]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - drivers/staging/lustre/lustre/mdc/mdc_request.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'iwlwifi-next-for-kalle-2016-07-06' of git://git.kernel.org/pub/scm/linux...
[mirror_ubuntu-zesty-kernel.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 mdc_update_max_ea_from_body(exp, body);
146 if (body->eadatasize != 0) {
147 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
148 body->eadatasize);
149 if (!eadata)
150 return -EPROTO;
151 }
152
153 if (body->valid & OBD_MD_FLRMTPERM) {
154 struct mdt_remote_perm *perm;
155
156 LASSERT(client_is_remote(exp));
157 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
158 lustre_swab_mdt_remote_perm);
159 if (!perm)
160 return -EPROTO;
161 }
162
163 return 0;
164 }
165
166 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
167 struct ptlrpc_request **request)
168 {
169 struct ptlrpc_request *req;
170 int rc;
171
172 /* Single MDS without an LMV case */
173 if (op_data->op_flags & MF_GET_MDT_IDX) {
174 op_data->op_mds = 0;
175 return 0;
176 }
177 *request = NULL;
178 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
179 if (!req)
180 return -ENOMEM;
181
182 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
183 if (rc) {
184 ptlrpc_request_free(req);
185 return rc;
186 }
187
188 mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
189 op_data->op_mode, -1, 0);
190
191 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
192 op_data->op_mode);
193 if (op_data->op_valid & OBD_MD_FLRMTPERM) {
194 LASSERT(client_is_remote(exp));
195 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
196 sizeof(struct mdt_remote_perm));
197 }
198 ptlrpc_request_set_replen(req);
199
200 rc = mdc_getattr_common(exp, req);
201 if (rc)
202 ptlrpc_req_finished(req);
203 else
204 *request = req;
205 return rc;
206 }
207
208 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
209 struct ptlrpc_request **request)
210 {
211 struct ptlrpc_request *req;
212 int rc;
213
214 *request = NULL;
215 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
216 &RQF_MDS_GETATTR_NAME);
217 if (!req)
218 return -ENOMEM;
219
220 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
221 op_data->op_namelen + 1);
222
223 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
224 if (rc) {
225 ptlrpc_request_free(req);
226 return rc;
227 }
228
229 mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
230 op_data->op_mode, op_data->op_suppgids[0], 0);
231
232 if (op_data->op_name) {
233 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
234
235 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
236 op_data->op_namelen);
237 memcpy(name, op_data->op_name, op_data->op_namelen);
238 }
239
240 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
241 op_data->op_mode);
242 ptlrpc_request_set_replen(req);
243
244 rc = mdc_getattr_common(exp, req);
245 if (rc)
246 ptlrpc_req_finished(req);
247 else
248 *request = req;
249 return rc;
250 }
251
252 static int mdc_is_subdir(struct obd_export *exp,
253 const struct lu_fid *pfid,
254 const struct lu_fid *cfid,
255 struct ptlrpc_request **request)
256 {
257 struct ptlrpc_request *req;
258 int rc;
259
260 *request = NULL;
261 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
262 &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
263 MDS_IS_SUBDIR);
264 if (!req)
265 return -ENOMEM;
266
267 mdc_is_subdir_pack(req, pfid, cfid, 0);
268 ptlrpc_request_set_replen(req);
269
270 rc = ptlrpc_queue_wait(req);
271 if (rc && rc != -EREMOTE)
272 ptlrpc_req_finished(req);
273 else
274 *request = req;
275 return rc;
276 }
277
278 static int mdc_xattr_common(struct obd_export *exp,
279 const struct req_format *fmt,
280 const struct lu_fid *fid,
281 int opcode, u64 valid,
282 const char *xattr_name, const char *input,
283 int input_size, int output_size, int flags,
284 __u32 suppgid, struct ptlrpc_request **request)
285 {
286 struct ptlrpc_request *req;
287 int xattr_namelen = 0;
288 char *tmp;
289 int rc;
290
291 *request = NULL;
292 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
293 if (!req)
294 return -ENOMEM;
295
296 if (xattr_name) {
297 xattr_namelen = strlen(xattr_name) + 1;
298 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
299 xattr_namelen);
300 }
301 if (input_size) {
302 LASSERT(input);
303 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
304 input_size);
305 }
306
307 /* Flush local XATTR locks to get rid of a possible cancel RPC */
308 if (opcode == MDS_REINT && fid_is_sane(fid) &&
309 exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
310 LIST_HEAD(cancels);
311 int count;
312
313 /* Without that packing would fail */
314 if (input_size == 0)
315 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
316 RCL_CLIENT, 0);
317
318 count = mdc_resource_get_unused(exp, fid,
319 &cancels, LCK_EX,
320 MDS_INODELOCK_XATTR);
321
322 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
323 if (rc) {
324 ptlrpc_request_free(req);
325 return rc;
326 }
327 } else {
328 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
329 if (rc) {
330 ptlrpc_request_free(req);
331 return rc;
332 }
333 }
334
335 if (opcode == MDS_REINT) {
336 struct mdt_rec_setxattr *rec;
337
338 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
339 sizeof(struct mdt_rec_reint));
340 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
341 rec->sx_opcode = REINT_SETXATTR;
342 rec->sx_fsuid = from_kuid(&init_user_ns, current_fsuid());
343 rec->sx_fsgid = from_kgid(&init_user_ns, current_fsgid());
344 rec->sx_cap = cfs_curproc_cap_pack();
345 rec->sx_suppgid1 = suppgid;
346 rec->sx_suppgid2 = -1;
347 rec->sx_fid = *fid;
348 rec->sx_valid = valid | OBD_MD_FLCTIME;
349 rec->sx_time = ktime_get_real_seconds();
350 rec->sx_size = output_size;
351 rec->sx_flags = flags;
352
353 } else {
354 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
355 }
356
357 if (xattr_name) {
358 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
359 memcpy(tmp, xattr_name, xattr_namelen);
360 }
361 if (input_size) {
362 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
363 memcpy(tmp, input, input_size);
364 }
365
366 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
367 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
368 RCL_SERVER, output_size);
369 ptlrpc_request_set_replen(req);
370
371 /* make rpc */
372 if (opcode == MDS_REINT)
373 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
374
375 rc = ptlrpc_queue_wait(req);
376
377 if (opcode == MDS_REINT)
378 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
379
380 if (rc)
381 ptlrpc_req_finished(req);
382 else
383 *request = req;
384 return rc;
385 }
386
387 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
388 u64 valid, const char *xattr_name,
389 const char *input, int input_size, int output_size,
390 int flags, __u32 suppgid,
391 struct ptlrpc_request **request)
392 {
393 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
394 fid, MDS_REINT, valid, xattr_name,
395 input, input_size, output_size, flags,
396 suppgid, request);
397 }
398
399 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
400 u64 valid, const char *xattr_name,
401 const char *input, int input_size, int output_size,
402 int flags, struct ptlrpc_request **request)
403 {
404 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
405 fid, MDS_GETXATTR, valid, xattr_name,
406 input, input_size, output_size, flags,
407 -1, request);
408 }
409
410 #ifdef CONFIG_FS_POSIX_ACL
411 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
412 {
413 struct req_capsule *pill = &req->rq_pill;
414 struct mdt_body *body = md->body;
415 struct posix_acl *acl;
416 void *buf;
417 int rc;
418
419 if (!body->aclsize)
420 return 0;
421
422 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
423
424 if (!buf)
425 return -EPROTO;
426
427 acl = posix_acl_from_xattr(&init_user_ns, buf, body->aclsize);
428 if (!acl)
429 return 0;
430
431 if (IS_ERR(acl)) {
432 rc = PTR_ERR(acl);
433 CERROR("convert xattr to acl: %d\n", rc);
434 return rc;
435 }
436
437 rc = posix_acl_valid(acl);
438 if (rc) {
439 CERROR("validate acl: %d\n", rc);
440 posix_acl_release(acl);
441 return rc;
442 }
443
444 md->posix_acl = acl;
445 return 0;
446 }
447 #else
448 #define mdc_unpack_acl(req, md) 0
449 #endif
450
451 static int mdc_get_lustre_md(struct obd_export *exp,
452 struct ptlrpc_request *req,
453 struct obd_export *dt_exp,
454 struct obd_export *md_exp,
455 struct lustre_md *md)
456 {
457 struct req_capsule *pill = &req->rq_pill;
458 int rc;
459
460 LASSERT(md);
461 memset(md, 0, sizeof(*md));
462
463 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
464
465 if (md->body->valid & OBD_MD_FLEASIZE) {
466 int lmmsize;
467 struct lov_mds_md *lmm;
468
469 if (!S_ISREG(md->body->mode)) {
470 CDEBUG(D_INFO,
471 "OBD_MD_FLEASIZE set, should be a regular file, but is not\n");
472 rc = -EPROTO;
473 goto out;
474 }
475
476 if (md->body->eadatasize == 0) {
477 CDEBUG(D_INFO,
478 "OBD_MD_FLEASIZE set, but eadatasize 0\n");
479 rc = -EPROTO;
480 goto out;
481 }
482 lmmsize = md->body->eadatasize;
483 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
484 if (!lmm) {
485 rc = -EPROTO;
486 goto out;
487 }
488
489 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
490 if (rc < 0)
491 goto out;
492
493 if (rc < sizeof(*md->lsm)) {
494 CDEBUG(D_INFO,
495 "lsm size too small: rc < sizeof (*md->lsm) (%d < %d)\n",
496 rc, (int)sizeof(*md->lsm));
497 rc = -EPROTO;
498 goto out;
499 }
500
501 } else if (md->body->valid & OBD_MD_FLDIREA) {
502 int lmvsize;
503 struct lov_mds_md *lmv;
504
505 if (!S_ISDIR(md->body->mode)) {
506 CDEBUG(D_INFO,
507 "OBD_MD_FLDIREA set, should be a directory, but is not\n");
508 rc = -EPROTO;
509 goto out;
510 }
511
512 if (md->body->eadatasize == 0) {
513 CDEBUG(D_INFO,
514 "OBD_MD_FLDIREA is set, but eadatasize 0\n");
515 return -EPROTO;
516 }
517 if (md->body->valid & OBD_MD_MEA) {
518 lmvsize = md->body->eadatasize;
519 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
520 lmvsize);
521 if (!lmv) {
522 rc = -EPROTO;
523 goto out;
524 }
525
526 rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
527 lmvsize);
528 if (rc < 0)
529 goto out;
530
531 if (rc < sizeof(*md->mea)) {
532 CDEBUG(D_INFO,
533 "size too small: rc < sizeof(*md->mea) (%d < %d)\n",
534 rc, (int)sizeof(*md->mea));
535 rc = -EPROTO;
536 goto out;
537 }
538 }
539 }
540 rc = 0;
541
542 if (md->body->valid & OBD_MD_FLRMTPERM) {
543 /* remote permission */
544 LASSERT(client_is_remote(exp));
545 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
546 lustre_swab_mdt_remote_perm);
547 if (!md->remote_perm) {
548 rc = -EPROTO;
549 goto out;
550 }
551 } else if (md->body->valid & OBD_MD_FLACL) {
552 /* for ACL, it's possible that FLACL is set but aclsize is zero.
553 * only when aclsize != 0 there's an actual segment for ACL
554 * in reply buffer.
555 */
556 if (md->body->aclsize) {
557 rc = mdc_unpack_acl(req, md);
558 if (rc)
559 goto out;
560 #ifdef CONFIG_FS_POSIX_ACL
561 } else {
562 md->posix_acl = NULL;
563 #endif
564 }
565 }
566
567 out:
568 if (rc) {
569 #ifdef CONFIG_FS_POSIX_ACL
570 posix_acl_release(md->posix_acl);
571 #endif
572 if (md->lsm)
573 obd_free_memmd(dt_exp, &md->lsm);
574 }
575 return rc;
576 }
577
578 static int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
579 {
580 return 0;
581 }
582
583 /**
584 * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
585 * RPC chains.
586 */
587 void mdc_replay_open(struct ptlrpc_request *req)
588 {
589 struct md_open_data *mod = req->rq_cb_data;
590 struct ptlrpc_request *close_req;
591 struct obd_client_handle *och;
592 struct lustre_handle old;
593 struct mdt_body *body;
594
595 if (!mod) {
596 DEBUG_REQ(D_ERROR, req,
597 "Can't properly replay without open data.");
598 return;
599 }
600
601 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
602
603 och = mod->mod_och;
604 if (och) {
605 struct lustre_handle *file_fh;
606
607 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
608
609 file_fh = &och->och_fh;
610 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
611 file_fh->cookie, body->handle.cookie);
612 old = *file_fh;
613 *file_fh = body->handle;
614 }
615 close_req = mod->mod_close_req;
616 if (close_req) {
617 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
618 struct mdt_ioepoch *epoch;
619
620 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
621 epoch = req_capsule_client_get(&close_req->rq_pill,
622 &RMF_MDT_EPOCH);
623 LASSERT(epoch);
624
625 if (och)
626 LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
627 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
628 epoch->handle = body->handle;
629 }
630 }
631
632 void mdc_commit_open(struct ptlrpc_request *req)
633 {
634 struct md_open_data *mod = req->rq_cb_data;
635
636 if (!mod)
637 return;
638
639 /**
640 * No need to touch md_open_data::mod_och, it holds a reference on
641 * \var mod and will zero references to each other, \var mod will be
642 * freed after that when md_open_data::mod_och will put the reference.
643 */
644
645 /**
646 * Do not let open request to disappear as it still may be needed
647 * for close rpc to happen (it may happen on evict only, otherwise
648 * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
649 * called), just mark this rpc as committed to distinguish these 2
650 * cases, see mdc_close() for details. The open request reference will
651 * be put along with freeing \var mod.
652 */
653 ptlrpc_request_addref(req);
654 spin_lock(&req->rq_lock);
655 req->rq_committed = 1;
656 spin_unlock(&req->rq_lock);
657 req->rq_cb_data = NULL;
658 obd_mod_put(mod);
659 }
660
661 int mdc_set_open_replay_data(struct obd_export *exp,
662 struct obd_client_handle *och,
663 struct lookup_intent *it)
664 {
665 struct md_open_data *mod;
666 struct mdt_rec_create *rec;
667 struct mdt_body *body;
668 struct ptlrpc_request *open_req = it->d.lustre.it_data;
669 struct obd_import *imp = open_req->rq_import;
670
671 if (!open_req->rq_replay)
672 return 0;
673
674 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
675 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
676 LASSERT(rec);
677 /* Incoming message in my byte order (it's been swabbed). */
678 /* Outgoing messages always in my byte order. */
679 LASSERT(body);
680
681 /* Only if the import is replayable, we set replay_open data */
682 if (och && imp->imp_replayable) {
683 mod = obd_mod_alloc();
684 if (!mod) {
685 DEBUG_REQ(D_ERROR, open_req,
686 "Can't allocate md_open_data");
687 return 0;
688 }
689
690 /**
691 * Take a reference on \var mod, to be freed on mdc_close().
692 * It protects \var mod from being freed on eviction (commit
693 * callback is called despite rq_replay flag).
694 * Another reference for \var och.
695 */
696 obd_mod_get(mod);
697 obd_mod_get(mod);
698
699 spin_lock(&open_req->rq_lock);
700 och->och_mod = mod;
701 mod->mod_och = och;
702 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
703 it_disposition(it, DISP_OPEN_STRIPE);
704 mod->mod_open_req = open_req;
705 open_req->rq_cb_data = mod;
706 open_req->rq_commit_cb = mdc_commit_open;
707 spin_unlock(&open_req->rq_lock);
708 }
709
710 rec->cr_fid2 = body->fid1;
711 rec->cr_ioepoch = body->ioepoch;
712 rec->cr_old_handle.cookie = body->handle.cookie;
713 open_req->rq_replay_cb = mdc_replay_open;
714 if (!fid_is_sane(&body->fid1)) {
715 DEBUG_REQ(D_ERROR, open_req,
716 "Saving replay request with insane fid");
717 LBUG();
718 }
719
720 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
721 return 0;
722 }
723
724 static void mdc_free_open(struct md_open_data *mod)
725 {
726 int committed = 0;
727
728 if (mod->mod_is_create == 0 &&
729 imp_connect_disp_stripe(mod->mod_open_req->rq_import))
730 committed = 1;
731
732 LASSERT(mod->mod_open_req->rq_replay == 0);
733
734 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
735
736 ptlrpc_request_committed(mod->mod_open_req, committed);
737 if (mod->mod_close_req)
738 ptlrpc_request_committed(mod->mod_close_req, committed);
739 }
740
741 static int mdc_clear_open_replay_data(struct obd_export *exp,
742 struct obd_client_handle *och)
743 {
744 struct md_open_data *mod = och->och_mod;
745
746 /**
747 * It is possible to not have \var mod in a case of eviction between
748 * lookup and ll_file_open().
749 **/
750 if (!mod)
751 return 0;
752
753 LASSERT(mod != LP_POISON);
754 LASSERT(mod->mod_open_req);
755 mdc_free_open(mod);
756
757 mod->mod_och = NULL;
758 och->och_mod = NULL;
759 obd_mod_put(mod);
760
761 return 0;
762 }
763
764 /* Prepares the request for the replay by the given reply */
765 static void mdc_close_handle_reply(struct ptlrpc_request *req,
766 struct md_op_data *op_data, int rc) {
767 struct mdt_body *repbody;
768 struct mdt_ioepoch *epoch;
769
770 if (req && rc == -EAGAIN) {
771 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
772 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
773
774 epoch->flags |= MF_SOM_AU;
775 if (repbody->valid & OBD_MD_FLGETATTRLOCK)
776 op_data->op_flags |= MF_GETATTR_LOCK;
777 }
778 }
779
780 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
781 struct md_open_data *mod, struct ptlrpc_request **request)
782 {
783 struct obd_device *obd = class_exp2obd(exp);
784 struct ptlrpc_request *req;
785 struct req_format *req_fmt;
786 int rc;
787 int saved_rc = 0;
788
789 req_fmt = &RQF_MDS_CLOSE;
790 if (op_data->op_bias & MDS_HSM_RELEASE) {
791 req_fmt = &RQF_MDS_RELEASE_CLOSE;
792
793 /* allocate a FID for volatile file */
794 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
795 if (rc < 0) {
796 CERROR("%s: "DFID" failed to allocate FID: %d\n",
797 obd->obd_name, PFID(&op_data->op_fid1), rc);
798 /* save the errcode and proceed to close */
799 saved_rc = rc;
800 }
801 }
802
803 *request = NULL;
804 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
805 if (!req)
806 return -ENOMEM;
807
808 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
809 if (rc) {
810 ptlrpc_request_free(req);
811 return rc;
812 }
813
814 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
815 * portal whose threads are not taking any DLM locks and are therefore
816 * always progressing
817 */
818 req->rq_request_portal = MDS_READPAGE_PORTAL;
819 ptlrpc_at_set_req_timeout(req);
820
821 /* Ensure that this close's handle is fixed up during replay. */
822 if (likely(mod)) {
823 LASSERTF(mod->mod_open_req &&
824 mod->mod_open_req->rq_type != LI_POISON,
825 "POISONED open %p!\n", mod->mod_open_req);
826
827 mod->mod_close_req = req;
828
829 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
830 /* We no longer want to preserve this open for replay even
831 * though the open was committed. b=3632, b=3633
832 */
833 spin_lock(&mod->mod_open_req->rq_lock);
834 mod->mod_open_req->rq_replay = 0;
835 spin_unlock(&mod->mod_open_req->rq_lock);
836 } else {
837 CDEBUG(D_HA,
838 "couldn't find open req; expecting close error\n");
839 }
840
841 mdc_close_pack(req, op_data);
842
843 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
844 obd->u.cli.cl_default_mds_easize);
845 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
846 obd->u.cli.cl_default_mds_cookiesize);
847
848 ptlrpc_request_set_replen(req);
849
850 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
851 rc = ptlrpc_queue_wait(req);
852 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
853
854 if (!req->rq_repmsg) {
855 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
856 req->rq_status);
857 if (rc == 0)
858 rc = req->rq_status ?: -EIO;
859 } else if (rc == 0 || rc == -EAGAIN) {
860 struct mdt_body *body;
861
862 rc = lustre_msg_get_status(req->rq_repmsg);
863 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
864 DEBUG_REQ(D_ERROR, req,
865 "type == PTL_RPC_MSG_ERR, err = %d", rc);
866 if (rc > 0)
867 rc = -rc;
868 }
869 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
870 if (!body)
871 rc = -EPROTO;
872 } else if (rc == -ESTALE) {
873 /**
874 * it can be allowed error after 3633 if open was committed and
875 * server failed before close was sent. Let's check if mod
876 * exists and return no error in that case
877 */
878 if (mod) {
879 DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
880 if (mod->mod_open_req->rq_committed)
881 rc = 0;
882 }
883 }
884
885 if (mod) {
886 if (rc != 0)
887 mod->mod_close_req = NULL;
888 /* Since now, mod is accessed through open_req only,
889 * thus close req does not keep a reference on mod anymore.
890 */
891 obd_mod_put(mod);
892 }
893 *request = req;
894 mdc_close_handle_reply(req, op_data, rc);
895 return rc < 0 ? rc : saved_rc;
896 }
897
898 static int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
899 struct md_open_data *mod)
900 {
901 struct obd_device *obd = class_exp2obd(exp);
902 struct ptlrpc_request *req;
903 int rc;
904
905 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
906 &RQF_MDS_DONE_WRITING);
907 if (!req)
908 return -ENOMEM;
909
910 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
911 if (rc) {
912 ptlrpc_request_free(req);
913 return rc;
914 }
915
916 if (mod) {
917 LASSERTF(mod->mod_open_req &&
918 mod->mod_open_req->rq_type != LI_POISON,
919 "POISONED setattr %p!\n", mod->mod_open_req);
920
921 mod->mod_close_req = req;
922 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
923 /* We no longer want to preserve this setattr for replay even
924 * though the open was committed. b=3632, b=3633
925 */
926 spin_lock(&mod->mod_open_req->rq_lock);
927 mod->mod_open_req->rq_replay = 0;
928 spin_unlock(&mod->mod_open_req->rq_lock);
929 }
930
931 mdc_close_pack(req, op_data);
932 ptlrpc_request_set_replen(req);
933
934 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
935 rc = ptlrpc_queue_wait(req);
936 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
937
938 if (rc == -ESTALE) {
939 /**
940 * it can be allowed error after 3633 if open or setattr were
941 * committed and server failed before close was sent.
942 * Let's check if mod exists and return no error in that case
943 */
944 if (mod) {
945 if (mod->mod_open_req->rq_committed)
946 rc = 0;
947 }
948 }
949
950 if (mod) {
951 if (rc != 0)
952 mod->mod_close_req = NULL;
953 LASSERT(mod->mod_open_req);
954 mdc_free_open(mod);
955
956 /* Since now, mod is accessed through setattr req only,
957 * thus DW req does not keep a reference on mod anymore.
958 */
959 obd_mod_put(mod);
960 }
961
962 mdc_close_handle_reply(req, op_data, rc);
963 ptlrpc_req_finished(req);
964 return rc;
965 }
966
967 static int mdc_readpage(struct obd_export *exp, struct md_op_data *op_data,
968 struct page **pages, struct ptlrpc_request **request)
969 {
970 struct ptlrpc_request *req;
971 struct ptlrpc_bulk_desc *desc;
972 int i;
973 wait_queue_head_t waitq;
974 int resends = 0;
975 struct l_wait_info lwi;
976 int rc;
977
978 *request = NULL;
979 init_waitqueue_head(&waitq);
980
981 restart_bulk:
982 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
983 if (!req)
984 return -ENOMEM;
985
986 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
987 if (rc) {
988 ptlrpc_request_free(req);
989 return rc;
990 }
991
992 req->rq_request_portal = MDS_READPAGE_PORTAL;
993 ptlrpc_at_set_req_timeout(req);
994
995 desc = ptlrpc_prep_bulk_imp(req, op_data->op_npages, 1, BULK_PUT_SINK,
996 MDS_BULK_PORTAL);
997 if (!desc) {
998 ptlrpc_request_free(req);
999 return -ENOMEM;
1000 }
1001
1002 /* NB req now owns desc and will free it when it gets freed */
1003 for (i = 0; i < op_data->op_npages; i++)
1004 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_SIZE);
1005
1006 mdc_readdir_pack(req, op_data->op_offset,
1007 PAGE_SIZE * op_data->op_npages,
1008 &op_data->op_fid1);
1009
1010 ptlrpc_request_set_replen(req);
1011 rc = ptlrpc_queue_wait(req);
1012 if (rc) {
1013 ptlrpc_req_finished(req);
1014 if (rc != -ETIMEDOUT)
1015 return rc;
1016
1017 resends++;
1018 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1019 CERROR("too many resend retries, returning error\n");
1020 return -EIO;
1021 }
1022 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends),
1023 NULL, NULL, NULL);
1024 l_wait_event(waitq, 0, &lwi);
1025
1026 goto restart_bulk;
1027 }
1028
1029 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1030 req->rq_bulk->bd_nob_transferred);
1031 if (rc < 0) {
1032 ptlrpc_req_finished(req);
1033 return rc;
1034 }
1035
1036 if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1037 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1038 req->rq_bulk->bd_nob_transferred,
1039 PAGE_SIZE * op_data->op_npages);
1040 ptlrpc_req_finished(req);
1041 return -EPROTO;
1042 }
1043
1044 *request = req;
1045 return 0;
1046 }
1047
1048 static int mdc_statfs(const struct lu_env *env,
1049 struct obd_export *exp, struct obd_statfs *osfs,
1050 __u64 max_age, __u32 flags)
1051 {
1052 struct obd_device *obd = class_exp2obd(exp);
1053 struct ptlrpc_request *req;
1054 struct obd_statfs *msfs;
1055 struct obd_import *imp = NULL;
1056 int rc;
1057
1058 /*
1059 * Since the request might also come from lprocfs, so we need
1060 * sync this with client_disconnect_export Bug15684
1061 */
1062 down_read(&obd->u.cli.cl_sem);
1063 if (obd->u.cli.cl_import)
1064 imp = class_import_get(obd->u.cli.cl_import);
1065 up_read(&obd->u.cli.cl_sem);
1066 if (!imp)
1067 return -ENODEV;
1068
1069 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1070 LUSTRE_MDS_VERSION, MDS_STATFS);
1071 if (!req) {
1072 rc = -ENOMEM;
1073 goto output;
1074 }
1075
1076 ptlrpc_request_set_replen(req);
1077
1078 if (flags & OBD_STATFS_NODELAY) {
1079 /* procfs requests not want stay in wait for avoid deadlock */
1080 req->rq_no_resend = 1;
1081 req->rq_no_delay = 1;
1082 }
1083
1084 rc = ptlrpc_queue_wait(req);
1085 if (rc) {
1086 /* check connection error first */
1087 if (imp->imp_connect_error)
1088 rc = imp->imp_connect_error;
1089 goto out;
1090 }
1091
1092 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1093 if (!msfs) {
1094 rc = -EPROTO;
1095 goto out;
1096 }
1097
1098 *osfs = *msfs;
1099 out:
1100 ptlrpc_req_finished(req);
1101 output:
1102 class_import_put(imp);
1103 return rc;
1104 }
1105
1106 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1107 {
1108 __u32 keylen, vallen;
1109 void *key;
1110 int rc;
1111
1112 if (gf->gf_pathlen > PATH_MAX)
1113 return -ENAMETOOLONG;
1114 if (gf->gf_pathlen < 2)
1115 return -EOVERFLOW;
1116
1117 /* Key is KEY_FID2PATH + getinfo_fid2path description */
1118 keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1119 key = kzalloc(keylen, GFP_NOFS);
1120 if (!key)
1121 return -ENOMEM;
1122 memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1123 memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1124
1125 CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1126 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1127
1128 if (!fid_is_sane(&gf->gf_fid)) {
1129 rc = -EINVAL;
1130 goto out;
1131 }
1132
1133 /* Val is struct getinfo_fid2path result plus path */
1134 vallen = sizeof(*gf) + gf->gf_pathlen;
1135
1136 rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1137 if (rc != 0 && rc != -EREMOTE)
1138 goto out;
1139
1140 if (vallen <= sizeof(*gf)) {
1141 rc = -EPROTO;
1142 goto out;
1143 } else if (vallen > sizeof(*gf) + gf->gf_pathlen) {
1144 rc = -EOVERFLOW;
1145 goto out;
1146 }
1147
1148 CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n%s\n",
1149 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1150
1151 out:
1152 kfree(key);
1153 return rc;
1154 }
1155
1156 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1157 struct hsm_progress_kernel *hpk)
1158 {
1159 struct obd_import *imp = class_exp2cliimp(exp);
1160 struct hsm_progress_kernel *req_hpk;
1161 struct ptlrpc_request *req;
1162 int rc;
1163
1164 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1165 LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1166 if (!req) {
1167 rc = -ENOMEM;
1168 goto out;
1169 }
1170
1171 mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1172
1173 /* Copy hsm_progress struct */
1174 req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1175 if (!req_hpk) {
1176 rc = -EPROTO;
1177 goto out;
1178 }
1179
1180 *req_hpk = *hpk;
1181 req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1182
1183 ptlrpc_request_set_replen(req);
1184
1185 rc = mdc_queue_wait(req);
1186 out:
1187 ptlrpc_req_finished(req);
1188 return rc;
1189 }
1190
1191 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1192 {
1193 __u32 *archive_mask;
1194 struct ptlrpc_request *req;
1195 int rc;
1196
1197 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1198 LUSTRE_MDS_VERSION,
1199 MDS_HSM_CT_REGISTER);
1200 if (!req) {
1201 rc = -ENOMEM;
1202 goto out;
1203 }
1204
1205 mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1206
1207 /* Copy hsm_progress struct */
1208 archive_mask = req_capsule_client_get(&req->rq_pill,
1209 &RMF_MDS_HSM_ARCHIVE);
1210 if (!archive_mask) {
1211 rc = -EPROTO;
1212 goto out;
1213 }
1214
1215 *archive_mask = archives;
1216
1217 ptlrpc_request_set_replen(req);
1218
1219 rc = mdc_queue_wait(req);
1220 out:
1221 ptlrpc_req_finished(req);
1222 return rc;
1223 }
1224
1225 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1226 struct md_op_data *op_data)
1227 {
1228 struct hsm_current_action *hca = op_data->op_data;
1229 struct hsm_current_action *req_hca;
1230 struct ptlrpc_request *req;
1231 int rc;
1232
1233 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1234 &RQF_MDS_HSM_ACTION);
1235 if (!req)
1236 return -ENOMEM;
1237
1238 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1239 if (rc) {
1240 ptlrpc_request_free(req);
1241 return rc;
1242 }
1243
1244 mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1245 op_data->op_suppgids[0], 0);
1246
1247 ptlrpc_request_set_replen(req);
1248
1249 rc = mdc_queue_wait(req);
1250 if (rc)
1251 goto out;
1252
1253 req_hca = req_capsule_server_get(&req->rq_pill,
1254 &RMF_MDS_HSM_CURRENT_ACTION);
1255 if (!req_hca) {
1256 rc = -EPROTO;
1257 goto out;
1258 }
1259
1260 *hca = *req_hca;
1261
1262 out:
1263 ptlrpc_req_finished(req);
1264 return rc;
1265 }
1266
1267 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1268 {
1269 struct ptlrpc_request *req;
1270 int rc;
1271
1272 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1273 LUSTRE_MDS_VERSION,
1274 MDS_HSM_CT_UNREGISTER);
1275 if (!req) {
1276 rc = -ENOMEM;
1277 goto out;
1278 }
1279
1280 mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1281
1282 ptlrpc_request_set_replen(req);
1283
1284 rc = mdc_queue_wait(req);
1285 out:
1286 ptlrpc_req_finished(req);
1287 return rc;
1288 }
1289
1290 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1291 struct md_op_data *op_data)
1292 {
1293 struct hsm_user_state *hus = op_data->op_data;
1294 struct hsm_user_state *req_hus;
1295 struct ptlrpc_request *req;
1296 int rc;
1297
1298 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1299 &RQF_MDS_HSM_STATE_GET);
1300 if (!req)
1301 return -ENOMEM;
1302
1303 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1304 if (rc != 0) {
1305 ptlrpc_request_free(req);
1306 return rc;
1307 }
1308
1309 mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1310 op_data->op_suppgids[0], 0);
1311
1312 ptlrpc_request_set_replen(req);
1313
1314 rc = mdc_queue_wait(req);
1315 if (rc)
1316 goto out;
1317
1318 req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1319 if (!req_hus) {
1320 rc = -EPROTO;
1321 goto out;
1322 }
1323
1324 *hus = *req_hus;
1325
1326 out:
1327 ptlrpc_req_finished(req);
1328 return rc;
1329 }
1330
1331 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1332 struct md_op_data *op_data)
1333 {
1334 struct hsm_state_set *hss = op_data->op_data;
1335 struct hsm_state_set *req_hss;
1336 struct ptlrpc_request *req;
1337 int rc;
1338
1339 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1340 &RQF_MDS_HSM_STATE_SET);
1341 if (!req)
1342 return -ENOMEM;
1343
1344 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1345 if (rc) {
1346 ptlrpc_request_free(req);
1347 return rc;
1348 }
1349
1350 mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1351 op_data->op_suppgids[0], 0);
1352
1353 /* Copy states */
1354 req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1355 if (!req_hss) {
1356 rc = -EPROTO;
1357 goto out;
1358 }
1359 *req_hss = *hss;
1360
1361 ptlrpc_request_set_replen(req);
1362
1363 rc = mdc_queue_wait(req);
1364 out:
1365 ptlrpc_req_finished(req);
1366 return rc;
1367 }
1368
1369 static int mdc_ioc_hsm_request(struct obd_export *exp,
1370 struct hsm_user_request *hur)
1371 {
1372 struct obd_import *imp = class_exp2cliimp(exp);
1373 struct ptlrpc_request *req;
1374 struct hsm_request *req_hr;
1375 struct hsm_user_item *req_hui;
1376 char *req_opaque;
1377 int rc;
1378
1379 req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1380 if (!req) {
1381 rc = -ENOMEM;
1382 goto out;
1383 }
1384
1385 req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1386 hur->hur_request.hr_itemcount
1387 * sizeof(struct hsm_user_item));
1388 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1389 hur->hur_request.hr_data_len);
1390
1391 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1392 if (rc) {
1393 ptlrpc_request_free(req);
1394 return rc;
1395 }
1396
1397 mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1398
1399 /* Copy hsm_request struct */
1400 req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1401 if (!req_hr) {
1402 rc = -EPROTO;
1403 goto out;
1404 }
1405 *req_hr = hur->hur_request;
1406
1407 /* Copy hsm_user_item structs */
1408 req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1409 if (!req_hui) {
1410 rc = -EPROTO;
1411 goto out;
1412 }
1413 memcpy(req_hui, hur->hur_user_item,
1414 hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1415
1416 /* Copy opaque field */
1417 req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1418 if (!req_opaque) {
1419 rc = -EPROTO;
1420 goto out;
1421 }
1422 memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1423
1424 ptlrpc_request_set_replen(req);
1425
1426 rc = mdc_queue_wait(req);
1427 out:
1428 ptlrpc_req_finished(req);
1429 return rc;
1430 }
1431
1432 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
1433 {
1434 struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1435
1436 LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1437
1438 lh->kuc_magic = KUC_MAGIC;
1439 lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1440 lh->kuc_flags = flags;
1441 lh->kuc_msgtype = CL_RECORD;
1442 lh->kuc_msglen = len;
1443 return lh;
1444 }
1445
1446 #define D_CHANGELOG 0
1447
1448 struct changelog_show {
1449 __u64 cs_startrec;
1450 __u32 cs_flags;
1451 struct file *cs_fp;
1452 char *cs_buf;
1453 struct obd_device *cs_obd;
1454 };
1455
1456 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1457 struct llog_rec_hdr *hdr, void *data)
1458 {
1459 struct changelog_show *cs = data;
1460 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1461 struct kuc_hdr *lh;
1462 int len, rc;
1463
1464 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1465 rc = -EINVAL;
1466 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1467 cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
1468 rec->cr.cr_type, rc);
1469 return rc;
1470 }
1471
1472 if (rec->cr.cr_index < cs->cs_startrec) {
1473 /* Skip entries earlier than what we are interested in */
1474 CDEBUG(D_CHANGELOG, "rec=%llu start=%llu\n",
1475 rec->cr.cr_index, cs->cs_startrec);
1476 return 0;
1477 }
1478
1479 CDEBUG(D_CHANGELOG, "%llu %02d%-5s %llu 0x%x t="DFID" p="DFID
1480 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
1481 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1482 rec->cr.cr_flags & CLF_FLAGMASK,
1483 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1484 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1485
1486 len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1487
1488 /* Set up the message */
1489 lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1490 memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1491
1492 rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1493 CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len, rc);
1494
1495 return rc;
1496 }
1497
1498 static int mdc_changelog_send_thread(void *csdata)
1499 {
1500 struct changelog_show *cs = csdata;
1501 struct llog_ctxt *ctxt = NULL;
1502 struct llog_handle *llh = NULL;
1503 struct kuc_hdr *kuch;
1504 int rc;
1505
1506 CDEBUG(D_CHANGELOG, "changelog to fp=%p start %llu\n",
1507 cs->cs_fp, cs->cs_startrec);
1508
1509 cs->cs_buf = kzalloc(KUC_CHANGELOG_MSG_MAXSIZE, GFP_NOFS);
1510 if (!cs->cs_buf) {
1511 rc = -ENOMEM;
1512 goto out;
1513 }
1514
1515 /* Set up the remote catalog handle */
1516 ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1517 if (!ctxt) {
1518 rc = -ENOENT;
1519 goto out;
1520 }
1521 rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1522 LLOG_OPEN_EXISTS);
1523 if (rc) {
1524 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1525 cs->cs_obd->obd_name, rc);
1526 goto out;
1527 }
1528 rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
1529 if (rc) {
1530 CERROR("llog_init_handle failed %d\n", rc);
1531 goto out;
1532 }
1533
1534 rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1535
1536 /* Send EOF no matter what our result */
1537 kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags);
1538 if (kuch) {
1539 kuch->kuc_msgtype = CL_EOF;
1540 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1541 }
1542
1543 out:
1544 fput(cs->cs_fp);
1545 if (llh)
1546 llog_cat_close(NULL, llh);
1547 if (ctxt)
1548 llog_ctxt_put(ctxt);
1549 kfree(cs->cs_buf);
1550 kfree(cs);
1551 return rc;
1552 }
1553
1554 static int mdc_ioc_changelog_send(struct obd_device *obd,
1555 struct ioc_changelog *icc)
1556 {
1557 struct changelog_show *cs;
1558 struct task_struct *task;
1559 int rc;
1560
1561 /* Freed in mdc_changelog_send_thread */
1562 cs = kzalloc(sizeof(*cs), GFP_NOFS);
1563 if (!cs)
1564 return -ENOMEM;
1565
1566 cs->cs_obd = obd;
1567 cs->cs_startrec = icc->icc_recno;
1568 /* matching fput in mdc_changelog_send_thread */
1569 cs->cs_fp = fget(icc->icc_id);
1570 cs->cs_flags = icc->icc_flags;
1571
1572 /*
1573 * New thread because we should return to user app before
1574 * writing into our pipe
1575 */
1576 task = kthread_run(mdc_changelog_send_thread, cs,
1577 "mdc_clg_send_thread");
1578 if (IS_ERR(task)) {
1579 rc = PTR_ERR(task);
1580 CERROR("%s: can't start changelog thread: rc = %d\n",
1581 obd->obd_name, rc);
1582 kfree(cs);
1583 } else {
1584 rc = 0;
1585 CDEBUG(D_CHANGELOG, "%s: started changelog thread\n",
1586 obd->obd_name);
1587 }
1588
1589 CERROR("Failed to start changelog thread: %d\n", rc);
1590 return rc;
1591 }
1592
1593 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1594 struct lustre_kernelcomm *lk);
1595
1596 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
1597 struct obd_quotactl *oqctl)
1598 {
1599 struct client_obd *cli = &exp->exp_obd->u.cli;
1600 struct ptlrpc_request *req;
1601 struct obd_quotactl *body;
1602 int rc;
1603
1604 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1605 &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
1606 MDS_QUOTACHECK);
1607 if (!req)
1608 return -ENOMEM;
1609
1610 body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1611 *body = *oqctl;
1612
1613 ptlrpc_request_set_replen(req);
1614
1615 /* the next poll will find -ENODATA, that means quotacheck is
1616 * going on
1617 */
1618 cli->cl_qchk_stat = -ENODATA;
1619 rc = ptlrpc_queue_wait(req);
1620 if (rc)
1621 cli->cl_qchk_stat = rc;
1622 ptlrpc_req_finished(req);
1623 return rc;
1624 }
1625
1626 static int mdc_quota_poll_check(struct obd_export *exp,
1627 struct if_quotacheck *qchk)
1628 {
1629 struct client_obd *cli = &exp->exp_obd->u.cli;
1630 int rc;
1631
1632 qchk->obd_uuid = cli->cl_target_uuid;
1633 memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
1634
1635 rc = cli->cl_qchk_stat;
1636 /* the client is not the previous one */
1637 if (rc == CL_NOT_QUOTACHECKED)
1638 rc = -EINTR;
1639 return rc;
1640 }
1641
1642 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1643 struct obd_quotactl *oqctl)
1644 {
1645 struct ptlrpc_request *req;
1646 struct obd_quotactl *oqc;
1647 int rc;
1648
1649 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1650 &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1651 MDS_QUOTACTL);
1652 if (!req)
1653 return -ENOMEM;
1654
1655 oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1656 *oqc = *oqctl;
1657
1658 ptlrpc_request_set_replen(req);
1659 ptlrpc_at_set_req_timeout(req);
1660 req->rq_no_resend = 1;
1661
1662 rc = ptlrpc_queue_wait(req);
1663 if (rc)
1664 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1665
1666 if (req->rq_repmsg) {
1667 oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1668 if (oqc) {
1669 *oqctl = *oqc;
1670 } else if (!rc) {
1671 CERROR("Can't unpack obd_quotactl\n");
1672 rc = -EPROTO;
1673 }
1674 } else if (!rc) {
1675 CERROR("Can't unpack obd_quotactl\n");
1676 rc = -EPROTO;
1677 }
1678 ptlrpc_req_finished(req);
1679
1680 return rc;
1681 }
1682
1683 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1684 struct md_op_data *op_data)
1685 {
1686 LIST_HEAD(cancels);
1687 struct ptlrpc_request *req;
1688 int rc, count;
1689 struct mdc_swap_layouts *msl, *payload;
1690
1691 msl = op_data->op_data;
1692
1693 /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1694 * first thing it will do is to cancel the 2 layout
1695 * locks hold by this client.
1696 * So the client must cancel its layout locks on the 2 fids
1697 * with the request RPC to avoid extra RPC round trips
1698 */
1699 count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1700 LCK_CR, MDS_INODELOCK_LAYOUT);
1701 count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1702 LCK_CR, MDS_INODELOCK_LAYOUT);
1703
1704 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1705 &RQF_MDS_SWAP_LAYOUTS);
1706 if (!req) {
1707 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1708 return -ENOMEM;
1709 }
1710
1711 rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1712 if (rc) {
1713 ptlrpc_request_free(req);
1714 return rc;
1715 }
1716
1717 mdc_swap_layouts_pack(req, op_data);
1718
1719 payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1720 LASSERT(payload);
1721
1722 *payload = *msl;
1723
1724 ptlrpc_request_set_replen(req);
1725
1726 rc = ptlrpc_queue_wait(req);
1727
1728 ptlrpc_req_finished(req);
1729 return rc;
1730 }
1731
1732 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1733 void *karg, void __user *uarg)
1734 {
1735 struct obd_device *obd = exp->exp_obd;
1736 struct obd_ioctl_data *data = karg;
1737 struct obd_import *imp = obd->u.cli.cl_import;
1738 int rc;
1739
1740 if (!try_module_get(THIS_MODULE)) {
1741 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
1742 module_name(THIS_MODULE));
1743 return -EINVAL;
1744 }
1745 switch (cmd) {
1746 case OBD_IOC_CHANGELOG_SEND:
1747 rc = mdc_ioc_changelog_send(obd, karg);
1748 goto out;
1749 case OBD_IOC_CHANGELOG_CLEAR: {
1750 struct ioc_changelog *icc = karg;
1751 struct changelog_setinfo cs = {
1752 .cs_recno = icc->icc_recno,
1753 .cs_id = icc->icc_id
1754 };
1755
1756 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
1757 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1758 NULL);
1759 goto out;
1760 }
1761 case OBD_IOC_FID2PATH:
1762 rc = mdc_ioc_fid2path(exp, karg);
1763 goto out;
1764 case LL_IOC_HSM_CT_START:
1765 rc = mdc_ioc_hsm_ct_start(exp, karg);
1766 /* ignore if it was already registered on this MDS. */
1767 if (rc == -EEXIST)
1768 rc = 0;
1769 goto out;
1770 case LL_IOC_HSM_PROGRESS:
1771 rc = mdc_ioc_hsm_progress(exp, karg);
1772 goto out;
1773 case LL_IOC_HSM_STATE_GET:
1774 rc = mdc_ioc_hsm_state_get(exp, karg);
1775 goto out;
1776 case LL_IOC_HSM_STATE_SET:
1777 rc = mdc_ioc_hsm_state_set(exp, karg);
1778 goto out;
1779 case LL_IOC_HSM_ACTION:
1780 rc = mdc_ioc_hsm_current_action(exp, karg);
1781 goto out;
1782 case LL_IOC_HSM_REQUEST:
1783 rc = mdc_ioc_hsm_request(exp, karg);
1784 goto out;
1785 case OBD_IOC_CLIENT_RECOVER:
1786 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1787 if (rc < 0)
1788 goto out;
1789 rc = 0;
1790 goto out;
1791 case IOC_OSC_SET_ACTIVE:
1792 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1793 goto out;
1794 case OBD_IOC_POLL_QUOTACHECK:
1795 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
1796 goto out;
1797 case OBD_IOC_PING_TARGET:
1798 rc = ptlrpc_obd_ping(obd);
1799 goto out;
1800 /*
1801 * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1802 * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1803 * there'd be no LMV layer thus we might be called here. Eventually
1804 * this code should be removed.
1805 * bz20731, LU-592.
1806 */
1807 case IOC_OBD_STATFS: {
1808 struct obd_statfs stat_buf = {0};
1809
1810 if (*((__u32 *) data->ioc_inlbuf2) != 0) {
1811 rc = -ENODEV;
1812 goto out;
1813 }
1814
1815 /* copy UUID */
1816 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1817 min_t(size_t, data->ioc_plen2,
1818 sizeof(struct obd_uuid)))) {
1819 rc = -EFAULT;
1820 goto out;
1821 }
1822
1823 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1824 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1825 0);
1826 if (rc != 0)
1827 goto out;
1828
1829 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1830 min_t(size_t, data->ioc_plen1,
1831 sizeof(stat_buf)))) {
1832 rc = -EFAULT;
1833 goto out;
1834 }
1835
1836 rc = 0;
1837 goto out;
1838 }
1839 case OBD_IOC_QUOTACTL: {
1840 struct if_quotactl *qctl = karg;
1841 struct obd_quotactl *oqctl;
1842
1843 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1844 if (!oqctl) {
1845 rc = -ENOMEM;
1846 goto out;
1847 }
1848
1849 QCTL_COPY(oqctl, qctl);
1850 rc = obd_quotactl(exp, oqctl);
1851 if (rc == 0) {
1852 QCTL_COPY(qctl, oqctl);
1853 qctl->qc_valid = QC_MDTIDX;
1854 qctl->obd_uuid = obd->u.cli.cl_target_uuid;
1855 }
1856
1857 kfree(oqctl);
1858 goto out;
1859 }
1860 case LL_IOC_GET_CONNECT_FLAGS:
1861 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
1862 sizeof(*exp_connect_flags_ptr(exp)))) {
1863 rc = -EFAULT;
1864 goto out;
1865 }
1866
1867 rc = 0;
1868 goto out;
1869 case LL_IOC_LOV_SWAP_LAYOUTS:
1870 rc = mdc_ioc_swap_layouts(exp, karg);
1871 goto out;
1872 default:
1873 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
1874 rc = -ENOTTY;
1875 goto out;
1876 }
1877 out:
1878 module_put(THIS_MODULE);
1879
1880 return rc;
1881 }
1882
1883 static int mdc_get_info_rpc(struct obd_export *exp,
1884 u32 keylen, void *key,
1885 int vallen, void *val)
1886 {
1887 struct obd_import *imp = class_exp2cliimp(exp);
1888 struct ptlrpc_request *req;
1889 char *tmp;
1890 int rc = -EINVAL;
1891
1892 req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1893 if (!req)
1894 return -ENOMEM;
1895
1896 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1897 RCL_CLIENT, keylen);
1898 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1899 RCL_CLIENT, sizeof(__u32));
1900
1901 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1902 if (rc) {
1903 ptlrpc_request_free(req);
1904 return rc;
1905 }
1906
1907 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
1908 memcpy(tmp, key, keylen);
1909 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
1910 memcpy(tmp, &vallen, sizeof(__u32));
1911
1912 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1913 RCL_SERVER, vallen);
1914 ptlrpc_request_set_replen(req);
1915
1916 rc = ptlrpc_queue_wait(req);
1917 /* -EREMOTE means the get_info result is partial, and it needs to
1918 * continue on another MDT, see fid2path part in lmv_iocontrol
1919 */
1920 if (rc == 0 || rc == -EREMOTE) {
1921 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
1922 memcpy(val, tmp, vallen);
1923 if (ptlrpc_rep_need_swab(req)) {
1924 if (KEY_IS(KEY_FID2PATH))
1925 lustre_swab_fid2path(val);
1926 }
1927 }
1928 ptlrpc_req_finished(req);
1929
1930 return rc;
1931 }
1932
1933 static void lustre_swab_hai(struct hsm_action_item *h)
1934 {
1935 __swab32s(&h->hai_len);
1936 __swab32s(&h->hai_action);
1937 lustre_swab_lu_fid(&h->hai_fid);
1938 lustre_swab_lu_fid(&h->hai_dfid);
1939 __swab64s(&h->hai_cookie);
1940 __swab64s(&h->hai_extent.offset);
1941 __swab64s(&h->hai_extent.length);
1942 __swab64s(&h->hai_gid);
1943 }
1944
1945 static void lustre_swab_hal(struct hsm_action_list *h)
1946 {
1947 struct hsm_action_item *hai;
1948 int i;
1949
1950 __swab32s(&h->hal_version);
1951 __swab32s(&h->hal_count);
1952 __swab32s(&h->hal_archive_id);
1953 __swab64s(&h->hal_flags);
1954 hai = hai_first(h);
1955 for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
1956 lustre_swab_hai(hai);
1957 }
1958
1959 static void lustre_swab_kuch(struct kuc_hdr *l)
1960 {
1961 __swab16s(&l->kuc_magic);
1962 /* __u8 l->kuc_transport */
1963 __swab16s(&l->kuc_msgtype);
1964 __swab16s(&l->kuc_msglen);
1965 }
1966
1967 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1968 struct lustre_kernelcomm *lk)
1969 {
1970 struct obd_import *imp = class_exp2cliimp(exp);
1971 __u32 archive = lk->lk_data;
1972 int rc = 0;
1973
1974 if (lk->lk_group != KUC_GRP_HSM) {
1975 CERROR("Bad copytool group %d\n", lk->lk_group);
1976 return -EINVAL;
1977 }
1978
1979 CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
1980 lk->lk_uid, lk->lk_group, lk->lk_flags);
1981
1982 if (lk->lk_flags & LK_FLG_STOP) {
1983 /* Unregister with the coordinator */
1984 rc = mdc_ioc_hsm_ct_unregister(imp);
1985 } else {
1986 rc = mdc_ioc_hsm_ct_register(imp, archive);
1987 }
1988
1989 return rc;
1990 }
1991
1992 /**
1993 * Send a message to any listening copytools
1994 * @param val KUC message (kuc_hdr + hsm_action_list)
1995 * @param len total length of message
1996 */
1997 static int mdc_hsm_copytool_send(int len, void *val)
1998 {
1999 struct kuc_hdr *lh = (struct kuc_hdr *)val;
2000 struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
2001
2002 if (len < sizeof(*lh) + sizeof(*hal)) {
2003 CERROR("Short HSM message %d < %d\n", len,
2004 (int) (sizeof(*lh) + sizeof(*hal)));
2005 return -EPROTO;
2006 }
2007 if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2008 lustre_swab_kuch(lh);
2009 lustre_swab_hal(hal);
2010 } else if (lh->kuc_magic != KUC_MAGIC) {
2011 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2012 return -EPROTO;
2013 }
2014
2015 CDEBUG(D_HSM,
2016 "Received message mg=%x t=%d m=%d l=%d actions=%d on %s\n",
2017 lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2018 lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2019
2020 /* Broadcast to HSM listeners */
2021 return libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2022 }
2023
2024 /**
2025 * callback function passed to kuc for re-registering each HSM copytool
2026 * running on MDC, after MDT shutdown/recovery.
2027 * @param data copytool registration data
2028 * @param cb_arg callback argument (obd_import)
2029 */
2030 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2031 {
2032 struct kkuc_ct_data *kcd = data;
2033 struct obd_import *imp = (struct obd_import *)cb_arg;
2034 int rc;
2035
2036 if (!kcd || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2037 return -EPROTO;
2038
2039 if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2040 return 0;
2041
2042 CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2043 imp->imp_obd->obd_name, kcd->kcd_archive);
2044 rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2045
2046 /* ignore error if the copytool is already registered */
2047 return (rc == -EEXIST) ? 0 : rc;
2048 }
2049
2050 static int mdc_set_info_async(const struct lu_env *env,
2051 struct obd_export *exp,
2052 u32 keylen, void *key,
2053 u32 vallen, void *val,
2054 struct ptlrpc_request_set *set)
2055 {
2056 struct obd_import *imp = class_exp2cliimp(exp);
2057 int rc;
2058
2059 if (KEY_IS(KEY_READ_ONLY)) {
2060 if (vallen != sizeof(int))
2061 return -EINVAL;
2062
2063 spin_lock(&imp->imp_lock);
2064 if (*((int *)val)) {
2065 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2066 imp->imp_connect_data.ocd_connect_flags |=
2067 OBD_CONNECT_RDONLY;
2068 } else {
2069 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2070 imp->imp_connect_data.ocd_connect_flags &=
2071 ~OBD_CONNECT_RDONLY;
2072 }
2073 spin_unlock(&imp->imp_lock);
2074
2075 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2076 keylen, key, vallen, val, set);
2077 return rc;
2078 }
2079 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2080 sptlrpc_conf_client_adapt(exp->exp_obd);
2081 return 0;
2082 }
2083 if (KEY_IS(KEY_FLUSH_CTX)) {
2084 sptlrpc_import_flush_my_ctx(imp);
2085 return 0;
2086 }
2087 if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2088 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2089 keylen, key, vallen, val, set);
2090 return rc;
2091 }
2092 if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2093 rc = mdc_hsm_copytool_send(vallen, val);
2094 return rc;
2095 }
2096
2097 CERROR("Unknown key %s\n", (char *)key);
2098 return -EINVAL;
2099 }
2100
2101 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2102 __u32 keylen, void *key, __u32 *vallen, void *val,
2103 struct lov_stripe_md *lsm)
2104 {
2105 int rc = -EINVAL;
2106
2107 if (KEY_IS(KEY_MAX_EASIZE)) {
2108 int mdsize, *max_easize;
2109
2110 if (*vallen != sizeof(int))
2111 return -EINVAL;
2112 mdsize = *(int *)val;
2113 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2114 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2115 max_easize = val;
2116 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2117 return 0;
2118 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2119 int *default_easize;
2120
2121 if (*vallen != sizeof(int))
2122 return -EINVAL;
2123 default_easize = val;
2124 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2125 return 0;
2126 } else if (KEY_IS(KEY_CONN_DATA)) {
2127 struct obd_import *imp = class_exp2cliimp(exp);
2128 struct obd_connect_data *data = val;
2129
2130 if (*vallen != sizeof(*data))
2131 return -EINVAL;
2132
2133 *data = imp->imp_connect_data;
2134 return 0;
2135 } else if (KEY_IS(KEY_TGT_COUNT)) {
2136 *((int *)val) = 1;
2137 return 0;
2138 }
2139
2140 rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2141
2142 return rc;
2143 }
2144
2145 static int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
2146 struct ptlrpc_request **request)
2147 {
2148 struct ptlrpc_request *req;
2149 int rc;
2150
2151 *request = NULL;
2152 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2153 if (!req)
2154 return -ENOMEM;
2155
2156 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2157 if (rc) {
2158 ptlrpc_request_free(req);
2159 return rc;
2160 }
2161
2162 mdc_pack_body(req, fid, 0, 0, -1, 0);
2163
2164 ptlrpc_request_set_replen(req);
2165
2166 rc = ptlrpc_queue_wait(req);
2167 if (rc)
2168 ptlrpc_req_finished(req);
2169 else
2170 *request = req;
2171 return rc;
2172 }
2173
2174 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2175 enum obd_import_event event)
2176 {
2177 int rc = 0;
2178
2179 LASSERT(imp->imp_obd == obd);
2180
2181 switch (event) {
2182 case IMP_EVENT_DISCON: {
2183 #if 0
2184 /* XXX Pass event up to OBDs stack. used only for FLD now */
2185 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2186 #endif
2187 break;
2188 }
2189 case IMP_EVENT_INACTIVE: {
2190 struct client_obd *cli = &obd->u.cli;
2191 /*
2192 * Flush current sequence to make client obtain new one
2193 * from server in case of disconnect/reconnect.
2194 */
2195 if (cli->cl_seq)
2196 seq_client_flush(cli->cl_seq);
2197
2198 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2199 break;
2200 }
2201 case IMP_EVENT_INVALIDATE: {
2202 struct ldlm_namespace *ns = obd->obd_namespace;
2203
2204 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2205
2206 break;
2207 }
2208 case IMP_EVENT_ACTIVE:
2209 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2210 /* redo the kuc registration after reconnecting */
2211 if (rc == 0)
2212 /* re-register HSM agents */
2213 rc = libcfs_kkuc_group_foreach(KUC_GRP_HSM,
2214 mdc_hsm_ct_reregister,
2215 (void *)imp);
2216 break;
2217 case IMP_EVENT_OCD:
2218 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2219 break;
2220 case IMP_EVENT_DEACTIVATE:
2221 case IMP_EVENT_ACTIVATE:
2222 break;
2223 default:
2224 CERROR("Unknown import event %x\n", event);
2225 LBUG();
2226 }
2227 return rc;
2228 }
2229
2230 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
2231 struct md_op_data *op_data)
2232 {
2233 struct client_obd *cli = &exp->exp_obd->u.cli;
2234 struct lu_client_seq *seq = cli->cl_seq;
2235
2236 return seq_client_alloc_fid(NULL, seq, fid);
2237 }
2238
2239 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2240 {
2241 struct client_obd *cli = &exp->exp_obd->u.cli;
2242
2243 return &cli->cl_target_uuid;
2244 }
2245
2246 /**
2247 * Determine whether the lock can be canceled before replaying it during
2248 * recovery, non zero value will be return if the lock can be canceled,
2249 * or zero returned for not
2250 */
2251 static int mdc_cancel_weight(struct ldlm_lock *lock)
2252 {
2253 if (lock->l_resource->lr_type != LDLM_IBITS)
2254 return 0;
2255
2256 /* FIXME: if we ever get into a situation where there are too many
2257 * opened files with open locks on a single node, then we really
2258 * should replay these open locks to reget it
2259 */
2260 if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2261 return 0;
2262
2263 return 1;
2264 }
2265
2266 static int mdc_resource_inode_free(struct ldlm_resource *res)
2267 {
2268 if (res->lr_lvb_inode)
2269 res->lr_lvb_inode = NULL;
2270
2271 return 0;
2272 }
2273
2274 static struct ldlm_valblock_ops inode_lvbo = {
2275 .lvbo_free = mdc_resource_inode_free,
2276 };
2277
2278 static int mdc_llog_init(struct obd_device *obd)
2279 {
2280 struct obd_llog_group *olg = &obd->obd_olg;
2281 struct llog_ctxt *ctxt;
2282 int rc;
2283
2284 rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2285 &llog_client_ops);
2286 if (rc)
2287 return rc;
2288
2289 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2290 llog_initiator_connect(ctxt);
2291 llog_ctxt_put(ctxt);
2292
2293 return 0;
2294 }
2295
2296 static void mdc_llog_finish(struct obd_device *obd)
2297 {
2298 struct llog_ctxt *ctxt;
2299
2300 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2301 if (ctxt)
2302 llog_cleanup(NULL, ctxt);
2303 }
2304
2305 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2306 {
2307 struct client_obd *cli = &obd->u.cli;
2308 struct lprocfs_static_vars lvars = { NULL };
2309 int rc;
2310
2311 cli->cl_rpc_lock = kzalloc(sizeof(*cli->cl_rpc_lock), GFP_NOFS);
2312 if (!cli->cl_rpc_lock)
2313 return -ENOMEM;
2314 mdc_init_rpc_lock(cli->cl_rpc_lock);
2315
2316 rc = ptlrpcd_addref();
2317 if (rc < 0)
2318 goto err_rpc_lock;
2319
2320 cli->cl_close_lock = kzalloc(sizeof(*cli->cl_close_lock), GFP_NOFS);
2321 if (!cli->cl_close_lock) {
2322 rc = -ENOMEM;
2323 goto err_ptlrpcd_decref;
2324 }
2325 mdc_init_rpc_lock(cli->cl_close_lock);
2326
2327 rc = client_obd_setup(obd, cfg);
2328 if (rc)
2329 goto err_close_lock;
2330 lprocfs_mdc_init_vars(&lvars);
2331 lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars);
2332 sptlrpc_lprocfs_cliobd_attach(obd);
2333 ptlrpc_lprocfs_register_obd(obd);
2334
2335 ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2336
2337 obd->obd_namespace->ns_lvbo = &inode_lvbo;
2338
2339 rc = mdc_llog_init(obd);
2340 if (rc) {
2341 mdc_cleanup(obd);
2342 CERROR("failed to setup llogging subsystems\n");
2343 }
2344
2345 return rc;
2346
2347 err_close_lock:
2348 kfree(cli->cl_close_lock);
2349 err_ptlrpcd_decref:
2350 ptlrpcd_decref();
2351 err_rpc_lock:
2352 kfree(cli->cl_rpc_lock);
2353 return rc;
2354 }
2355
2356 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
2357 * us to make MDS RPCs with large enough reply buffers to hold a default
2358 * sized EA and cookie without having to calculate this (via a call into the
2359 * LOV + OSCs) each time we make an RPC. The maximum size is also tracked
2360 * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2361 * a large number of stripes is possible. If a larger reply buffer is
2362 * required it will be reallocated in the ptlrpc layer due to overflow.
2363 */
2364 static int mdc_init_ea_size(struct obd_export *exp, int easize,
2365 int def_easize, int cookiesize, int def_cookiesize)
2366 {
2367 struct obd_device *obd = exp->exp_obd;
2368 struct client_obd *cli = &obd->u.cli;
2369
2370 if (cli->cl_max_mds_easize < easize)
2371 cli->cl_max_mds_easize = easize;
2372
2373 if (cli->cl_default_mds_easize < def_easize)
2374 cli->cl_default_mds_easize = def_easize;
2375
2376 if (cli->cl_max_mds_cookiesize < cookiesize)
2377 cli->cl_max_mds_cookiesize = cookiesize;
2378
2379 if (cli->cl_default_mds_cookiesize < def_cookiesize)
2380 cli->cl_default_mds_cookiesize = def_cookiesize;
2381
2382 return 0;
2383 }
2384
2385 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2386 {
2387 switch (stage) {
2388 case OBD_CLEANUP_EARLY:
2389 break;
2390 case OBD_CLEANUP_EXPORTS:
2391 /* Failsafe, ok if racy */
2392 if (obd->obd_type->typ_refcnt <= 1)
2393 libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2394
2395 obd_cleanup_client_import(obd);
2396 ptlrpc_lprocfs_unregister_obd(obd);
2397 lprocfs_obd_cleanup(obd);
2398
2399 mdc_llog_finish(obd);
2400 break;
2401 }
2402 return 0;
2403 }
2404
2405 static int mdc_cleanup(struct obd_device *obd)
2406 {
2407 struct client_obd *cli = &obd->u.cli;
2408
2409 kfree(cli->cl_rpc_lock);
2410 kfree(cli->cl_close_lock);
2411
2412 ptlrpcd_decref();
2413
2414 return client_obd_cleanup(obd);
2415 }
2416
2417 static int mdc_process_config(struct obd_device *obd, u32 len, void *buf)
2418 {
2419 struct lustre_cfg *lcfg = buf;
2420 struct lprocfs_static_vars lvars = { NULL };
2421 int rc = 0;
2422
2423 lprocfs_mdc_init_vars(&lvars);
2424 switch (lcfg->lcfg_command) {
2425 default:
2426 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
2427 lcfg, obd);
2428 if (rc > 0)
2429 rc = 0;
2430 break;
2431 }
2432 return rc;
2433 }
2434
2435 /* get remote permission for current user on fid */
2436 static int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2437 __u32 suppgid, struct ptlrpc_request **request)
2438 {
2439 struct ptlrpc_request *req;
2440 int rc;
2441
2442 LASSERT(client_is_remote(exp));
2443
2444 *request = NULL;
2445 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2446 if (!req)
2447 return -ENOMEM;
2448
2449 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2450 if (rc) {
2451 ptlrpc_request_free(req);
2452 return rc;
2453 }
2454
2455 mdc_pack_body(req, fid, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2456
2457 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2458 sizeof(struct mdt_remote_perm));
2459
2460 ptlrpc_request_set_replen(req);
2461
2462 rc = ptlrpc_queue_wait(req);
2463 if (rc)
2464 ptlrpc_req_finished(req);
2465 else
2466 *request = req;
2467 return rc;
2468 }
2469
2470 static struct obd_ops mdc_obd_ops = {
2471 .owner = THIS_MODULE,
2472 .setup = mdc_setup,
2473 .precleanup = mdc_precleanup,
2474 .cleanup = mdc_cleanup,
2475 .add_conn = client_import_add_conn,
2476 .del_conn = client_import_del_conn,
2477 .connect = client_connect_import,
2478 .disconnect = client_disconnect_export,
2479 .iocontrol = mdc_iocontrol,
2480 .set_info_async = mdc_set_info_async,
2481 .statfs = mdc_statfs,
2482 .fid_init = client_fid_init,
2483 .fid_fini = client_fid_fini,
2484 .fid_alloc = mdc_fid_alloc,
2485 .import_event = mdc_import_event,
2486 .get_info = mdc_get_info,
2487 .process_config = mdc_process_config,
2488 .get_uuid = mdc_get_uuid,
2489 .quotactl = mdc_quotactl,
2490 .quotacheck = mdc_quotacheck
2491 };
2492
2493 static struct md_ops mdc_md_ops = {
2494 .getstatus = mdc_getstatus,
2495 .null_inode = mdc_null_inode,
2496 .find_cbdata = mdc_find_cbdata,
2497 .close = mdc_close,
2498 .create = mdc_create,
2499 .done_writing = mdc_done_writing,
2500 .enqueue = mdc_enqueue,
2501 .getattr = mdc_getattr,
2502 .getattr_name = mdc_getattr_name,
2503 .intent_lock = mdc_intent_lock,
2504 .link = mdc_link,
2505 .is_subdir = mdc_is_subdir,
2506 .rename = mdc_rename,
2507 .setattr = mdc_setattr,
2508 .setxattr = mdc_setxattr,
2509 .getxattr = mdc_getxattr,
2510 .sync = mdc_sync,
2511 .readpage = mdc_readpage,
2512 .unlink = mdc_unlink,
2513 .cancel_unused = mdc_cancel_unused,
2514 .init_ea_size = mdc_init_ea_size,
2515 .set_lock_data = mdc_set_lock_data,
2516 .lock_match = mdc_lock_match,
2517 .get_lustre_md = mdc_get_lustre_md,
2518 .free_lustre_md = mdc_free_lustre_md,
2519 .set_open_replay_data = mdc_set_open_replay_data,
2520 .clear_open_replay_data = mdc_clear_open_replay_data,
2521 .get_remote_perm = mdc_get_remote_perm,
2522 .intent_getattr_async = mdc_intent_getattr_async,
2523 .revalidate_lock = mdc_revalidate_lock
2524 };
2525
2526 static int __init mdc_init(void)
2527 {
2528 struct lprocfs_static_vars lvars = { NULL };
2529
2530 lprocfs_mdc_init_vars(&lvars);
2531
2532 return class_register_type(&mdc_obd_ops, &mdc_md_ops,
2533 LUSTRE_MDC_NAME, NULL);
2534 }
2535
2536 static void /*__exit*/ mdc_exit(void)
2537 {
2538 class_unregister_type(LUSTRE_MDC_NAME);
2539 }
2540
2541 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2542 MODULE_DESCRIPTION("Lustre Metadata Client");
2543 MODULE_VERSION(LUSTRE_VERSION_STRING);
2544 MODULE_LICENSE("GPL");
2545
2546 module_init(mdc_init);
2547 module_exit(mdc_exit);
ubuntu-zesty-kernel mirror