projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'powerpc-4.6-5' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
[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.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, 2015, 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 "../include/lustre_acl.h"
46 #include "../include/obd_class.h"
47 #include "../include/lustre_fid.h"
48 #include "../include/lprocfs_status.h"
49 #include "../include/lustre_param.h"
50 #include "../include/lustre_log.h"
51 #include "../include/lustre_kernelcomm.h"
52
53 #include "mdc_internal.h"
54
55 #define REQUEST_MINOR 244
56
57 static int mdc_cleanup(struct obd_device *obd);
58
59 static inline int mdc_queue_wait(struct ptlrpc_request *req)
60 {
61 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
62 int rc;
63
64 /* mdc_enter_request() ensures that this client has no more
65 * than cl_max_rpcs_in_flight RPCs simultaneously inf light
66 * against an MDT.
67 */
68 rc = mdc_enter_request(cli);
69 if (rc != 0)
70 return rc;
71
72 rc = ptlrpc_queue_wait(req);
73 mdc_exit_request(cli);
74
75 return rc;
76 }
77
78 static int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid)
79 {
80 struct ptlrpc_request *req;
81 struct mdt_body *body;
82 int rc;
83
84 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
85 &RQF_MDS_GETSTATUS,
86 LUSTRE_MDS_VERSION, MDS_GETSTATUS);
87 if (!req)
88 return -ENOMEM;
89
90 mdc_pack_body(req, NULL, 0, 0, -1, 0);
91 req->rq_send_state = LUSTRE_IMP_FULL;
92
93 ptlrpc_request_set_replen(req);
94
95 rc = ptlrpc_queue_wait(req);
96 if (rc)
97 goto out;
98
99 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
100 if (!body) {
101 rc = -EPROTO;
102 goto out;
103 }
104
105 *rootfid = body->fid1;
106 CDEBUG(D_NET,
107 "root fid="DFID", last_committed=%llu\n",
108 PFID(rootfid),
109 lustre_msg_get_last_committed(req->rq_repmsg));
110 out:
111 ptlrpc_req_finished(req);
112 return rc;
113 }
114
115 /*
116 * This function now is known to always saying that it will receive 4 buffers
117 * from server. Even for cases when acl_size and md_size is zero, RPC header
118 * will contain 4 fields and RPC itself will contain zero size fields. This is
119 * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
120 * and thus zero, it shrinks it, making zero size. The same story about
121 * md_size. And this is course of problem when client waits for smaller number
122 * of fields. This issue will be fixed later when client gets aware of RPC
123 * layouts. --umka
124 */
125 static int mdc_getattr_common(struct obd_export *exp,
126 struct ptlrpc_request *req)
127 {
128 struct req_capsule *pill = &req->rq_pill;
129 struct mdt_body *body;
130 void *eadata;
131 int rc;
132
133 /* Request message already built. */
134 rc = ptlrpc_queue_wait(req);
135 if (rc != 0)
136 return rc;
137
138 /* sanity check for the reply */
139 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
140 if (!body)
141 return -EPROTO;
142
143 CDEBUG(D_NET, "mode: %o\n", body->mode);
144
145 if (body->eadatasize != 0) {
146 mdc_update_max_ea_from_body(exp, body);
147
148 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
149 body->eadatasize);
150 if (!eadata)
151 return -EPROTO;
152 }
153
154 if (body->valid & OBD_MD_FLRMTPERM) {
155 struct mdt_remote_perm *perm;
156
157 LASSERT(client_is_remote(exp));
158 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
159 lustre_swab_mdt_remote_perm);
160 if (!perm)
161 return -EPROTO;
162 }
163
164 return 0;
165 }
166
167 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
168 struct ptlrpc_request **request)
169 {
170 struct ptlrpc_request *req;
171 int rc;
172
173 /* Single MDS without an LMV case */
174 if (op_data->op_flags & MF_GET_MDT_IDX) {
175 op_data->op_mds = 0;
176 return 0;
177 }
178 *request = NULL;
179 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
180 if (!req)
181 return -ENOMEM;
182
183 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
184 if (rc) {
185 ptlrpc_request_free(req);
186 return rc;
187 }
188
189 mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
190 op_data->op_mode, -1, 0);
191
192 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
193 op_data->op_mode);
194 if (op_data->op_valid & OBD_MD_FLRMTPERM) {
195 LASSERT(client_is_remote(exp));
196 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
197 sizeof(struct mdt_remote_perm));
198 }
199 ptlrpc_request_set_replen(req);
200
201 rc = mdc_getattr_common(exp, req);
202 if (rc)
203 ptlrpc_req_finished(req);
204 else
205 *request = req;
206 return rc;
207 }
208
209 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
210 struct ptlrpc_request **request)
211 {
212 struct ptlrpc_request *req;
213 int rc;
214
215 *request = NULL;
216 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
217 &RQF_MDS_GETATTR_NAME);
218 if (!req)
219 return -ENOMEM;
220
221 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
222 op_data->op_namelen + 1);
223
224 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
225 if (rc) {
226 ptlrpc_request_free(req);
227 return rc;
228 }
229
230 mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
231 op_data->op_mode, op_data->op_suppgids[0], 0);
232
233 if (op_data->op_name) {
234 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
235
236 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
237 op_data->op_namelen);
238 memcpy(name, op_data->op_name, op_data->op_namelen);
239 }
240
241 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
242 op_data->op_mode);
243 ptlrpc_request_set_replen(req);
244
245 rc = mdc_getattr_common(exp, req);
246 if (rc)
247 ptlrpc_req_finished(req);
248 else
249 *request = req;
250 return rc;
251 }
252
253 static int mdc_is_subdir(struct obd_export *exp,
254 const struct lu_fid *pfid,
255 const struct lu_fid *cfid,
256 struct ptlrpc_request **request)
257 {
258 struct ptlrpc_request *req;
259 int rc;
260
261 *request = NULL;
262 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
263 &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
264 MDS_IS_SUBDIR);
265 if (!req)
266 return -ENOMEM;
267
268 mdc_is_subdir_pack(req, pfid, cfid, 0);
269 ptlrpc_request_set_replen(req);
270
271 rc = ptlrpc_queue_wait(req);
272 if (rc && rc != -EREMOTE)
273 ptlrpc_req_finished(req);
274 else
275 *request = req;
276 return rc;
277 }
278
279 static int mdc_xattr_common(struct obd_export *exp,
280 const struct req_format *fmt,
281 const struct lu_fid *fid,
282 int opcode, u64 valid,
283 const char *xattr_name, const char *input,
284 int input_size, int output_size, int flags,
285 __u32 suppgid, struct ptlrpc_request **request)
286 {
287 struct ptlrpc_request *req;
288 int xattr_namelen = 0;
289 char *tmp;
290 int rc;
291
292 *request = NULL;
293 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
294 if (!req)
295 return -ENOMEM;
296
297 if (xattr_name) {
298 xattr_namelen = strlen(xattr_name) + 1;
299 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
300 xattr_namelen);
301 }
302 if (input_size) {
303 LASSERT(input);
304 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
305 input_size);
306 }
307
308 /* Flush local XATTR locks to get rid of a possible cancel RPC */
309 if (opcode == MDS_REINT && fid_is_sane(fid) &&
310 exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
311 LIST_HEAD(cancels);
312 int count;
313
314 /* Without that packing would fail */
315 if (input_size == 0)
316 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
317 RCL_CLIENT, 0);
318
319 count = mdc_resource_get_unused(exp, fid,
320 &cancels, LCK_EX,
321 MDS_INODELOCK_XATTR);
322
323 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
324 if (rc) {
325 ptlrpc_request_free(req);
326 return rc;
327 }
328 } else {
329 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
330 if (rc) {
331 ptlrpc_request_free(req);
332 return rc;
333 }
334 }
335
336 if (opcode == MDS_REINT) {
337 struct mdt_rec_setxattr *rec;
338
339 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
340 sizeof(struct mdt_rec_reint));
341 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
342 rec->sx_opcode = REINT_SETXATTR;
343 rec->sx_fsuid = from_kuid(&init_user_ns, current_fsuid());
344 rec->sx_fsgid = from_kgid(&init_user_ns, current_fsgid());
345 rec->sx_cap = cfs_curproc_cap_pack();
346 rec->sx_suppgid1 = suppgid;
347 rec->sx_suppgid2 = -1;
348 rec->sx_fid = *fid;
349 rec->sx_valid = valid | OBD_MD_FLCTIME;
350 rec->sx_time = ktime_get_real_seconds();
351 rec->sx_size = output_size;
352 rec->sx_flags = flags;
353
354 } else {
355 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
356 }
357
358 if (xattr_name) {
359 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
360 memcpy(tmp, xattr_name, xattr_namelen);
361 }
362 if (input_size) {
363 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
364 memcpy(tmp, input, input_size);
365 }
366
367 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
368 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
369 RCL_SERVER, output_size);
370 ptlrpc_request_set_replen(req);
371
372 /* make rpc */
373 if (opcode == MDS_REINT)
374 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
375
376 rc = ptlrpc_queue_wait(req);
377
378 if (opcode == MDS_REINT)
379 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
380
381 if (rc)
382 ptlrpc_req_finished(req);
383 else
384 *request = req;
385 return rc;
386 }
387
388 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
389 u64 valid, const char *xattr_name,
390 const char *input, int input_size, int output_size,
391 int flags, __u32 suppgid,
392 struct ptlrpc_request **request)
393 {
394 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
395 fid, MDS_REINT, valid, xattr_name,
396 input, input_size, output_size, flags,
397 suppgid, request);
398 }
399
400 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
401 u64 valid, const char *xattr_name,
402 const char *input, int input_size, int output_size,
403 int flags, struct ptlrpc_request **request)
404 {
405 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
406 fid, MDS_GETXATTR, valid, xattr_name,
407 input, input_size, output_size, flags,
408 -1, request);
409 }
410
411 #ifdef CONFIG_FS_POSIX_ACL
412 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
413 {
414 struct req_capsule *pill = &req->rq_pill;
415 struct mdt_body *body = md->body;
416 struct posix_acl *acl;
417 void *buf;
418 int rc;
419
420 if (!body->aclsize)
421 return 0;
422
423 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
424
425 if (!buf)
426 return -EPROTO;
427
428 acl = posix_acl_from_xattr(&init_user_ns, buf, body->aclsize);
429 if (!acl)
430 return 0;
431
432 if (IS_ERR(acl)) {
433 rc = PTR_ERR(acl);
434 CERROR("convert xattr to acl: %d\n", rc);
435 return rc;
436 }
437
438 rc = posix_acl_valid(acl);
439 if (rc) {
440 CERROR("validate acl: %d\n", rc);
441 posix_acl_release(acl);
442 return rc;
443 }
444
445 md->posix_acl = acl;
446 return 0;
447 }
448 #else
449 #define mdc_unpack_acl(req, md) 0
450 #endif
451
452 static int mdc_get_lustre_md(struct obd_export *exp,
453 struct ptlrpc_request *req,
454 struct obd_export *dt_exp,
455 struct obd_export *md_exp,
456 struct lustre_md *md)
457 {
458 struct req_capsule *pill = &req->rq_pill;
459 int rc;
460
461 LASSERT(md);
462 memset(md, 0, sizeof(*md));
463
464 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
465
466 if (md->body->valid & OBD_MD_FLEASIZE) {
467 int lmmsize;
468 struct lov_mds_md *lmm;
469
470 if (!S_ISREG(md->body->mode)) {
471 CDEBUG(D_INFO,
472 "OBD_MD_FLEASIZE set, should be a regular file, but is not\n");
473 rc = -EPROTO;
474 goto out;
475 }
476
477 if (md->body->eadatasize == 0) {
478 CDEBUG(D_INFO,
479 "OBD_MD_FLEASIZE set, but eadatasize 0\n");
480 rc = -EPROTO;
481 goto out;
482 }
483 lmmsize = md->body->eadatasize;
484 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
485 if (!lmm) {
486 rc = -EPROTO;
487 goto out;
488 }
489
490 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
491 if (rc < 0)
492 goto out;
493
494 if (rc < sizeof(*md->lsm)) {
495 CDEBUG(D_INFO,
496 "lsm size too small: rc < sizeof (*md->lsm) (%d < %d)\n",
497 rc, (int)sizeof(*md->lsm));
498 rc = -EPROTO;
499 goto out;
500 }
501
502 } else if (md->body->valid & OBD_MD_FLDIREA) {
503 int lmvsize;
504 struct lov_mds_md *lmv;
505
506 if (!S_ISDIR(md->body->mode)) {
507 CDEBUG(D_INFO,
508 "OBD_MD_FLDIREA set, should be a directory, but is not\n");
509 rc = -EPROTO;
510 goto out;
511 }
512
513 if (md->body->eadatasize == 0) {
514 CDEBUG(D_INFO,
515 "OBD_MD_FLDIREA is set, but eadatasize 0\n");
516 return -EPROTO;
517 }
518 if (md->body->valid & OBD_MD_MEA) {
519 lmvsize = md->body->eadatasize;
520 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
521 lmvsize);
522 if (!lmv) {
523 rc = -EPROTO;
524 goto out;
525 }
526
527 rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
528 lmvsize);
529 if (rc < 0)
530 goto out;
531
532 if (rc < sizeof(*md->mea)) {
533 CDEBUG(D_INFO,
534 "size too small: rc < sizeof(*md->mea) (%d < %d)\n",
535 rc, (int)sizeof(*md->mea));
536 rc = -EPROTO;
537 goto out;
538 }
539 }
540 }
541 rc = 0;
542
543 if (md->body->valid & OBD_MD_FLRMTPERM) {
544 /* remote permission */
545 LASSERT(client_is_remote(exp));
546 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
547 lustre_swab_mdt_remote_perm);
548 if (!md->remote_perm) {
549 rc = -EPROTO;
550 goto out;
551 }
552 } else if (md->body->valid & OBD_MD_FLACL) {
553 /* for ACL, it's possible that FLACL is set but aclsize is zero.
554 * only when aclsize != 0 there's an actual segment for ACL
555 * in reply buffer.
556 */
557 if (md->body->aclsize) {
558 rc = mdc_unpack_acl(req, md);
559 if (rc)
560 goto out;
561 #ifdef CONFIG_FS_POSIX_ACL
562 } else {
563 md->posix_acl = NULL;
564 #endif
565 }
566 }
567
568 out:
569 if (rc) {
570 #ifdef CONFIG_FS_POSIX_ACL
571 posix_acl_release(md->posix_acl);
572 #endif
573 if (md->lsm)
574 obd_free_memmd(dt_exp, &md->lsm);
575 }
576 return rc;
577 }
578
579 static int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
580 {
581 return 0;
582 }
583
584 /**
585 * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
586 * RPC chains.
587 */
588 void mdc_replay_open(struct ptlrpc_request *req)
589 {
590 struct md_open_data *mod = req->rq_cb_data;
591 struct ptlrpc_request *close_req;
592 struct obd_client_handle *och;
593 struct lustre_handle old;
594 struct mdt_body *body;
595
596 if (!mod) {
597 DEBUG_REQ(D_ERROR, req,
598 "Can't properly replay without open data.");
599 return;
600 }
601
602 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
603
604 och = mod->mod_och;
605 if (och) {
606 struct lustre_handle *file_fh;
607
608 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
609
610 file_fh = &och->och_fh;
611 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
612 file_fh->cookie, body->handle.cookie);
613 old = *file_fh;
614 *file_fh = body->handle;
615 }
616 close_req = mod->mod_close_req;
617 if (close_req) {
618 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
619 struct mdt_ioepoch *epoch;
620
621 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
622 epoch = req_capsule_client_get(&close_req->rq_pill,
623 &RMF_MDT_EPOCH);
624 LASSERT(epoch);
625
626 if (och)
627 LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
628 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
629 epoch->handle = body->handle;
630 }
631 }
632
633 void mdc_commit_open(struct ptlrpc_request *req)
634 {
635 struct md_open_data *mod = req->rq_cb_data;
636
637 if (!mod)
638 return;
639
640 /**
641 * No need to touch md_open_data::mod_och, it holds a reference on
642 * \var mod and will zero references to each other, \var mod will be
643 * freed after that when md_open_data::mod_och will put the reference.
644 */
645
646 /**
647 * Do not let open request to disappear as it still may be needed
648 * for close rpc to happen (it may happen on evict only, otherwise
649 * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
650 * called), just mark this rpc as committed to distinguish these 2
651 * cases, see mdc_close() for details. The open request reference will
652 * be put along with freeing \var mod.
653 */
654 ptlrpc_request_addref(req);
655 spin_lock(&req->rq_lock);
656 req->rq_committed = 1;
657 spin_unlock(&req->rq_lock);
658 req->rq_cb_data = NULL;
659 obd_mod_put(mod);
660 }
661
662 int mdc_set_open_replay_data(struct obd_export *exp,
663 struct obd_client_handle *och,
664 struct lookup_intent *it)
665 {
666 struct md_open_data *mod;
667 struct mdt_rec_create *rec;
668 struct mdt_body *body;
669 struct ptlrpc_request *open_req = it->d.lustre.it_data;
670 struct obd_import *imp = open_req->rq_import;
671
672 if (!open_req->rq_replay)
673 return 0;
674
675 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
676 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
677 LASSERT(rec);
678 /* Incoming message in my byte order (it's been swabbed). */
679 /* Outgoing messages always in my byte order. */
680 LASSERT(body);
681
682 /* Only if the import is replayable, we set replay_open data */
683 if (och && imp->imp_replayable) {
684 mod = obd_mod_alloc();
685 if (!mod) {
686 DEBUG_REQ(D_ERROR, open_req,
687 "Can't allocate md_open_data");
688 return 0;
689 }
690
691 /**
692 * Take a reference on \var mod, to be freed on mdc_close().
693 * It protects \var mod from being freed on eviction (commit
694 * callback is called despite rq_replay flag).
695 * Another reference for \var och.
696 */
697 obd_mod_get(mod);
698 obd_mod_get(mod);
699
700 spin_lock(&open_req->rq_lock);
701 och->och_mod = mod;
702 mod->mod_och = och;
703 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
704 it_disposition(it, DISP_OPEN_STRIPE);
705 mod->mod_open_req = open_req;
706 open_req->rq_cb_data = mod;
707 open_req->rq_commit_cb = mdc_commit_open;
708 spin_unlock(&open_req->rq_lock);
709 }
710
711 rec->cr_fid2 = body->fid1;
712 rec->cr_ioepoch = body->ioepoch;
713 rec->cr_old_handle.cookie = body->handle.cookie;
714 open_req->rq_replay_cb = mdc_replay_open;
715 if (!fid_is_sane(&body->fid1)) {
716 DEBUG_REQ(D_ERROR, open_req,
717 "Saving replay request with insane fid");
718 LBUG();
719 }
720
721 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
722 return 0;
723 }
724
725 static void mdc_free_open(struct md_open_data *mod)
726 {
727 int committed = 0;
728
729 if (mod->mod_is_create == 0 &&
730 imp_connect_disp_stripe(mod->mod_open_req->rq_import))
731 committed = 1;
732
733 LASSERT(mod->mod_open_req->rq_replay == 0);
734
735 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
736
737 ptlrpc_request_committed(mod->mod_open_req, committed);
738 if (mod->mod_close_req)
739 ptlrpc_request_committed(mod->mod_close_req, committed);
740 }
741
742 static int mdc_clear_open_replay_data(struct obd_export *exp,
743 struct obd_client_handle *och)
744 {
745 struct md_open_data *mod = och->och_mod;
746
747 /**
748 * It is possible to not have \var mod in a case of eviction between
749 * lookup and ll_file_open().
750 **/
751 if (!mod)
752 return 0;
753
754 LASSERT(mod != LP_POISON);
755 LASSERT(mod->mod_open_req);
756 mdc_free_open(mod);
757
758 mod->mod_och = NULL;
759 och->och_mod = NULL;
760 obd_mod_put(mod);
761
762 return 0;
763 }
764
765 /* Prepares the request for the replay by the given reply */
766 static void mdc_close_handle_reply(struct ptlrpc_request *req,
767 struct md_op_data *op_data, int rc) {
768 struct mdt_body *repbody;
769 struct mdt_ioepoch *epoch;
770
771 if (req && rc == -EAGAIN) {
772 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
773 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
774
775 epoch->flags |= MF_SOM_AU;
776 if (repbody->valid & OBD_MD_FLGETATTRLOCK)
777 op_data->op_flags |= MF_GETATTR_LOCK;
778 }
779 }
780
781 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
782 struct md_open_data *mod, struct ptlrpc_request **request)
783 {
784 struct obd_device *obd = class_exp2obd(exp);
785 struct ptlrpc_request *req;
786 struct req_format *req_fmt;
787 int rc;
788 int saved_rc = 0;
789
790 req_fmt = &RQF_MDS_CLOSE;
791 if (op_data->op_bias & MDS_HSM_RELEASE) {
792 req_fmt = &RQF_MDS_RELEASE_CLOSE;
793
794 /* allocate a FID for volatile file */
795 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
796 if (rc < 0) {
797 CERROR("%s: "DFID" failed to allocate FID: %d\n",
798 obd->obd_name, PFID(&op_data->op_fid1), rc);
799 /* save the errcode and proceed to close */
800 saved_rc = rc;
801 }
802 }
803
804 *request = NULL;
805 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
806 if (!req)
807 return -ENOMEM;
808
809 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
810 if (rc) {
811 ptlrpc_request_free(req);
812 return rc;
813 }
814
815 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
816 * portal whose threads are not taking any DLM locks and are therefore
817 * always progressing
818 */
819 req->rq_request_portal = MDS_READPAGE_PORTAL;
820 ptlrpc_at_set_req_timeout(req);
821
822 /* Ensure that this close's handle is fixed up during replay. */
823 if (likely(mod)) {
824 LASSERTF(mod->mod_open_req &&
825 mod->mod_open_req->rq_type != LI_POISON,
826 "POISONED open %p!\n", mod->mod_open_req);
827
828 mod->mod_close_req = req;
829
830 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
831 /* We no longer want to preserve this open for replay even
832 * though the open was committed. b=3632, b=3633
833 */
834 spin_lock(&mod->mod_open_req->rq_lock);
835 mod->mod_open_req->rq_replay = 0;
836 spin_unlock(&mod->mod_open_req->rq_lock);
837 } else {
838 CDEBUG(D_HA,
839 "couldn't find open req; expecting close error\n");
840 }
841
842 mdc_close_pack(req, op_data);
843
844 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
845 obd->u.cli.cl_default_mds_easize);
846 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
847 obd->u.cli.cl_default_mds_cookiesize);
848
849 ptlrpc_request_set_replen(req);
850
851 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
852 rc = ptlrpc_queue_wait(req);
853 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
854
855 if (!req->rq_repmsg) {
856 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
857 req->rq_status);
858 if (rc == 0)
859 rc = req->rq_status ?: -EIO;
860 } else if (rc == 0 || rc == -EAGAIN) {
861 struct mdt_body *body;
862
863 rc = lustre_msg_get_status(req->rq_repmsg);
864 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
865 DEBUG_REQ(D_ERROR, req,
866 "type == PTL_RPC_MSG_ERR, err = %d", rc);
867 if (rc > 0)
868 rc = -rc;
869 }
870 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
871 if (!body)
872 rc = -EPROTO;
873 } else if (rc == -ESTALE) {
874 /**
875 * it can be allowed error after 3633 if open was committed and
876 * server failed before close was sent. Let's check if mod
877 * exists and return no error in that case
878 */
879 if (mod) {
880 DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
881 if (mod->mod_open_req->rq_committed)
882 rc = 0;
883 }
884 }
885
886 if (mod) {
887 if (rc != 0)
888 mod->mod_close_req = NULL;
889 /* Since now, mod is accessed through open_req only,
890 * thus close req does not keep a reference on mod anymore.
891 */
892 obd_mod_put(mod);
893 }
894 *request = req;
895 mdc_close_handle_reply(req, op_data, rc);
896 return rc < 0 ? rc : saved_rc;
897 }
898
899 static int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
900 struct md_open_data *mod)
901 {
902 struct obd_device *obd = class_exp2obd(exp);
903 struct ptlrpc_request *req;
904 int rc;
905
906 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
907 &RQF_MDS_DONE_WRITING);
908 if (!req)
909 return -ENOMEM;
910
911 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
912 if (rc) {
913 ptlrpc_request_free(req);
914 return rc;
915 }
916
917 if (mod) {
918 LASSERTF(mod->mod_open_req &&
919 mod->mod_open_req->rq_type != LI_POISON,
920 "POISONED setattr %p!\n", mod->mod_open_req);
921
922 mod->mod_close_req = req;
923 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
924 /* We no longer want to preserve this setattr for replay even
925 * though the open was committed. b=3632, b=3633
926 */
927 spin_lock(&mod->mod_open_req->rq_lock);
928 mod->mod_open_req->rq_replay = 0;
929 spin_unlock(&mod->mod_open_req->rq_lock);
930 }
931
932 mdc_close_pack(req, op_data);
933 ptlrpc_request_set_replen(req);
934
935 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
936 rc = ptlrpc_queue_wait(req);
937 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
938
939 if (rc == -ESTALE) {
940 /**
941 * it can be allowed error after 3633 if open or setattr were
942 * committed and server failed before close was sent.
943 * Let's check if mod exists and return no error in that case
944 */
945 if (mod) {
946 if (mod->mod_open_req->rq_committed)
947 rc = 0;
948 }
949 }
950
951 if (mod) {
952 if (rc != 0)
953 mod->mod_close_req = NULL;
954 LASSERT(mod->mod_open_req);
955 mdc_free_open(mod);
956
957 /* Since now, mod is accessed through setattr req only,
958 * thus DW req does not keep a reference on mod anymore.
959 */
960 obd_mod_put(mod);
961 }
962
963 mdc_close_handle_reply(req, op_data, rc);
964 ptlrpc_req_finished(req);
965 return rc;
966 }
967
968 static int mdc_readpage(struct obd_export *exp, struct md_op_data *op_data,
969 struct page **pages, struct ptlrpc_request **request)
970 {
971 struct ptlrpc_request *req;
972 struct ptlrpc_bulk_desc *desc;
973 int i;
974 wait_queue_head_t waitq;
975 int resends = 0;
976 struct l_wait_info lwi;
977 int rc;
978
979 *request = NULL;
980 init_waitqueue_head(&waitq);
981
982 restart_bulk:
983 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
984 if (!req)
985 return -ENOMEM;
986
987 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
988 if (rc) {
989 ptlrpc_request_free(req);
990 return rc;
991 }
992
993 req->rq_request_portal = MDS_READPAGE_PORTAL;
994 ptlrpc_at_set_req_timeout(req);
995
996 desc = ptlrpc_prep_bulk_imp(req, op_data->op_npages, 1, BULK_PUT_SINK,
997 MDS_BULK_PORTAL);
998 if (!desc) {
999 ptlrpc_request_free(req);
1000 return -ENOMEM;
1001 }
1002
1003 /* NB req now owns desc and will free it when it gets freed */
1004 for (i = 0; i < op_data->op_npages; i++)
1005 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_SIZE);
1006
1007 mdc_readdir_pack(req, op_data->op_offset,
1008 PAGE_SIZE * op_data->op_npages,
1009 &op_data->op_fid1);
1010
1011 ptlrpc_request_set_replen(req);
1012 rc = ptlrpc_queue_wait(req);
1013 if (rc) {
1014 ptlrpc_req_finished(req);
1015 if (rc != -ETIMEDOUT)
1016 return rc;
1017
1018 resends++;
1019 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1020 CERROR("too many resend retries, returning error\n");
1021 return -EIO;
1022 }
1023 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends),
1024 NULL, NULL, NULL);
1025 l_wait_event(waitq, 0, &lwi);
1026
1027 goto restart_bulk;
1028 }
1029
1030 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1031 req->rq_bulk->bd_nob_transferred);
1032 if (rc < 0) {
1033 ptlrpc_req_finished(req);
1034 return rc;
1035 }
1036
1037 if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1038 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1039 req->rq_bulk->bd_nob_transferred,
1040 PAGE_SIZE * op_data->op_npages);
1041 ptlrpc_req_finished(req);
1042 return -EPROTO;
1043 }
1044
1045 *request = req;
1046 return 0;
1047 }
1048
1049 static int mdc_statfs(const struct lu_env *env,
1050 struct obd_export *exp, struct obd_statfs *osfs,
1051 __u64 max_age, __u32 flags)
1052 {
1053 struct obd_device *obd = class_exp2obd(exp);
1054 struct ptlrpc_request *req;
1055 struct obd_statfs *msfs;
1056 struct obd_import *imp = NULL;
1057 int rc;
1058
1059 /*
1060 * Since the request might also come from lprocfs, so we need
1061 * sync this with client_disconnect_export Bug15684
1062 */
1063 down_read(&obd->u.cli.cl_sem);
1064 if (obd->u.cli.cl_import)
1065 imp = class_import_get(obd->u.cli.cl_import);
1066 up_read(&obd->u.cli.cl_sem);
1067 if (!imp)
1068 return -ENODEV;
1069
1070 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1071 LUSTRE_MDS_VERSION, MDS_STATFS);
1072 if (!req) {
1073 rc = -ENOMEM;
1074 goto output;
1075 }
1076
1077 ptlrpc_request_set_replen(req);
1078
1079 if (flags & OBD_STATFS_NODELAY) {
1080 /* procfs requests not want stay in wait for avoid deadlock */
1081 req->rq_no_resend = 1;
1082 req->rq_no_delay = 1;
1083 }
1084
1085 rc = ptlrpc_queue_wait(req);
1086 if (rc) {
1087 /* check connection error first */
1088 if (imp->imp_connect_error)
1089 rc = imp->imp_connect_error;
1090 goto out;
1091 }
1092
1093 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1094 if (!msfs) {
1095 rc = -EPROTO;
1096 goto out;
1097 }
1098
1099 *osfs = *msfs;
1100 out:
1101 ptlrpc_req_finished(req);
1102 output:
1103 class_import_put(imp);
1104 return rc;
1105 }
1106
1107 static 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 key = kzalloc(keylen, GFP_NOFS);
1121 if (!key)
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 %llu #%d\n",
1127 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1128
1129 if (!fid_is_sane(&gf->gf_fid)) {
1130 rc = -EINVAL;
1131 goto out;
1132 }
1133
1134 /* Val is struct getinfo_fid2path result plus path */
1135 vallen = sizeof(*gf) + gf->gf_pathlen;
1136
1137 rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1138 if (rc != 0 && rc != -EREMOTE)
1139 goto out;
1140
1141 if (vallen <= sizeof(*gf)) {
1142 rc = -EPROTO;
1143 goto out;
1144 } else if (vallen > sizeof(*gf) + gf->gf_pathlen) {
1145 rc = -EOVERFLOW;
1146 goto out;
1147 }
1148
1149 CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n%s\n",
1150 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1151
1152 out:
1153 kfree(key);
1154 return rc;
1155 }
1156
1157 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1158 struct hsm_progress_kernel *hpk)
1159 {
1160 struct obd_import *imp = class_exp2cliimp(exp);
1161 struct hsm_progress_kernel *req_hpk;
1162 struct ptlrpc_request *req;
1163 int rc;
1164
1165 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1166 LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1167 if (!req) {
1168 rc = -ENOMEM;
1169 goto out;
1170 }
1171
1172 mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1173
1174 /* Copy hsm_progress struct */
1175 req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1176 if (!req_hpk) {
1177 rc = -EPROTO;
1178 goto out;
1179 }
1180
1181 *req_hpk = *hpk;
1182 req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1183
1184 ptlrpc_request_set_replen(req);
1185
1186 rc = mdc_queue_wait(req);
1187 out:
1188 ptlrpc_req_finished(req);
1189 return rc;
1190 }
1191
1192 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1193 {
1194 __u32 *archive_mask;
1195 struct ptlrpc_request *req;
1196 int rc;
1197
1198 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1199 LUSTRE_MDS_VERSION,
1200 MDS_HSM_CT_REGISTER);
1201 if (!req) {
1202 rc = -ENOMEM;
1203 goto out;
1204 }
1205
1206 mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1207
1208 /* Copy hsm_progress struct */
1209 archive_mask = req_capsule_client_get(&req->rq_pill,
1210 &RMF_MDS_HSM_ARCHIVE);
1211 if (!archive_mask) {
1212 rc = -EPROTO;
1213 goto out;
1214 }
1215
1216 *archive_mask = archives;
1217
1218 ptlrpc_request_set_replen(req);
1219
1220 rc = mdc_queue_wait(req);
1221 out:
1222 ptlrpc_req_finished(req);
1223 return rc;
1224 }
1225
1226 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1227 struct md_op_data *op_data)
1228 {
1229 struct hsm_current_action *hca = op_data->op_data;
1230 struct hsm_current_action *req_hca;
1231 struct ptlrpc_request *req;
1232 int rc;
1233
1234 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1235 &RQF_MDS_HSM_ACTION);
1236 if (!req)
1237 return -ENOMEM;
1238
1239 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1240 if (rc) {
1241 ptlrpc_request_free(req);
1242 return rc;
1243 }
1244
1245 mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1246 op_data->op_suppgids[0], 0);
1247
1248 ptlrpc_request_set_replen(req);
1249
1250 rc = mdc_queue_wait(req);
1251 if (rc)
1252 goto out;
1253
1254 req_hca = req_capsule_server_get(&req->rq_pill,
1255 &RMF_MDS_HSM_CURRENT_ACTION);
1256 if (!req_hca) {
1257 rc = -EPROTO;
1258 goto out;
1259 }
1260
1261 *hca = *req_hca;
1262
1263 out:
1264 ptlrpc_req_finished(req);
1265 return rc;
1266 }
1267
1268 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1269 {
1270 struct ptlrpc_request *req;
1271 int rc;
1272
1273 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1274 LUSTRE_MDS_VERSION,
1275 MDS_HSM_CT_UNREGISTER);
1276 if (!req) {
1277 rc = -ENOMEM;
1278 goto out;
1279 }
1280
1281 mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1282
1283 ptlrpc_request_set_replen(req);
1284
1285 rc = mdc_queue_wait(req);
1286 out:
1287 ptlrpc_req_finished(req);
1288 return rc;
1289 }
1290
1291 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1292 struct md_op_data *op_data)
1293 {
1294 struct hsm_user_state *hus = op_data->op_data;
1295 struct hsm_user_state *req_hus;
1296 struct ptlrpc_request *req;
1297 int rc;
1298
1299 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1300 &RQF_MDS_HSM_STATE_GET);
1301 if (!req)
1302 return -ENOMEM;
1303
1304 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1305 if (rc != 0) {
1306 ptlrpc_request_free(req);
1307 return rc;
1308 }
1309
1310 mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1311 op_data->op_suppgids[0], 0);
1312
1313 ptlrpc_request_set_replen(req);
1314
1315 rc = mdc_queue_wait(req);
1316 if (rc)
1317 goto out;
1318
1319 req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1320 if (!req_hus) {
1321 rc = -EPROTO;
1322 goto out;
1323 }
1324
1325 *hus = *req_hus;
1326
1327 out:
1328 ptlrpc_req_finished(req);
1329 return rc;
1330 }
1331
1332 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1333 struct md_op_data *op_data)
1334 {
1335 struct hsm_state_set *hss = op_data->op_data;
1336 struct hsm_state_set *req_hss;
1337 struct ptlrpc_request *req;
1338 int rc;
1339
1340 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1341 &RQF_MDS_HSM_STATE_SET);
1342 if (!req)
1343 return -ENOMEM;
1344
1345 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1346 if (rc) {
1347 ptlrpc_request_free(req);
1348 return rc;
1349 }
1350
1351 mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1352 op_data->op_suppgids[0], 0);
1353
1354 /* Copy states */
1355 req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1356 if (!req_hss) {
1357 rc = -EPROTO;
1358 goto out;
1359 }
1360 *req_hss = *hss;
1361
1362 ptlrpc_request_set_replen(req);
1363
1364 rc = mdc_queue_wait(req);
1365 out:
1366 ptlrpc_req_finished(req);
1367 return rc;
1368 }
1369
1370 static int mdc_ioc_hsm_request(struct obd_export *exp,
1371 struct hsm_user_request *hur)
1372 {
1373 struct obd_import *imp = class_exp2cliimp(exp);
1374 struct ptlrpc_request *req;
1375 struct hsm_request *req_hr;
1376 struct hsm_user_item *req_hui;
1377 char *req_opaque;
1378 int rc;
1379
1380 req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1381 if (!req) {
1382 rc = -ENOMEM;
1383 goto out;
1384 }
1385
1386 req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1387 hur->hur_request.hr_itemcount
1388 * sizeof(struct hsm_user_item));
1389 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1390 hur->hur_request.hr_data_len);
1391
1392 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1393 if (rc) {
1394 ptlrpc_request_free(req);
1395 return rc;
1396 }
1397
1398 mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1399
1400 /* Copy hsm_request struct */
1401 req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1402 if (!req_hr) {
1403 rc = -EPROTO;
1404 goto out;
1405 }
1406 *req_hr = hur->hur_request;
1407
1408 /* Copy hsm_user_item structs */
1409 req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1410 if (!req_hui) {
1411 rc = -EPROTO;
1412 goto out;
1413 }
1414 memcpy(req_hui, hur->hur_user_item,
1415 hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1416
1417 /* Copy opaque field */
1418 req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1419 if (!req_opaque) {
1420 rc = -EPROTO;
1421 goto out;
1422 }
1423 memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1424
1425 ptlrpc_request_set_replen(req);
1426
1427 rc = mdc_queue_wait(req);
1428 out:
1429 ptlrpc_req_finished(req);
1430 return rc;
1431 }
1432
1433 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
1434 {
1435 struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1436
1437 LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1438
1439 lh->kuc_magic = KUC_MAGIC;
1440 lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1441 lh->kuc_flags = flags;
1442 lh->kuc_msgtype = CL_RECORD;
1443 lh->kuc_msglen = len;
1444 return lh;
1445 }
1446
1447 #define D_CHANGELOG 0
1448
1449 struct changelog_show {
1450 __u64 cs_startrec;
1451 __u32 cs_flags;
1452 struct file *cs_fp;
1453 char *cs_buf;
1454 struct obd_device *cs_obd;
1455 };
1456
1457 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1458 struct llog_rec_hdr *hdr, void *data)
1459 {
1460 struct changelog_show *cs = data;
1461 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1462 struct kuc_hdr *lh;
1463 int len, rc;
1464
1465 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1466 rc = -EINVAL;
1467 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1468 cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
1469 rec->cr.cr_type, rc);
1470 return rc;
1471 }
1472
1473 if (rec->cr.cr_index < cs->cs_startrec) {
1474 /* Skip entries earlier than what we are interested in */
1475 CDEBUG(D_CHANGELOG, "rec=%llu start=%llu\n",
1476 rec->cr.cr_index, cs->cs_startrec);
1477 return 0;
1478 }
1479
1480 CDEBUG(D_CHANGELOG, "%llu %02d%-5s %llu 0x%x t="DFID" p="DFID
1481 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
1482 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1483 rec->cr.cr_flags & CLF_FLAGMASK,
1484 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1485 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1486
1487 len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1488
1489 /* Set up the message */
1490 lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1491 memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1492
1493 rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1494 CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len, rc);
1495
1496 return rc;
1497 }
1498
1499 static int mdc_changelog_send_thread(void *csdata)
1500 {
1501 struct changelog_show *cs = csdata;
1502 struct llog_ctxt *ctxt = NULL;
1503 struct llog_handle *llh = NULL;
1504 struct kuc_hdr *kuch;
1505 int rc;
1506
1507 CDEBUG(D_CHANGELOG, "changelog to fp=%p start %llu\n",
1508 cs->cs_fp, cs->cs_startrec);
1509
1510 cs->cs_buf = kzalloc(KUC_CHANGELOG_MSG_MAXSIZE, GFP_NOFS);
1511 if (!cs->cs_buf) {
1512 rc = -ENOMEM;
1513 goto out;
1514 }
1515
1516 /* Set up the remote catalog handle */
1517 ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1518 if (!ctxt) {
1519 rc = -ENOENT;
1520 goto out;
1521 }
1522 rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1523 LLOG_OPEN_EXISTS);
1524 if (rc) {
1525 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1526 cs->cs_obd->obd_name, rc);
1527 goto out;
1528 }
1529 rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
1530 if (rc) {
1531 CERROR("llog_init_handle failed %d\n", rc);
1532 goto out;
1533 }
1534
1535 rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1536
1537 /* Send EOF no matter what our result */
1538 kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags);
1539 if (kuch) {
1540 kuch->kuc_msgtype = CL_EOF;
1541 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1542 }
1543
1544 out:
1545 fput(cs->cs_fp);
1546 if (llh)
1547 llog_cat_close(NULL, llh);
1548 if (ctxt)
1549 llog_ctxt_put(ctxt);
1550 kfree(cs->cs_buf);
1551 kfree(cs);
1552 return rc;
1553 }
1554
1555 static int mdc_ioc_changelog_send(struct obd_device *obd,
1556 struct ioc_changelog *icc)
1557 {
1558 struct changelog_show *cs;
1559 struct task_struct *task;
1560 int rc;
1561
1562 /* Freed in mdc_changelog_send_thread */
1563 cs = kzalloc(sizeof(*cs), GFP_NOFS);
1564 if (!cs)
1565 return -ENOMEM;
1566
1567 cs->cs_obd = obd;
1568 cs->cs_startrec = icc->icc_recno;
1569 /* matching fput in mdc_changelog_send_thread */
1570 cs->cs_fp = fget(icc->icc_id);
1571 cs->cs_flags = icc->icc_flags;
1572
1573 /*
1574 * New thread because we should return to user app before
1575 * writing into our pipe
1576 */
1577 task = kthread_run(mdc_changelog_send_thread, cs,
1578 "mdc_clg_send_thread");
1579 if (IS_ERR(task)) {
1580 rc = PTR_ERR(task);
1581 CERROR("%s: can't start changelog thread: rc = %d\n",
1582 obd->obd_name, rc);
1583 kfree(cs);
1584 } else {
1585 rc = 0;
1586 CDEBUG(D_CHANGELOG, "%s: started changelog thread\n",
1587 obd->obd_name);
1588 }
1589
1590 CERROR("Failed to start changelog thread: %d\n", rc);
1591 return rc;
1592 }
1593
1594 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1595 struct lustre_kernelcomm *lk);
1596
1597 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
1598 struct obd_quotactl *oqctl)
1599 {
1600 struct client_obd *cli = &exp->exp_obd->u.cli;
1601 struct ptlrpc_request *req;
1602 struct obd_quotactl *body;
1603 int rc;
1604
1605 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1606 &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
1607 MDS_QUOTACHECK);
1608 if (!req)
1609 return -ENOMEM;
1610
1611 body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1612 *body = *oqctl;
1613
1614 ptlrpc_request_set_replen(req);
1615
1616 /* the next poll will find -ENODATA, that means quotacheck is
1617 * going on
1618 */
1619 cli->cl_qchk_stat = -ENODATA;
1620 rc = ptlrpc_queue_wait(req);
1621 if (rc)
1622 cli->cl_qchk_stat = rc;
1623 ptlrpc_req_finished(req);
1624 return rc;
1625 }
1626
1627 static int mdc_quota_poll_check(struct obd_export *exp,
1628 struct if_quotacheck *qchk)
1629 {
1630 struct client_obd *cli = &exp->exp_obd->u.cli;
1631 int rc;
1632
1633 qchk->obd_uuid = cli->cl_target_uuid;
1634 memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
1635
1636 rc = cli->cl_qchk_stat;
1637 /* the client is not the previous one */
1638 if (rc == CL_NOT_QUOTACHECKED)
1639 rc = -EINTR;
1640 return rc;
1641 }
1642
1643 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1644 struct obd_quotactl *oqctl)
1645 {
1646 struct ptlrpc_request *req;
1647 struct obd_quotactl *oqc;
1648 int rc;
1649
1650 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1651 &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1652 MDS_QUOTACTL);
1653 if (!req)
1654 return -ENOMEM;
1655
1656 oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1657 *oqc = *oqctl;
1658
1659 ptlrpc_request_set_replen(req);
1660 ptlrpc_at_set_req_timeout(req);
1661 req->rq_no_resend = 1;
1662
1663 rc = ptlrpc_queue_wait(req);
1664 if (rc)
1665 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1666
1667 if (req->rq_repmsg) {
1668 oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1669 if (oqc) {
1670 *oqctl = *oqc;
1671 } else if (!rc) {
1672 CERROR("Can't unpack obd_quotactl\n");
1673 rc = -EPROTO;
1674 }
1675 } else if (!rc) {
1676 CERROR("Can't unpack obd_quotactl\n");
1677 rc = -EPROTO;
1678 }
1679 ptlrpc_req_finished(req);
1680
1681 return rc;
1682 }
1683
1684 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1685 struct md_op_data *op_data)
1686 {
1687 LIST_HEAD(cancels);
1688 struct ptlrpc_request *req;
1689 int rc, count;
1690 struct mdc_swap_layouts *msl, *payload;
1691
1692 msl = op_data->op_data;
1693
1694 /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1695 * first thing it will do is to cancel the 2 layout
1696 * locks hold by this client.
1697 * So the client must cancel its layout locks on the 2 fids
1698 * with the request RPC to avoid extra RPC round trips
1699 */
1700 count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1701 LCK_CR, MDS_INODELOCK_LAYOUT);
1702 count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1703 LCK_CR, MDS_INODELOCK_LAYOUT);
1704
1705 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1706 &RQF_MDS_SWAP_LAYOUTS);
1707 if (!req) {
1708 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1709 return -ENOMEM;
1710 }
1711
1712 rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1713 if (rc) {
1714 ptlrpc_request_free(req);
1715 return rc;
1716 }
1717
1718 mdc_swap_layouts_pack(req, op_data);
1719
1720 payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1721 LASSERT(payload);
1722
1723 *payload = *msl;
1724
1725 ptlrpc_request_set_replen(req);
1726
1727 rc = ptlrpc_queue_wait(req);
1728
1729 ptlrpc_req_finished(req);
1730 return rc;
1731 }
1732
1733 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1734 void *karg, void __user *uarg)
1735 {
1736 struct obd_device *obd = exp->exp_obd;
1737 struct obd_ioctl_data *data = karg;
1738 struct obd_import *imp = obd->u.cli.cl_import;
1739 int rc;
1740
1741 if (!try_module_get(THIS_MODULE)) {
1742 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
1743 module_name(THIS_MODULE));
1744 return -EINVAL;
1745 }
1746 switch (cmd) {
1747 case OBD_IOC_CHANGELOG_SEND:
1748 rc = mdc_ioc_changelog_send(obd, karg);
1749 goto out;
1750 case OBD_IOC_CHANGELOG_CLEAR: {
1751 struct ioc_changelog *icc = karg;
1752 struct changelog_setinfo cs = {
1753 .cs_recno = icc->icc_recno,
1754 .cs_id = icc->icc_id
1755 };
1756
1757 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
1758 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1759 NULL);
1760 goto out;
1761 }
1762 case OBD_IOC_FID2PATH:
1763 rc = mdc_ioc_fid2path(exp, karg);
1764 goto out;
1765 case LL_IOC_HSM_CT_START:
1766 rc = mdc_ioc_hsm_ct_start(exp, karg);
1767 /* ignore if it was already registered on this MDS. */
1768 if (rc == -EEXIST)
1769 rc = 0;
1770 goto out;
1771 case LL_IOC_HSM_PROGRESS:
1772 rc = mdc_ioc_hsm_progress(exp, karg);
1773 goto out;
1774 case LL_IOC_HSM_STATE_GET:
1775 rc = mdc_ioc_hsm_state_get(exp, karg);
1776 goto out;
1777 case LL_IOC_HSM_STATE_SET:
1778 rc = mdc_ioc_hsm_state_set(exp, karg);
1779 goto out;
1780 case LL_IOC_HSM_ACTION:
1781 rc = mdc_ioc_hsm_current_action(exp, karg);
1782 goto out;
1783 case LL_IOC_HSM_REQUEST:
1784 rc = mdc_ioc_hsm_request(exp, karg);
1785 goto out;
1786 case OBD_IOC_CLIENT_RECOVER:
1787 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1788 if (rc < 0)
1789 goto out;
1790 rc = 0;
1791 goto out;
1792 case IOC_OSC_SET_ACTIVE:
1793 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1794 goto out;
1795 case OBD_IOC_POLL_QUOTACHECK:
1796 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
1797 goto out;
1798 case OBD_IOC_PING_TARGET:
1799 rc = ptlrpc_obd_ping(obd);
1800 goto out;
1801 /*
1802 * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1803 * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1804 * there'd be no LMV layer thus we might be called here. Eventually
1805 * this code should be removed.
1806 * bz20731, LU-592.
1807 */
1808 case IOC_OBD_STATFS: {
1809 struct obd_statfs stat_buf = {0};
1810
1811 if (*((__u32 *) data->ioc_inlbuf2) != 0) {
1812 rc = -ENODEV;
1813 goto out;
1814 }
1815
1816 /* copy UUID */
1817 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1818 min_t(size_t, data->ioc_plen2,
1819 sizeof(struct obd_uuid)))) {
1820 rc = -EFAULT;
1821 goto out;
1822 }
1823
1824 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1825 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1826 0);
1827 if (rc != 0)
1828 goto out;
1829
1830 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1831 min_t(size_t, data->ioc_plen1,
1832 sizeof(stat_buf)))) {
1833 rc = -EFAULT;
1834 goto out;
1835 }
1836
1837 rc = 0;
1838 goto out;
1839 }
1840 case OBD_IOC_QUOTACTL: {
1841 struct if_quotactl *qctl = karg;
1842 struct obd_quotactl *oqctl;
1843
1844 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1845 if (!oqctl) {
1846 rc = -ENOMEM;
1847 goto out;
1848 }
1849
1850 QCTL_COPY(oqctl, qctl);
1851 rc = obd_quotactl(exp, oqctl);
1852 if (rc == 0) {
1853 QCTL_COPY(qctl, oqctl);
1854 qctl->qc_valid = QC_MDTIDX;
1855 qctl->obd_uuid = obd->u.cli.cl_target_uuid;
1856 }
1857
1858 kfree(oqctl);
1859 goto out;
1860 }
1861 case LL_IOC_GET_CONNECT_FLAGS:
1862 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
1863 sizeof(*exp_connect_flags_ptr(exp)))) {
1864 rc = -EFAULT;
1865 goto out;
1866 }
1867
1868 rc = 0;
1869 goto out;
1870 case LL_IOC_LOV_SWAP_LAYOUTS:
1871 rc = mdc_ioc_swap_layouts(exp, karg);
1872 goto out;
1873 default:
1874 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
1875 rc = -ENOTTY;
1876 goto out;
1877 }
1878 out:
1879 module_put(THIS_MODULE);
1880
1881 return rc;
1882 }
1883
1884 static int mdc_get_info_rpc(struct obd_export *exp,
1885 u32 keylen, void *key,
1886 int vallen, void *val)
1887 {
1888 struct obd_import *imp = class_exp2cliimp(exp);
1889 struct ptlrpc_request *req;
1890 char *tmp;
1891 int rc = -EINVAL;
1892
1893 req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1894 if (!req)
1895 return -ENOMEM;
1896
1897 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1898 RCL_CLIENT, keylen);
1899 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1900 RCL_CLIENT, sizeof(__u32));
1901
1902 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1903 if (rc) {
1904 ptlrpc_request_free(req);
1905 return rc;
1906 }
1907
1908 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
1909 memcpy(tmp, key, keylen);
1910 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
1911 memcpy(tmp, &vallen, sizeof(__u32));
1912
1913 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1914 RCL_SERVER, vallen);
1915 ptlrpc_request_set_replen(req);
1916
1917 rc = ptlrpc_queue_wait(req);
1918 /* -EREMOTE means the get_info result is partial, and it needs to
1919 * continue on another MDT, see fid2path part in lmv_iocontrol
1920 */
1921 if (rc == 0 || rc == -EREMOTE) {
1922 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
1923 memcpy(val, tmp, vallen);
1924 if (ptlrpc_rep_need_swab(req)) {
1925 if (KEY_IS(KEY_FID2PATH))
1926 lustre_swab_fid2path(val);
1927 }
1928 }
1929 ptlrpc_req_finished(req);
1930
1931 return rc;
1932 }
1933
1934 static void lustre_swab_hai(struct hsm_action_item *h)
1935 {
1936 __swab32s(&h->hai_len);
1937 __swab32s(&h->hai_action);
1938 lustre_swab_lu_fid(&h->hai_fid);
1939 lustre_swab_lu_fid(&h->hai_dfid);
1940 __swab64s(&h->hai_cookie);
1941 __swab64s(&h->hai_extent.offset);
1942 __swab64s(&h->hai_extent.length);
1943 __swab64s(&h->hai_gid);
1944 }
1945
1946 static void lustre_swab_hal(struct hsm_action_list *h)
1947 {
1948 struct hsm_action_item *hai;
1949 int i;
1950
1951 __swab32s(&h->hal_version);
1952 __swab32s(&h->hal_count);
1953 __swab32s(&h->hal_archive_id);
1954 __swab64s(&h->hal_flags);
1955 hai = hai_zero(h);
1956 for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
1957 lustre_swab_hai(hai);
1958 }
1959
1960 static void lustre_swab_kuch(struct kuc_hdr *l)
1961 {
1962 __swab16s(&l->kuc_magic);
1963 /* __u8 l->kuc_transport */
1964 __swab16s(&l->kuc_msgtype);
1965 __swab16s(&l->kuc_msglen);
1966 }
1967
1968 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1969 struct lustre_kernelcomm *lk)
1970 {
1971 struct obd_import *imp = class_exp2cliimp(exp);
1972 __u32 archive = lk->lk_data;
1973 int rc = 0;
1974
1975 if (lk->lk_group != KUC_GRP_HSM) {
1976 CERROR("Bad copytool group %d\n", lk->lk_group);
1977 return -EINVAL;
1978 }
1979
1980 CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
1981 lk->lk_uid, lk->lk_group, lk->lk_flags);
1982
1983 if (lk->lk_flags & LK_FLG_STOP) {
1984 /* Unregister with the coordinator */
1985 rc = mdc_ioc_hsm_ct_unregister(imp);
1986 } else {
1987 rc = mdc_ioc_hsm_ct_register(imp, archive);
1988 }
1989
1990 return rc;
1991 }
1992
1993 /**
1994 * Send a message to any listening copytools
1995 * @param val KUC message (kuc_hdr + hsm_action_list)
1996 * @param len total length of message
1997 */
1998 static int mdc_hsm_copytool_send(int len, void *val)
1999 {
2000 struct kuc_hdr *lh = (struct kuc_hdr *)val;
2001 struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
2002
2003 if (len < sizeof(*lh) + sizeof(*hal)) {
2004 CERROR("Short HSM message %d < %d\n", len,
2005 (int) (sizeof(*lh) + sizeof(*hal)));
2006 return -EPROTO;
2007 }
2008 if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2009 lustre_swab_kuch(lh);
2010 lustre_swab_hal(hal);
2011 } else if (lh->kuc_magic != KUC_MAGIC) {
2012 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2013 return -EPROTO;
2014 }
2015
2016 CDEBUG(D_HSM,
2017 "Received message mg=%x t=%d m=%d l=%d actions=%d on %s\n",
2018 lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2019 lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2020
2021 /* Broadcast to HSM listeners */
2022 return libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2023 }
2024
2025 /**
2026 * callback function passed to kuc for re-registering each HSM copytool
2027 * running on MDC, after MDT shutdown/recovery.
2028 * @param data copytool registration data
2029 * @param cb_arg callback argument (obd_import)
2030 */
2031 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2032 {
2033 struct kkuc_ct_data *kcd = data;
2034 struct obd_import *imp = (struct obd_import *)cb_arg;
2035 int rc;
2036
2037 if (!kcd || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2038 return -EPROTO;
2039
2040 if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2041 return 0;
2042
2043 CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2044 imp->imp_obd->obd_name, kcd->kcd_archive);
2045 rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2046
2047 /* ignore error if the copytool is already registered */
2048 return (rc == -EEXIST) ? 0 : rc;
2049 }
2050
2051 static int mdc_set_info_async(const struct lu_env *env,
2052 struct obd_export *exp,
2053 u32 keylen, void *key,
2054 u32 vallen, void *val,
2055 struct ptlrpc_request_set *set)
2056 {
2057 struct obd_import *imp = class_exp2cliimp(exp);
2058 int rc;
2059
2060 if (KEY_IS(KEY_READ_ONLY)) {
2061 if (vallen != sizeof(int))
2062 return -EINVAL;
2063
2064 spin_lock(&imp->imp_lock);
2065 if (*((int *)val)) {
2066 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2067 imp->imp_connect_data.ocd_connect_flags |=
2068 OBD_CONNECT_RDONLY;
2069 } else {
2070 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2071 imp->imp_connect_data.ocd_connect_flags &=
2072 ~OBD_CONNECT_RDONLY;
2073 }
2074 spin_unlock(&imp->imp_lock);
2075
2076 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2077 keylen, key, vallen, val, set);
2078 return rc;
2079 }
2080 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2081 sptlrpc_conf_client_adapt(exp->exp_obd);
2082 return 0;
2083 }
2084 if (KEY_IS(KEY_FLUSH_CTX)) {
2085 sptlrpc_import_flush_my_ctx(imp);
2086 return 0;
2087 }
2088 if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2089 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2090 keylen, key, vallen, val, set);
2091 return rc;
2092 }
2093 if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2094 rc = mdc_hsm_copytool_send(vallen, val);
2095 return rc;
2096 }
2097
2098 CERROR("Unknown key %s\n", (char *)key);
2099 return -EINVAL;
2100 }
2101
2102 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2103 __u32 keylen, void *key, __u32 *vallen, void *val,
2104 struct lov_stripe_md *lsm)
2105 {
2106 int rc = -EINVAL;
2107
2108 if (KEY_IS(KEY_MAX_EASIZE)) {
2109 int mdsize, *max_easize;
2110
2111 if (*vallen != sizeof(int))
2112 return -EINVAL;
2113 mdsize = *(int *)val;
2114 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2115 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2116 max_easize = val;
2117 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2118 return 0;
2119 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2120 int *default_easize;
2121
2122 if (*vallen != sizeof(int))
2123 return -EINVAL;
2124 default_easize = val;
2125 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2126 return 0;
2127 } else if (KEY_IS(KEY_CONN_DATA)) {
2128 struct obd_import *imp = class_exp2cliimp(exp);
2129 struct obd_connect_data *data = val;
2130
2131 if (*vallen != sizeof(*data))
2132 return -EINVAL;
2133
2134 *data = imp->imp_connect_data;
2135 return 0;
2136 } else if (KEY_IS(KEY_TGT_COUNT)) {
2137 *((int *)val) = 1;
2138 return 0;
2139 }
2140
2141 rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2142
2143 return rc;
2144 }
2145
2146 static int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
2147 struct ptlrpc_request **request)
2148 {
2149 struct ptlrpc_request *req;
2150 int rc;
2151
2152 *request = NULL;
2153 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2154 if (!req)
2155 return -ENOMEM;
2156
2157 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2158 if (rc) {
2159 ptlrpc_request_free(req);
2160 return rc;
2161 }
2162
2163 mdc_pack_body(req, fid, 0, 0, -1, 0);
2164
2165 ptlrpc_request_set_replen(req);
2166
2167 rc = ptlrpc_queue_wait(req);
2168 if (rc)
2169 ptlrpc_req_finished(req);
2170 else
2171 *request = req;
2172 return rc;
2173 }
2174
2175 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2176 enum obd_import_event event)
2177 {
2178 int rc = 0;
2179
2180 LASSERT(imp->imp_obd == obd);
2181
2182 switch (event) {
2183 case IMP_EVENT_DISCON: {
2184 #if 0
2185 /* XXX Pass event up to OBDs stack. used only for FLD now */
2186 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2187 #endif
2188 break;
2189 }
2190 case IMP_EVENT_INACTIVE: {
2191 struct client_obd *cli = &obd->u.cli;
2192 /*
2193 * Flush current sequence to make client obtain new one
2194 * from server in case of disconnect/reconnect.
2195 */
2196 if (cli->cl_seq)
2197 seq_client_flush(cli->cl_seq);
2198
2199 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2200 break;
2201 }
2202 case IMP_EVENT_INVALIDATE: {
2203 struct ldlm_namespace *ns = obd->obd_namespace;
2204
2205 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2206
2207 break;
2208 }
2209 case IMP_EVENT_ACTIVE:
2210 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2211 /* redo the kuc registration after reconnecting */
2212 if (rc == 0)
2213 /* re-register HSM agents */
2214 rc = libcfs_kkuc_group_foreach(KUC_GRP_HSM,
2215 mdc_hsm_ct_reregister,
2216 (void *)imp);
2217 break;
2218 case IMP_EVENT_OCD:
2219 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2220 break;
2221 case IMP_EVENT_DEACTIVATE:
2222 case IMP_EVENT_ACTIVATE:
2223 break;
2224 default:
2225 CERROR("Unknown import event %x\n", event);
2226 LBUG();
2227 }
2228 return rc;
2229 }
2230
2231 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
2232 struct md_op_data *op_data)
2233 {
2234 struct client_obd *cli = &exp->exp_obd->u.cli;
2235 struct lu_client_seq *seq = cli->cl_seq;
2236
2237 return seq_client_alloc_fid(NULL, seq, fid);
2238 }
2239
2240 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2241 {
2242 struct client_obd *cli = &exp->exp_obd->u.cli;
2243
2244 return &cli->cl_target_uuid;
2245 }
2246
2247 /**
2248 * Determine whether the lock can be canceled before replaying it during
2249 * recovery, non zero value will be return if the lock can be canceled,
2250 * or zero returned for not
2251 */
2252 static int mdc_cancel_for_recovery(struct ldlm_lock *lock)
2253 {
2254 if (lock->l_resource->lr_type != LDLM_IBITS)
2255 return 0;
2256
2257 /* FIXME: if we ever get into a situation where there are too many
2258 * opened files with open locks on a single node, then we really
2259 * should replay these open locks to reget it
2260 */
2261 if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2262 return 0;
2263
2264 return 1;
2265 }
2266
2267 static int mdc_resource_inode_free(struct ldlm_resource *res)
2268 {
2269 if (res->lr_lvb_inode)
2270 res->lr_lvb_inode = NULL;
2271
2272 return 0;
2273 }
2274
2275 static struct ldlm_valblock_ops inode_lvbo = {
2276 .lvbo_free = mdc_resource_inode_free,
2277 };
2278
2279 static int mdc_llog_init(struct obd_device *obd)
2280 {
2281 struct obd_llog_group *olg = &obd->obd_olg;
2282 struct llog_ctxt *ctxt;
2283 int rc;
2284
2285 rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2286 &llog_client_ops);
2287 if (rc)
2288 return rc;
2289
2290 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2291 llog_initiator_connect(ctxt);
2292 llog_ctxt_put(ctxt);
2293
2294 return 0;
2295 }
2296
2297 static void mdc_llog_finish(struct obd_device *obd)
2298 {
2299 struct llog_ctxt *ctxt;
2300
2301 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2302 if (ctxt)
2303 llog_cleanup(NULL, ctxt);
2304 }
2305
2306 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2307 {
2308 struct client_obd *cli = &obd->u.cli;
2309 struct lprocfs_static_vars lvars = { NULL };
2310 int rc;
2311
2312 cli->cl_rpc_lock = kzalloc(sizeof(*cli->cl_rpc_lock), GFP_NOFS);
2313 if (!cli->cl_rpc_lock)
2314 return -ENOMEM;
2315 mdc_init_rpc_lock(cli->cl_rpc_lock);
2316
2317 ptlrpcd_addref();
2318
2319 cli->cl_close_lock = kzalloc(sizeof(*cli->cl_close_lock), GFP_NOFS);
2320 if (!cli->cl_close_lock) {
2321 rc = -ENOMEM;
2322 goto err_rpc_lock;
2323 }
2324 mdc_init_rpc_lock(cli->cl_close_lock);
2325
2326 rc = client_obd_setup(obd, cfg);
2327 if (rc)
2328 goto err_close_lock;
2329 lprocfs_mdc_init_vars(&lvars);
2330 lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars);
2331 sptlrpc_lprocfs_cliobd_attach(obd);
2332 ptlrpc_lprocfs_register_obd(obd);
2333
2334 ns_register_cancel(obd->obd_namespace, mdc_cancel_for_recovery);
2335
2336 obd->obd_namespace->ns_lvbo = &inode_lvbo;
2337
2338 rc = mdc_llog_init(obd);
2339 if (rc) {
2340 mdc_cleanup(obd);
2341 CERROR("failed to setup llogging subsystems\n");
2342 }
2343
2344 return rc;
2345
2346 err_close_lock:
2347 kfree(cli->cl_close_lock);
2348 err_rpc_lock:
2349 kfree(cli->cl_rpc_lock);
2350 ptlrpcd_decref();
2351 return rc;
2352 }
2353
2354 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
2355 * us to make MDS RPCs with large enough reply buffers to hold a default
2356 * sized EA and cookie without having to calculate this (via a call into the
2357 * LOV + OSCs) each time we make an RPC. The maximum size is also tracked
2358 * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2359 * a large number of stripes is possible. If a larger reply buffer is
2360 * required it will be reallocated in the ptlrpc layer due to overflow.
2361 */
2362 static int mdc_init_ea_size(struct obd_export *exp, int easize,
2363 int def_easize, int cookiesize, int def_cookiesize)
2364 {
2365 struct obd_device *obd = exp->exp_obd;
2366 struct client_obd *cli = &obd->u.cli;
2367
2368 if (cli->cl_max_mds_easize < easize)
2369 cli->cl_max_mds_easize = easize;
2370
2371 if (cli->cl_default_mds_easize < def_easize)
2372 cli->cl_default_mds_easize = def_easize;
2373
2374 if (cli->cl_max_mds_cookiesize < cookiesize)
2375 cli->cl_max_mds_cookiesize = cookiesize;
2376
2377 if (cli->cl_default_mds_cookiesize < def_cookiesize)
2378 cli->cl_default_mds_cookiesize = def_cookiesize;
2379
2380 return 0;
2381 }
2382
2383 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2384 {
2385 switch (stage) {
2386 case OBD_CLEANUP_EARLY:
2387 break;
2388 case OBD_CLEANUP_EXPORTS:
2389 /* Failsafe, ok if racy */
2390 if (obd->obd_type->typ_refcnt <= 1)
2391 libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2392
2393 obd_cleanup_client_import(obd);
2394 ptlrpc_lprocfs_unregister_obd(obd);
2395 lprocfs_obd_cleanup(obd);
2396
2397 mdc_llog_finish(obd);
2398 break;
2399 }
2400 return 0;
2401 }
2402
2403 static int mdc_cleanup(struct obd_device *obd)
2404 {
2405 struct client_obd *cli = &obd->u.cli;
2406
2407 kfree(cli->cl_rpc_lock);
2408 kfree(cli->cl_close_lock);
2409
2410 ptlrpcd_decref();
2411
2412 return client_obd_cleanup(obd);
2413 }
2414
2415 static int mdc_process_config(struct obd_device *obd, u32 len, void *buf)
2416 {
2417 struct lustre_cfg *lcfg = buf;
2418 struct lprocfs_static_vars lvars = { NULL };
2419 int rc = 0;
2420
2421 lprocfs_mdc_init_vars(&lvars);
2422 switch (lcfg->lcfg_command) {
2423 default:
2424 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
2425 lcfg, obd);
2426 if (rc > 0)
2427 rc = 0;
2428 break;
2429 }
2430 return rc;
2431 }
2432
2433 /* get remote permission for current user on fid */
2434 static int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2435 __u32 suppgid, struct ptlrpc_request **request)
2436 {
2437 struct ptlrpc_request *req;
2438 int rc;
2439
2440 LASSERT(client_is_remote(exp));
2441
2442 *request = NULL;
2443 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2444 if (!req)
2445 return -ENOMEM;
2446
2447 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2448 if (rc) {
2449 ptlrpc_request_free(req);
2450 return rc;
2451 }
2452
2453 mdc_pack_body(req, fid, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2454
2455 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2456 sizeof(struct mdt_remote_perm));
2457
2458 ptlrpc_request_set_replen(req);
2459
2460 rc = ptlrpc_queue_wait(req);
2461 if (rc)
2462 ptlrpc_req_finished(req);
2463 else
2464 *request = req;
2465 return rc;
2466 }
2467
2468 static struct obd_ops mdc_obd_ops = {
2469 .owner = THIS_MODULE,
2470 .setup = mdc_setup,
2471 .precleanup = mdc_precleanup,
2472 .cleanup = mdc_cleanup,
2473 .add_conn = client_import_add_conn,
2474 .del_conn = client_import_del_conn,
2475 .connect = client_connect_import,
2476 .disconnect = client_disconnect_export,
2477 .iocontrol = mdc_iocontrol,
2478 .set_info_async = mdc_set_info_async,
2479 .statfs = mdc_statfs,
2480 .fid_init = client_fid_init,
2481 .fid_fini = client_fid_fini,
2482 .fid_alloc = mdc_fid_alloc,
2483 .import_event = mdc_import_event,
2484 .get_info = mdc_get_info,
2485 .process_config = mdc_process_config,
2486 .get_uuid = mdc_get_uuid,
2487 .quotactl = mdc_quotactl,
2488 .quotacheck = mdc_quotacheck
2489 };
2490
2491 static struct md_ops mdc_md_ops = {
2492 .getstatus = mdc_getstatus,
2493 .null_inode = mdc_null_inode,
2494 .find_cbdata = mdc_find_cbdata,
2495 .close = mdc_close,
2496 .create = mdc_create,
2497 .done_writing = mdc_done_writing,
2498 .enqueue = mdc_enqueue,
2499 .getattr = mdc_getattr,
2500 .getattr_name = mdc_getattr_name,
2501 .intent_lock = mdc_intent_lock,
2502 .link = mdc_link,
2503 .is_subdir = mdc_is_subdir,
2504 .rename = mdc_rename,
2505 .setattr = mdc_setattr,
2506 .setxattr = mdc_setxattr,
2507 .getxattr = mdc_getxattr,
2508 .sync = mdc_sync,
2509 .readpage = mdc_readpage,
2510 .unlink = mdc_unlink,
2511 .cancel_unused = mdc_cancel_unused,
2512 .init_ea_size = mdc_init_ea_size,
2513 .set_lock_data = mdc_set_lock_data,
2514 .lock_match = mdc_lock_match,
2515 .get_lustre_md = mdc_get_lustre_md,
2516 .free_lustre_md = mdc_free_lustre_md,
2517 .set_open_replay_data = mdc_set_open_replay_data,
2518 .clear_open_replay_data = mdc_clear_open_replay_data,
2519 .get_remote_perm = mdc_get_remote_perm,
2520 .intent_getattr_async = mdc_intent_getattr_async,
2521 .revalidate_lock = mdc_revalidate_lock
2522 };
2523
2524 static int __init mdc_init(void)
2525 {
2526 struct lprocfs_static_vars lvars = { NULL };
2527
2528 lprocfs_mdc_init_vars(&lvars);
2529
2530 return class_register_type(&mdc_obd_ops, &mdc_md_ops,
2531 LUSTRE_MDC_NAME, NULL);
2532 }
2533
2534 static void /*__exit*/ mdc_exit(void)
2535 {
2536 class_unregister_type(LUSTRE_MDC_NAME);
2537 }
2538
2539 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2540 MODULE_DESCRIPTION("Lustre Metadata Client");
2541 MODULE_VERSION(LUSTRE_VERSION_STRING);
2542 MODULE_LICENSE("GPL");
2543
2544 module_init(mdc_init);
2545 module_exit(mdc_exit);
Linux kernel - Deliverables
This page took 0.085659 seconds and 5 git commands to generate.