *req;
781 struct req_format *req_fmt;
782 int rc;
783 int saved_rc = 0;
784
785 req_fmt = &RQF_MDS_CLOSE;
786 if (op_data->op_bias & MDS_HSM_RELEASE) {
787 req_fmt = &RQF_MDS_RELEASE_CLOSE;
788
789 /* allocate a FID for volatile file */
790 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
791 if (rc < 0) {
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 */
795 saved_rc = rc;
796 }
797 }
798
799 *request = NULL;
800 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
801 if (!req)
802 return -ENOMEM;
803
804 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
805 if (rc) {
806 ptlrpc_request_free(req);
807 return rc;
808 }
809
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
812 * always progressing
813 */
814 req->rq_request_portal = MDS_READPAGE_PORTAL;
815 ptlrpc_at_set_req_timeout(req);
816
817 /* Ensure that this close's handle is fixed up during replay. */
818 if (likely(mod)) {
819 LASSERTF(mod->mod_open_req &&
820 mod->mod_open_req->rq_type != LI_POISON,
821 "POISONED open %p!\n", mod->mod_open_req);
822
823 mod->mod_close_req = req;
824
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
828 */
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);
832 } else {
833 CDEBUG(D_HA,
834 "couldn't find open req; expecting close error\n");
835 }
836
837 mdc_close_pack(req, op_data);
838
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);
843
844 ptlrpc_request_set_replen(req);
845
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);
849
850 if (!req->rq_repmsg) {
851 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
852 req->rq_status);
853 if (rc == 0)
854 rc = req->rq_status ?: -EIO;
855 } else if (rc == 0 || rc == -EAGAIN) {
856 struct mdt_body *body;
857
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);
862 if (rc > 0)
863 rc = -rc;
864 }
865 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
866 if (!body)
867 rc = -EPROTO;
868 } else if (rc == -ESTALE) {
869 /**
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
873 */
874 if (mod) {
875 DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
876 if (mod->mod_open_req->rq_committed)
877 rc = 0;
878 }
879 }
880
881 if (mod) {
882 if (rc != 0)
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.
886 */
887 obd_mod_put(mod);
888 }
889 *request = req;
890 mdc_close_handle_reply(req, op_data, rc);
891 return rc < 0 ? rc : saved_rc;
892 }
893
894 static int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
895 struct md_open_data *mod)
896 {
897 struct obd_device *obd = class_exp2obd(exp);
898 struct ptlrpc_request *req;
899 int rc;
900
901 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
902 &RQF_MDS_DONE_WRITING);
903 if (!req)
904 return -ENOMEM;
905
906 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
907 if (rc) {
908 ptlrpc_request_free(req);
909 return rc;
910 }
911
912 if (mod) {
913 LASSERTF(mod->mod_open_req &&
914 mod->mod_open_req->rq_type != LI_POISON,
915 "POISONED setattr %p!\n", mod->mod_open_req);
916
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
921 */
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);
925 }
926
927 mdc_close_pack(req, op_data);
928 ptlrpc_request_set_replen(req);
929
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);
933
934 if (rc == -ESTALE) {
935 /**
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
939 */
940 if (mod) {
941 if (mod->mod_open_req->rq_committed)
942 rc = 0;
943 }
944 }
945
946 if (mod) {
947 if (rc != 0)
948 mod->mod_close_req = NULL;
949 LASSERT(mod->mod_open_req);
950 mdc_free_open(mod);
951
952 /* Since now, mod is accessed through setattr req only,
953 * thus DW req does not keep a reference on mod anymore.
954 */
955 obd_mod_put(mod);
956 }
957
958 mdc_close_handle_reply(req, op_data, rc);
959 ptlrpc_req_finished(req);
960 return rc;
961 }
962
963 static int mdc_readpage(struct obd_export *exp, struct md_op_data *op_data,
964 struct page **pages, struct ptlrpc_request **request)
965 {
966 struct ptlrpc_request *req;
967 struct ptlrpc_bulk_desc *desc;
968 int i;
969 wait_queue_head_t waitq;
970 int resends = 0;
971 struct l_wait_info lwi;
972 int rc;
973
974 *request = NULL;
975 init_waitqueue_head(&waitq);
976
977 restart_bulk:
978 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
979 if (!req)
980 return -ENOMEM;
981
982 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
983 if (rc) {
984 ptlrpc_request_free(req);
985 return rc;
986 }
987
988 req->rq_request_portal = MDS_READPAGE_PORTAL;
989 ptlrpc_at_set_req_timeout(req);
990
991 desc = ptlrpc_prep_bulk_imp(req, op_data->op_npages, 1, BULK_PUT_SINK,
992 MDS_BULK_PORTAL);
993 if (!desc) {
994 ptlrpc_request_free(req);
995 return -ENOMEM;
996 }
997
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);
1001
1002 mdc_readdir_pack(req, op_data->op_offset,
1003 PAGE_SIZE * op_data->op_npages,
1004 &op_data->op_fid1);
1005
1006 ptlrpc_request_set_replen(req);
1007 rc = ptlrpc_queue_wait(req);
1008 if (rc) {
1009 ptlrpc_req_finished(req);
1010 if (rc != -ETIMEDOUT)
1011 return rc;
1012
1013 resends++;
1014 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1015 CERROR("too many resend retries, returning error\n");
1016 return -EIO;
1017 }
1018 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends),
1019 NULL, NULL, NULL);
1020 l_wait_event(waitq, 0, &lwi);
1021
1022 goto restart_bulk;
1023 }
1024
1025 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1026 req->rq_bulk->bd_nob_transferred);
1027 if (rc < 0) {
1028 ptlrpc_req_finished(req);
1029 return rc;
1030 }
1031
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);
1037 return -EPROTO;
1038 }
1039
1040 *request = req;
1041 return 0;
1042 }
1043
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)
1047 {
1048 struct obd_device *obd = class_exp2obd(exp);
1049 struct ptlrpc_request *req;
1050 struct obd_statfs *msfs;
1051 struct obd_import *imp = NULL;
1052 int rc;
1053
1054 /*
1055 * Since the request might also come from lprocfs, so we need
1056 * sync this with client_disconnect_export Bug15684
1057 */
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);
1062 if (!imp)
1063 return -ENODEV;
1064
1065 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1066 LUSTRE_MDS_VERSION, MDS_STATFS);
1067 if (!req) {
1068 rc = -ENOMEM;
1069 goto output;
1070 }
1071
1072 ptlrpc_request_set_replen(req);
1073
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;
1078 }
1079
1080 rc = ptlrpc_queue_wait(req);
1081 if (rc) {
1082 /* check connection error first */
1083 if (imp->imp_connect_error)
1084 rc = imp->imp_connect_error;
1085 goto out;
1086 }
1087
1088 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1089 if (!msfs) {
1090 rc = -EPROTO;
1091 goto out;
1092 }
1093
1094 *osfs = *msfs;
1095 out:
1096 ptlrpc_req_finished(req);
1097 output:
1098 class_import_put(imp);
1099 return rc;
1100 }
1101
1102 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1103 {
1104 __u32 keylen, vallen;
1105 void *key;
1106 int rc;
1107
1108 if (gf->gf_pathlen > PATH_MAX)
1109 return -ENAMETOOLONG;
1110 if (gf->gf_pathlen < 2)
1111 return -EOVERFLOW;
1112
1113 /* Key is KEY_FID2PATH + getinfo_fid2path description */
1114 keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1115 key = kzalloc(keylen, GFP_NOFS);
1116 if (!key)
1117 return -ENOMEM;
1118 memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1119 memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1120
1121 CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1122 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1123
1124 if (!fid_is_sane(&gf->gf_fid)) {
1125 rc = -EINVAL;
1126 goto out;
1127 }
1128
1129 /* Val is struct getinfo_fid2path result plus path */
1130 vallen = sizeof(*gf) + gf->gf_pathlen;
1131
1132 rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1133 if (rc != 0 && rc != -EREMOTE)
1134 goto out;
1135
1136 if (vallen <= sizeof(*gf)) {
1137 rc = -EPROTO;
1138 goto out;
1139 } else if (vallen > sizeof(*gf) + gf->gf_pathlen) {
1140 rc = -EOVERFLOW;
1141 goto out;
1142 }
1143
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);
1146
1147 out:
1148 kfree(key);
1149 return rc;
1150 }
1151
1152 static 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;
1159
1160 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1161 LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1162 if (!req) {
1163 rc = -ENOMEM;
1164 goto out;
1165 }
1166
1167 mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1168
1169 /* Copy hsm_progress struct */
1170 req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1171 if (!req_hpk) {
1172 rc = -EPROTO;
1173 goto out;
1174 }
1175
1176 *req_hpk = *hpk;
1177 req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1178
1179 ptlrpc_request_set_replen(req);
1180
1181 rc = mdc_queue_wait(req);
1182 out:
1183 ptlrpc_req_finished(req);
1184 return rc;
1185 }
1186
1187 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1188 {
1189 __u32 *archive_mask;
1190 struct ptlrpc_request *req;
1191 int rc;
1192
1193 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1194 LUSTRE_MDS_VERSION,
1195 MDS_HSM_CT_REGISTER);
1196 if (!req) {
1197 rc = -ENOMEM;
1198 goto out;
1199 }
1200
1201 mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1202
1203 /* Copy hsm_progress struct */
1204 archive_mask = req_capsule_client_get(&req->rq_pill,
1205 &RMF_MDS_HSM_ARCHIVE);
1206 if (!archive_mask) {
1207 rc = -EPROTO;
1208 goto out;
1209 }
1210
1211 *archive_mask = archives;
1212
1213 ptlrpc_request_set_replen(req);
1214
1215 rc = mdc_queue_wait(req);
1216 out:
1217 ptlrpc_req_finished(req);
1218 return rc;
1219 }
1220
1221 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1222 struct md_op_data *op_data)
1223 {
1224 struct hsm_current_action *hca = op_data->op_data;
1225 struct hsm_current_action *req_hca;
1226 struct ptlrpc_request *req;
1227 int rc;
1228
1229 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1230 &RQF_MDS_HSM_ACTION);
1231 if (!req)
1232 return -ENOMEM;
1233
1234 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1235 if (rc) {
1236 ptlrpc_request_free(req);
1237 return rc;
1238 }
1239
1240 mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1241 op_data->op_suppgids[0], 0);
1242
1243 ptlrpc_request_set_replen(req);
1244
1245 rc = mdc_queue_wait(req);
1246 if (rc)
1247 goto out;
1248
1249 req_hca = req_capsule_server_get(&req->rq_pill,
1250 &RMF_MDS_HSM_CURRENT_ACTION);
1251 if (!req_hca) {
1252 rc = -EPROTO;
1253 goto out;
1254 }
1255
1256 *hca = *req_hca;
1257
1258 out:
1259 ptlrpc_req_finished(req);
1260 return rc;
1261 }
1262
1263 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1264 {
1265 struct ptlrpc_request *req;
1266 int rc;
1267
1268 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1269 LUSTRE_MDS_VERSION,
1270 MDS_HSM_CT_UNREGISTER);
1271 if (!req) {
1272 rc = -ENOMEM;
1273 goto out;
1274 }
1275
1276 mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1277
1278 ptlrpc_request_set_replen(req);
1279
1280 rc = mdc_queue_wait(req);
1281 out:
1282 ptlrpc_req_finished(req);
1283 return rc;
1284 }
1285
1286 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1287 struct md_op_data *op_data)
1288 {
1289 struct hsm_user_state *hus = op_data->op_data;
1290 struct hsm_user_state *req_hus;
1291 struct ptlrpc_request *req;
1292 int rc;
1293
1294 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1295 &RQF_MDS_HSM_STATE_GET);
1296 if (!req)
1297 return -ENOMEM;
1298
1299 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1300 if (rc != 0) {
1301 ptlrpc_request_free(req);
1302 return rc;
1303 }
1304
1305 mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1306 op_data->op_suppgids[0], 0);
1307
1308 ptlrpc_request_set_replen(req);
1309
1310 rc = mdc_queue_wait(req);
1311 if (rc)
1312 goto out;
1313
1314 req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1315 if (!req_hus) {
1316 rc = -EPROTO;
1317 goto out;
1318 }
1319
1320 *hus = *req_hus;
1321
1322 out:
1323 ptlrpc_req_finished(req);
1324 return rc;
1325 }
1326
1327 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1328 struct md_op_data *op_data)
1329 {
1330 struct hsm_state_set *hss = op_data->op_data;
1331 struct hsm_state_set *req_hss;
1332 struct ptlrpc_request *req;
1333 int rc;
1334
1335 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1336 &RQF_MDS_HSM_STATE_SET);
1337 if (!req)
1338 return -ENOMEM;
1339
1340 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1341 if (rc) {
1342 ptlrpc_request_free(req);
1343 return rc;
1344 }
1345
1346 mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1347 op_data->op_suppgids[0], 0);
1348
1349 /* Copy states */
1350 req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1351 if (!req_hss) {
1352 rc = -EPROTO;
1353 goto out;
1354 }
1355 *req_hss = *hss;
1356
1357 ptlrpc_request_set_replen(req);
1358
1359 rc = mdc_queue_wait(req);
1360 out:
1361 ptlrpc_req_finished(req);
1362 return rc;
1363 }
1364
1365 static int mdc_ioc_hsm_request(struct obd_export *exp,
1366 struct hsm_user_request *hur)
1367 {
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;
1372 char *req_opaque;
1373 int rc;
1374
1375 req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1376 if (!req) {
1377 rc = -ENOMEM;
1378 goto out;
1379 }
1380
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);
1386
1387 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1388 if (rc) {
1389 ptlrpc_request_free(req);
1390 return rc;
1391 }
1392
1393 mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1394
1395 /* Copy hsm_request struct */
1396 req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1397 if (!req_hr) {
1398 rc = -EPROTO;
1399 goto out;
1400 }
1401 *req_hr = hur->hur_request;
1402
1403 /* Copy hsm_user_item structs */
1404 req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1405 if (!req_hui) {
1406 rc = -EPROTO;
1407 goto out;
1408 }
1409 memcpy(req_hui, hur->hur_user_item,
1410 hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1411
1412 /* Copy opaque field */
1413 req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1414 if (!req_opaque) {
1415 rc = -EPROTO;
1416 goto out;
1417 }
1418 memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1419
1420 ptlrpc_request_set_replen(req);
1421
1422 rc = mdc_queue_wait(req);
1423 out:
1424 ptlrpc_req_finished(req);
1425 return rc;
1426 }
1427
1428 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
1429 {
1430 struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1431
1432 LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1433
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;
1439 return lh;
1440 }
1441
1442 #define D_CHANGELOG 0
1443
1444 struct changelog_show {
1445 __u64 cs_startrec;
1446 __u32 cs_flags;
1447 struct file *cs_fp;
1448 char *cs_buf;
1449 struct obd_device *cs_obd;
1450 };
1451
1452 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1453 struct llog_rec_hdr *hdr, void *data)
1454 {
1455 struct changelog_show *cs = data;
1456 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1457 struct kuc_hdr *lh;
1458 int len, rc;
1459
1460 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1461 rc = -EINVAL;
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);
1465 return rc;
1466 }
1467
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);
1472 return 0;
1473 }
1474
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));
1481
1482 len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1483
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));
1487
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);
1490
1491 return rc;
1492 }
1493
1494 static int mdc_changelog_send_thread(void *csdata)
1495 {
1496 struct changelog_show *cs = csdata;
1497 struct llog_ctxt *ctxt = NULL;
1498 struct llog_handle *llh = NULL;
1499 struct kuc_hdr *kuch;
1500 int rc;
1501
1502 CDEBUG(D_CHANGELOG, "changelog to fp=%p start %llu\n",
1503 cs->cs_fp, cs->cs_startrec);
1504
1505 cs->cs_buf = kzalloc(KUC_CHANGELOG_MSG_MAXSIZE, GFP_NOFS);
1506 if (!cs->cs_buf) {
1507 rc = -ENOMEM;
1508 goto out;
1509 }
1510
1511 /* Set up the remote catalog handle */
1512 ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1513 if (!ctxt) {
1514 rc = -ENOENT;
1515 goto out;
1516 }
1517 rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1518 LLOG_OPEN_EXISTS);
1519 if (rc) {
1520 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1521 cs->cs_obd->obd_name, rc);
1522 goto out;
1523 }
1524 rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
1525 if (rc) {
1526 CERROR("llog_init_handle failed %d\n", rc);
1527 goto out;
1528 }
1529
1530 rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1531
1532 /* Send EOF no matter what our result */
1533 kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags);
1534 if (kuch) {
1535 kuch->kuc_msgtype = CL_EOF;
1536 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1537 }
1538
1539 out:
1540 fput(cs->cs_fp);
1541 if (llh)
1542 llog_cat_close(NULL, llh);
1543 if (ctxt)
1544 llog_ctxt_put(ctxt);
1545 kfree(cs->cs_buf);
1546 kfree(cs);
1547 return rc;
1548 }
1549
1550 static int mdc_ioc_changelog_send(struct obd_device *obd,
1551 struct ioc_changelog *icc)
1552 {
1553 struct changelog_show *cs;
1554 struct task_struct *task;
1555 int rc;
1556
1557 /* Freed in mdc_changelog_send_thread */
1558 cs = kzalloc(sizeof(*cs), GFP_NOFS);
1559 if (!cs)
1560 return -ENOMEM;
1561
1562 cs->cs_obd = obd;
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;
1567
1568 /*
1569 * New thread because we should return to user app before
1570 * writing into our pipe
1571 */
1572 task = kthread_run(mdc_changelog_send_thread, cs,
1573 "mdc_clg_send_thread");
1574 if (IS_ERR(task)) {
1575 rc = PTR_ERR(task);
1576 CERROR("%s: can't start changelog thread: rc = %d\n",
1577 obd->obd_name, rc);
1578 kfree(cs);
1579 } else {
1580 rc = 0;
1581 CDEBUG(D_CHANGELOG, "%s: started changelog thread\n",
1582 obd->obd_name);
1583 }
1584
1585 CERROR("Failed to start changelog thread: %d\n", rc);
1586 return rc;
1587 }
1588
1589 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1590 struct lustre_kernelcomm *lk);
1591
1592 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
1593 struct obd_quotactl *oqctl)
1594 {
1595 struct client_obd *cli = &exp->exp_obd->u.cli;
1596 struct ptlrpc_request *req;
1597 struct obd_quotactl *body;
1598 int rc;
1599
1600 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1601 &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
1602 MDS_QUOTACHECK);
1603 if (!req)
1604 return -ENOMEM;
1605
1606 body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1607 *body = *oqctl;
1608
1609 ptlrpc_request_set_replen(req);
1610
1611 /* the next poll will find -ENODATA, that means quotacheck is
1612 * going on
1613 */
1614 cli->cl_qchk_stat = -ENODATA;
1615 rc = ptlrpc_queue_wait(req);
1616 if (rc)
1617 cli->cl_qchk_stat = rc;
1618 ptlrpc_req_finished(req);
1619 return rc;
1620 }
1621
1622 static int mdc_quota_poll_check(struct obd_export *exp,
1623 struct if_quotacheck *qchk)
1624 {
1625 struct client_obd *cli = &exp->exp_obd->u.cli;
1626 int rc;
1627
1628 qchk->obd_uuid = cli->cl_target_uuid;
1629 memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
1630
1631 rc = cli->cl_qchk_stat;
1632 /* the client is not the previous one */
1633 if (rc == CL_NOT_QUOTACHECKED)
1634 rc = -EINTR;
1635 return rc;
1636 }
1637
1638 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1639 struct obd_quotactl *oqctl)
1640 {
1641 struct ptlrpc_request *req;
1642 struct obd_quotactl *oqc;
1643 int rc;
1644
1645 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1646 &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1647 MDS_QUOTACTL);
1648 if (!req)
1649 return -ENOMEM;
1650
1651 oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1652 *oqc = *oqctl;
1653
1654 ptlrpc_request_set_replen(req);
1655 ptlrpc_at_set_req_timeout(req);
1656 req->rq_no_resend = 1;
1657
1658 rc = ptlrpc_queue_wait(req);
1659 if (rc)
1660 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1661
1662 if (req->rq_repmsg) {
1663 oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1664 if (oqc) {
1665 *oqctl = *oqc;
1666 } else if (!rc) {
1667 CERROR("Can't unpack obd_quotactl\n");
1668 rc = -EPROTO;
1669 }
1670 } else if (!rc) {
1671 CERROR("Can't unpack obd_quotactl\n");
1672 rc = -EPROTO;
1673 }
1674 ptlrpc_req_finished(req);
1675
1676 return rc;
1677 }
1678
1679 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1680 struct md_op_data *op_data)
1681 {
1682 LIST_HEAD(cancels);
1683 struct ptlrpc_request *req;
1684 int rc, count;
1685 struct mdc_swap_layouts *msl, *payload;
1686
1687 msl = op_data->op_data;
1688
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
1694 */
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);
1699
1700 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1701 &RQF_MDS_SWAP_LAYOUTS);
1702 if (!req) {
1703 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1704 return -ENOMEM;
1705 }
1706
1707 rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1708 if (rc) {
1709 ptlrpc_request_free(req);
1710 return rc;
1711 }
1712
1713 mdc_swap_layouts_pack(req, op_data);
1714
1715 payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1716 LASSERT(payload);
1717
1718 *payload = *msl;
1719
1720 ptlrpc_request_set_replen(req);
1721
1722 rc = ptlrpc_queue_wait(req);
1723
1724 ptlrpc_req_finished(req);
1725 return rc;
1726 }
1727
1728 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1729 void *karg, void __user *uarg)
1730 {
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;
1734 int rc;
1735
1736 if (!try_module_get(THIS_MODULE)) {
1737 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
1738 module_name(THIS_MODULE));
1739 return -EINVAL;
1740 }
1741 switch (cmd) {
1742 case OBD_IOC_CHANGELOG_SEND:
1743 rc = mdc_ioc_changelog_send(obd, karg);
1744 goto out;
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
1750 };
1751
1752 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
1753 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1754 NULL);
1755 goto out;
1756 }
1757 case OBD_IOC_FID2PATH:
1758 rc = mdc_ioc_fid2path(exp, karg);
1759 goto out;
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. */
1763 if (rc == -EEXIST)
1764 rc = 0;
1765 goto out;
1766 case LL_IOC_HSM_PROGRESS:
1767 rc = mdc_ioc_hsm_progress(exp, karg);
1768 goto out;
1769 case LL_IOC_HSM_STATE_GET:
1770 rc = mdc_ioc_hsm_state_get(exp, karg);
1771 goto out;
1772 case LL_IOC_HSM_STATE_SET:
1773 rc = mdc_ioc_hsm_state_set(exp, karg);
1774 goto out;
1775 case LL_IOC_HSM_ACTION:
1776 rc = mdc_ioc_hsm_current_action(exp, karg);
1777 goto out;
1778 case LL_IOC_HSM_REQUEST:
1779 rc = mdc_ioc_hsm_request(exp, karg);
1780 goto out;
1781 case OBD_IOC_CLIENT_RECOVER:
1782 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1783 if (rc < 0)
1784 goto out;
1785 rc = 0;
1786 goto out;
1787 case IOC_OSC_SET_ACTIVE:
1788 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1789 goto out;
1790 case OBD_IOC_POLL_QUOTACHECK:
1791 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
1792 goto out;
1793 case OBD_IOC_PING_TARGET:
1794 rc = ptlrpc_obd_ping(obd);
1795 goto out;
1796 /*
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.
1801 * bz20731, LU-592.
1802 */
1803 case IOC_OBD_STATFS: {
1804 struct obd_statfs stat_buf = {0};
1805
1806 if (*((__u32 *) data->ioc_inlbuf2) != 0) {
1807 rc = -ENODEV;
1808 goto out;
1809 }
1810
1811 /* copy UUID */
1812 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1813 min_t(size_t, data->ioc_plen2,
1814 sizeof(struct obd_uuid)))) {
1815 rc = -EFAULT;
1816 goto out;
1817 }
1818
1819 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1820 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1821 0);
1822 if (rc != 0)
1823 goto out;
1824
1825 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1826 min_t(size_t, data->ioc_plen1,
1827 sizeof(stat_buf)))) {
1828 rc = -EFAULT;
1829 goto out;
1830 }
1831
1832 rc = 0;
1833 goto out;
1834 }
1835 case OBD_IOC_QUOTACTL: {
1836 struct if_quotactl *qctl = karg;
1837 struct obd_quotactl *oqctl;
1838
1839 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1840 if (!oqctl) {
1841 rc = -ENOMEM;
1842 goto out;
1843 }
1844
1845 QCTL_COPY(oqctl, qctl);
1846 rc = obd_quotactl(exp, oqctl);
1847 if (rc == 0) {
1848 QCTL_COPY(qctl, oqctl);
1849 qctl->qc_valid = QC_MDTIDX;
1850 qctl->obd_uuid = obd->u.cli.cl_target_uuid;
1851 }
1852
1853 kfree(oqctl);
1854 goto out;
1855 }
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)))) {
1859 rc = -EFAULT;
1860 goto out;
1861 }
1862
1863 rc = 0;
1864 goto out;
1865 case LL_IOC_LOV_SWAP_LAYOUTS:
1866 rc = mdc_ioc_swap_layouts(exp, karg);
1867 goto out;
1868 default:
1869 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
1870 rc = -ENOTTY;
1871 goto out;
1872 }
1873 out:
1874 module_put(THIS_MODULE);
1875
1876 return rc;
1877 }
1878
1879 static int mdc_get_info_rpc(struct obd_export *exp,
1880 u32 keylen, void *key,
1881 int vallen, void *val)
1882 {
1883 struct obd_import *imp = class_exp2cliimp(exp);
1884 struct ptlrpc_request *req;
1885 char *tmp;
1886 int rc = -EINVAL;
1887
1888 req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1889 if (!req)
1890 return -ENOMEM;
1891
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));
1896
1897 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1898 if (rc) {
1899 ptlrpc_request_free(req);
1900 return rc;
1901 }
1902
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));
1907
1908 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1909 RCL_SERVER, vallen);
1910 ptlrpc_request_set_replen(req);
1911
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
1915 */
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);
1922 }
1923 }
1924 ptlrpc_req_finished(req);
1925
1926 return rc;
1927 }
1928
1929 static void lustre_swab_hai(struct hsm_action_item *h)
1930 {
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);
1939 }
1940
1941 static void lustre_swab_hal(struct hsm_action_list *h)
1942 {
1943 struct hsm_action_item *hai;
1944 int i;
1945
1946 __swab32s(&h->hal_version);
1947 __swab32s(&h->hal_count);
1948 __swab32s(&h->hal_archive_id);
1949 __swab64s(&h->hal_flags);
1950 hai = hai_first(h);
1951 for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
1952 lustre_swab_hai(hai);
1953 }
1954
1955 static void lustre_swab_kuch(struct kuc_hdr *l)
1956 {
1957 __swab16s(&l->kuc_magic);
1958 /* __u8 l->kuc_transport */
1959 __swab16s(&l->kuc_msgtype);
1960 __swab16s(&l->kuc_msglen);
1961 }
1962
1963 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1964 struct lustre_kernelcomm *lk)
1965 {
1966 struct obd_import *imp = class_exp2cliimp(exp);
1967 __u32 archive = lk->lk_data;
1968 int rc = 0;
1969
1970 if (lk->lk_group != KUC_GRP_HSM) {
1971 CERROR("Bad copytool group %d\n", lk->lk_group);
1972 return -EINVAL;
1973 }
1974
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);
1977
1978 if (lk->lk_flags & LK_FLG_STOP) {
1979 /* Unregister with the coordinator */
1980 rc = mdc_ioc_hsm_ct_unregister(imp);
1981 } else {
1982 rc = mdc_ioc_hsm_ct_register(imp, archive);
1983 }
1984
1985 return rc;
1986 }
1987
1988 /**
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
1992 */
1993 static int mdc_hsm_copytool_send(int len, void *val)
1994 {
1995 struct kuc_hdr *lh = (struct kuc_hdr *)val;
1996 struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
1997
1998 if (len < sizeof(*lh) + sizeof(*hal)) {
1999 CERROR("Short HSM message %d < %d\n", len,
2000 (int) (sizeof(*lh) + sizeof(*hal)));
2001 return -EPROTO;
2002 }
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);
2008 return -EPROTO;
2009 }
2010
2011 CDEBUG(D_HSM,
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);
2015
2016 /* Broadcast to HSM listeners */
2017 return libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2018 }
2019
2020 /**
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)
2025 */
2026 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2027 {
2028 struct kkuc_ct_data *kcd = data;
2029 struct obd_import *imp = (struct obd_import *)cb_arg;
2030 int rc;
2031
2032 if (!kcd || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2033 return -EPROTO;
2034
2035 if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2036 return 0;
2037
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);
2041
2042 /* ignore error if the copytool is already registered */
2043 return (rc == -EEXIST) ? 0 : rc;
2044 }
2045
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)
2051 {
2052 struct obd_import *imp = class_exp2cliimp(exp);
2053 int rc;
2054
2055 if (KEY_IS(KEY_READ_ONLY)) {
2056 if (vallen != sizeof(int))
2057 return -EINVAL;
2058
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 |=
2063 OBD_CONNECT_RDONLY;
2064 } else {
2065 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2066 imp->imp_connect_data.ocd_connect_flags &=
2067 ~OBD_CONNECT_RDONLY;
2068 }
2069 spin_unlock(&imp->imp_lock);
2070
2071 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2072 keylen, key, vallen, val, set);
2073 return rc;
2074 }
2075 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2076 sptlrpc_conf_client_adapt(exp->exp_obd);
2077 return 0;
2078 }
2079 if (KEY_IS(KEY_FLUSH_CTX)) {
2080 sptlrpc_import_flush_my_ctx(imp);
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
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)
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_DEFAULT_EASIZE)) {
2115 int *default_easize;
2116
2117 if (*vallen != sizeof(int))
2118 return -EINVAL;
2119 default_easize = val;
2120 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2121 return 0;
2122 } else if (KEY_IS(KEY_CONN_DATA)) {
2123 struct obd_import *imp = class_exp2cliimp(exp);
2124 struct obd_connect_data *data = val;
2125
2126 if (*vallen != sizeof(*data))
2127 return -EINVAL;
2128
2129 *data = imp->imp_connect_data;
2130 return 0;
2131 } else if (KEY_IS(KEY_TGT_COUNT)) {
2132 *((int *)val) = 1;
2133 return 0;
2134 }
2135
2136 rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2137
2138 return rc;
2139 }
2140
2141 static int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
2142 struct ptlrpc_request **request)
2143 {
2144 struct ptlrpc_request *req;
2145 int rc;
2146
2147 *request = NULL;
2148 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2149 if (!req)
2150 return -ENOMEM;
2151
2152 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2153 if (rc) {
2154 ptlrpc_request_free(req);
2155 return rc;
2156 }
2157
2158 mdc_pack_body(req, fid, 0, 0, -1, 0);
2159
2160 ptlrpc_request_set_replen(req);
2161
2162 rc = ptlrpc_queue_wait(req);
2163 if (rc)
2164 ptlrpc_req_finished(req);
2165 else
2166 *request = req;
2167 return rc;
2168 }
2169
2170 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2171 enum obd_import_event event)
2172 {
2173 int rc = 0;
2174
2175 LASSERT(imp->imp_obd == obd);
2176
2177 switch (event) {
2178 case IMP_EVENT_DISCON: {
2179 #if 0
2180 /* XXX Pass event up to OBDs stack. used only for FLD now */
2181 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2182 #endif
2183 break;
2184 }
2185 case IMP_EVENT_INACTIVE: {
2186 struct client_obd *cli = &obd->u.cli;
2187 /*
2188 * Flush current sequence to make client obtain new one
2189 * from server in case of disconnect/reconnect.
2190 */
2191 if (cli->cl_seq)
2192 seq_client_flush(cli->cl_seq);
2193
2194 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2195 break;
2196 }
2197 case IMP_EVENT_INVALIDATE: {
2198 struct ldlm_namespace *ns = obd->obd_namespace;
2199
2200 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2201
2202 break;
2203 }
2204 case IMP_EVENT_ACTIVE:
2205 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2206 /* redo the kuc registration after reconnecting */
2207 if (rc == 0)
2208 /* re-register HSM agents */
2209 rc = libcfs_kkuc_group_foreach(KUC_GRP_HSM,
2210 mdc_hsm_ct_reregister,
2211 (void *)imp);
2212 break;
2213 case IMP_EVENT_OCD:
2214 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2215 break;
2216 case IMP_EVENT_DEACTIVATE:
2217 case IMP_EVENT_ACTIVATE:
2218 break;
2219 default:
2220 CERROR("Unknown import event %x\n", event);
2221 LBUG();
2222 }
2223 return rc;
2224 }
2225
2226 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
2227 struct md_op_data *op_data)
2228 {
2229 struct client_obd *cli = &exp->exp_obd->u.cli;
2230 struct lu_client_seq *seq = cli->cl_seq;
2231
2232 return seq_client_alloc_fid(NULL, seq, fid);
2233 }
2234
2235 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2236 {
2237 struct client_obd *cli = &exp->exp_obd->u.cli;
2238
2239 return &cli->cl_target_uuid;
2240 }
2241
2242 /**
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
2246 */
2247 static int mdc_cancel_weight(struct ldlm_lock *lock)
2248 {
2249 if (lock->l_resource->lr_type != LDLM_IBITS)
2250 return 0;
2251
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
2255 */
2256 if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2257 return 0;
2258
2259 return 1;
2260 }
2261
2262 static int mdc_resource_inode_free(struct ldlm_resource *res)
2263 {
2264 if (res->lr_lvb_inode)
2265 res->lr_lvb_inode = NULL;
2266
2267 return 0;
2268 }
2269
2270 static struct ldlm_valblock_ops inode_lvbo = {
2271 .lvbo_free = mdc_resource_inode_free,
2272 };
2273
2274 static int mdc_llog_init(struct obd_device *obd)
2275 {
2276 struct obd_llog_group *olg = &obd->obd_olg;
2277 struct llog_ctxt *ctxt;
2278 int rc;
2279
2280 rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2281 &llog_client_ops);
2282 if (rc)
2283 return rc;
2284
2285 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2286 llog_initiator_connect(ctxt);
2287 llog_ctxt_put(ctxt);
2288
2289 return 0;
2290 }
2291
2292 static void mdc_llog_finish(struct obd_device *obd)
2293 {
2294 struct llog_ctxt *ctxt;
2295
2296 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2297 if (ctxt)
2298 llog_cleanup(NULL, ctxt);
2299 }
2300
2301 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2302 {
2303 struct client_obd *cli = &obd->u.cli;
2304 struct lprocfs_static_vars lvars = { NULL };
2305 int rc;
2306
2307 cli->cl_rpc_lock = kzalloc(sizeof(*cli->cl_rpc_lock), GFP_NOFS);
2308 if (!cli->cl_rpc_lock)
2309 return -ENOMEM;
2310 mdc_init_rpc_lock(cli->cl_rpc_lock);
2311
2312 rc = ptlrpcd_addref();
2313 if (rc < 0)
2314 goto err_rpc_lock;
2315
2316 cli->cl_close_lock = kzalloc(sizeof(*cli->cl_close_lock), GFP_NOFS);
2317 if (!cli->cl_close_lock) {
2318 rc = -ENOMEM;
2319 goto err_ptlrpcd_decref;
2320 }
2321 mdc_init_rpc_lock(cli->cl_close_lock);
2322
2323 rc = client_obd_setup(obd, cfg);
2324 if (rc)
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);
2330
2331 ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2332
2333 obd->obd_namespace->ns_lvbo = &inode_lvbo;
2334
2335 rc = mdc_llog_init(obd);
2336 if (rc) {
2337 mdc_cleanup(obd);
2338 CERROR("failed to setup llogging subsystems\n");
2339 }
2340
2341 return rc;
2342
2343 err_close_lock:
2344 kfree(cli->cl_close_lock);
2345 err_ptlrpcd_decref:
2346 ptlrpcd_decref();
2347 err_rpc_lock:
2348 kfree(cli->cl_rpc_lock);
2349 return rc;
2350 }
2351
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.
2359 */
2360 static int mdc_init_ea_size(struct obd_export *exp, int easize,
2361 int def_easize, int cookiesize, int def_cookiesize)
2362 {
2363 struct obd_device *obd = exp->exp_obd;
2364 struct client_obd *cli = &obd->u.cli;
2365
2366 if (cli->cl_max_mds_easize < easize)
2367 cli->cl_max_mds_easize = easize;
2368
2369 if (cli->cl_default_mds_easize < def_easize)
2370 cli->cl_default_mds_easize = def_easize;
2371
2372 if (cli->cl_max_mds_cookiesize < cookiesize)
2373 cli->cl_max_mds_cookiesize = cookiesize;
2374
2375 if (cli->cl_default_mds_cookiesize < def_cookiesize)
2376 cli->cl_default_mds_cookiesize = def_cookiesize;
2377
2378 return 0;
2379 }
2380
2381 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2382 {
2383 switch (stage) {
2384 case OBD_CLEANUP_EARLY:
2385 break;
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);
2390
2391 obd_cleanup_client_import(obd);
2392 ptlrpc_lprocfs_unregister_obd(obd);
2393 lprocfs_obd_cleanup(obd);
2394
2395 mdc_llog_finish(obd);
2396 break;
2397 }
2398 return 0;
2399 }
2400
2401 static int mdc_cleanup(struct obd_device *obd)
2402 {
2403 struct client_obd *cli = &obd->u.cli;
2404
2405 kfree(cli->cl_rpc_lock);
2406 kfree(cli->cl_close_lock);
2407
2408 ptlrpcd_decref();
2409
2410 return client_obd_cleanup(obd);
2411 }
2412
2413 static int mdc_process_config(struct obd_device *obd, u32 len, void *buf)
2414 {
2415 struct lustre_cfg *lcfg = buf;
2416 struct lprocfs_static_vars lvars = { NULL };
2417 int rc = 0;
2418
2419 lprocfs_mdc_init_vars(&lvars);
2420 switch (lcfg->lcfg_command) {
2421 default:
2422 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
2423 lcfg, obd);
2424 if (rc > 0)
2425 rc = 0;
2426 break;
2427 }
2428 return rc;
2429 }
2430
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)
2434 {
2435 struct ptlrpc_request *req;
2436 int rc;
2437
2438 LASSERT(client_is_remote(exp));
2439
2440 *request = NULL;
2441 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2442 if (!req)
2443 return -ENOMEM;
2444
2445 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2446 if (rc) {
2447 ptlrpc_request_free(req);
2448 return rc;
2449 }
2450
2451 mdc_pack_body(req, fid, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2452
2453 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2454 sizeof(struct mdt_remote_perm));
2455
2456 ptlrpc_request_set_replen(req);
2457
2458 rc = ptlrpc_queue_wait(req);
2459 if (rc)
2460 ptlrpc_req_finished(req);
2461 else
2462 *request = req;
2463 return rc;
2464 }
2465
2466 static struct obd_ops mdc_obd_ops = {
2467 .owner = THIS_MODULE,
2468 .setup = mdc_setup,
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
2487 };
2488
2489 static struct md_ops mdc_md_ops = {
2490 .getstatus = mdc_getstatus,
2491 .null_inode = mdc_null_inode,
2492 .find_cbdata = mdc_find_cbdata,
2493 .close = mdc_close,
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,
2500 .link = mdc_link,
2501 .is_subdir = mdc_is_subdir,
2502 .rename = mdc_rename,
2503 .setattr = mdc_setattr,
2504 .setxattr = mdc_setxattr,
2505 .getxattr = mdc_getxattr,
2506 .sync = mdc_sync,
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
2520 };
2521
2522 static int __init mdc_init(void)
2523 {
2524 struct lprocfs_static_vars lvars = { NULL };
2525
2526 lprocfs_mdc_init_vars(&lvars);
2527
2528 return class_register_type(&mdc_obd_ops, &mdc_md_ops,
2529 LUSTRE_MDC_NAME, NULL);
2530 }
2531
2532 static void /*__exit*/ mdc_exit(void)
2533 {
2534 class_unregister_type(LUSTRE_MDC_NAME);
2535 }
2536
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");
2541
2542 module_init(mdc_init);
2543 module_exit(mdc_exit);
Linux kernel - Deliverables
This page took 0.085497 seconds and 6 git commands to generate.