Version:  2.0.40 2.2.26 2.4.37 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2

Linux/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, 2012, Intel Corporation.
 31  */
 32 /*
 33  * This file is part of Lustre, http://www.lustre.org/
 34  * Lustre is a trademark of Sun Microsystems, Inc.
 35  */
 36 
 37 #define DEBUG_SUBSYSTEM S_MDC
 38 
 39 # include <linux/module.h>
 40 # include <linux/pagemap.h>
 41 # include <linux/miscdevice.h>
 42 # include <linux/init.h>
 43 # include <linux/utsname.h>
 44 
 45 #include "../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 
 52 #include "mdc_internal.h"
 53 
 54 #define REQUEST_MINOR 244
 55 
 56 struct mdc_renew_capa_args {
 57         struct obd_capa *ra_oc;
 58         renew_capa_cb_t  ra_cb;
 59 };
 60 
 61 static int mdc_cleanup(struct obd_device *obd);
 62 
 63 static int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
 64                     const struct req_msg_field *field, struct obd_capa **oc)
 65 {
 66         struct lustre_capa *capa;
 67         struct obd_capa *c;
 68 
 69         /* swabbed already in mdc_enqueue */
 70         capa = req_capsule_server_get(&req->rq_pill, field);
 71         if (capa == NULL)
 72                 return -EPROTO;
 73 
 74         c = alloc_capa(CAPA_SITE_CLIENT);
 75         if (IS_ERR(c)) {
 76                 CDEBUG(D_INFO, "alloc capa failed!\n");
 77                 return PTR_ERR(c);
 78         }
 79 
 80         c->c_capa = *capa;
 81         *oc = c;
 82         return 0;
 83 }
 84 
 85 static inline int mdc_queue_wait(struct ptlrpc_request *req)
 86 {
 87         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
 88         int rc;
 89 
 90         /* mdc_enter_request() ensures that this client has no more
 91          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
 92          * against an MDT. */
 93         rc = mdc_enter_request(cli);
 94         if (rc != 0)
 95                 return rc;
 96 
 97         rc = ptlrpc_queue_wait(req);
 98         mdc_exit_request(cli);
 99 
100         return rc;
101 }
102 
103 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
104 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
105 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
106                           struct obd_capa **pc, int level, int msg_flags)
107 {
108         struct ptlrpc_request *req;
109         struct mdt_body       *body;
110         int                 rc;
111 
112         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
113                                         LUSTRE_MDS_VERSION, MDS_GETSTATUS);
114         if (req == NULL)
115                 return -ENOMEM;
116 
117         mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
118         lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
119         req->rq_send_state = level;
120 
121         ptlrpc_request_set_replen(req);
122 
123         rc = ptlrpc_queue_wait(req);
124         if (rc)
125                 goto out;
126 
127         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
128         if (body == NULL) {
129                 rc = -EPROTO;
130                 goto out;
131         }
132 
133         if (body->valid & OBD_MD_FLMDSCAPA) {
134                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
135                 if (rc)
136                         goto out;
137         }
138 
139         *rootfid = body->fid1;
140         CDEBUG(D_NET,
141                "root fid="DFID", last_committed=%llu\n",
142                PFID(rootfid),
143                lustre_msg_get_last_committed(req->rq_repmsg));
144 out:
145         ptlrpc_req_finished(req);
146         return rc;
147 }
148 
149 /* This should be mdc_get_info("rootfid") */
150 static int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
151                   struct obd_capa **pc)
152 {
153         return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
154                               LUSTRE_IMP_FULL, 0);
155 }
156 
157 /*
158  * This function now is known to always saying that it will receive 4 buffers
159  * from server. Even for cases when acl_size and md_size is zero, RPC header
160  * will contain 4 fields and RPC itself will contain zero size fields. This is
161  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
162  * and thus zero, it shrinks it, making zero size. The same story about
163  * md_size. And this is course of problem when client waits for smaller number
164  * of fields. This issue will be fixed later when client gets aware of RPC
165  * layouts.  --umka
166  */
167 static int mdc_getattr_common(struct obd_export *exp,
168                               struct ptlrpc_request *req)
169 {
170         struct req_capsule *pill = &req->rq_pill;
171         struct mdt_body    *body;
172         void           *eadata;
173         int              rc;
174 
175         /* Request message already built. */
176         rc = ptlrpc_queue_wait(req);
177         if (rc != 0)
178                 return rc;
179 
180         /* sanity check for the reply */
181         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
182         if (body == NULL)
183                 return -EPROTO;
184 
185         CDEBUG(D_NET, "mode: %o\n", body->mode);
186 
187         if (body->eadatasize != 0) {
188                 mdc_update_max_ea_from_body(exp, body);
189 
190                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
191                                                       body->eadatasize);
192                 if (eadata == NULL)
193                         return -EPROTO;
194         }
195 
196         if (body->valid & OBD_MD_FLRMTPERM) {
197                 struct mdt_remote_perm *perm;
198 
199                 LASSERT(client_is_remote(exp));
200                 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
201                                                 lustre_swab_mdt_remote_perm);
202                 if (perm == NULL)
203                         return -EPROTO;
204         }
205 
206         if (body->valid & OBD_MD_FLMDSCAPA) {
207                 struct lustre_capa *capa;
208 
209                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
210                 if (capa == NULL)
211                         return -EPROTO;
212         }
213 
214         return 0;
215 }
216 
217 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
218                 struct ptlrpc_request **request)
219 {
220         struct ptlrpc_request *req;
221         int                 rc;
222 
223         /* Single MDS without an LMV case */
224         if (op_data->op_flags & MF_GET_MDT_IDX) {
225                 op_data->op_mds = 0;
226                 return 0;
227         }
228         *request = NULL;
229         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
230         if (req == NULL)
231                 return -ENOMEM;
232 
233         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
234 
235         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
236         if (rc) {
237                 ptlrpc_request_free(req);
238                 return rc;
239         }
240 
241         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
242                       op_data->op_valid, op_data->op_mode, -1, 0);
243 
244         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
245                              op_data->op_mode);
246         if (op_data->op_valid & OBD_MD_FLRMTPERM) {
247                 LASSERT(client_is_remote(exp));
248                 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
249                                      sizeof(struct mdt_remote_perm));
250         }
251         ptlrpc_request_set_replen(req);
252 
253         rc = mdc_getattr_common(exp, req);
254         if (rc)
255                 ptlrpc_req_finished(req);
256         else
257                 *request = req;
258         return rc;
259 }
260 
261 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
262                      struct ptlrpc_request **request)
263 {
264         struct ptlrpc_request *req;
265         int                 rc;
266 
267         *request = NULL;
268         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
269                                    &RQF_MDS_GETATTR_NAME);
270         if (req == NULL)
271                 return -ENOMEM;
272 
273         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
274         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
275                              op_data->op_namelen + 1);
276 
277         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
278         if (rc) {
279                 ptlrpc_request_free(req);
280                 return rc;
281         }
282 
283         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
284                       op_data->op_valid, op_data->op_mode,
285                       op_data->op_suppgids[0], 0);
286 
287         if (op_data->op_name) {
288                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
289 
290                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
291                                 op_data->op_namelen);
292                 memcpy(name, op_data->op_name, op_data->op_namelen);
293         }
294 
295         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
296                              op_data->op_mode);
297         ptlrpc_request_set_replen(req);
298 
299         rc = mdc_getattr_common(exp, req);
300         if (rc)
301                 ptlrpc_req_finished(req);
302         else
303                 *request = req;
304         return rc;
305 }
306 
307 static int mdc_is_subdir(struct obd_export *exp,
308                          const struct lu_fid *pfid,
309                          const struct lu_fid *cfid,
310                          struct ptlrpc_request **request)
311 {
312         struct ptlrpc_request  *req;
313         int                  rc;
314 
315         *request = NULL;
316         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
317                                         &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
318                                         MDS_IS_SUBDIR);
319         if (req == NULL)
320                 return -ENOMEM;
321 
322         mdc_is_subdir_pack(req, pfid, cfid, 0);
323         ptlrpc_request_set_replen(req);
324 
325         rc = ptlrpc_queue_wait(req);
326         if (rc && rc != -EREMOTE)
327                 ptlrpc_req_finished(req);
328         else
329                 *request = req;
330         return rc;
331 }
332 
333 static int mdc_xattr_common(struct obd_export *exp,
334                             const struct req_format *fmt,
335                             const struct lu_fid *fid,
336                             struct obd_capa *oc, int opcode, u64 valid,
337                             const char *xattr_name, const char *input,
338                             int input_size, int output_size, int flags,
339                             __u32 suppgid, struct ptlrpc_request **request)
340 {
341         struct ptlrpc_request *req;
342         int   xattr_namelen = 0;
343         char *tmp;
344         int   rc;
345 
346         *request = NULL;
347         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
348         if (req == NULL)
349                 return -ENOMEM;
350 
351         mdc_set_capa_size(req, &RMF_CAPA1, oc);
352         if (xattr_name) {
353                 xattr_namelen = strlen(xattr_name) + 1;
354                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
355                                      xattr_namelen);
356         }
357         if (input_size) {
358                 LASSERT(input);
359                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
360                                      input_size);
361         }
362 
363         /* Flush local XATTR locks to get rid of a possible cancel RPC */
364         if (opcode == MDS_REINT && fid_is_sane(fid) &&
365             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
366                 LIST_HEAD(cancels);
367                 int count;
368 
369                 /* Without that packing would fail */
370                 if (input_size == 0)
371                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
372                                              RCL_CLIENT, 0);
373 
374                 count = mdc_resource_get_unused(exp, fid,
375                                                 &cancels, LCK_EX,
376                                                 MDS_INODELOCK_XATTR);
377 
378                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
379                 if (rc) {
380                         ptlrpc_request_free(req);
381                         return rc;
382                 }
383         } else {
384                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
385                 if (rc) {
386                         ptlrpc_request_free(req);
387                         return rc;
388                 }
389         }
390 
391         if (opcode == MDS_REINT) {
392                 struct mdt_rec_setxattr *rec;
393 
394                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
395                          sizeof(struct mdt_rec_reint));
396                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
397                 rec->sx_opcode = REINT_SETXATTR;
398                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
399                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
400                 rec->sx_cap    = cfs_curproc_cap_pack();
401                 rec->sx_suppgid1 = suppgid;
402                 rec->sx_suppgid2 = -1;
403                 rec->sx_fid    = *fid;
404                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
405                 rec->sx_time   = get_seconds();
406                 rec->sx_size   = output_size;
407                 rec->sx_flags  = flags;
408 
409                 mdc_pack_capa(req, &RMF_CAPA1, oc);
410         } else {
411                 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
412         }
413 
414         if (xattr_name) {
415                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
416                 memcpy(tmp, xattr_name, xattr_namelen);
417         }
418         if (input_size) {
419                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
420                 memcpy(tmp, input, input_size);
421         }
422 
423         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
424                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
425                                      RCL_SERVER, output_size);
426         ptlrpc_request_set_replen(req);
427 
428         /* make rpc */
429         if (opcode == MDS_REINT)
430                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
431 
432         rc = ptlrpc_queue_wait(req);
433 
434         if (opcode == MDS_REINT)
435                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
436 
437         if (rc)
438                 ptlrpc_req_finished(req);
439         else
440                 *request = req;
441         return rc;
442 }
443 
444 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
445                  struct obd_capa *oc, u64 valid, const char *xattr_name,
446                  const char *input, int input_size, int output_size,
447                  int flags, __u32 suppgid, struct ptlrpc_request **request)
448 {
449         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
450                                 fid, oc, MDS_REINT, valid, xattr_name,
451                                 input, input_size, output_size, flags,
452                                 suppgid, request);
453 }
454 
455 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
456                  struct obd_capa *oc, u64 valid, const char *xattr_name,
457                  const char *input, int input_size, int output_size,
458                  int flags, struct ptlrpc_request **request)
459 {
460         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
461                                 fid, oc, MDS_GETXATTR, valid, xattr_name,
462                                 input, input_size, output_size, flags,
463                                 -1, request);
464 }
465 
466 #ifdef CONFIG_FS_POSIX_ACL
467 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
468 {
469         struct req_capsule     *pill = &req->rq_pill;
470         struct mdt_body *body = md->body;
471         struct posix_acl       *acl;
472         void               *buf;
473         int                  rc;
474 
475         if (!body->aclsize)
476                 return 0;
477 
478         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
479 
480         if (!buf)
481                 return -EPROTO;
482 
483         acl = posix_acl_from_xattr(&init_user_ns, buf, body->aclsize);
484         if (acl == NULL)
485                 return 0;
486 
487         if (IS_ERR(acl)) {
488                 rc = PTR_ERR(acl);
489                 CERROR("convert xattr to acl: %d\n", rc);
490                 return rc;
491         }
492 
493         rc = posix_acl_valid(acl);
494         if (rc) {
495                 CERROR("validate acl: %d\n", rc);
496                 posix_acl_release(acl);
497                 return rc;
498         }
499 
500         md->posix_acl = acl;
501         return 0;
502 }
503 #else
504 #define mdc_unpack_acl(req, md) 0
505 #endif
506 
507 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
508                       struct obd_export *dt_exp, struct obd_export *md_exp,
509                       struct lustre_md *md)
510 {
511         struct req_capsule *pill = &req->rq_pill;
512         int rc;
513 
514         LASSERT(md);
515         memset(md, 0, sizeof(*md));
516 
517         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
518         LASSERT(md->body != NULL);
519 
520         if (md->body->valid & OBD_MD_FLEASIZE) {
521                 int lmmsize;
522                 struct lov_mds_md *lmm;
523 
524                 if (!S_ISREG(md->body->mode)) {
525                         CDEBUG(D_INFO,
526                                "OBD_MD_FLEASIZE set, should be a regular file, but is not\n");
527                         rc = -EPROTO;
528                         goto out;
529                 }
530 
531                 if (md->body->eadatasize == 0) {
532                         CDEBUG(D_INFO,
533                                "OBD_MD_FLEASIZE set, but eadatasize 0\n");
534                         rc = -EPROTO;
535                         goto out;
536                 }
537                 lmmsize = md->body->eadatasize;
538                 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
539                 if (!lmm) {
540                         rc = -EPROTO;
541                         goto out;
542                 }
543 
544                 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
545                 if (rc < 0)
546                         goto out;
547 
548                 if (rc < sizeof(*md->lsm)) {
549                         CDEBUG(D_INFO,
550                                "lsm size too small: rc < sizeof (*md->lsm) (%d < %d)\n",
551                                rc, (int)sizeof(*md->lsm));
552                         rc = -EPROTO;
553                         goto out;
554                 }
555 
556         } else if (md->body->valid & OBD_MD_FLDIREA) {
557                 int lmvsize;
558                 struct lov_mds_md *lmv;
559 
560                 if (!S_ISDIR(md->body->mode)) {
561                         CDEBUG(D_INFO,
562                                "OBD_MD_FLDIREA set, should be a directory, but is not\n");
563                         rc = -EPROTO;
564                         goto out;
565                 }
566 
567                 if (md->body->eadatasize == 0) {
568                         CDEBUG(D_INFO,
569                                "OBD_MD_FLDIREA is set, but eadatasize 0\n");
570                         return -EPROTO;
571                 }
572                 if (md->body->valid & OBD_MD_MEA) {
573                         lmvsize = md->body->eadatasize;
574                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
575                                                            lmvsize);
576                         if (!lmv) {
577                                 rc = -EPROTO;
578                                 goto out;
579                         }
580 
581                         rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
582                                           lmvsize);
583                         if (rc < 0)
584                                 goto out;
585 
586                         if (rc < sizeof(*md->mea)) {
587                                 CDEBUG(D_INFO,
588                                        "size too small: rc < sizeof(*md->mea) (%d < %d)\n",
589                                         rc, (int)sizeof(*md->mea));
590                                 rc = -EPROTO;
591                                 goto out;
592                         }
593                 }
594         }
595         rc = 0;
596 
597         if (md->body->valid & OBD_MD_FLRMTPERM) {
598                 /* remote permission */
599                 LASSERT(client_is_remote(exp));
600                 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
601                                                 lustre_swab_mdt_remote_perm);
602                 if (!md->remote_perm) {
603                         rc = -EPROTO;
604                         goto out;
605                 }
606         } else if (md->body->valid & OBD_MD_FLACL) {
607                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
608                  * only when aclsize != 0 there's an actual segment for ACL
609                  * in reply buffer.
610                  */
611                 if (md->body->aclsize) {
612                         rc = mdc_unpack_acl(req, md);
613                         if (rc)
614                                 goto out;
615 #ifdef CONFIG_FS_POSIX_ACL
616                 } else {
617                         md->posix_acl = NULL;
618 #endif
619                 }
620         }
621         if (md->body->valid & OBD_MD_FLMDSCAPA) {
622                 struct obd_capa *oc = NULL;
623 
624                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
625                 if (rc)
626                         goto out;
627                 md->mds_capa = oc;
628         }
629 
630         if (md->body->valid & OBD_MD_FLOSSCAPA) {
631                 struct obd_capa *oc = NULL;
632 
633                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
634                 if (rc)
635                         goto out;
636                 md->oss_capa = oc;
637         }
638 
639 out:
640         if (rc) {
641                 if (md->oss_capa) {
642                         capa_put(md->oss_capa);
643                         md->oss_capa = NULL;
644                 }
645                 if (md->mds_capa) {
646                         capa_put(md->mds_capa);
647                         md->mds_capa = NULL;
648                 }
649 #ifdef CONFIG_FS_POSIX_ACL
650                 posix_acl_release(md->posix_acl);
651 #endif
652                 if (md->lsm)
653                         obd_free_memmd(dt_exp, &md->lsm);
654         }
655         return rc;
656 }
657 
658 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
659 {
660         return 0;
661 }
662 
663 /**
664  * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
665  * RPC chains.
666  */
667 void mdc_replay_open(struct ptlrpc_request *req)
668 {
669         struct md_open_data *mod = req->rq_cb_data;
670         struct ptlrpc_request *close_req;
671         struct obd_client_handle *och;
672         struct lustre_handle old;
673         struct mdt_body *body;
674 
675         if (mod == NULL) {
676                 DEBUG_REQ(D_ERROR, req,
677                           "Can't properly replay without open data.");
678                 return;
679         }
680 
681         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
682         LASSERT(body != NULL);
683 
684         och = mod->mod_och;
685         if (och != NULL) {
686                 struct lustre_handle *file_fh;
687 
688                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
689 
690                 file_fh = &och->och_fh;
691                 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
692                        file_fh->cookie, body->handle.cookie);
693                 old = *file_fh;
694                 *file_fh = body->handle;
695         }
696         close_req = mod->mod_close_req;
697         if (close_req != NULL) {
698                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
699                 struct mdt_ioepoch *epoch;
700 
701                 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
702                 epoch = req_capsule_client_get(&close_req->rq_pill,
703                                                &RMF_MDT_EPOCH);
704                 LASSERT(epoch);
705 
706                 if (och != NULL)
707                         LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
708                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
709                 epoch->handle = body->handle;
710         }
711 }
712 
713 void mdc_commit_open(struct ptlrpc_request *req)
714 {
715         struct md_open_data *mod = req->rq_cb_data;
716 
717         if (mod == NULL)
718                 return;
719 
720         /**
721          * No need to touch md_open_data::mod_och, it holds a reference on
722          * \var mod and will zero references to each other, \var mod will be
723          * freed after that when md_open_data::mod_och will put the reference.
724          */
725 
726         /**
727          * Do not let open request to disappear as it still may be needed
728          * for close rpc to happen (it may happen on evict only, otherwise
729          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
730          * called), just mark this rpc as committed to distinguish these 2
731          * cases, see mdc_close() for details. The open request reference will
732          * be put along with freeing \var mod.
733          */
734         ptlrpc_request_addref(req);
735         spin_lock(&req->rq_lock);
736         req->rq_committed = 1;
737         spin_unlock(&req->rq_lock);
738         req->rq_cb_data = NULL;
739         obd_mod_put(mod);
740 }
741 
742 int mdc_set_open_replay_data(struct obd_export *exp,
743                              struct obd_client_handle *och,
744                              struct lookup_intent *it)
745 {
746         struct md_open_data   *mod;
747         struct mdt_rec_create *rec;
748         struct mdt_body       *body;
749         struct ptlrpc_request *open_req = it->d.lustre.it_data;
750         struct obd_import     *imp = open_req->rq_import;
751 
752         if (!open_req->rq_replay)
753                 return 0;
754 
755         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
756         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
757         LASSERT(rec != NULL);
758         /* Incoming message in my byte order (it's been swabbed). */
759         /* Outgoing messages always in my byte order. */
760         LASSERT(body != NULL);
761 
762         /* Only if the import is replayable, we set replay_open data */
763         if (och && imp->imp_replayable) {
764                 mod = obd_mod_alloc();
765                 if (mod == NULL) {
766                         DEBUG_REQ(D_ERROR, open_req,
767                                   "Can't allocate md_open_data");
768                         return 0;
769                 }
770 
771                 /**
772                  * Take a reference on \var mod, to be freed on mdc_close().
773                  * It protects \var mod from being freed on eviction (commit
774                  * callback is called despite rq_replay flag).
775                  * Another reference for \var och.
776                  */
777                 obd_mod_get(mod);
778                 obd_mod_get(mod);
779 
780                 spin_lock(&open_req->rq_lock);
781                 och->och_mod = mod;
782                 mod->mod_och = och;
783                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
784                                      it_disposition(it, DISP_OPEN_STRIPE);
785                 mod->mod_open_req = open_req;
786                 open_req->rq_cb_data = mod;
787                 open_req->rq_commit_cb = mdc_commit_open;
788                 spin_unlock(&open_req->rq_lock);
789         }
790 
791         rec->cr_fid2 = body->fid1;
792         rec->cr_ioepoch = body->ioepoch;
793         rec->cr_old_handle.cookie = body->handle.cookie;
794         open_req->rq_replay_cb = mdc_replay_open;
795         if (!fid_is_sane(&body->fid1)) {
796                 DEBUG_REQ(D_ERROR, open_req,
797                           "Saving replay request with insane fid");
798                 LBUG();
799         }
800 
801         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
802         return 0;
803 }
804 
805 static void mdc_free_open(struct md_open_data *mod)
806 {
807         int committed = 0;
808 
809         if (mod->mod_is_create == 0 &&
810             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
811                 committed = 1;
812 
813         LASSERT(mod->mod_open_req->rq_replay == 0);
814 
815         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
816 
817         ptlrpc_request_committed(mod->mod_open_req, committed);
818         if (mod->mod_close_req)
819                 ptlrpc_request_committed(mod->mod_close_req, committed);
820 }
821 
822 int mdc_clear_open_replay_data(struct obd_export *exp,
823                                struct obd_client_handle *och)
824 {
825         struct md_open_data *mod = och->och_mod;
826 
827         /**
828          * It is possible to not have \var mod in a case of eviction between
829          * lookup and ll_file_open().
830          **/
831         if (mod == NULL)
832                 return 0;
833 
834         LASSERT(mod != LP_POISON);
835         LASSERT(mod->mod_open_req != NULL);
836         mdc_free_open(mod);
837 
838         mod->mod_och = NULL;
839         och->och_mod = NULL;
840         obd_mod_put(mod);
841 
842         return 0;
843 }
844 
845 /* Prepares the request for the replay by the given reply */
846 static void mdc_close_handle_reply(struct ptlrpc_request *req,
847                                    struct md_op_data *op_data, int rc) {
848         struct mdt_body  *repbody;
849         struct mdt_ioepoch *epoch;
850 
851         if (req && rc == -EAGAIN) {
852                 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
853                 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
854 
855                 epoch->flags |= MF_SOM_AU;
856                 if (repbody->valid & OBD_MD_FLGETATTRLOCK)
857                         op_data->op_flags |= MF_GETATTR_LOCK;
858         }
859 }
860 
861 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
862                      struct md_open_data *mod, struct ptlrpc_request **request)
863 {
864         struct obd_device     *obd = class_exp2obd(exp);
865         struct ptlrpc_request *req;
866         struct req_format     *req_fmt;
867         int                    rc;
868         int                    saved_rc = 0;
869 
870 
871         req_fmt = &RQF_MDS_CLOSE;
872         if (op_data->op_bias & MDS_HSM_RELEASE) {
873                 req_fmt = &RQF_MDS_RELEASE_CLOSE;
874 
875                 /* allocate a FID for volatile file */
876                 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
877                 if (rc < 0) {
878                         CERROR("%s: "DFID" failed to allocate FID: %d\n",
879                                obd->obd_name, PFID(&op_data->op_fid1), rc);
880                         /* save the errcode and proceed to close */
881                         saved_rc = rc;
882                 }
883         }
884 
885         *request = NULL;
886         req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
887         if (req == NULL)
888                 return -ENOMEM;
889 
890         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
891 
892         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
893         if (rc) {
894                 ptlrpc_request_free(req);
895                 return rc;
896         }
897 
898         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
899          * portal whose threads are not taking any DLM locks and are therefore
900          * always progressing */
901         req->rq_request_portal = MDS_READPAGE_PORTAL;
902         ptlrpc_at_set_req_timeout(req);
903 
904         /* Ensure that this close's handle is fixed up during replay. */
905         if (likely(mod != NULL)) {
906                 LASSERTF(mod->mod_open_req != NULL &&
907                          mod->mod_open_req->rq_type != LI_POISON,
908                          "POISONED open %p!\n", mod->mod_open_req);
909 
910                 mod->mod_close_req = req;
911 
912                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
913                 /* We no longer want to preserve this open for replay even
914                  * though the open was committed. b=3632, b=3633 */
915                 spin_lock(&mod->mod_open_req->rq_lock);
916                 mod->mod_open_req->rq_replay = 0;
917                 spin_unlock(&mod->mod_open_req->rq_lock);
918         } else {
919                  CDEBUG(D_HA,
920                         "couldn't find open req; expecting close error\n");
921         }
922 
923         mdc_close_pack(req, op_data);
924 
925         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
926                              obd->u.cli.cl_default_mds_easize);
927         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
928                              obd->u.cli.cl_default_mds_cookiesize);
929 
930         ptlrpc_request_set_replen(req);
931 
932         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
933         rc = ptlrpc_queue_wait(req);
934         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
935 
936         if (req->rq_repmsg == NULL) {
937                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
938                        req->rq_status);
939                 if (rc == 0)
940                         rc = req->rq_status ?: -EIO;
941         } else if (rc == 0 || rc == -EAGAIN) {
942                 struct mdt_body *body;
943 
944                 rc = lustre_msg_get_status(req->rq_repmsg);
945                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
946                         DEBUG_REQ(D_ERROR, req,
947                                   "type == PTL_RPC_MSG_ERR, err = %d", rc);
948                         if (rc > 0)
949                                 rc = -rc;
950                 }
951                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
952                 if (body == NULL)
953                         rc = -EPROTO;
954         } else if (rc == -ESTALE) {
955                 /**
956                  * it can be allowed error after 3633 if open was committed and
957                  * server failed before close was sent. Let's check if mod
958                  * exists and return no error in that case
959                  */
960                 if (mod) {
961                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
962                         LASSERT(mod->mod_open_req != NULL);
963                         if (mod->mod_open_req->rq_committed)
964                                 rc = 0;
965                 }
966         }
967 
968         if (mod) {
969                 if (rc != 0)
970                         mod->mod_close_req = NULL;
971                 /* Since now, mod is accessed through open_req only,
972                  * thus close req does not keep a reference on mod anymore. */
973                 obd_mod_put(mod);
974         }
975         *request = req;
976         mdc_close_handle_reply(req, op_data, rc);
977         return rc < 0 ? rc : saved_rc;
978 }
979 
980 static int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
981                             struct md_open_data *mod)
982 {
983         struct obd_device     *obd = class_exp2obd(exp);
984         struct ptlrpc_request *req;
985         int                 rc;
986 
987         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
988                                    &RQF_MDS_DONE_WRITING);
989         if (req == NULL)
990                 return -ENOMEM;
991 
992         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
993         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
994         if (rc) {
995                 ptlrpc_request_free(req);
996                 return rc;
997         }
998 
999         if (mod != NULL) {
1000                 LASSERTF(mod->mod_open_req != NULL &&
1001                          mod->mod_open_req->rq_type != LI_POISON,
1002                          "POISONED setattr %p!\n", mod->mod_open_req);
1003 
1004                 mod->mod_close_req = req;
1005                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
1006                 /* We no longer want to preserve this setattr for replay even
1007                  * though the open was committed. b=3632, b=3633 */
1008                 spin_lock(&mod->mod_open_req->rq_lock);
1009                 mod->mod_open_req->rq_replay = 0;
1010                 spin_unlock(&mod->mod_open_req->rq_lock);
1011         }
1012 
1013         mdc_close_pack(req, op_data);
1014         ptlrpc_request_set_replen(req);
1015 
1016         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
1017         rc = ptlrpc_queue_wait(req);
1018         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
1019 
1020         if (rc == -ESTALE) {
1021                 /**
1022                  * it can be allowed error after 3633 if open or setattr were
1023                  * committed and server failed before close was sent.
1024                  * Let's check if mod exists and return no error in that case
1025                  */
1026                 if (mod) {
1027                         LASSERT(mod->mod_open_req != NULL);
1028                         if (mod->mod_open_req->rq_committed)
1029                                 rc = 0;
1030                 }
1031         }
1032 
1033         if (mod) {
1034                 if (rc != 0)
1035                         mod->mod_close_req = NULL;
1036                 LASSERT(mod->mod_open_req != NULL);
1037                 mdc_free_open(mod);
1038 
1039                 /* Since now, mod is accessed through setattr req only,
1040                  * thus DW req does not keep a reference on mod anymore. */
1041                 obd_mod_put(mod);
1042         }
1043 
1044         mdc_close_handle_reply(req, op_data, rc);
1045         ptlrpc_req_finished(req);
1046         return rc;
1047 }
1048 
1049 
1050 static int mdc_readpage(struct obd_export *exp, struct md_op_data *op_data,
1051                         struct page **pages, struct ptlrpc_request **request)
1052 {
1053         struct ptlrpc_request   *req;
1054         struct ptlrpc_bulk_desc *desc;
1055         int                   i;
1056         wait_queue_head_t             waitq;
1057         int                   resends = 0;
1058         struct l_wait_info       lwi;
1059         int                   rc;
1060 
1061         *request = NULL;
1062         init_waitqueue_head(&waitq);
1063 
1064 restart_bulk:
1065         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1066         if (req == NULL)
1067                 return -ENOMEM;
1068 
1069         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1070 
1071         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1072         if (rc) {
1073                 ptlrpc_request_free(req);
1074                 return rc;
1075         }
1076 
1077         req->rq_request_portal = MDS_READPAGE_PORTAL;
1078         ptlrpc_at_set_req_timeout(req);
1079 
1080         desc = ptlrpc_prep_bulk_imp(req, op_data->op_npages, 1, BULK_PUT_SINK,
1081                                     MDS_BULK_PORTAL);
1082         if (desc == NULL) {
1083                 ptlrpc_request_free(req);
1084                 return -ENOMEM;
1085         }
1086 
1087         /* NB req now owns desc and will free it when it gets freed */
1088         for (i = 0; i < op_data->op_npages; i++)
1089                 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
1090 
1091         mdc_readdir_pack(req, op_data->op_offset,
1092                          PAGE_CACHE_SIZE * op_data->op_npages,
1093                          &op_data->op_fid1, op_data->op_capa1);
1094 
1095         ptlrpc_request_set_replen(req);
1096         rc = ptlrpc_queue_wait(req);
1097         if (rc) {
1098                 ptlrpc_req_finished(req);
1099                 if (rc != -ETIMEDOUT)
1100                         return rc;
1101 
1102                 resends++;
1103                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1104                         CERROR("too many resend retries, returning error\n");
1105                         return -EIO;
1106                 }
1107                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends),
1108                                        NULL, NULL, NULL);
1109                 l_wait_event(waitq, 0, &lwi);
1110 
1111                 goto restart_bulk;
1112         }
1113 
1114         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1115                                           req->rq_bulk->bd_nob_transferred);
1116         if (rc < 0) {
1117                 ptlrpc_req_finished(req);
1118                 return rc;
1119         }
1120 
1121         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1122                 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1123                         req->rq_bulk->bd_nob_transferred,
1124                         PAGE_CACHE_SIZE * op_data->op_npages);
1125                 ptlrpc_req_finished(req);
1126                 return -EPROTO;
1127         }
1128 
1129         *request = req;
1130         return 0;
1131 }
1132 
1133 static int mdc_statfs(const struct lu_env *env,
1134                       struct obd_export *exp, struct obd_statfs *osfs,
1135                       __u64 max_age, __u32 flags)
1136 {
1137         struct obd_device     *obd = class_exp2obd(exp);
1138         struct ptlrpc_request *req;
1139         struct obd_statfs     *msfs;
1140         struct obd_import     *imp = NULL;
1141         int                 rc;
1142 
1143         /*
1144          * Since the request might also come from lprocfs, so we need
1145          * sync this with client_disconnect_export Bug15684
1146          */
1147         down_read(&obd->u.cli.cl_sem);
1148         if (obd->u.cli.cl_import)
1149                 imp = class_import_get(obd->u.cli.cl_import);
1150         up_read(&obd->u.cli.cl_sem);
1151         if (!imp)
1152                 return -ENODEV;
1153 
1154         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1155                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1156         if (req == NULL) {
1157                 rc = -ENOMEM;
1158                 goto output;
1159         }
1160 
1161         ptlrpc_request_set_replen(req);
1162 
1163         if (flags & OBD_STATFS_NODELAY) {
1164                 /* procfs requests not want stay in wait for avoid deadlock */
1165                 req->rq_no_resend = 1;
1166                 req->rq_no_delay = 1;
1167         }
1168 
1169         rc = ptlrpc_queue_wait(req);
1170         if (rc) {
1171                 /* check connection error first */
1172                 if (imp->imp_connect_error)
1173                         rc = imp->imp_connect_error;
1174                 goto out;
1175         }
1176 
1177         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1178         if (msfs == NULL) {
1179                 rc = -EPROTO;
1180                 goto out;
1181         }
1182 
1183         *osfs = *msfs;
1184 out:
1185         ptlrpc_req_finished(req);
1186 output:
1187         class_import_put(imp);
1188         return rc;
1189 }
1190 
1191 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1192 {
1193         __u32 keylen, vallen;
1194         void *key;
1195         int rc;
1196 
1197         if (gf->gf_pathlen > PATH_MAX)
1198                 return -ENAMETOOLONG;
1199         if (gf->gf_pathlen < 2)
1200                 return -EOVERFLOW;
1201 
1202         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1203         keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1204         key = kzalloc(keylen, GFP_NOFS);
1205         if (key == NULL)
1206                 return -ENOMEM;
1207         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1208         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1209 
1210         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1211                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1212 
1213         if (!fid_is_sane(&gf->gf_fid)) {
1214                 rc = -EINVAL;
1215                 goto out;
1216         }
1217 
1218         /* Val is struct getinfo_fid2path result plus path */
1219         vallen = sizeof(*gf) + gf->gf_pathlen;
1220 
1221         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1222         if (rc != 0 && rc != -EREMOTE)
1223                 goto out;
1224 
1225         if (vallen <= sizeof(*gf)) {
1226                 rc = -EPROTO;
1227                 goto out;
1228         } else if (vallen > sizeof(*gf) + gf->gf_pathlen) {
1229                 rc = -EOVERFLOW;
1230                 goto out;
1231         }
1232 
1233         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n%s\n",
1234                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1235 
1236 out:
1237         kfree(key);
1238         return rc;
1239 }
1240 
1241 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1242                                 struct hsm_progress_kernel *hpk)
1243 {
1244         struct obd_import               *imp = class_exp2cliimp(exp);
1245         struct hsm_progress_kernel      *req_hpk;
1246         struct ptlrpc_request           *req;
1247         int                              rc;
1248 
1249         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1250                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1251         if (req == NULL) {
1252                 rc = -ENOMEM;
1253                 goto out;
1254         }
1255 
1256         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1257 
1258         /* Copy hsm_progress struct */
1259         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1260         if (req_hpk == NULL) {
1261                 rc = -EPROTO;
1262                 goto out;
1263         }
1264 
1265         *req_hpk = *hpk;
1266         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1267 
1268         ptlrpc_request_set_replen(req);
1269 
1270         rc = mdc_queue_wait(req);
1271         goto out;
1272 out:
1273         ptlrpc_req_finished(req);
1274         return rc;
1275 }
1276 
1277 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1278 {
1279         __u32                   *archive_mask;
1280         struct ptlrpc_request   *req;
1281         int                      rc;
1282 
1283         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1284                                         LUSTRE_MDS_VERSION,
1285                                         MDS_HSM_CT_REGISTER);
1286         if (req == NULL) {
1287                 rc = -ENOMEM;
1288                 goto out;
1289         }
1290 
1291         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1292 
1293         /* Copy hsm_progress struct */
1294         archive_mask = req_capsule_client_get(&req->rq_pill,
1295                                               &RMF_MDS_HSM_ARCHIVE);
1296         if (archive_mask == NULL) {
1297                 rc = -EPROTO;
1298                 goto out;
1299         }
1300 
1301         *archive_mask = archives;
1302 
1303         ptlrpc_request_set_replen(req);
1304 
1305         rc = mdc_queue_wait(req);
1306         goto out;
1307 out:
1308         ptlrpc_req_finished(req);
1309         return rc;
1310 }
1311 
1312 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1313                                       struct md_op_data *op_data)
1314 {
1315         struct hsm_current_action       *hca = op_data->op_data;
1316         struct hsm_current_action       *req_hca;
1317         struct ptlrpc_request           *req;
1318         int                              rc;
1319 
1320         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1321                                    &RQF_MDS_HSM_ACTION);
1322         if (req == NULL)
1323                 return -ENOMEM;
1324 
1325         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1326 
1327         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1328         if (rc) {
1329                 ptlrpc_request_free(req);
1330                 return rc;
1331         }
1332 
1333         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1334                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1335 
1336         ptlrpc_request_set_replen(req);
1337 
1338         rc = mdc_queue_wait(req);
1339         if (rc)
1340                 goto out;
1341 
1342         req_hca = req_capsule_server_get(&req->rq_pill,
1343                                          &RMF_MDS_HSM_CURRENT_ACTION);
1344         if (req_hca == NULL) {
1345                 rc = -EPROTO;
1346                 goto out;
1347         }
1348 
1349         *hca = *req_hca;
1350 
1351 out:
1352         ptlrpc_req_finished(req);
1353         return rc;
1354 }
1355 
1356 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1357 {
1358         struct ptlrpc_request   *req;
1359         int                      rc;
1360 
1361         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1362                                         LUSTRE_MDS_VERSION,
1363                                         MDS_HSM_CT_UNREGISTER);
1364         if (req == NULL) {
1365                 rc = -ENOMEM;
1366                 goto out;
1367         }
1368 
1369         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1370 
1371         ptlrpc_request_set_replen(req);
1372 
1373         rc = mdc_queue_wait(req);
1374         goto out;
1375 out:
1376         ptlrpc_req_finished(req);
1377         return rc;
1378 }
1379 
1380 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1381                                  struct md_op_data *op_data)
1382 {
1383         struct hsm_user_state   *hus = op_data->op_data;
1384         struct hsm_user_state   *req_hus;
1385         struct ptlrpc_request   *req;
1386         int                      rc;
1387 
1388         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1389                                    &RQF_MDS_HSM_STATE_GET);
1390         if (req == NULL)
1391                 return -ENOMEM;
1392 
1393         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1394 
1395         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1396         if (rc != 0) {
1397                 ptlrpc_request_free(req);
1398                 return rc;
1399         }
1400 
1401         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1402                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1403 
1404         ptlrpc_request_set_replen(req);
1405 
1406         rc = mdc_queue_wait(req);
1407         if (rc)
1408                 goto out;
1409 
1410         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1411         if (req_hus == NULL) {
1412                 rc = -EPROTO;
1413                 goto out;
1414         }
1415 
1416         *hus = *req_hus;
1417 
1418 out:
1419         ptlrpc_req_finished(req);
1420         return rc;
1421 }
1422 
1423 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1424                                  struct md_op_data *op_data)
1425 {
1426         struct hsm_state_set    *hss = op_data->op_data;
1427         struct hsm_state_set    *req_hss;
1428         struct ptlrpc_request   *req;
1429         int                      rc;
1430 
1431         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1432                                    &RQF_MDS_HSM_STATE_SET);
1433         if (req == NULL)
1434                 return -ENOMEM;
1435 
1436         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1437 
1438         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1439         if (rc) {
1440                 ptlrpc_request_free(req);
1441                 return rc;
1442         }
1443 
1444         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1445                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1446 
1447         /* Copy states */
1448         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1449         if (req_hss == NULL) {
1450                 rc = -EPROTO;
1451                 goto out;
1452         }
1453         *req_hss = *hss;
1454 
1455         ptlrpc_request_set_replen(req);
1456 
1457         rc = mdc_queue_wait(req);
1458         goto out;
1459 
1460 out:
1461         ptlrpc_req_finished(req);
1462         return rc;
1463 }
1464 
1465 static int mdc_ioc_hsm_request(struct obd_export *exp,
1466                                struct hsm_user_request *hur)
1467 {
1468         struct obd_import       *imp = class_exp2cliimp(exp);
1469         struct ptlrpc_request   *req;
1470         struct hsm_request      *req_hr;
1471         struct hsm_user_item    *req_hui;
1472         char                    *req_opaque;
1473         int                      rc;
1474 
1475         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1476         if (req == NULL) {
1477                 rc = -ENOMEM;
1478                 goto out;
1479         }
1480 
1481         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1482                              hur->hur_request.hr_itemcount
1483                              * sizeof(struct hsm_user_item));
1484         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1485                              hur->hur_request.hr_data_len);
1486 
1487         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1488         if (rc) {
1489                 ptlrpc_request_free(req);
1490                 return rc;
1491         }
1492 
1493         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1494 
1495         /* Copy hsm_request struct */
1496         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1497         if (req_hr == NULL) {
1498                 rc = -EPROTO;
1499                 goto out;
1500         }
1501         *req_hr = hur->hur_request;
1502 
1503         /* Copy hsm_user_item structs */
1504         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1505         if (req_hui == NULL) {
1506                 rc = -EPROTO;
1507                 goto out;
1508         }
1509         memcpy(req_hui, hur->hur_user_item,
1510                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1511 
1512         /* Copy opaque field */
1513         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1514         if (req_opaque == NULL) {
1515                 rc = -EPROTO;
1516                 goto out;
1517         }
1518         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1519 
1520         ptlrpc_request_set_replen(req);
1521 
1522         rc = mdc_queue_wait(req);
1523         goto out;
1524 
1525 out:
1526         ptlrpc_req_finished(req);
1527         return rc;
1528 }
1529 
1530 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
1531 {
1532         struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1533 
1534         LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1535 
1536         lh->kuc_magic = KUC_MAGIC;
1537         lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1538         lh->kuc_flags = flags;
1539         lh->kuc_msgtype = CL_RECORD;
1540         lh->kuc_msglen = len;
1541         return lh;
1542 }
1543 
1544 #define D_CHANGELOG 0
1545 
1546 struct changelog_show {
1547         __u64           cs_startrec;
1548         __u32           cs_flags;
1549         struct file     *cs_fp;
1550         char            *cs_buf;
1551         struct obd_device *cs_obd;
1552 };
1553 
1554 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1555                              struct llog_rec_hdr *hdr, void *data)
1556 {
1557         struct changelog_show *cs = data;
1558         struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1559         struct kuc_hdr *lh;
1560         int len, rc;
1561 
1562         if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1563                 rc = -EINVAL;
1564                 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1565                        cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
1566                        rec->cr.cr_type, rc);
1567                 return rc;
1568         }
1569 
1570         if (rec->cr.cr_index < cs->cs_startrec) {
1571                 /* Skip entries earlier than what we are interested in */
1572                 CDEBUG(D_CHANGELOG, "rec=%llu start=%llu\n",
1573                        rec->cr.cr_index, cs->cs_startrec);
1574                 return 0;
1575         }
1576 
1577         CDEBUG(D_CHANGELOG, "%llu %02d%-5s %llu 0x%x t="DFID" p="DFID
1578                 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
1579                 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1580                 rec->cr.cr_flags & CLF_FLAGMASK,
1581                 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1582                 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1583 
1584         len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1585 
1586         /* Set up the message */
1587         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1588         memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1589 
1590         rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1591         CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len, rc);
1592 
1593         return rc;
1594 }
1595 
1596 static int mdc_changelog_send_thread(void *csdata)
1597 {
1598         struct changelog_show *cs = csdata;
1599         struct llog_ctxt *ctxt = NULL;
1600         struct llog_handle *llh = NULL;
1601         struct kuc_hdr *kuch;
1602         int rc;
1603 
1604         CDEBUG(D_CHANGELOG, "changelog to fp=%p start %llu\n",
1605                cs->cs_fp, cs->cs_startrec);
1606 
1607         cs->cs_buf = kzalloc(KUC_CHANGELOG_MSG_MAXSIZE, GFP_NOFS);
1608         if (cs->cs_buf == NULL) {
1609                 rc = -ENOMEM;
1610                 goto out;
1611         }
1612 
1613         /* Set up the remote catalog handle */
1614         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1615         if (ctxt == NULL) {
1616                 rc = -ENOENT;
1617                 goto out;
1618         }
1619         rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1620                        LLOG_OPEN_EXISTS);
1621         if (rc) {
1622                 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1623                        cs->cs_obd->obd_name, rc);
1624                 goto out;
1625         }
1626         rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
1627         if (rc) {
1628                 CERROR("llog_init_handle failed %d\n", rc);
1629                 goto out;
1630         }
1631 
1632         rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1633 
1634         /* Send EOF no matter what our result */
1635         kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags);
1636         if (kuch) {
1637                 kuch->kuc_msgtype = CL_EOF;
1638                 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1639         }
1640 
1641 out:
1642         fput(cs->cs_fp);
1643         if (llh)
1644                 llog_cat_close(NULL, llh);
1645         if (ctxt)
1646                 llog_ctxt_put(ctxt);
1647         kfree(cs->cs_buf);
1648         kfree(cs);
1649         return rc;
1650 }
1651 
1652 static int mdc_ioc_changelog_send(struct obd_device *obd,
1653                                   struct ioc_changelog *icc)
1654 {
1655         struct changelog_show *cs;
1656         int rc;
1657 
1658         /* Freed in mdc_changelog_send_thread */
1659         cs = kzalloc(sizeof(*cs), GFP_NOFS);
1660         if (!cs)
1661                 return -ENOMEM;
1662 
1663         cs->cs_obd = obd;
1664         cs->cs_startrec = icc->icc_recno;
1665         /* matching fput in mdc_changelog_send_thread */
1666         cs->cs_fp = fget(icc->icc_id);
1667         cs->cs_flags = icc->icc_flags;
1668 
1669         /*
1670          * New thread because we should return to user app before
1671          * writing into our pipe
1672          */
1673         rc = PTR_ERR(kthread_run(mdc_changelog_send_thread, cs,
1674                                  "mdc_clg_send_thread"));
1675         if (!IS_ERR_VALUE(rc)) {
1676                 CDEBUG(D_CHANGELOG, "start changelog thread\n");
1677                 return 0;
1678         }
1679 
1680         CERROR("Failed to start changelog thread: %d\n", rc);
1681         kfree(cs);
1682         return rc;
1683 }
1684 
1685 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1686                                 struct lustre_kernelcomm *lk);
1687 
1688 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
1689                           struct obd_quotactl *oqctl)
1690 {
1691         struct client_obd       *cli = &exp->exp_obd->u.cli;
1692         struct ptlrpc_request   *req;
1693         struct obd_quotactl     *body;
1694         int                   rc;
1695 
1696         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1697                                         &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
1698                                         MDS_QUOTACHECK);
1699         if (req == NULL)
1700                 return -ENOMEM;
1701 
1702         body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1703         *body = *oqctl;
1704 
1705         ptlrpc_request_set_replen(req);
1706 
1707         /* the next poll will find -ENODATA, that means quotacheck is
1708          * going on */
1709         cli->cl_qchk_stat = -ENODATA;
1710         rc = ptlrpc_queue_wait(req);
1711         if (rc)
1712                 cli->cl_qchk_stat = rc;
1713         ptlrpc_req_finished(req);
1714         return rc;
1715 }
1716 
1717 static int mdc_quota_poll_check(struct obd_export *exp,
1718                                 struct if_quotacheck *qchk)
1719 {
1720         struct client_obd *cli = &exp->exp_obd->u.cli;
1721         int rc;
1722 
1723         qchk->obd_uuid = cli->cl_target_uuid;
1724         memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
1725 
1726         rc = cli->cl_qchk_stat;
1727         /* the client is not the previous one */
1728         if (rc == CL_NOT_QUOTACHECKED)
1729                 rc = -EINTR;
1730         return rc;
1731 }
1732 
1733 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1734                         struct obd_quotactl *oqctl)
1735 {
1736         struct ptlrpc_request   *req;
1737         struct obd_quotactl     *oqc;
1738         int                   rc;
1739 
1740         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1741                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1742                                         MDS_QUOTACTL);
1743         if (req == NULL)
1744                 return -ENOMEM;
1745 
1746         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1747         *oqc = *oqctl;
1748 
1749         ptlrpc_request_set_replen(req);
1750         ptlrpc_at_set_req_timeout(req);
1751         req->rq_no_resend = 1;
1752 
1753         rc = ptlrpc_queue_wait(req);
1754         if (rc)
1755                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1756 
1757         if (req->rq_repmsg) {
1758                 oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1759                 if (oqc) {
1760                         *oqctl = *oqc;
1761                 } else if (!rc) {
1762                         CERROR("Can't unpack obd_quotactl\n");
1763                         rc = -EPROTO;
1764                 }
1765         } else if (!rc) {
1766                 CERROR("Can't unpack obd_quotactl\n");
1767                 rc = -EPROTO;
1768         }
1769         ptlrpc_req_finished(req);
1770 
1771         return rc;
1772 }
1773 
1774 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1775                                 struct md_op_data *op_data)
1776 {
1777         LIST_HEAD(cancels);
1778         struct ptlrpc_request   *req;
1779         int                      rc, count;
1780         struct mdc_swap_layouts *msl, *payload;
1781 
1782         msl = op_data->op_data;
1783 
1784         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1785          * first thing it will do is to cancel the 2 layout
1786          * locks hold by this client.
1787          * So the client must cancel its layout locks on the 2 fids
1788          * with the request RPC to avoid extra RPC round trips
1789          */
1790         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1791                                         LCK_CR, MDS_INODELOCK_LAYOUT);
1792         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1793                                          LCK_CR, MDS_INODELOCK_LAYOUT);
1794 
1795         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1796                                    &RQF_MDS_SWAP_LAYOUTS);
1797         if (req == NULL) {
1798                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1799                 return -ENOMEM;
1800         }
1801 
1802         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1803         mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
1804 
1805         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1806         if (rc) {
1807                 ptlrpc_request_free(req);
1808                 return rc;
1809         }
1810 
1811         mdc_swap_layouts_pack(req, op_data);
1812 
1813         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1814         LASSERT(payload);
1815 
1816         *payload = *msl;
1817 
1818         ptlrpc_request_set_replen(req);
1819 
1820         rc = ptlrpc_queue_wait(req);
1821 
1822         ptlrpc_req_finished(req);
1823         return rc;
1824 }
1825 
1826 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1827                          void *karg, void *uarg)
1828 {
1829         struct obd_device *obd = exp->exp_obd;
1830         struct obd_ioctl_data *data = karg;
1831         struct obd_import *imp = obd->u.cli.cl_import;
1832         int rc;
1833 
1834         if (!try_module_get(THIS_MODULE)) {
1835                 CERROR("Can't get module. Is it alive?");
1836                 return -EINVAL;
1837         }
1838         switch (cmd) {
1839         case OBD_IOC_CHANGELOG_SEND:
1840                 rc = mdc_ioc_changelog_send(obd, karg);
1841                 goto out;
1842         case OBD_IOC_CHANGELOG_CLEAR: {
1843                 struct ioc_changelog *icc = karg;
1844                 struct changelog_setinfo cs = {
1845                         .cs_recno = icc->icc_recno,
1846                         .cs_id = icc->icc_id
1847                 };
1848 
1849                 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
1850                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1851                                         NULL);
1852                 goto out;
1853         }
1854         case OBD_IOC_FID2PATH:
1855                 rc = mdc_ioc_fid2path(exp, karg);
1856                 goto out;
1857         case LL_IOC_HSM_CT_START:
1858                 rc = mdc_ioc_hsm_ct_start(exp, karg);
1859                 /* ignore if it was already registered on this MDS. */
1860                 if (rc == -EEXIST)
1861                         rc = 0;
1862                 goto out;
1863         case LL_IOC_HSM_PROGRESS:
1864                 rc = mdc_ioc_hsm_progress(exp, karg);
1865                 goto out;
1866         case LL_IOC_HSM_STATE_GET:
1867                 rc = mdc_ioc_hsm_state_get(exp, karg);
1868                 goto out;
1869         case LL_IOC_HSM_STATE_SET:
1870                 rc = mdc_ioc_hsm_state_set(exp, karg);
1871                 goto out;
1872         case LL_IOC_HSM_ACTION:
1873                 rc = mdc_ioc_hsm_current_action(exp, karg);
1874                 goto out;
1875         case LL_IOC_HSM_REQUEST:
1876                 rc = mdc_ioc_hsm_request(exp, karg);
1877                 goto out;
1878         case OBD_IOC_CLIENT_RECOVER:
1879                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1880                 if (rc < 0)
1881                         goto out;
1882                 rc = 0;
1883                 goto out;
1884         case IOC_OSC_SET_ACTIVE:
1885                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1886                 goto out;
1887         case OBD_IOC_POLL_QUOTACHECK:
1888                 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
1889                 goto out;
1890         case OBD_IOC_PING_TARGET:
1891                 rc = ptlrpc_obd_ping(obd);
1892                 goto out;
1893         /*
1894          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1895          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1896          * there'd be no LMV layer thus we might be called here. Eventually
1897          * this code should be removed.
1898          * bz20731, LU-592.
1899          */
1900         case IOC_OBD_STATFS: {
1901                 struct obd_statfs stat_buf = {0};
1902 
1903                 if (*((__u32 *) data->ioc_inlbuf2) != 0) {
1904                         rc = -ENODEV;
1905                         goto out;
1906                 }
1907 
1908                 /* copy UUID */
1909                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1910                                  min_t(size_t, data->ioc_plen2,
1911                                                sizeof(struct obd_uuid)))) {
1912                         rc = -EFAULT;
1913                         goto out;
1914                 }
1915 
1916                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1917                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1918                                 0);
1919                 if (rc != 0)
1920                         goto out;
1921 
1922                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1923                                  min_t(size_t, data->ioc_plen1,
1924                                                sizeof(stat_buf)))) {
1925                         rc = -EFAULT;
1926                         goto out;
1927                 }
1928 
1929                 rc = 0;
1930                 goto out;
1931         }
1932         case OBD_IOC_QUOTACTL: {
1933                 struct if_quotactl *qctl = karg;
1934                 struct obd_quotactl *oqctl;
1935 
1936                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1937                 if (oqctl == NULL) {
1938                         rc = -ENOMEM;
1939                         goto out;
1940                 }
1941 
1942                 QCTL_COPY(oqctl, qctl);
1943                 rc = obd_quotactl(exp, oqctl);
1944                 if (rc == 0) {
1945                         QCTL_COPY(qctl, oqctl);
1946                         qctl->qc_valid = QC_MDTIDX;
1947                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
1948                 }
1949 
1950                 kfree(oqctl);
1951                 goto out;
1952         }
1953         case LL_IOC_GET_CONNECT_FLAGS:
1954                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
1955                                  sizeof(*exp_connect_flags_ptr(exp)))) {
1956                         rc = -EFAULT;
1957                         goto out;
1958                 }
1959 
1960                 rc = 0;
1961                 goto out;
1962         case LL_IOC_LOV_SWAP_LAYOUTS:
1963                 rc = mdc_ioc_swap_layouts(exp, karg);
1964                 goto out;
1965         default:
1966                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
1967                 rc = -ENOTTY;
1968                 goto out;
1969         }
1970 out:
1971         module_put(THIS_MODULE);
1972 
1973         return rc;
1974 }
1975 
1976 static int mdc_get_info_rpc(struct obd_export *exp,
1977                             u32 keylen, void *key,
1978                             int vallen, void *val)
1979 {
1980         struct obd_import      *imp = class_exp2cliimp(exp);
1981         struct ptlrpc_request  *req;
1982         char               *tmp;
1983         int                  rc = -EINVAL;
1984 
1985         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1986         if (req == NULL)
1987                 return -ENOMEM;
1988 
1989         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1990                              RCL_CLIENT, keylen);
1991         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1992                              RCL_CLIENT, sizeof(__u32));
1993 
1994         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1995         if (rc) {
1996                 ptlrpc_request_free(req);
1997                 return rc;
1998         }
1999 
2000         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2001         memcpy(tmp, key, keylen);
2002         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2003         memcpy(tmp, &vallen, sizeof(__u32));
2004 
2005         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2006                              RCL_SERVER, vallen);
2007         ptlrpc_request_set_replen(req);
2008 
2009         rc = ptlrpc_queue_wait(req);
2010         /* -EREMOTE means the get_info result is partial, and it needs to
2011          * continue on another MDT, see fid2path part in lmv_iocontrol */
2012         if (rc == 0 || rc == -EREMOTE) {
2013                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2014                 memcpy(val, tmp, vallen);
2015                 if (ptlrpc_rep_need_swab(req)) {
2016                         if (KEY_IS(KEY_FID2PATH))
2017                                 lustre_swab_fid2path(val);
2018                 }
2019         }
2020         ptlrpc_req_finished(req);
2021 
2022         return rc;
2023 }
2024 
2025 static void lustre_swab_hai(struct hsm_action_item *h)
2026 {
2027         __swab32s(&h->hai_len);
2028         __swab32s(&h->hai_action);
2029         lustre_swab_lu_fid(&h->hai_fid);
2030         lustre_swab_lu_fid(&h->hai_dfid);
2031         __swab64s(&h->hai_cookie);
2032         __swab64s(&h->hai_extent.offset);
2033         __swab64s(&h->hai_extent.length);
2034         __swab64s(&h->hai_gid);
2035 }
2036 
2037 static void lustre_swab_hal(struct hsm_action_list *h)
2038 {
2039         struct hsm_action_item  *hai;
2040         int                      i;
2041 
2042         __swab32s(&h->hal_version);
2043         __swab32s(&h->hal_count);
2044         __swab32s(&h->hal_archive_id);
2045         __swab64s(&h->hal_flags);
2046         hai = hai_zero(h);
2047         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2048                 lustre_swab_hai(hai);
2049 }
2050 
2051 static void lustre_swab_kuch(struct kuc_hdr *l)
2052 {
2053         __swab16s(&l->kuc_magic);
2054         /* __u8 l->kuc_transport */
2055         __swab16s(&l->kuc_msgtype);
2056         __swab16s(&l->kuc_msglen);
2057 }
2058 
2059 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2060                                 struct lustre_kernelcomm *lk)
2061 {
2062         struct obd_import  *imp = class_exp2cliimp(exp);
2063         __u32               archive = lk->lk_data;
2064         int                 rc = 0;
2065 
2066         if (lk->lk_group != KUC_GRP_HSM) {
2067                 CERROR("Bad copytool group %d\n", lk->lk_group);
2068                 return -EINVAL;
2069         }
2070 
2071         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2072                lk->lk_uid, lk->lk_group, lk->lk_flags);
2073 
2074         if (lk->lk_flags & LK_FLG_STOP) {
2075                 /* Unregister with the coordinator */
2076                 rc = mdc_ioc_hsm_ct_unregister(imp);
2077         } else {
2078                 rc = mdc_ioc_hsm_ct_register(imp, archive);
2079         }
2080 
2081         return rc;
2082 }
2083 
2084 /**
2085  * Send a message to any listening copytools
2086  * @param val KUC message (kuc_hdr + hsm_action_list)
2087  * @param len total length of message
2088  */
2089 static int mdc_hsm_copytool_send(int len, void *val)
2090 {
2091         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2092         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2093 
2094         if (len < sizeof(*lh) + sizeof(*hal)) {
2095                 CERROR("Short HSM message %d < %d\n", len,
2096                        (int) (sizeof(*lh) + sizeof(*hal)));
2097                 return -EPROTO;
2098         }
2099         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2100                 lustre_swab_kuch(lh);
2101                 lustre_swab_hal(hal);
2102         } else if (lh->kuc_magic != KUC_MAGIC) {
2103                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2104                 return -EPROTO;
2105         }
2106 
2107         CDEBUG(D_HSM,
2108                "Received message mg=%x t=%d m=%d l=%d actions=%d on %s\n",
2109                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2110                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2111 
2112         /* Broadcast to HSM listeners */
2113         return libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2114 }
2115 
2116 /**
2117  * callback function passed to kuc for re-registering each HSM copytool
2118  * running on MDC, after MDT shutdown/recovery.
2119  * @param data archive id served by the copytool
2120  * @param cb_arg callback argument (obd_import)
2121  */
2122 static int mdc_hsm_ct_reregister(__u32 data, void *cb_arg)
2123 {
2124         struct obd_import       *imp = (struct obd_import *)cb_arg;
2125         __u32                    archive = data;
2126         int                      rc;
2127 
2128         CDEBUG(D_HA, "recover copytool registration to MDT (archive=%#x)\n",
2129                archive);
2130         rc = mdc_ioc_hsm_ct_register(imp, archive);
2131 
2132         /* ignore error if the copytool is already registered */
2133         return ((rc != 0) && (rc != -EEXIST)) ? rc : 0;
2134 }
2135 
2136 /**
2137  * Re-establish all kuc contexts with MDT
2138  * after MDT shutdown/recovery.
2139  */
2140 static int mdc_kuc_reregister(struct obd_import *imp)
2141 {
2142         /* re-register HSM agents */
2143         return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2144                                          (void *)imp);
2145 }
2146 
2147 static int mdc_set_info_async(const struct lu_env *env,
2148                               struct obd_export *exp,
2149                               u32 keylen, void *key,
2150                               u32 vallen, void *val,
2151                               struct ptlrpc_request_set *set)
2152 {
2153         struct obd_import       *imp = class_exp2cliimp(exp);
2154         int                      rc;
2155 
2156         if (KEY_IS(KEY_READ_ONLY)) {
2157                 if (vallen != sizeof(int))
2158                         return -EINVAL;
2159 
2160                 spin_lock(&imp->imp_lock);
2161                 if (*((int *)val)) {
2162                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2163                         imp->imp_connect_data.ocd_connect_flags |=
2164                                                         OBD_CONNECT_RDONLY;
2165                 } else {
2166                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2167                         imp->imp_connect_data.ocd_connect_flags &=
2168                                                         ~OBD_CONNECT_RDONLY;
2169                 }
2170                 spin_unlock(&imp->imp_lock);
2171 
2172                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2173                                        keylen, key, vallen, val, set);
2174                 return rc;
2175         }
2176         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2177                 sptlrpc_conf_client_adapt(exp->exp_obd);
2178                 return 0;
2179         }
2180         if (KEY_IS(KEY_FLUSH_CTX)) {
2181                 sptlrpc_import_flush_my_ctx(imp);
2182                 return 0;
2183         }
2184         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2185                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2186                                        keylen, key, vallen, val, set);
2187                 return rc;
2188         }
2189         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2190                 rc = mdc_hsm_copytool_send(vallen, val);
2191                 return rc;
2192         }
2193 
2194         CERROR("Unknown key %s\n", (char *)key);
2195         return -EINVAL;
2196 }
2197 
2198 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2199                         __u32 keylen, void *key, __u32 *vallen, void *val,
2200                         struct lov_stripe_md *lsm)
2201 {
2202         int rc = -EINVAL;
2203 
2204         if (KEY_IS(KEY_MAX_EASIZE)) {
2205                 int mdsize, *max_easize;
2206 
2207                 if (*vallen != sizeof(int))
2208                         return -EINVAL;
2209                 mdsize = *(int *)val;
2210                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2211                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2212                 max_easize = val;
2213                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2214                 return 0;
2215         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2216                 int *default_easize;
2217 
2218                 if (*vallen != sizeof(int))
2219                         return -EINVAL;
2220                 default_easize = val;
2221                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2222                 return 0;
2223         } else if (KEY_IS(KEY_MAX_COOKIESIZE)) {
2224                 int mdsize, *max_cookiesize;
2225 
2226                 if (*vallen != sizeof(int))
2227                         return -EINVAL;
2228                 mdsize = *(int *)val;
2229                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_cookiesize)
2230                         exp->exp_obd->u.cli.cl_max_mds_cookiesize = mdsize;
2231                 max_cookiesize = val;
2232                 *max_cookiesize = exp->exp_obd->u.cli.cl_max_mds_cookiesize;
2233                 return 0;
2234         } else if (KEY_IS(KEY_DEFAULT_COOKIESIZE)) {
2235                 int *default_cookiesize;
2236 
2237                 if (*vallen != sizeof(int))
2238                         return -EINVAL;
2239                 default_cookiesize = val;
2240                 *default_cookiesize =
2241                         exp->exp_obd->u.cli.cl_default_mds_cookiesize;
2242                 return 0;
2243         } else if (KEY_IS(KEY_CONN_DATA)) {
2244                 struct obd_import *imp = class_exp2cliimp(exp);
2245                 struct obd_connect_data *data = val;
2246 
2247                 if (*vallen != sizeof(*data))
2248                         return -EINVAL;
2249 
2250                 *data = imp->imp_connect_data;
2251                 return 0;
2252         } else if (KEY_IS(KEY_TGT_COUNT)) {
2253                 *((int *)val) = 1;
2254                 return 0;
2255         }
2256 
2257         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2258 
2259         return rc;
2260 }
2261 
2262 static int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
2263                     struct obd_capa *oc, struct ptlrpc_request **request)
2264 {
2265         struct ptlrpc_request *req;
2266         int                 rc;
2267 
2268         *request = NULL;
2269         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2270         if (req == NULL)
2271                 return -ENOMEM;
2272 
2273         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2274 
2275         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2276         if (rc) {
2277                 ptlrpc_request_free(req);
2278                 return rc;
2279         }
2280 
2281         mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
2282 
2283         ptlrpc_request_set_replen(req);
2284 
2285         rc = ptlrpc_queue_wait(req);
2286         if (rc)
2287                 ptlrpc_req_finished(req);
2288         else
2289                 *request = req;
2290         return rc;
2291 }
2292 
2293 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2294                             enum obd_import_event event)
2295 {
2296         int rc = 0;
2297 
2298         LASSERT(imp->imp_obd == obd);
2299 
2300         switch (event) {
2301         case IMP_EVENT_DISCON: {
2302 #if 0
2303                 /* XXX Pass event up to OBDs stack. used only for FLD now */
2304                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2305 #endif
2306                 break;
2307         }
2308         case IMP_EVENT_INACTIVE: {
2309                 struct client_obd *cli = &obd->u.cli;
2310                 /*
2311                  * Flush current sequence to make client obtain new one
2312                  * from server in case of disconnect/reconnect.
2313                  */
2314                 if (cli->cl_seq != NULL)
2315                         seq_client_flush(cli->cl_seq);
2316 
2317                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2318                 break;
2319         }
2320         case IMP_EVENT_INVALIDATE: {
2321                 struct ldlm_namespace *ns = obd->obd_namespace;
2322 
2323                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2324 
2325                 break;
2326         }
2327         case IMP_EVENT_ACTIVE:
2328                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2329                 /* redo the kuc registration after reconnecting */
2330                 if (rc == 0)
2331                         rc = mdc_kuc_reregister(imp);
2332                 break;
2333         case IMP_EVENT_OCD:
2334                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2335                 break;
2336         case IMP_EVENT_DEACTIVATE:
2337         case IMP_EVENT_ACTIVATE:
2338                 break;
2339         default:
2340                 CERROR("Unknown import event %x\n", event);
2341                 LBUG();
2342         }
2343         return rc;
2344 }
2345 
2346 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
2347                   struct md_op_data *op_data)
2348 {
2349         struct client_obd *cli = &exp->exp_obd->u.cli;
2350         struct lu_client_seq *seq = cli->cl_seq;
2351 
2352         return seq_client_alloc_fid(NULL, seq, fid);
2353 }
2354 
2355 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2356 {
2357         struct client_obd *cli = &exp->exp_obd->u.cli;
2358 
2359         return &cli->cl_target_uuid;
2360 }
2361 
2362 /**
2363  * Determine whether the lock can be canceled before replaying it during
2364  * recovery, non zero value will be return if the lock can be canceled,
2365  * or zero returned for not
2366  */
2367 static int mdc_cancel_for_recovery(struct ldlm_lock *lock)
2368 {
2369         if (lock->l_resource->lr_type != LDLM_IBITS)
2370                 return 0;
2371 
2372         /* FIXME: if we ever get into a situation where there are too many
2373          * opened files with open locks on a single node, then we really
2374          * should replay these open locks to reget it */
2375         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2376                 return 0;
2377 
2378         return 1;
2379 }
2380 
2381 static int mdc_resource_inode_free(struct ldlm_resource *res)
2382 {
2383         if (res->lr_lvb_inode)
2384                 res->lr_lvb_inode = NULL;
2385 
2386         return 0;
2387 }
2388 
2389 static struct ldlm_valblock_ops inode_lvbo = {
2390         .lvbo_free = mdc_resource_inode_free,
2391 };
2392 
2393 static int mdc_llog_init(struct obd_device *obd)
2394 {
2395         struct obd_llog_group   *olg = &obd->obd_olg;
2396         struct llog_ctxt        *ctxt;
2397         int                      rc;
2398 
2399         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2400                         &llog_client_ops);
2401         if (rc)
2402                 return rc;
2403 
2404         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2405         llog_initiator_connect(ctxt);
2406         llog_ctxt_put(ctxt);
2407 
2408         return 0;
2409 }
2410 
2411 static void mdc_llog_finish(struct obd_device *obd)
2412 {
2413         struct llog_ctxt *ctxt;
2414 
2415         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2416         if (ctxt)
2417                 llog_cleanup(NULL, ctxt);
2418 }
2419 
2420 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2421 {
2422         struct client_obd *cli = &obd->u.cli;
2423         struct lprocfs_static_vars lvars = { NULL };
2424         int rc;
2425 
2426         cli->cl_rpc_lock = kzalloc(sizeof(*cli->cl_rpc_lock), GFP_NOFS);
2427         if (!cli->cl_rpc_lock)
2428                 return -ENOMEM;
2429         mdc_init_rpc_lock(cli->cl_rpc_lock);
2430 
2431         ptlrpcd_addref();
2432 
2433         cli->cl_close_lock = kzalloc(sizeof(*cli->cl_close_lock), GFP_NOFS);
2434         if (!cli->cl_close_lock) {
2435                 rc = -ENOMEM;
2436                 goto err_rpc_lock;
2437         }
2438         mdc_init_rpc_lock(cli->cl_close_lock);
2439 
2440         rc = client_obd_setup(obd, cfg);
2441         if (rc)
2442                 goto err_close_lock;
2443         lprocfs_mdc_init_vars(&lvars);
2444         lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars);
2445         sptlrpc_lprocfs_cliobd_attach(obd);
2446         ptlrpc_lprocfs_register_obd(obd);
2447 
2448         ns_register_cancel(obd->obd_namespace, mdc_cancel_for_recovery);
2449 
2450         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2451 
2452         rc = mdc_llog_init(obd);
2453         if (rc) {
2454                 mdc_cleanup(obd);
2455                 CERROR("failed to setup llogging subsystems\n");
2456         }
2457 
2458         return rc;
2459 
2460 err_close_lock:
2461         kfree(cli->cl_close_lock);
2462 err_rpc_lock:
2463         kfree(cli->cl_rpc_lock);
2464         ptlrpcd_decref();
2465         return rc;
2466 }
2467 
2468 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
2469  * us to make MDS RPCs with large enough reply buffers to hold a default
2470  * sized EA and cookie without having to calculate this (via a call into the
2471  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2472  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2473  * a large number of stripes is possible.  If a larger reply buffer is
2474  * required it will be reallocated in the ptlrpc layer due to overflow.
2475  */
2476 static int mdc_init_ea_size(struct obd_export *exp, int easize,
2477                             int def_easize, int cookiesize, int def_cookiesize)
2478 {
2479         struct obd_device *obd = exp->exp_obd;
2480         struct client_obd *cli = &obd->u.cli;
2481 
2482         if (cli->cl_max_mds_easize < easize)
2483                 cli->cl_max_mds_easize = easize;
2484 
2485         if (cli->cl_default_mds_easize < def_easize)
2486                 cli->cl_default_mds_easize = def_easize;
2487 
2488         if (cli->cl_max_mds_cookiesize < cookiesize)
2489                 cli->cl_max_mds_cookiesize = cookiesize;
2490 
2491         if (cli->cl_default_mds_cookiesize < def_cookiesize)
2492                 cli->cl_default_mds_cookiesize = def_cookiesize;
2493 
2494         return 0;
2495 }
2496 
2497 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2498 {
2499         switch (stage) {
2500         case OBD_CLEANUP_EARLY:
2501                 break;
2502         case OBD_CLEANUP_EXPORTS:
2503                 /* Failsafe, ok if racy */
2504                 if (obd->obd_type->typ_refcnt <= 1)
2505                         libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2506 
2507                 obd_cleanup_client_import(obd);
2508                 ptlrpc_lprocfs_unregister_obd(obd);
2509                 lprocfs_obd_cleanup(obd);
2510 
2511                 mdc_llog_finish(obd);
2512                 break;
2513         }
2514         return 0;
2515 }
2516 
2517 static int mdc_cleanup(struct obd_device *obd)
2518 {
2519         struct client_obd *cli = &obd->u.cli;
2520 
2521         kfree(cli->cl_rpc_lock);
2522         kfree(cli->cl_close_lock);
2523 
2524         ptlrpcd_decref();
2525 
2526         return client_obd_cleanup(obd);
2527 }
2528 
2529 static int mdc_process_config(struct obd_device *obd, u32 len, void *buf)
2530 {
2531         struct lustre_cfg *lcfg = buf;
2532         struct lprocfs_static_vars lvars = { NULL };
2533         int rc = 0;
2534 
2535         lprocfs_mdc_init_vars(&lvars);
2536         switch (lcfg->lcfg_command) {
2537         default:
2538                 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
2539                                               lcfg, obd);
2540                 if (rc > 0)
2541                         rc = 0;
2542                 break;
2543         }
2544         return rc;
2545 }
2546 
2547 
2548 /* get remote permission for current user on fid */
2549 static int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2550                                struct obd_capa *oc, __u32 suppgid,
2551                                struct ptlrpc_request **request)
2552 {
2553         struct ptlrpc_request  *req;
2554         int                 rc;
2555 
2556         LASSERT(client_is_remote(exp));
2557 
2558         *request = NULL;
2559         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2560         if (req == NULL)
2561                 return -ENOMEM;
2562 
2563         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2564 
2565         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2566         if (rc) {
2567                 ptlrpc_request_free(req);
2568                 return rc;
2569         }
2570 
2571         mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2572 
2573         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2574                              sizeof(struct mdt_remote_perm));
2575 
2576         ptlrpc_request_set_replen(req);
2577 
2578         rc = ptlrpc_queue_wait(req);
2579         if (rc)
2580                 ptlrpc_req_finished(req);
2581         else
2582                 *request = req;
2583         return rc;
2584 }
2585 
2586 static int mdc_interpret_renew_capa(const struct lu_env *env,
2587                                     struct ptlrpc_request *req, void *args,
2588                                     int status)
2589 {
2590         struct mdc_renew_capa_args *ra = args;
2591         struct mdt_body *body = NULL;
2592         struct lustre_capa *capa;
2593 
2594         if (status) {
2595                 capa = ERR_PTR(status);
2596                 goto out;
2597         }
2598 
2599         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2600         if (body == NULL) {
2601                 capa = ERR_PTR(-EFAULT);
2602                 goto out;
2603         }
2604 
2605         if ((body->valid & OBD_MD_FLOSSCAPA) == 0) {
2606                 capa = ERR_PTR(-ENOENT);
2607                 goto out;
2608         }
2609 
2610         capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
2611         if (!capa) {
2612                 capa = ERR_PTR(-EFAULT);
2613                 goto out;
2614         }
2615 out:
2616         ra->ra_cb(ra->ra_oc, capa);
2617         return 0;
2618 }
2619 
2620 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2621                           renew_capa_cb_t cb)
2622 {
2623         struct ptlrpc_request *req;
2624         struct mdc_renew_capa_args *ra;
2625 
2626         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
2627                                         LUSTRE_MDS_VERSION, MDS_GETATTR);
2628         if (req == NULL)
2629                 return -ENOMEM;
2630 
2631         /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
2632          * capa to renew is oss capa.
2633          */
2634         mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
2635         ptlrpc_request_set_replen(req);
2636 
2637         CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
2638         ra = ptlrpc_req_async_args(req);
2639         ra->ra_oc = oc;
2640         ra->ra_cb = cb;
2641         req->rq_interpret_reply = mdc_interpret_renew_capa;
2642         ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
2643         return 0;
2644 }
2645 
2646 static struct obd_ops mdc_obd_ops = {
2647         .o_owner            = THIS_MODULE,
2648         .o_setup            = mdc_setup,
2649         .o_precleanup       = mdc_precleanup,
2650         .o_cleanup        = mdc_cleanup,
2651         .o_add_conn      = client_import_add_conn,
2652         .o_del_conn      = client_import_del_conn,
2653         .o_connect          = client_connect_import,
2654         .o_disconnect       = client_disconnect_export,
2655         .o_iocontrol    = mdc_iocontrol,
2656         .o_set_info_async   = mdc_set_info_async,
2657         .o_statfs          = mdc_statfs,
2658         .o_fid_init         = client_fid_init,
2659         .o_fid_fini         = client_fid_fini,
2660         .o_fid_alloc    = mdc_fid_alloc,
2661         .o_import_event     = mdc_import_event,
2662         .o_get_info      = mdc_get_info,
2663         .o_process_config   = mdc_process_config,
2664         .o_get_uuid      = mdc_get_uuid,
2665         .o_quotactl      = mdc_quotactl,
2666         .o_quotacheck       = mdc_quotacheck
2667 };
2668 
2669 static struct md_ops mdc_md_ops = {
2670         .m_getstatus    = mdc_getstatus,
2671         .m_null_inode       = mdc_null_inode,
2672         .m_find_cbdata      = mdc_find_cbdata,
2673         .m_close            = mdc_close,
2674         .m_create          = mdc_create,
2675         .m_done_writing     = mdc_done_writing,
2676         .m_enqueue        = mdc_enqueue,
2677         .m_getattr        = mdc_getattr,
2678         .m_getattr_name     = mdc_getattr_name,
2679         .m_intent_lock      = mdc_intent_lock,
2680         .m_link      = mdc_link,
2681         .m_is_subdir    = mdc_is_subdir,
2682         .m_rename          = mdc_rename,
2683         .m_setattr        = mdc_setattr,
2684         .m_setxattr      = mdc_setxattr,
2685         .m_getxattr      = mdc_getxattr,
2686         .m_sync      = mdc_sync,
2687         .m_readpage      = mdc_readpage,
2688         .m_unlink          = mdc_unlink,
2689         .m_cancel_unused    = mdc_cancel_unused,
2690         .m_init_ea_size     = mdc_init_ea_size,
2691         .m_set_lock_data    = mdc_set_lock_data,
2692         .m_lock_match       = mdc_lock_match,
2693         .m_get_lustre_md    = mdc_get_lustre_md,
2694         .m_free_lustre_md   = mdc_free_lustre_md,
2695         .m_set_open_replay_data = mdc_set_open_replay_data,
2696         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2697         .m_renew_capa       = mdc_renew_capa,
2698         .m_unpack_capa      = mdc_unpack_capa,
2699         .m_get_remote_perm  = mdc_get_remote_perm,
2700         .m_intent_getattr_async = mdc_intent_getattr_async,
2701         .m_revalidate_lock      = mdc_revalidate_lock
2702 };
2703 
2704 static int __init mdc_init(void)
2705 {
2706         struct lprocfs_static_vars lvars = { NULL };
2707 
2708         lprocfs_mdc_init_vars(&lvars);
2709 
2710         return class_register_type(&mdc_obd_ops, &mdc_md_ops,
2711                                  LUSTRE_MDC_NAME, NULL);
2712 }
2713 
2714 static void /*__exit*/ mdc_exit(void)
2715 {
2716         class_unregister_type(LUSTRE_MDC_NAME);
2717 }
2718 
2719 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2720 MODULE_DESCRIPTION("Lustre Metadata Client");
2721 MODULE_LICENSE("GPL");
2722 
2723 module_init(mdc_init);
2724 module_exit(mdc_exit);
2725 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us