Whamcloud - gitweb
71784631e5c5700fb01996175e70156b493a2ab0
[fs/lustre-release.git] / lustre / mdt / mdt_handler.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  * GPL HEADER START
5  *
6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License version 2 only,
10  * as published by the Free Software Foundation.
11  *
12  * This program is distributed in the hope that it will be useful, but
13  * WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * General Public License version 2 for more details (a copy is included
16  * in the LICENSE file that accompanied this code).
17  *
18  * You should have received a copy of the GNU General Public License
19  * version 2 along with this program; If not, see
20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
21  *
22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23  * CA 95054 USA or visit www.sun.com if you need additional information or
24  * have any questions.
25  *
26  * GPL HEADER END
27  */
28 /*
29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
30  * Use is subject to license terms.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/mdt/mdt_handler.c
37  *
38  * Lustre Metadata Target (mdt) request handler
39  *
40  * Author: Peter Braam <braam@clusterfs.com>
41  * Author: Andreas Dilger <adilger@clusterfs.com>
42  * Author: Phil Schwan <phil@clusterfs.com>
43  * Author: Mike Shaver <shaver@clusterfs.com>
44  * Author: Nikita Danilov <nikita@clusterfs.com>
45  * Author: Huang Hua <huanghua@clusterfs.com>
46  * Author: Yury Umanets <umka@clusterfs.com>
47  */
48
49 #ifndef EXPORT_SYMTAB
50 # define EXPORT_SYMTAB
51 #endif
52 #define DEBUG_SUBSYSTEM S_MDS
53
54 #include <linux/module.h>
55 /*
56  * struct OBD_{ALLOC,FREE}*()
57  */
58 #include <obd_support.h>
59 /* struct ptlrpc_request */
60 #include <lustre_net.h>
61 /* struct obd_export */
62 #include <lustre_export.h>
63 /* struct obd_device */
64 #include <obd.h>
65 /* lu2dt_dev() */
66 #include <dt_object.h>
67 #include <lustre_mds.h>
68 #include <lustre_mdt.h>
69 #include "mdt_internal.h"
70 #ifdef HAVE_QUOTA_SUPPORT
71 # include <lustre_quota.h>
72 #endif
73 #include <lustre_acl.h>
74 #include <lustre_param.h>
75 #include <lustre_fsfilt.h>
76
77 mdl_mode_t mdt_mdl_lock_modes[] = {
78         [LCK_MINMODE] = MDL_MINMODE,
79         [LCK_EX]      = MDL_EX,
80         [LCK_PW]      = MDL_PW,
81         [LCK_PR]      = MDL_PR,
82         [LCK_CW]      = MDL_CW,
83         [LCK_CR]      = MDL_CR,
84         [LCK_NL]      = MDL_NL,
85         [LCK_GROUP]   = MDL_GROUP
86 };
87
88 ldlm_mode_t mdt_dlm_lock_modes[] = {
89         [MDL_MINMODE] = LCK_MINMODE,
90         [MDL_EX]      = LCK_EX,
91         [MDL_PW]      = LCK_PW,
92         [MDL_PR]      = LCK_PR,
93         [MDL_CW]      = LCK_CW,
94         [MDL_CR]      = LCK_CR,
95         [MDL_NL]      = LCK_NL,
96         [MDL_GROUP]   = LCK_GROUP
97 };
98
99 /*
100  * Initialized in mdt_mod_init().
101  */
102 unsigned long mdt_num_threads;
103
104 /* ptlrpc request handler for MDT. All handlers are
105  * grouped into several slices - struct mdt_opc_slice,
106  * and stored in an array - mdt_handlers[].
107  */
108 struct mdt_handler {
109         /* The name of this handler. */
110         const char *mh_name;
111         /* Fail id for this handler, checked at the beginning of this handler*/
112         int         mh_fail_id;
113         /* Operation code for this handler */
114         __u32       mh_opc;
115         /* flags are listed in enum mdt_handler_flags below. */
116         __u32       mh_flags;
117         /* The actual handler function to execute. */
118         int (*mh_act)(struct mdt_thread_info *info);
119         /* Request format for this request. */
120         const struct req_format *mh_fmt;
121 };
122
123 enum mdt_handler_flags {
124         /*
125          * struct mdt_body is passed in the incoming message, and object
126          * identified by this fid exists on disk.
127          *
128          * "habeo corpus" == "I have a body"
129          */
130         HABEO_CORPUS = (1 << 0),
131         /*
132          * struct ldlm_request is passed in the incoming message.
133          *
134          * "habeo clavis" == "I have a key"
135          */
136         HABEO_CLAVIS = (1 << 1),
137         /*
138          * this request has fixed reply format, so that reply message can be
139          * packed by generic code.
140          *
141          * "habeo refero" == "I have a reply"
142          */
143         HABEO_REFERO = (1 << 2),
144         /*
145          * this request will modify something, so check whether the filesystem
146          * is readonly or not, then return -EROFS to client asap if necessary.
147          *
148          * "mutabor" == "I shall modify"
149          */
150         MUTABOR      = (1 << 3)
151 };
152
153 struct mdt_opc_slice {
154         __u32               mos_opc_start;
155         int                 mos_opc_end;
156         struct mdt_handler *mos_hs;
157 };
158
159 static struct mdt_opc_slice mdt_regular_handlers[];
160 static struct mdt_opc_slice mdt_readpage_handlers[];
161 static struct mdt_opc_slice mdt_xmds_handlers[];
162 static struct mdt_opc_slice mdt_seq_handlers[];
163 static struct mdt_opc_slice mdt_fld_handlers[];
164
165 static struct mdt_device *mdt_dev(struct lu_device *d);
166 static int mdt_regular_handle(struct ptlrpc_request *req);
167 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
168 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
169                         struct getinfo_fid2path *fp);
170
171 static const struct lu_object_operations mdt_obj_ops;
172
173 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
174 {
175         if (!rep)
176                 return 0;
177         return (rep->lock_policy_res1 & flag);
178 }
179
180 void mdt_clear_disposition(struct mdt_thread_info *info,
181                            struct ldlm_reply *rep, int flag)
182 {
183         if (info)
184                 info->mti_opdata &= ~flag;
185         if (rep)
186                 rep->lock_policy_res1 &= ~flag;
187 }
188
189 void mdt_set_disposition(struct mdt_thread_info *info,
190                          struct ldlm_reply *rep, int flag)
191 {
192         if (info)
193                 info->mti_opdata |= flag;
194         if (rep)
195                 rep->lock_policy_res1 |= flag;
196 }
197
198 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
199 {
200         lh->mlh_pdo_hash = 0;
201         lh->mlh_reg_mode = lm;
202         lh->mlh_type = MDT_REG_LOCK;
203 }
204
205 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
206                        const char *name, int namelen)
207 {
208         lh->mlh_reg_mode = lm;
209         lh->mlh_type = MDT_PDO_LOCK;
210
211         if (name != NULL) {
212                 LASSERT(namelen > 0);
213                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
214         } else {
215                 LASSERT(namelen == 0);
216                 lh->mlh_pdo_hash = 0ull;
217         }
218 }
219
220 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
221                               struct mdt_lock_handle *lh)
222 {
223         mdl_mode_t mode;
224         ENTRY;
225
226         /*
227          * Any dir access needs couple of locks:
228          *
229          * 1) on part of dir we gonna take lookup/modify;
230          *
231          * 2) on whole dir to protect it from concurrent splitting and/or to
232          * flush client's cache for readdir().
233          *
234          * so, for a given mode and object this routine decides what lock mode
235          * to use for lock #2:
236          *
237          * 1) if caller's gonna lookup in dir then we need to protect dir from
238          * being splitted only - LCK_CR
239          *
240          * 2) if caller's gonna modify dir then we need to protect dir from
241          * being splitted and to flush cache - LCK_CW
242          *
243          * 3) if caller's gonna modify dir and that dir seems ready for
244          * splitting then we need to protect it from any type of access
245          * (lookup/modify/split) - LCK_EX --bzzz
246          */
247
248         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
249         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
250
251         /*
252          * Ask underlaying level its opinion about preferable PDO lock mode
253          * having access type passed as regular lock mode:
254          *
255          * - MDL_MINMODE means that lower layer does not want to specify lock
256          * mode;
257          *
258          * - MDL_NL means that no PDO lock should be taken. This is used in some
259          * cases. Say, for non-splittable directories no need to use PDO locks
260          * at all.
261          */
262         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
263                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
264
265         if (mode != MDL_MINMODE) {
266                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
267         } else {
268                 /*
269                  * Lower layer does not want to specify locking mode. We do it
270                  * our selves. No special protection is needed, just flush
271                  * client's cache on modification and allow concurrent
272                  * mondification.
273                  */
274                 switch (lh->mlh_reg_mode) {
275                 case LCK_EX:
276                         lh->mlh_pdo_mode = LCK_EX;
277                         break;
278                 case LCK_PR:
279                         lh->mlh_pdo_mode = LCK_CR;
280                         break;
281                 case LCK_PW:
282                         lh->mlh_pdo_mode = LCK_CW;
283                         break;
284                 default:
285                         CERROR("Not expected lock type (0x%x)\n",
286                                (int)lh->mlh_reg_mode);
287                         LBUG();
288                 }
289         }
290
291         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
292         EXIT;
293 }
294
295 static int mdt_getstatus(struct mdt_thread_info *info)
296 {
297         struct mdt_device *mdt  = info->mti_mdt;
298         struct md_device  *next = mdt->mdt_child;
299         struct mdt_body   *repbody;
300         int                rc;
301
302         ENTRY;
303
304         rc = mdt_check_ucred(info);
305         if (rc)
306                 RETURN(err_serious(rc));
307
308         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
309                 RETURN(err_serious(-ENOMEM));
310
311         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
312         rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
313         if (rc != 0)
314                 RETURN(rc);
315
316         repbody->valid |= OBD_MD_FLID;
317
318         if (mdt->mdt_opts.mo_mds_capa &&
319             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
320                 struct mdt_object  *root;
321                 struct lustre_capa *capa;
322
323                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
324                 if (IS_ERR(root))
325                         RETURN(PTR_ERR(root));
326
327                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
328                 LASSERT(capa);
329                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
330                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
331                                  0);
332                 mdt_object_put(info->mti_env, root);
333                 if (rc == 0)
334                         repbody->valid |= OBD_MD_FLMDSCAPA;
335         }
336
337         RETURN(rc);
338 }
339
340 static int mdt_statfs(struct mdt_thread_info *info)
341 {
342         struct md_device      *next  = info->mti_mdt->mdt_child;
343         struct ptlrpc_service *svc;
344         struct obd_statfs     *osfs;
345         int                    rc;
346
347         ENTRY;
348
349         svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
350
351         /* This will trigger a watchdog timeout */
352         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
353                          (MDT_SERVICE_WATCHDOG_FACTOR *
354                           at_get(&svc->srv_at_estimate)) + 1);
355
356         rc = mdt_check_ucred(info);
357         if (rc)
358                 RETURN(err_serious(rc));
359
360         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
361                 rc = err_serious(-ENOMEM);
362         } else {
363                 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
364                 rc = next->md_ops->mdo_statfs(info->mti_env, next,
365                                               &info->mti_u.ksfs);
366                 statfs_pack(osfs, &info->mti_u.ksfs);
367         }
368         RETURN(rc);
369 }
370
371 void mdt_pack_size2body(struct mdt_thread_info *info, struct mdt_object *o)
372 {
373         struct mdt_body *b;
374         struct lu_attr *attr = &info->mti_attr.ma_attr;
375
376         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
377
378         /* Check if Size-on-MDS is enabled. */
379         if ((mdt_conn_flags(info) & OBD_CONNECT_SOM) &&
380             S_ISREG(attr->la_mode) && mdt_sizeonmds_enabled(o)) {
381                 b->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
382                 b->size = attr->la_size;
383                 b->blocks = attr->la_blocks;
384         }
385 }
386
387 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
388                         const struct lu_attr *attr, const struct lu_fid *fid)
389 {
390         /*XXX should pack the reply body according to lu_valid*/
391         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
392                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
393                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
394                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
395
396         if (!S_ISREG(attr->la_mode))
397                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
398
399         b->atime      = attr->la_atime;
400         b->mtime      = attr->la_mtime;
401         b->ctime      = attr->la_ctime;
402         b->mode       = attr->la_mode;
403         b->size       = attr->la_size;
404         b->blocks     = attr->la_blocks;
405         b->uid        = attr->la_uid;
406         b->gid        = attr->la_gid;
407         b->flags      = attr->la_flags;
408         b->nlink      = attr->la_nlink;
409         b->rdev       = attr->la_rdev;
410
411         if (fid) {
412                 b->fid1 = *fid;
413                 b->valid |= OBD_MD_FLID;
414
415                 /* FIXME: these should be fixed when new igif ready.*/
416                 b->ino  =  fid_oid(fid);       /* 1.6 compatibility */
417                 b->generation = fid_ver(fid);  /* 1.6 compatibility */
418                 b->valid |= OBD_MD_FLGENER;    /* 1.6 compatibility */
419
420                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
421                                 PFID(fid), b->nlink, b->mode, b->size);
422         }
423
424         if (info)
425                 mdt_body_reverse_idmap(info, b);
426 }
427
428 static inline int mdt_body_has_lov(const struct lu_attr *la,
429                                    const struct mdt_body *body)
430 {
431         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
432                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
433 }
434
435 static int mdt_getattr_internal(struct mdt_thread_info *info,
436                                 struct mdt_object *o)
437 {
438         struct md_object        *next = mdt_object_child(o);
439         const struct mdt_body   *reqbody = info->mti_body;
440         struct ptlrpc_request   *req = mdt_info_req(info);
441         struct md_attr          *ma = &info->mti_attr;
442         struct lu_attr          *la = &ma->ma_attr;
443         struct req_capsule      *pill = info->mti_pill;
444         const struct lu_env     *env = info->mti_env;
445         struct mdt_body         *repbody;
446         struct lu_buf           *buffer = &info->mti_buf;
447         int                     rc;
448         ENTRY;
449
450         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
451                 RETURN(err_serious(-ENOMEM));
452
453         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
454
455         ma->ma_valid = 0;
456
457         rc = mdt_object_exists(o);
458         if (rc < 0) {
459                 /* This object is located on remote node.*/
460                 repbody->fid1 = *mdt_object_fid(o);
461                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
462                 RETURN(0);
463         }
464
465         buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
466         buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
467
468         /* If it is dir object and client require MEA, then we got MEA */
469         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
470             reqbody->valid & OBD_MD_MEA) {
471                 /* Assumption: MDT_MD size is enough for lmv size. */
472                 ma->ma_lmv = buffer->lb_buf;
473                 ma->ma_lmv_size = buffer->lb_len;
474                 ma->ma_need = MA_LMV | MA_INODE;
475         } else {
476                 ma->ma_lmm = buffer->lb_buf;
477                 ma->ma_lmm_size = buffer->lb_len;
478                 ma->ma_need = MA_LOV | MA_INODE;
479         }
480
481         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
482             reqbody->valid & OBD_MD_FLDIREA  &&
483             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
484                 /* get default stripe info for this dir. */
485                 ma->ma_need |= MA_LOV_DEF;
486         }
487         rc = mo_attr_get(env, next, ma);
488         if (unlikely(rc)) {
489                 CERROR("getattr error for "DFID": %d\n",
490                         PFID(mdt_object_fid(o)), rc);
491                 RETURN(rc);
492         }
493
494         if (likely(ma->ma_valid & MA_INODE))
495                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
496         else
497                 RETURN(-EFAULT);
498
499         if (mdt_body_has_lov(la, reqbody)) {
500                 if (ma->ma_valid & MA_LOV) {
501                         LASSERT(ma->ma_lmm_size);
502                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
503                         repbody->eadatasize = ma->ma_lmm_size;
504                         if (S_ISDIR(la->la_mode))
505                                 repbody->valid |= OBD_MD_FLDIREA;
506                         else
507                                 repbody->valid |= OBD_MD_FLEASIZE;
508                 }
509                 if (ma->ma_valid & MA_LMV) {
510                         LASSERT(S_ISDIR(la->la_mode));
511                         repbody->eadatasize = ma->ma_lmv_size;
512                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
513                 }
514                 if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
515                         repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
516                 }
517         } else if (S_ISLNK(la->la_mode) &&
518                    reqbody->valid & OBD_MD_LINKNAME) {
519                 buffer->lb_buf = ma->ma_lmm;
520                 buffer->lb_len = reqbody->eadatasize;
521                 rc = mo_readlink(env, next, buffer);
522                 if (unlikely(rc <= 0)) {
523                         CERROR("readlink failed: %d\n", rc);
524                         rc = -EFAULT;
525                 } else {
526                         repbody->valid |= OBD_MD_LINKNAME;
527                         repbody->eadatasize = rc;
528                         /* NULL terminate */
529                         ((char*)ma->ma_lmm)[rc - 1] = 0;
530                         CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
531                                (char*)ma->ma_lmm, rc);
532                         rc = 0;
533                 }
534         }
535
536         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
537                 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
538                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
539                 repbody->valid |= OBD_MD_FLMODEASIZE;
540                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
541                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
542                        repbody->max_cookiesize);
543         }
544
545         if (exp_connect_rmtclient(info->mti_exp) &&
546             reqbody->valid & OBD_MD_FLRMTPERM) {
547                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
548
549                 /* mdt_getattr_lock only */
550                 rc = mdt_pack_remote_perm(info, o, buf);
551                 if (rc) {
552                         repbody->valid &= ~OBD_MD_FLRMTPERM;
553                         repbody->aclsize = 0;
554                         RETURN(rc);
555                 } else {
556                         repbody->valid |= OBD_MD_FLRMTPERM;
557                         repbody->aclsize = sizeof(struct mdt_remote_perm);
558                 }
559         }
560 #ifdef CONFIG_FS_POSIX_ACL
561         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
562                  (reqbody->valid & OBD_MD_FLACL)) {
563                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
564                 buffer->lb_len = req_capsule_get_size(pill,
565                                                       &RMF_ACL, RCL_SERVER);
566                 if (buffer->lb_len > 0) {
567                         rc = mo_xattr_get(env, next, buffer,
568                                           XATTR_NAME_ACL_ACCESS);
569                         if (rc < 0) {
570                                 if (rc == -ENODATA) {
571                                         repbody->aclsize = 0;
572                                         repbody->valid |= OBD_MD_FLACL;
573                                         rc = 0;
574                                 } else if (rc == -EOPNOTSUPP) {
575                                         rc = 0;
576                                 } else {
577                                         CERROR("got acl size: %d\n", rc);
578                                 }
579                         } else {
580                                 repbody->aclsize = rc;
581                                 repbody->valid |= OBD_MD_FLACL;
582                                 rc = 0;
583                         }
584                 }
585         }
586 #endif
587
588         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
589             info->mti_mdt->mdt_opts.mo_mds_capa &&
590             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
591                 struct lustre_capa *capa;
592
593                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
594                 LASSERT(capa);
595                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
596                 rc = mo_capa_get(env, next, capa, 0);
597                 if (rc)
598                         RETURN(rc);
599                 repbody->valid |= OBD_MD_FLMDSCAPA;
600         }
601         RETURN(rc);
602 }
603
604 static int mdt_renew_capa(struct mdt_thread_info *info)
605 {
606         struct mdt_object  *obj = info->mti_object;
607         struct mdt_body    *body;
608         struct lustre_capa *capa, *c;
609         int rc;
610         ENTRY;
611
612         /* if object doesn't exist, or server has disabled capability,
613          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
614          * flag not set.
615          */
616         if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
617             !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
618                 RETURN(0);
619
620         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
621         LASSERT(body != NULL);
622
623         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
624         LASSERT(c);
625
626         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
627         LASSERT(capa);
628
629         *capa = *c;
630         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
631         if (rc == 0)
632                 body->valid |= OBD_MD_FLOSSCAPA;
633         RETURN(rc);
634 }
635
636 static int mdt_getattr(struct mdt_thread_info *info)
637 {
638         struct mdt_object       *obj = info->mti_object;
639         struct req_capsule      *pill = info->mti_pill;
640         struct mdt_body         *reqbody;
641         struct mdt_body         *repbody;
642         mode_t                   mode;
643         int                      md_size;
644         int rc;
645         ENTRY;
646
647         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
648         LASSERT(reqbody);
649
650         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
651                 rc = req_capsule_server_pack(pill);
652                 if (unlikely(rc))
653                         RETURN(err_serious(rc));
654                 rc = mdt_renew_capa(info);
655                 GOTO(out_shrink, rc);
656         }
657
658         LASSERT(obj != NULL);
659         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
660
661         mode = lu_object_attr(&obj->mot_obj.mo_lu);
662         if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
663             (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
664                 md_size = reqbody->eadatasize;
665         else
666                 md_size = info->mti_mdt->mdt_max_mdsize;
667
668         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
669
670         rc = req_capsule_server_pack(pill);
671         if (unlikely(rc != 0))
672                 RETURN(err_serious(rc));
673
674         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
675         LASSERT(repbody != NULL);
676         repbody->eadatasize = 0;
677         repbody->aclsize = 0;
678
679         if (reqbody->valid & OBD_MD_FLRMTPERM)
680                 rc = mdt_init_ucred(info, reqbody);
681         else
682                 rc = mdt_check_ucred(info);
683         if (unlikely(rc))
684                 GOTO(out_shrink, rc);
685
686         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
687         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
688
689         /*
690          * Don't check capability at all, because rename might getattr for
691          * remote obj, and at that time no capability is available.
692          */
693         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
694         rc = mdt_getattr_internal(info, obj);
695         if (reqbody->valid & OBD_MD_FLRMTPERM)
696                 mdt_exit_ucred(info);
697         EXIT;
698 out_shrink:
699         mdt_shrink_reply(info);
700         return rc;
701 }
702
703 static int mdt_is_subdir(struct mdt_thread_info *info)
704 {
705         struct mdt_object     *o = info->mti_object;
706         struct req_capsule    *pill = info->mti_pill;
707         const struct mdt_body *body = info->mti_body;
708         struct mdt_body       *repbody;
709         int                    rc;
710         ENTRY;
711
712         LASSERT(o != NULL);
713
714         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
715
716         /*
717          * We save last checked parent fid to @repbody->fid1 for remote
718          * directory case.
719          */
720         LASSERT(fid_is_sane(&body->fid2));
721         LASSERT(mdt_object_exists(o) > 0);
722         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
723                            &body->fid2, &repbody->fid1);
724         if (rc == 0 || rc == -EREMOTE)
725                 repbody->valid |= OBD_MD_FLID;
726
727         RETURN(rc);
728 }
729
730 static int mdt_raw_lookup(struct mdt_thread_info *info,
731                           struct mdt_object *parent,
732                           const struct lu_name *lname,
733                           struct ldlm_reply *ldlm_rep)
734 {
735         struct md_object *next = mdt_object_child(info->mti_object);
736         const struct mdt_body *reqbody = info->mti_body;
737         struct lu_fid *child_fid = &info->mti_tmp_fid1;
738         struct mdt_body *repbody;
739         int rc;
740         ENTRY;
741
742         if (reqbody->valid != OBD_MD_FLID)
743                 RETURN(0);
744
745         LASSERT(!info->mti_cross_ref);
746
747         /* Only got the fid of this obj by name */
748         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
749                         &info->mti_spec);
750 #if 0
751         /* XXX is raw_lookup possible as intent operation? */
752         if (rc != 0) {
753                 if (rc == -ENOENT)
754                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
755                 RETURN(rc);
756         } else
757                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
758
759         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
760 #endif
761         if (rc == 0) {
762                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
763                 repbody->fid1 = *child_fid;
764                 repbody->valid = OBD_MD_FLID;
765         }
766         RETURN(1);
767 }
768
769 /*
770  * UPDATE lock should be taken against parent, and be release before exit;
771  * child_bits lock should be taken against child, and be returned back:
772  *            (1)normal request should release the child lock;
773  *            (2)intent request will grant the lock to client.
774  */
775 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
776                                  struct mdt_lock_handle *lhc,
777                                  __u64 child_bits,
778                                  struct ldlm_reply *ldlm_rep)
779 {
780         struct ptlrpc_request  *req       = mdt_info_req(info);
781         struct mdt_body        *reqbody   = NULL;
782         struct mdt_object      *parent    = info->mti_object;
783         struct mdt_object      *child;
784         struct md_object       *next      = mdt_object_child(parent);
785         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
786         struct lu_name         *lname     = NULL;
787         const char             *name      = NULL;
788         int                     namelen   = 0;
789         struct mdt_lock_handle *lhp;
790         struct ldlm_lock       *lock;
791         struct ldlm_res_id     *res_id;
792         int                     is_resent;
793         int                     rc;
794
795         ENTRY;
796
797         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
798         LASSERT(ergo(is_resent,
799                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
800
801         LASSERT(parent != NULL);
802         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
803         if (name == NULL)
804                 RETURN(err_serious(-EFAULT));
805
806         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
807                                        RCL_CLIENT) - 1;
808         if (!info->mti_cross_ref) {
809                 /*
810                  * XXX: Check for "namelen == 0" is for getattr by fid
811                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
812                  * that is the name must contain at least one character and
813                  * the terminating '\0'
814                  */
815                 if (namelen == 0) {
816                         reqbody = req_capsule_client_get(info->mti_pill,
817                                                          &RMF_MDT_BODY);
818                         LASSERT(fid_is_sane(&reqbody->fid2));
819                         name = NULL;
820
821                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
822                                "ldlm_rep = %p\n",
823                                PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
824                                ldlm_rep);
825                 } else {
826                         lname = mdt_name(info->mti_env, (char *)name, namelen);
827                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
828                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
829                                name, ldlm_rep);
830                 }
831         }
832         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
833
834         rc = mdt_object_exists(parent);
835         if (unlikely(rc == 0)) {
836                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
837                                 &parent->mot_obj.mo_lu,
838                                 "Parent doesn't exist!\n");
839                 RETURN(-ESTALE);
840         } else if (!info->mti_cross_ref) {
841                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
842                          PFID(mdt_object_fid(parent)));
843         }
844         if (lname) {
845                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
846                 if (rc != 0) {
847                         if (rc > 0)
848                                 rc = 0;
849                         RETURN(rc);
850                 }
851         }
852
853         if (info->mti_cross_ref) {
854                 /* Only getattr on the child. Parent is on another node. */
855                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
856                 child = parent;
857                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
858                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
859
860                 if (is_resent) {
861                         /* Do not take lock for resent case. */
862                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
863                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
864                                  lhc->mlh_reg_lh.cookie);
865                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
866                                                 &lock->l_resource->lr_name));
867                         LDLM_LOCK_PUT(lock);
868                         rc = 0;
869                 } else {
870                         mdt_lock_handle_init(lhc);
871                         mdt_lock_reg_init(lhc, LCK_PR);
872
873                         /*
874                          * Object's name is on another MDS, no lookup lock is
875                          * needed here but update is.
876                          */
877                         child_bits &= ~MDS_INODELOCK_LOOKUP;
878                         child_bits |= MDS_INODELOCK_UPDATE;
879
880                         rc = mdt_object_lock(info, child, lhc, child_bits,
881                                              MDT_LOCAL_LOCK);
882                 }
883                 if (rc == 0) {
884                         /* Finally, we can get attr for child. */
885                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
886                                          BYPASS_CAPA);
887                         rc = mdt_getattr_internal(info, child);
888                         if (unlikely(rc != 0))
889                                 mdt_object_unlock(info, child, lhc, 1);
890                 }
891                 RETURN(rc);
892         }
893
894         /* step 1: lock parent */
895         lhp = &info->mti_lh[MDT_LH_PARENT];
896         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
897         rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
898                              MDT_LOCAL_LOCK);
899
900         if (unlikely(rc != 0))
901                 RETURN(rc);
902
903         if (lname) {
904                 /* step 2: lookup child's fid by name */
905                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
906                                 &info->mti_spec);
907
908                 if (rc != 0) {
909                         if (rc == -ENOENT)
910                                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
911                         GOTO(out_parent, rc);
912                 } else
913                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
914         } else {
915                 *child_fid = reqbody->fid2;
916                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
917         }
918
919         /*
920          *step 3: find the child object by fid & lock it.
921          *        regardless if it is local or remote.
922          */
923         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
924
925         if (unlikely(IS_ERR(child)))
926                 GOTO(out_parent, rc = PTR_ERR(child));
927         if (is_resent) {
928                 /* Do not take lock for resent case. */
929                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
930                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
931                          lhc->mlh_reg_lh.cookie);
932
933                 res_id = &lock->l_resource->lr_name;
934                 if (!fid_res_name_eq(mdt_object_fid(child),
935                                     &lock->l_resource->lr_name)) {
936                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
937                                                  &lock->l_resource->lr_name),
938                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
939                                  (unsigned long)res_id->name[0],
940                                  (unsigned long)res_id->name[1],
941                                  (unsigned long)res_id->name[2],
942                                  PFID(mdt_object_fid(parent)));
943                           CWARN("Although resent, but still not get child lock"
944                                 "parent:"DFID" child:"DFID"\n",
945                                 PFID(mdt_object_fid(parent)),
946                                 PFID(mdt_object_fid(child)));
947                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
948                           LDLM_LOCK_PUT(lock);
949                           GOTO(relock, 0);
950                 }
951                 LDLM_LOCK_PUT(lock);
952                 rc = 0;
953         } else {
954                 struct md_attr *ma;
955 relock:
956                 ma = &info->mti_attr;
957
958                 mdt_lock_handle_init(lhc);
959                 mdt_lock_reg_init(lhc, LCK_PR);
960
961                 if (mdt_object_exists(child) == 0) {
962                         LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
963                                         &child->mot_obj.mo_lu,
964                                         "Object doesn't exist!\n");
965                         GOTO(out_child, rc = -ESTALE);
966                 }
967
968                 ma->ma_valid = 0;
969                 ma->ma_need = MA_INODE;
970                 rc = mo_attr_get(info->mti_env, next, ma);
971                 if (unlikely(rc != 0))
972                         GOTO(out_child, rc);
973
974                 /* If the file has not been changed for some time, we return
975                  * not only a LOOKUP lock, but also an UPDATE lock and this
976                  * might save us RPC on later STAT. For directories, it also
977                  * let negative dentry starts working for this dir. */
978                 if (ma->ma_valid & MA_INODE &&
979                     ma->ma_attr.la_valid & LA_CTIME &&
980                     info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
981                     ma->ma_attr.la_ctime < cfs_time_current_sec())
982                         child_bits |= MDS_INODELOCK_UPDATE;
983
984                 rc = mdt_object_lock(info, child, lhc, child_bits,
985                                      MDT_CROSS_LOCK);
986
987                 if (unlikely(rc != 0))
988                         GOTO(out_child, rc);
989         }
990
991         /* finally, we can get attr for child. */
992         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
993         rc = mdt_getattr_internal(info, child);
994         if (unlikely(rc != 0)) {
995                 mdt_object_unlock(info, child, lhc, 1);
996         } else {
997                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
998                 if (lock) {
999                         struct mdt_body *repbody;
1000
1001                         /* Debugging code. */
1002                         res_id = &lock->l_resource->lr_name;
1003                         LDLM_DEBUG(lock, "Returning lock to client\n");
1004                         LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1005                                                  &lock->l_resource->lr_name),
1006                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1007                                  (unsigned long)res_id->name[0],
1008                                  (unsigned long)res_id->name[1],
1009                                  (unsigned long)res_id->name[2],
1010                                  PFID(mdt_object_fid(child)));
1011                         /*
1012                          * Pack Size-on-MDS inode attributes to the body if
1013                          * update lock is given.
1014                          */
1015                         repbody = req_capsule_server_get(info->mti_pill,
1016                                                          &RMF_MDT_BODY);
1017                         if (lock->l_policy_data.l_inodebits.bits &
1018                             MDS_INODELOCK_UPDATE)
1019                                 mdt_pack_size2body(info, child);
1020                         LDLM_LOCK_PUT(lock);
1021                 }
1022         }
1023         EXIT;
1024 out_child:
1025         mdt_object_put(info->mti_env, child);
1026 out_parent:
1027         mdt_object_unlock(info, parent, lhp, 1);
1028         return rc;
1029 }
1030
1031 /* normal handler: should release the child lock */
1032 static int mdt_getattr_name(struct mdt_thread_info *info)
1033 {
1034         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1035         struct mdt_body        *reqbody;
1036         struct mdt_body        *repbody;
1037         int rc;
1038         ENTRY;
1039
1040         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1041         LASSERT(reqbody != NULL);
1042         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1043         LASSERT(repbody != NULL);
1044
1045         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1046         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1047         repbody->eadatasize = 0;
1048         repbody->aclsize = 0;
1049
1050         rc = mdt_init_ucred(info, reqbody);
1051         if (unlikely(rc))
1052                 GOTO(out_shrink, rc);
1053
1054         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1055         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1056                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1057                 lhc->mlh_reg_lh.cookie = 0;
1058         }
1059         mdt_exit_ucred(info);
1060         EXIT;
1061 out_shrink:
1062         mdt_shrink_reply(info);
1063         return rc;
1064 }
1065
1066 static const struct lu_device_operations mdt_lu_ops;
1067
1068 static int lu_device_is_mdt(struct lu_device *d)
1069 {
1070         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1071 }
1072
1073 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1074                          void *karg, void *uarg);
1075
1076 static int mdt_set_info(struct mdt_thread_info *info)
1077 {
1078         struct ptlrpc_request *req = mdt_info_req(info);
1079         char *key;
1080         void *val;
1081         int keylen, vallen, rc = 0;
1082         ENTRY;
1083
1084         rc = req_capsule_server_pack(info->mti_pill);
1085         if (rc)
1086                 RETURN(rc);
1087
1088         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1089         if (key == NULL) {
1090                 DEBUG_REQ(D_HA, req, "no set_info key");
1091                 RETURN(-EFAULT);
1092         }
1093
1094         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1095                                       RCL_CLIENT);
1096
1097         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1098         if (val == NULL) {
1099                 DEBUG_REQ(D_HA, req, "no set_info val");
1100                 RETURN(-EFAULT);
1101         }
1102
1103         vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1104                                       RCL_CLIENT);
1105
1106         if (KEY_IS(KEY_READ_ONLY)) {
1107                 req->rq_status = 0;
1108                 lustre_msg_set_status(req->rq_repmsg, 0);
1109
1110                 spin_lock(&req->rq_export->exp_lock);
1111                 if (*(__u32 *)val)
1112                         req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1113                 else
1114                         req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
1115                 spin_unlock(&req->rq_export->exp_lock);
1116
1117         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1118                 struct changelog_setinfo *cs =
1119                         (struct changelog_setinfo *)val;
1120                 if (vallen != sizeof(*cs)) {
1121                         CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1122                         RETURN(-EINVAL);
1123                 }
1124                 if (lustre_msg_swabbed(req->rq_reqmsg)) {
1125                         __swab64s(&cs->cs_recno);
1126                         __swab32s(&cs->cs_id);
1127                 }
1128
1129                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1130                                    vallen, val, NULL);
1131                 lustre_msg_set_status(req->rq_repmsg, rc);
1132
1133         } else {
1134                 RETURN(-EINVAL);
1135         }
1136         RETURN(0);
1137 }
1138
1139 static int mdt_connect(struct mdt_thread_info *info)
1140 {
1141         int rc;
1142         struct ptlrpc_request *req;
1143
1144         req = mdt_info_req(info);
1145         rc = target_handle_connect(req);
1146         if (rc == 0) {
1147                 LASSERT(req->rq_export != NULL);
1148                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1149                 rc = mdt_init_sec_level(info);
1150                 if (rc == 0)
1151                         rc = mdt_init_idmap(info);
1152                 if (rc != 0)
1153                         obd_disconnect(class_export_get(req->rq_export));
1154         } else {
1155                 rc = err_serious(rc);
1156         }
1157         return rc;
1158 }
1159
1160 static int mdt_disconnect(struct mdt_thread_info *info)
1161 {
1162         int rc;
1163         ENTRY;
1164
1165         rc = target_handle_disconnect(mdt_info_req(info));
1166         if (rc)
1167                 rc = err_serious(rc);
1168         RETURN(rc);
1169 }
1170
1171 static int mdt_sendpage(struct mdt_thread_info *info,
1172                         struct lu_rdpg *rdpg)
1173 {
1174         struct ptlrpc_request   *req = mdt_info_req(info);
1175         struct ptlrpc_bulk_desc *desc;
1176         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1177         int                      tmpcount;
1178         int                      tmpsize;
1179         int                      timeout;
1180         int                      i;
1181         int                      rc;
1182         ENTRY;
1183
1184         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1185                                     MDS_BULK_PORTAL);
1186         if (desc == NULL)
1187                 RETURN(-ENOMEM);
1188
1189         for (i = 0, tmpcount = rdpg->rp_count;
1190                 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1191                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1192                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1193         }
1194
1195         LASSERT(desc->bd_nob == rdpg->rp_count);
1196         rc = sptlrpc_svc_wrap_bulk(req, desc);
1197         if (rc)
1198                 GOTO(free_desc, rc);
1199
1200         rc = ptlrpc_start_bulk_transfer(desc);
1201         if (rc)
1202                 GOTO(free_desc, rc);
1203
1204         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1205                 GOTO(abort_bulk, rc = 0);
1206
1207         timeout = (int) req->rq_deadline - cfs_time_current_sec();
1208         if (timeout < 0)
1209                 CERROR("Req deadline already passed %lu (now: %lu)\n",
1210                        req->rq_deadline, cfs_time_current_sec());
1211         *lwi = LWI_TIMEOUT(cfs_time_seconds(max(timeout, 1)), NULL, NULL);
1212         rc = l_wait_event(desc->bd_waitq, !ptlrpc_server_bulk_active(desc), lwi);
1213         LASSERT (rc == 0 || rc == -ETIMEDOUT);
1214
1215         if (rc == 0) {
1216                 if (desc->bd_success &&
1217                     desc->bd_nob_transferred == rdpg->rp_count)
1218                         GOTO(free_desc, rc);
1219
1220                 rc = -ETIMEDOUT; /* XXX should this be a different errno? */
1221         }
1222
1223         DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1224                   (rc == -ETIMEDOUT) ? "timeout" : "network error",
1225                   desc->bd_nob_transferred, rdpg->rp_count,
1226                   req->rq_export->exp_client_uuid.uuid,
1227                   req->rq_export->exp_connection->c_remote_uuid.uuid);
1228
1229         class_fail_export(req->rq_export);
1230
1231         EXIT;
1232 abort_bulk:
1233         ptlrpc_abort_bulk(desc);
1234 free_desc:
1235         ptlrpc_free_bulk(desc);
1236         return rc;
1237 }
1238
1239 #ifdef HAVE_SPLIT_SUPPORT
1240 /*
1241  * Retrieve dir entry from the page and insert it to the slave object, actually,
1242  * this should be in osd layer, but since it will not in the final product, so
1243  * just do it here and do not define more moo api anymore for this.
1244  */
1245 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1246                               int size)
1247 {
1248         struct mdt_object *object = info->mti_object;
1249         struct lu_fid *lf = &info->mti_tmp_fid2;
1250         struct md_attr *ma = &info->mti_attr;
1251         struct lu_dirpage *dp;
1252         struct lu_dirent *ent;
1253         int rc = 0, offset = 0;
1254         ENTRY;
1255
1256         /* Make sure we have at least one entry. */
1257         if (size == 0)
1258                 RETURN(-EINVAL);
1259
1260         /*
1261          * Disable trans for this name insert, since it will include many trans
1262          * for this.
1263          */
1264         info->mti_no_need_trans = 1;
1265         /*
1266          * When write_dir_page, no need update parent's ctime,
1267          * and no permission check for name_insert.
1268          */
1269         ma->ma_attr.la_ctime = 0;
1270         ma->ma_attr.la_valid = LA_MODE;
1271         ma->ma_valid = MA_INODE;
1272
1273         cfs_kmap(page);
1274         dp = page_address(page);
1275         offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1276
1277         for (ent = lu_dirent_start(dp); ent != NULL;
1278              ent = lu_dirent_next(ent)) {
1279                 struct lu_name *lname;
1280                 char *name;
1281
1282                 if (le16_to_cpu(ent->lde_namelen) == 0)
1283                         continue;
1284
1285                 fid_le_to_cpu(lf, &ent->lde_fid);
1286                 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1287                         ma->ma_attr.la_mode = S_IFDIR;
1288                 else
1289                         ma->ma_attr.la_mode = 0;
1290                 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1291                 if (name == NULL)
1292                         GOTO(out, rc = -ENOMEM);
1293
1294                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1295                 lname = mdt_name(info->mti_env, name,
1296                                  le16_to_cpu(ent->lde_namelen));
1297                 ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE);
1298                 rc = mdo_name_insert(info->mti_env,
1299                                      md_object_next(&object->mot_obj),
1300                                      lname, lf, ma);
1301                 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1302                 if (rc) {
1303                         CERROR("Can't insert %*.*s, rc %d\n",
1304                                le16_to_cpu(ent->lde_namelen),
1305                                le16_to_cpu(ent->lde_namelen),
1306                                ent->lde_name, rc);
1307                         GOTO(out, rc);
1308                 }
1309
1310                 offset += lu_dirent_size(ent);
1311                 if (offset >= size)
1312                         break;
1313         }
1314         EXIT;
1315 out:
1316         cfs_kunmap(page);
1317         return rc;
1318 }
1319
1320 static int mdt_bulk_timeout(void *data)
1321 {
1322         ENTRY;
1323
1324         CERROR("mdt bulk transfer timeout \n");
1325
1326         RETURN(1);
1327 }
1328
1329 static int mdt_writepage(struct mdt_thread_info *info)
1330 {
1331         struct ptlrpc_request   *req = mdt_info_req(info);
1332         struct mdt_body         *reqbody;
1333         struct l_wait_info      *lwi;
1334         struct ptlrpc_bulk_desc *desc;
1335         struct page             *page;
1336         int                rc;
1337         ENTRY;
1338
1339
1340         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1341         if (reqbody == NULL)
1342                 RETURN(err_serious(-EFAULT));
1343
1344         desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1345         if (desc == NULL)
1346                 RETURN(err_serious(-ENOMEM));
1347
1348         /* allocate the page for the desc */
1349         page = cfs_alloc_page(CFS_ALLOC_STD);
1350         if (page == NULL)
1351                 GOTO(desc_cleanup, rc = -ENOMEM);
1352
1353         CDEBUG(D_INFO, "Received page offset %d size %d \n",
1354                (int)reqbody->size, (int)reqbody->nlink);
1355
1356         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1357                               (int)reqbody->nlink);
1358
1359         rc = sptlrpc_svc_prep_bulk(req, desc);
1360         if (rc != 0)
1361                 GOTO(cleanup_page, rc);
1362         /*
1363          * Check if client was evicted while we were doing i/o before touching
1364          * network.
1365          */
1366         OBD_ALLOC_PTR(lwi);
1367         if (!lwi)
1368                 GOTO(cleanup_page, rc = -ENOMEM);
1369
1370         if (desc->bd_export->exp_failed)
1371                 rc = -ENOTCONN;
1372         else
1373                 rc = ptlrpc_start_bulk_transfer (desc);
1374         if (rc == 0) {
1375                 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
1376                                             mdt_bulk_timeout, desc);
1377                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1378                                   desc->bd_export->exp_failed, lwi);
1379                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1380                 if (rc == -ETIMEDOUT) {
1381                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1382                         ptlrpc_abort_bulk(desc);
1383                 } else if (desc->bd_export->exp_failed) {
1384                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1385                         rc = -ENOTCONN;
1386                         ptlrpc_abort_bulk(desc);
1387                 } else if (!desc->bd_success ||
1388                            desc->bd_nob_transferred != desc->bd_nob) {
1389                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1390                                   desc->bd_success ?
1391                                   "truncated" : "network error on",
1392                                   desc->bd_nob_transferred, desc->bd_nob);
1393                         /* XXX should this be a different errno? */
1394                         rc = -ETIMEDOUT;
1395                 }
1396         } else {
1397                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1398         }
1399         if (rc)
1400                 GOTO(cleanup_lwi, rc);
1401         rc = mdt_write_dir_page(info, page, reqbody->nlink);
1402
1403 cleanup_lwi:
1404         OBD_FREE_PTR(lwi);
1405 cleanup_page:
1406         cfs_free_page(page);
1407 desc_cleanup:
1408         ptlrpc_free_bulk(desc);
1409         RETURN(rc);
1410 }
1411 #endif
1412
1413 static int mdt_readpage(struct mdt_thread_info *info)
1414 {
1415         struct mdt_object *object = info->mti_object;
1416         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1417         struct mdt_body   *reqbody;
1418         struct mdt_body   *repbody;
1419         int                rc;
1420         int                i;
1421         ENTRY;
1422
1423         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1424                 RETURN(err_serious(-ENOMEM));
1425
1426         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1427         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1428         if (reqbody == NULL || repbody == NULL)
1429                 RETURN(err_serious(-EFAULT));
1430
1431         /*
1432          * prepare @rdpg before calling lower layers and transfer itself. Here
1433          * reqbody->size contains offset of where to start to read and
1434          * reqbody->nlink contains number bytes to read.
1435          */
1436         rdpg->rp_hash = reqbody->size;
1437         if (rdpg->rp_hash != reqbody->size) {
1438                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1439                        rdpg->rp_hash, reqbody->size);
1440                 RETURN(-EFAULT);
1441         }
1442
1443         rdpg->rp_attrs = reqbody->mode;
1444         rdpg->rp_count  = reqbody->nlink;
1445         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1446         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1447         if (rdpg->rp_pages == NULL)
1448                 RETURN(-ENOMEM);
1449
1450         for (i = 0; i < rdpg->rp_npages; ++i) {
1451                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1452                 if (rdpg->rp_pages[i] == NULL)
1453                         GOTO(free_rdpg, rc = -ENOMEM);
1454         }
1455
1456         /* call lower layers to fill allocated pages with directory data */
1457         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1458         if (rc)
1459                 GOTO(free_rdpg, rc);
1460
1461         /* send pages to client */
1462         rc = mdt_sendpage(info, rdpg);
1463
1464         EXIT;
1465 free_rdpg:
1466
1467         for (i = 0; i < rdpg->rp_npages; i++)
1468                 if (rdpg->rp_pages[i] != NULL)
1469                         cfs_free_page(rdpg->rp_pages[i]);
1470         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1471
1472         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1473                 RETURN(0);
1474
1475         return rc;
1476 }
1477
1478 static int mdt_reint_internal(struct mdt_thread_info *info,
1479                               struct mdt_lock_handle *lhc,
1480                               __u32 op)
1481 {
1482         struct req_capsule      *pill = info->mti_pill;
1483         struct mdt_device       *mdt = info->mti_mdt;
1484         struct mdt_body         *repbody;
1485         int                      rc = 0;
1486         ENTRY;
1487
1488         /* pack reply */
1489         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1490                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1491                                      mdt->mdt_max_mdsize);
1492         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1493                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1494                                      mdt->mdt_max_cookiesize);
1495
1496         rc = req_capsule_server_pack(pill);
1497         if (rc != 0) {
1498                 CERROR("Can't pack response, rc %d\n", rc);
1499                 RETURN(err_serious(rc));
1500         }
1501
1502         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1503                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1504                 LASSERT(repbody);
1505                 repbody->eadatasize = 0;
1506                 repbody->aclsize = 0;
1507         }
1508
1509         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1510                 GOTO(out_shrink, rc = err_serious(-EFAULT));
1511
1512         rc = mdt_reint_unpack(info, op);
1513         if (rc != 0) {
1514                 CERROR("Can't unpack reint, rc %d\n", rc);
1515                 GOTO(out_shrink, rc = err_serious(rc));
1516         }
1517
1518         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1519
1520         /* for replay no cookkie / lmm need, because client have this already */
1521         if (info->mti_spec.no_create == 1)  {
1522                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1523                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1524
1525                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1526                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1527                                              0);
1528         }
1529
1530         rc = mdt_init_ucred_reint(info);
1531         if (rc)
1532                 GOTO(out_shrink, rc);
1533
1534         rc = mdt_fix_attr_ucred(info, op);
1535         if (rc != 0)
1536                 GOTO(out_ucred, rc = err_serious(rc));
1537
1538         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1539                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1540                 GOTO(out_ucred, rc);
1541         }
1542         rc = mdt_reint_rec(info, lhc);
1543         EXIT;
1544 out_ucred:
1545         mdt_exit_ucred(info);
1546 out_shrink:
1547         mdt_shrink_reply(info);
1548         return rc;
1549 }
1550
1551 static long mdt_reint_opcode(struct mdt_thread_info *info,
1552                              const struct req_format **fmt)
1553 {
1554         struct mdt_rec_reint *rec;
1555         long opc;
1556
1557         opc = err_serious(-EFAULT);
1558         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1559         if (rec != NULL) {
1560                 opc = rec->rr_opcode;
1561                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1562                 if (opc < REINT_MAX && fmt[opc] != NULL)
1563                         req_capsule_extend(info->mti_pill, fmt[opc]);
1564                 else {
1565                         CERROR("Unsupported opc: %ld\n", opc);
1566                         opc = err_serious(opc);
1567                 }
1568         }
1569         return opc;
1570 }
1571
1572 static int mdt_reint(struct mdt_thread_info *info)
1573 {
1574         long opc;
1575         int  rc;
1576
1577         static const struct req_format *reint_fmts[REINT_MAX] = {
1578                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1579                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1580                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1581                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1582                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1583                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1584                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1585         };
1586
1587         ENTRY;
1588
1589         opc = mdt_reint_opcode(info, reint_fmts);
1590         if (opc >= 0) {
1591                 /*
1592                  * No lock possible here from client to pass it to reint code
1593                  * path.
1594                  */
1595                 rc = mdt_reint_internal(info, NULL, opc);
1596         } else {
1597                 rc = opc;
1598         }
1599
1600         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1601         RETURN(rc);
1602 }
1603
1604 /* this should sync the whole device */
1605 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1606 {
1607         struct dt_device *dt = mdt->mdt_bottom;
1608         int rc;
1609         ENTRY;
1610
1611         rc = dt->dd_ops->dt_sync(env, dt);
1612         RETURN(rc);
1613 }
1614
1615 /* this should sync this object */
1616 static int mdt_object_sync(struct mdt_thread_info *info)
1617 {
1618         struct md_object *next;
1619         int rc;
1620         ENTRY;
1621
1622         if (!mdt_object_exists(info->mti_object)) {
1623                 CWARN("Non existing object  "DFID"!\n",
1624                       PFID(mdt_object_fid(info->mti_object)));
1625                 RETURN(-ESTALE);
1626         }
1627         next = mdt_object_child(info->mti_object);
1628         rc = mo_object_sync(info->mti_env, next);
1629
1630         RETURN(rc);
1631 }
1632
1633 static int mdt_sync(struct mdt_thread_info *info)
1634 {
1635         struct req_capsule *pill = info->mti_pill;
1636         struct mdt_body *body;
1637         int rc;
1638         ENTRY;
1639
1640         /* The fid may be zero, so we req_capsule_set manually */
1641         req_capsule_set(pill, &RQF_MDS_SYNC);
1642
1643         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1644         if (body == NULL)
1645                 RETURN(err_serious(-EINVAL));
1646
1647         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1648                 RETURN(err_serious(-ENOMEM));
1649
1650         if (fid_seq(&body->fid1) == 0) {
1651                 /* sync the whole device */
1652                 rc = req_capsule_server_pack(pill);
1653                 if (rc == 0)
1654                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1655                 else
1656                         rc = err_serious(rc);
1657         } else {
1658                 /* sync an object */
1659                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1660                 if (rc == 0) {
1661                         rc = mdt_object_sync(info);
1662                         if (rc == 0) {
1663                                 struct md_object *next;
1664                                 const struct lu_fid *fid;
1665                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1666
1667                                 next = mdt_object_child(info->mti_object);
1668                                 info->mti_attr.ma_need = MA_INODE;
1669                                 info->mti_attr.ma_valid = 0;
1670                                 rc = mo_attr_get(info->mti_env, next,
1671                                                  &info->mti_attr);
1672                                 if (rc == 0) {
1673                                         body = req_capsule_server_get(pill,
1674                                                                 &RMF_MDT_BODY);
1675                                         fid = mdt_object_fid(info->mti_object);
1676                                         mdt_pack_attr2body(info, body, la, fid);
1677                                 }
1678                         }
1679                 } else
1680                         rc = err_serious(rc);
1681         }
1682         RETURN(rc);
1683 }
1684
1685 #ifdef HAVE_QUOTA_SUPPORT
1686 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1687 {
1688         struct obd_quotactl *oqctl;
1689         struct req_capsule *pill = info->mti_pill;
1690         struct obd_export *exp = info->mti_exp;
1691         struct md_device *next = info->mti_mdt->mdt_child;
1692         int rc;
1693         ENTRY;
1694
1695         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_QUOTACHECK_NET))
1696                 RETURN(0);
1697
1698         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1699         if (oqctl == NULL)
1700                 RETURN(-EPROTO);
1701
1702         /* remote client has no permission for quotacheck */
1703         if (unlikely(exp_connect_rmtclient(exp)))
1704                 RETURN(-EPERM);
1705
1706         rc = req_capsule_server_pack(pill);
1707         if (rc)
1708                 RETURN(rc);
1709
1710         rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next, exp,
1711                                                oqctl->qc_type);
1712         RETURN(rc);
1713 }
1714
1715 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1716 {
1717         struct obd_quotactl *oqctl, *repoqc;
1718         struct req_capsule *pill = info->mti_pill;
1719         struct obd_export *exp = info->mti_exp;
1720         struct md_device *next = info->mti_mdt->mdt_child;
1721         const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
1722         int id, rc;
1723         ENTRY;
1724
1725         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_QUOTACTL_NET))
1726                 RETURN(0);
1727
1728         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1729         if (oqctl == NULL)
1730                 RETURN(-EPROTO);
1731
1732         id = oqctl->qc_id;
1733         if (exp_connect_rmtclient(exp)) {
1734                 struct ptlrpc_request *req = mdt_info_req(info);
1735                 struct mdt_export_data *med = mdt_req2med(req);
1736                 struct lustre_idmap_table *idmap = med->med_idmap;
1737
1738                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1739                              oqctl->qc_cmd != Q_GETINFO))
1740                         RETURN(-EPERM);
1741
1742
1743                 if (oqctl->qc_type == USRQUOTA)
1744                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1745                                                      oqctl->qc_id);
1746                 else if (oqctl->qc_type == GRPQUOTA)
1747                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1748                                                      oqctl->qc_id);
1749                 else
1750                         RETURN(-EINVAL);
1751
1752                 if (id == CFS_IDMAP_NOTFOUND) {
1753                         CDEBUG(D_QUOTA, "no mapping for id %u\n",
1754                                oqctl->qc_id);
1755                         RETURN(-EACCES);
1756                 }
1757         }
1758
1759         rc = req_capsule_server_pack(pill);
1760         if (rc)
1761                 RETURN(rc);
1762
1763         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1764         LASSERT(repoqc != NULL);
1765
1766         switch (oqctl->qc_cmd) {
1767         case Q_QUOTAON:
1768                 if (info->mti_mdt->mdt_som_conf) {
1769                         /* Quota cannot be used together with SOM while
1770                          * SOM stored blocks in i_blocks but not in SOM EA. */
1771                         LCONSOLE_ERROR("Fail to turn Quota on: SOM is enabled "
1772                                        "and temporary conflicts with quota.\n");
1773                         RETURN(-ENOTSUPP);
1774                 }
1775                 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
1776                 break;
1777         case Q_QUOTAOFF:
1778                 rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
1779                 break;
1780         case Q_SETINFO:
1781                 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
1782                                       &oqctl->qc_dqinfo);
1783                 break;
1784         case Q_GETINFO:
1785                 rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id,
1786                                       &oqctl->qc_dqinfo);
1787                 break;
1788         case Q_SETQUOTA:
1789                 rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id,
1790                                        &oqctl->qc_dqblk);
1791                 break;
1792         case Q_GETQUOTA:
1793                 rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id,
1794                                        &oqctl->qc_dqblk);
1795                 break;
1796         case Q_GETOINFO:
1797                 rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id,
1798                                        &oqctl->qc_dqinfo);
1799                 break;
1800         case Q_GETOQUOTA:
1801                 rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id,
1802                                         &oqctl->qc_dqblk);
1803                 break;
1804         case LUSTRE_Q_INVALIDATE:
1805                 rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type);
1806                 break;
1807         case LUSTRE_Q_FINVALIDATE:
1808                 rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type);
1809                 break;
1810         default:
1811                 CERROR("unsupported mdt_quotactl command: %d\n",
1812                        oqctl->qc_cmd);
1813                 RETURN(-EFAULT);
1814         }
1815
1816         *repoqc = *oqctl;
1817         RETURN(rc);
1818 }
1819 #endif
1820
1821
1822 /*
1823  * OBD PING and other handlers.
1824  */
1825 static int mdt_obd_ping(struct mdt_thread_info *info)
1826 {
1827         int rc;
1828         ENTRY;
1829
1830         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1831
1832         rc = target_handle_ping(mdt_info_req(info));
1833         if (rc < 0)
1834                 rc = err_serious(rc);
1835         RETURN(rc);
1836 }
1837
1838 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1839 {
1840         return err_serious(-EOPNOTSUPP);
1841 }
1842
1843 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1844 {
1845         return err_serious(-EOPNOTSUPP);
1846 }
1847
1848
1849 /*
1850  * LLOG handlers.
1851  */
1852
1853 /** clone llog ctxt from child (mdd)
1854  * This allows remote llog (replicator) access.
1855  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
1856  * context was originally set up, or we can handle them directly.
1857  * I choose the latter, but that means I need any llog
1858  * contexts set up by child to be accessable by the mdt.  So we clone the
1859  * context into our context list here.
1860  */
1861 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
1862                                int idx)
1863 {
1864         struct md_device  *next = mdt->mdt_child;
1865         struct llog_ctxt *ctxt;
1866         int rc;
1867
1868         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
1869                 return 0;
1870
1871         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
1872         if (rc || ctxt == NULL) {
1873                 CERROR("Can't get mdd ctxt %d\n", rc);
1874                 return rc;
1875         }
1876
1877         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
1878         if (rc)
1879                 CERROR("Can't set mdt ctxt %d\n", rc);
1880
1881         return rc;
1882 }
1883
1884 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
1885                                  struct mdt_device *mdt, int idx)
1886 {
1887         struct llog_ctxt *ctxt;
1888
1889         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
1890         if (ctxt == NULL)
1891                 return 0;
1892         /* Put once for the get we just did, and once for the clone */
1893         llog_ctxt_put(ctxt);
1894         llog_ctxt_put(ctxt);
1895         return 0;
1896 }
1897
1898 static int mdt_llog_create(struct mdt_thread_info *info)
1899 {
1900         int rc;
1901
1902         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
1903         rc = llog_origin_handle_create(mdt_info_req(info));
1904         return (rc < 0 ? err_serious(rc) : rc);
1905 }
1906
1907 static int mdt_llog_destroy(struct mdt_thread_info *info)
1908 {
1909         int rc;
1910
1911         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
1912         rc = llog_origin_handle_destroy(mdt_info_req(info));
1913         return (rc < 0 ? err_serious(rc) : rc);
1914 }
1915
1916 static int mdt_llog_read_header(struct mdt_thread_info *info)
1917 {
1918         int rc;
1919
1920         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
1921         rc = llog_origin_handle_read_header(mdt_info_req(info));
1922         return (rc < 0 ? err_serious(rc) : rc);
1923 }
1924
1925 static int mdt_llog_next_block(struct mdt_thread_info *info)
1926 {
1927         int rc;
1928
1929         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
1930         rc = llog_origin_handle_next_block(mdt_info_req(info));
1931         return (rc < 0 ? err_serious(rc) : rc);
1932 }
1933
1934 static int mdt_llog_prev_block(struct mdt_thread_info *info)
1935 {
1936         int rc;
1937
1938         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
1939         rc = llog_origin_handle_prev_block(mdt_info_req(info));
1940         return (rc < 0 ? err_serious(rc) : rc);
1941 }
1942
1943
1944 /*
1945  * DLM handlers.
1946  */
1947 static struct ldlm_callback_suite cbs = {
1948         .lcs_completion = ldlm_server_completion_ast,
1949         .lcs_blocking   = ldlm_server_blocking_ast,
1950         .lcs_glimpse    = NULL
1951 };
1952
1953 static int mdt_enqueue(struct mdt_thread_info *info)
1954 {
1955         struct ptlrpc_request *req;
1956         __u64 req_bits;
1957         int rc;
1958
1959         /*
1960          * info->mti_dlm_req already contains swapped and (if necessary)
1961          * converted dlm request.
1962          */
1963         LASSERT(info->mti_dlm_req != NULL);
1964
1965         req = mdt_info_req(info);
1966
1967         /*
1968          * Lock without inodebits makes no sense and will oops later in
1969          * ldlm. Let's check it now to see if we have wrong lock from client or
1970          * bits get corrupted somewhere in mdt_intent_policy().
1971          */
1972         req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1973         /* This is disabled because we need to support liblustre flock.
1974          * LASSERT(req_bits != 0);
1975          */
1976
1977         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1978                                   req, info->mti_dlm_req, &cbs);
1979         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1980         return rc ? err_serious(rc) : req->rq_status;
1981 }
1982
1983 static int mdt_convert(struct mdt_thread_info *info)
1984 {
1985         int rc;
1986         struct ptlrpc_request *req;
1987
1988         LASSERT(info->mti_dlm_req);
1989         req = mdt_info_req(info);
1990         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1991         return rc ? err_serious(rc) : req->rq_status;
1992 }
1993
1994 static int mdt_bl_callback(struct mdt_thread_info *info)
1995 {
1996         CERROR("bl callbacks should not happen on MDS\n");
1997         LBUG();
1998         return err_serious(-EOPNOTSUPP);
1999 }
2000
2001 static int mdt_cp_callback(struct mdt_thread_info *info)
2002 {
2003         CERROR("cp callbacks should not happen on MDS\n");
2004         LBUG();
2005         return err_serious(-EOPNOTSUPP);
2006 }
2007
2008 /*
2009  * sec context handlers
2010  */
2011 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2012 {
2013         int rc;
2014
2015         rc = mdt_handle_idmap(info);
2016
2017         if (unlikely(rc)) {
2018                 struct ptlrpc_request *req = mdt_info_req(info);
2019                 __u32                  opc;
2020
2021                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2022                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2023                         sptlrpc_svc_ctx_invalidate(req);
2024         }
2025
2026         OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
2027
2028         return rc;
2029 }
2030
2031 static struct mdt_object *mdt_obj(struct lu_object *o)
2032 {
2033         LASSERT(lu_device_is_mdt(o->lo_dev));
2034         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2035 }
2036
2037 struct mdt_object *mdt_object_find(const struct lu_env *env,
2038                                    struct mdt_device *d,
2039                                    const struct lu_fid *f)
2040 {
2041         struct lu_object *o;
2042         struct mdt_object *m;
2043         ENTRY;
2044
2045         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2046         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2047         if (unlikely(IS_ERR(o)))
2048                 m = (struct mdt_object *)o;
2049         else
2050                 m = mdt_obj(o);
2051         RETURN(m);
2052 }
2053
2054 /**
2055  * Asyncronous commit for mdt device.
2056  *
2057  * Pass asynchonous commit call down the MDS stack.
2058  *
2059  * \param env environment
2060  * \param mdt the mdt device
2061  */
2062 static void mdt_device_commit_async(const struct lu_env *env,
2063                                     struct mdt_device *mdt)
2064 {
2065         struct dt_device *dt = mdt->mdt_bottom;
2066         int rc;
2067
2068         rc = dt->dd_ops->dt_commit_async(env, dt);
2069         if (unlikely(rc != 0))
2070                 CWARN("async commit start failed with rc = %d", rc);
2071 }
2072
2073 /**
2074  * Mark the lock as "synchonous".
2075  *
2076  * Mark the lock to deffer transaction commit to the unlock time.
2077  *
2078  * \param lock the lock to mark as "synchonous"
2079  *
2080  * \see mdt_is_lock_sync
2081  * \see mdt_save_lock
2082  */
2083 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2084 {
2085         lock->l_ast_data = (void*)1;
2086 }
2087
2088 /**
2089  * Check whehter the lock "synchonous" or not.
2090  *
2091  * \param lock the lock to check
2092  * \retval 1 the lock is "synchonous"
2093  * \retval 0 the lock isn't "synchronous"
2094  *
2095  * \see mdt_set_lock_sync
2096  * \see mdt_save_lock
2097  */
2098 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2099 {
2100         return lock->l_ast_data != NULL;
2101 }
2102
2103 /**
2104  * Blocking AST for mdt locks.
2105  *
2106  * Starts transaction commit if in case of COS lock conflict or
2107  * deffers such a commit to the mdt_save_lock.
2108  *
2109  * \param lock the lock which blocks a request or cancelling lock
2110  * \param desc unused
2111  * \param data unused
2112  * \param flag indicates whether this cancelling or blocking callback
2113  * \retval 0
2114  * \see ldlm_blocking_ast_nocheck
2115  */
2116 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2117                      void *data, int flag)
2118 {
2119         struct obd_device *obd = lock->l_resource->lr_namespace->ns_obd;
2120         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2121         int rc;
2122         ENTRY;
2123
2124         if (flag == LDLM_CB_CANCELING)
2125                 RETURN(0);
2126         lock_res_and_lock(lock);
2127         if (lock->l_blocking_ast != mdt_blocking_ast) {
2128                 unlock_res_and_lock(lock);
2129                 RETURN(0);
2130         }
2131         if (mdt_cos_is_enabled(mdt) &&
2132             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2133             lock->l_blocking_lock != NULL &&
2134             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2135                 mdt_set_lock_sync(lock);
2136         }
2137         rc = ldlm_blocking_ast_nocheck(lock);
2138
2139         /* There is no lock conflict if l_blocking_lock == NULL,
2140          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2141          * when the last reference to a local lock was released */
2142         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2143                 struct lu_env env;
2144
2145                 rc = lu_env_init(&env, LCT_MD_THREAD);
2146                 if (unlikely(rc != 0))
2147                         CWARN("lu_env initialization failed with rc = %d,"
2148                               "cannot start asynchronous commit\n", rc);
2149                 else
2150                         mdt_device_commit_async(&env, mdt);
2151                 lu_env_fini(&env);
2152         }
2153         RETURN(rc);
2154 }
2155
2156 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2157                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2158 {
2159         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2160         ldlm_policy_data_t *policy = &info->mti_policy;
2161         struct ldlm_res_id *res_id = &info->mti_res_id;
2162         int rc;
2163         ENTRY;
2164
2165         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2166         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2167         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2168         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2169
2170         if (mdt_object_exists(o) < 0) {
2171                 if (locality == MDT_CROSS_LOCK) {
2172                         /* cross-ref object fix */
2173                         ibits &= ~MDS_INODELOCK_UPDATE;
2174                         ibits |= MDS_INODELOCK_LOOKUP;
2175                 } else {
2176                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2177                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2178                 }
2179                 /* No PDO lock on remote object */
2180                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2181         }
2182
2183         if (lh->mlh_type == MDT_PDO_LOCK) {
2184                 /* check for exists after object is locked */
2185                 if (mdt_object_exists(o) == 0) {
2186                         /* Non-existent object shouldn't have PDO lock */
2187                         RETURN(-ESTALE);
2188                 } else {
2189                         /* Non-dir object shouldn't have PDO lock */
2190                         LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
2191                 }
2192         }
2193
2194         memset(policy, 0, sizeof(*policy));
2195         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2196
2197         /*
2198          * Take PDO lock on whole directory and build correct @res_id for lock
2199          * on part of directory.
2200          */
2201         if (lh->mlh_pdo_hash != 0) {
2202                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2203                 mdt_lock_pdo_mode(info, o, lh);
2204                 if (lh->mlh_pdo_mode != LCK_NL) {
2205                         /*
2206                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2207                          * is never going to be sent to client and we do not
2208                          * want it slowed down due to possible cancels.
2209                          */
2210                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2211                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2212                                           policy, res_id, LDLM_FL_ATOMIC_CB,
2213                                           &info->mti_exp->exp_handle.h_cookie);
2214                         if (unlikely(rc))
2215                                 RETURN(rc);
2216                 }
2217
2218                 /*
2219                  * Finish res_id initializing by name hash marking part of
2220                  * directory which is taking modification.
2221                  */
2222                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2223         }
2224
2225         policy->l_inodebits.bits = ibits;
2226
2227         /*
2228          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2229          * going to be sent to client. If it is - mdt_intent_policy() path will
2230          * fix it up and turn FL_LOCAL flag off.
2231          */
2232         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2233                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
2234                           &info->mti_exp->exp_handle.h_cookie);
2235         if (rc)
2236                 GOTO(out, rc);
2237
2238 out:
2239         if (rc)
2240                 mdt_object_unlock(info, o, lh, 1);
2241
2242
2243         RETURN(rc);
2244 }
2245
2246 /**
2247  * Save a lock within request object.
2248  *
2249  * Keep the lock referenced until whether client ACK or transaction
2250  * commit happens or release the lock immediately depending on input
2251  * parameters. If COS is ON, a write lock is converted to COS lock
2252  * before saving.
2253  *
2254  * \param info thead info object
2255  * \param h lock handle
2256  * \param mode lock mode
2257  * \param decref force immediate lock releasing
2258  */
2259 static
2260 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2261                    ldlm_mode_t mode, int decref)
2262 {
2263         ENTRY;
2264
2265         if (lustre_handle_is_used(h)) {
2266                 if (decref || !info->mti_has_trans ||
2267                     !(mode & (LCK_PW | LCK_EX))){
2268                         mdt_fid_unlock(h, mode);
2269                 } else {
2270                         struct mdt_device *mdt = info->mti_mdt;
2271                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2272                         struct ptlrpc_request *req = mdt_info_req(info);
2273                         int no_ack = 0;
2274
2275                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2276                                  h->cookie);
2277                         CDEBUG(D_HA, "request = %p reply state = %p"
2278                                " transno = "LPD64"\n",
2279                                req, req->rq_reply_state, req->rq_transno);
2280                         if (mdt_cos_is_enabled(mdt)) {
2281                                 no_ack = 1;
2282                                 ldlm_lock_downgrade(lock, LCK_COS);
2283                                 mode = LCK_COS;
2284                         }
2285                         ptlrpc_save_lock(req, h, mode, no_ack);
2286                         if (mdt_is_lock_sync(lock)) {
2287                                 CDEBUG(D_HA, "found sync-lock,"
2288                                        " async commit started\n");
2289                                 mdt_device_commit_async(info->mti_env,
2290                                                         mdt);
2291                         }
2292                         LDLM_LOCK_PUT(lock);
2293                 }
2294                 h->cookie = 0ull;
2295         }
2296
2297         EXIT;
2298 }
2299
2300 /**
2301  * Unlock mdt object.
2302  *
2303  * Immeditely release the regular lock and the PDO lock or save the
2304  * lock in reqeuest and keep them referenced until client ACK or
2305  * transaction commit.
2306  *
2307  * \param info thread info object
2308  * \param o mdt object
2309  * \param h mdt lock handle referencing regular and PDO locks
2310  * \param decref force immediate lock releasing
2311  */
2312 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2313                        struct mdt_lock_handle *lh, int decref)
2314 {
2315         ENTRY;
2316
2317         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2318         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2319
2320         EXIT;
2321 }
2322
2323 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2324                                         const struct lu_fid *f,
2325                                         struct mdt_lock_handle *lh,
2326                                         __u64 ibits)
2327 {
2328         struct mdt_object *o;
2329
2330         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2331         if (!IS_ERR(o)) {
2332                 int rc;
2333
2334                 rc = mdt_object_lock(info, o, lh, ibits,
2335                                      MDT_LOCAL_LOCK);
2336                 if (rc != 0) {
2337                         mdt_object_put(info->mti_env, o);
2338                         o = ERR_PTR(rc);
2339                 }
2340         }
2341         return o;
2342 }
2343
2344 void mdt_object_unlock_put(struct mdt_thread_info * info,
2345                            struct mdt_object * o,
2346                            struct mdt_lock_handle *lh,
2347                            int decref)
2348 {
2349         mdt_object_unlock(info, o, lh, decref);
2350         mdt_object_put(info->mti_env, o);
2351 }
2352
2353 static struct mdt_handler *mdt_handler_find(__u32 opc,
2354                                             struct mdt_opc_slice *supported)
2355 {
2356         struct mdt_opc_slice *s;
2357         struct mdt_handler   *h;
2358
2359         h = NULL;
2360         for (s = supported; s->mos_hs != NULL; s++) {
2361                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2362                         h = s->mos_hs + (opc - s->mos_opc_start);
2363                         if (likely(h->mh_opc != 0))
2364                                 LASSERTF(h->mh_opc == opc,
2365                                          "opcode mismatch %d != %d\n",
2366                                          h->mh_opc, opc);
2367                         else
2368                                 h = NULL; /* unsupported opc */
2369                         break;
2370                 }
2371         }
2372         return h;
2373 }
2374
2375 static int mdt_lock_resname_compat(struct mdt_device *m,
2376                                    struct ldlm_request *req)
2377 {
2378         /* XXX something... later. */
2379         return 0;
2380 }
2381
2382 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2383 {
2384         /* XXX something... later. */
2385         return 0;
2386 }
2387
2388 /*
2389  * Generic code handling requests that have struct mdt_body passed in:
2390  *
2391  *  - extract mdt_body from request and save it in @info, if present;
2392  *
2393  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2394  *  @info;
2395  *
2396  *  - if HABEO_CORPUS flag is set for this request type check whether object
2397  *  actually exists on storage (lu_object_exists()).
2398  *
2399  */
2400 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2401 {
2402         const struct mdt_body    *body;
2403         struct mdt_object        *obj;
2404         const struct lu_env      *env;
2405         struct req_capsule       *pill;
2406         int                       rc;
2407         ENTRY;
2408
2409         env = info->mti_env;
2410         pill = info->mti_pill;
2411
2412         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2413         if (body == NULL)
2414                 RETURN(-EFAULT);
2415
2416         if (!(body->valid & OBD_MD_FLID))
2417                 RETURN(0);
2418
2419         if (!fid_is_sane(&body->fid1)) {
2420                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2421                 RETURN(-EINVAL);
2422         }
2423
2424         /*
2425          * Do not get size or any capa fields before we check that request
2426          * contains capa actually. There are some requests which do not, for
2427          * instance MDS_IS_SUBDIR.
2428          */
2429         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2430             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2431                 mdt_set_capainfo(info, 0, &body->fid1,
2432                                  req_capsule_client_get(pill, &RMF_CAPA1));
2433
2434         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2435         if (!IS_ERR(obj)) {
2436                 if ((flags & HABEO_CORPUS) &&
2437                     !mdt_object_exists(obj)) {
2438                         mdt_object_put(env, obj);
2439                         /* for capability renew ENOENT will be handled in
2440                          * mdt_renew_capa */
2441                         if (body->valid & OBD_MD_FLOSSCAPA)
2442                                 rc = 0;
2443                         else
2444                                 rc = -ENOENT;
2445                 } else {
2446                         info->mti_object = obj;
2447                         rc = 0;
2448                 }
2449         } else
2450                 rc = PTR_ERR(obj);
2451
2452         RETURN(rc);
2453 }
2454
2455 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2456 {
2457         struct req_capsule *pill = info->mti_pill;
2458         int rc;
2459         ENTRY;
2460
2461         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2462                 rc = mdt_body_unpack(info, flags);
2463         else
2464                 rc = 0;
2465
2466         if (rc == 0 && (flags & HABEO_REFERO)) {
2467                 struct mdt_device *mdt = info->mti_mdt;
2468
2469                 /* Pack reply. */
2470
2471                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2472                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2473                                              mdt->mdt_max_mdsize);
2474                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2475                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2476                                              mdt->mdt_max_cookiesize);
2477
2478                 rc = req_capsule_server_pack(pill);
2479         }
2480         RETURN(rc);
2481 }
2482
2483 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2484 {
2485         struct md_device *next = m->mdt_child;
2486
2487         return next->md_ops->mdo_init_capa_ctxt(env, next,
2488                                                 m->mdt_opts.mo_mds_capa,
2489                                                 m->mdt_capa_timeout,
2490                                                 m->mdt_capa_alg,
2491                                                 m->mdt_capa_keys);
2492 }
2493
2494 /*
2495  * Invoke handler for this request opc. Also do necessary preprocessing
2496  * (according to handler ->mh_flags), and post-processing (setting of
2497  * ->last_{xid,committed}).
2498  */
2499 static int mdt_req_handle(struct mdt_thread_info *info,
2500                           struct mdt_handler *h, struct ptlrpc_request *req)
2501 {
2502         int   rc, serious = 0;
2503         __u32 flags;
2504
2505         ENTRY;
2506
2507         LASSERT(h->mh_act != NULL);
2508         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2509         LASSERT(current->journal_info == NULL);
2510
2511         /*
2512          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2513          * to put same checks into handlers like mdt_close(), mdt_reint(),
2514          * etc., without talking to mdt authors first. Checking same thing
2515          * there again is useless and returning 0 error wihtout packing reply
2516          * is buggy! Handlers either pack reply or return error.
2517          *
2518          * We return 0 here and do not send any reply in order to emulate
2519          * network failure. Do not send any reply in case any of NET related
2520          * fail_id has occured.
2521          */
2522         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2523                 RETURN(0);
2524
2525         rc = 0;
2526         flags = h->mh_flags;
2527         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2528
2529         if (h->mh_fmt != NULL) {
2530                 req_capsule_set(info->mti_pill, h->mh_fmt);
2531                 rc = mdt_unpack_req_pack_rep(info, flags);
2532         }
2533
2534         if (rc == 0 && flags & MUTABOR &&
2535             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2536                 /* should it be rq_status? */
2537                 rc = -EROFS;
2538
2539         if (rc == 0 && flags & HABEO_CLAVIS) {
2540                 struct ldlm_request *dlm_req;
2541
2542                 LASSERT(h->mh_fmt != NULL);
2543
2544                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2545                 if (dlm_req != NULL) {
2546                         if (info->mti_mdt->mdt_opts.mo_compat_resname)
2547                                 rc = mdt_lock_resname_compat(info->mti_mdt,
2548                                                              dlm_req);
2549                         info->mti_dlm_req = dlm_req;
2550                 } else {
2551                         rc = -EFAULT;
2552                 }
2553         }
2554
2555         /* capability setting changed via /proc, needs reinitialize ctxt */
2556         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2557                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2558                 info->mti_mdt->mdt_capa_conf = 0;
2559         }
2560
2561         if (likely(rc == 0)) {
2562                 /*
2563                  * Process request, there can be two types of rc:
2564                  * 1) errors with msg unpack/pack, other failures outside the
2565                  * operation itself. This is counted as serious errors;
2566                  * 2) errors during fs operation, should be placed in rq_status
2567                  * only
2568                  */
2569                 rc = h->mh_act(info);
2570                 if (rc == 0 &&
2571                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2572                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2573                                   "pack reply and returned 0 error\n",
2574                                   h->mh_name);
2575                         LBUG();
2576                 }
2577                 serious = is_serious(rc);
2578                 rc = clear_serious(rc);
2579         } else
2580                 serious = 1;
2581
2582         req->rq_status = rc;
2583
2584         /*
2585          * ELDLM_* codes which > 0 should be in rq_status only as well as
2586          * all non-serious errors.
2587          */
2588         if (rc > 0 || !serious)
2589                 rc = 0;
2590
2591         LASSERT(current->journal_info == NULL);
2592
2593         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2594             info->mti_mdt->mdt_opts.mo_compat_resname) {
2595                 struct ldlm_reply *dlmrep;
2596
2597                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2598                 if (dlmrep != NULL)
2599                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2600         }
2601
2602         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2603         if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2604                 target_committed_to_req(req);
2605
2606         if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2607                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2608                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2609                 LBUG();
2610         }
2611
2612         target_send_reply(req, rc, info->mti_fail_id);
2613         RETURN(0);
2614 }
2615
2616 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2617 {
2618         lh->mlh_type = MDT_NUL_LOCK;
2619         lh->mlh_reg_lh.cookie = 0ull;
2620         lh->mlh_reg_mode = LCK_MINMODE;
2621         lh->mlh_pdo_lh.cookie = 0ull;
2622         lh->mlh_pdo_mode = LCK_MINMODE;
2623 }
2624
2625 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2626 {
2627         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2628         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2629 }
2630
2631 /*
2632  * Initialize fields of struct mdt_thread_info. Other fields are left in
2633  * uninitialized state, because it's too expensive to zero out whole
2634  * mdt_thread_info (> 1K) on each request arrival.
2635  */
2636 static void mdt_thread_info_init(struct ptlrpc_request *req,
2637                                  struct mdt_thread_info *info)
2638 {
2639         int i;
2640         struct md_capainfo *ci;
2641
2642         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2643         info->mti_pill = &req->rq_pill;
2644
2645         /* lock handle */
2646         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2647                 mdt_lock_handle_init(&info->mti_lh[i]);
2648
2649         /* mdt device: it can be NULL while CONNECT */
2650         if (req->rq_export) {
2651                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2652                 info->mti_exp = req->rq_export;
2653         } else
2654                 info->mti_mdt = NULL;
2655         info->mti_env = req->rq_svc_thread->t_env;
2656         ci = md_capainfo(info->mti_env);
2657         memset(ci, 0, sizeof *ci);
2658         if (req->rq_export) {
2659                 if (exp_connect_rmtclient(req->rq_export))
2660                         ci->mc_auth = LC_ID_CONVERT;
2661                 else if (req->rq_export->exp_connect_flags &
2662                          OBD_CONNECT_MDS_CAPA)
2663                         ci->mc_auth = LC_ID_PLAIN;
2664                 else
2665                         ci->mc_auth = LC_ID_NONE;
2666         }
2667
2668         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2669         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2670         info->mti_mos[0] = NULL;
2671         info->mti_mos[1] = NULL;
2672         info->mti_mos[2] = NULL;
2673         info->mti_mos[3] = NULL;
2674
2675         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2676         info->mti_body = NULL;
2677         info->mti_object = NULL;
2678         info->mti_dlm_req = NULL;
2679         info->mti_has_trans = 0;
2680         info->mti_no_need_trans = 0;
2681         info->mti_cross_ref = 0;
2682         info->mti_opdata = 0;
2683
2684         /* To not check for split by default. */
2685         info->mti_spec.sp_ck_split = 0;
2686         info->mti_spec.no_create = 0;
2687 }
2688
2689 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2690 {
2691         int i;
2692
2693         req_capsule_fini(info->mti_pill);
2694         if (info->mti_object != NULL) {
2695                 /*
2696                  * freeing an object may lead to OSD level transaction, do not
2697                  * let it mess with MDT. bz19385.
2698                  */
2699                 info->mti_no_need_trans = 1;
2700                 mdt_object_put(info->mti_env, info->mti_object);
2701                 info->mti_object = NULL;
2702         }
2703         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2704                 mdt_lock_handle_fini(&info->mti_lh[i]);
2705         info->mti_env = NULL;
2706 }
2707
2708 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2709                                        struct obd_device *obd, int *process)
2710 {
2711         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2712         case MDS_CONNECT: /* This will never get here, but for completeness. */
2713         case OST_CONNECT: /* This will never get here, but for completeness. */
2714         case MDS_DISCONNECT:
2715         case OST_DISCONNECT:
2716                *process = 1;
2717                RETURN(0);
2718
2719         case MDS_CLOSE:
2720         case MDS_DONE_WRITING:
2721         case MDS_SYNC: /* used in unmounting */
2722         case OBD_PING:
2723         case MDS_REINT:
2724         case SEQ_QUERY:
2725         case FLD_QUERY:
2726         case LDLM_ENQUEUE:
2727                 *process = target_queue_recovery_request(req, obd);
2728                 RETURN(0);
2729
2730         default:
2731                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2732                 *process = -EAGAIN;
2733                 RETURN(0);
2734         }
2735 }
2736
2737 /*
2738  * Handle recovery. Return:
2739  *        +1: continue request processing;
2740  *       -ve: abort immediately with the given error code;
2741  *         0: send reply with error code in req->rq_status;
2742  */
2743 static int mdt_recovery(struct mdt_thread_info *info)
2744 {
2745         struct ptlrpc_request *req = mdt_info_req(info);
2746         int recovering;
2747         struct obd_device *obd;
2748
2749         ENTRY;
2750
2751         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2752         case MDS_CONNECT:
2753         case SEC_CTX_INIT:
2754         case SEC_CTX_INIT_CONT:
2755         case SEC_CTX_FINI:
2756                 {
2757 #if 0
2758                         int rc;
2759
2760                         rc = mdt_handle_idmap(info);
2761                         if (rc)
2762                                 RETURN(rc);
2763                         else
2764 #endif
2765                                 RETURN(+1);
2766                 }
2767         }
2768
2769         if (unlikely(req->rq_export == NULL)) {
2770                 CERROR("operation %d on unconnected MDS from %s\n",
2771                        lustre_msg_get_opc(req->rq_reqmsg),
2772                        libcfs_id2str(req->rq_peer));
2773                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2774                  * mds_A will get -ENOTCONN(especially for ping req),
2775                  * which will cause that mds_A deactive timeout, then when
2776                  * mds_A cleanup, the cleanup process will be suspended since
2777                  * deactive timeout is not zero.
2778                  */
2779                 req->rq_status = -ENOTCONN;
2780                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2781                 RETURN(0);
2782         }
2783
2784         /* sanity check: if the xid matches, the request must be marked as a
2785          * resent or replayed */
2786         if (req_xid_is_last(req)) {
2787                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2788                       (MSG_RESENT | MSG_REPLAY))) {
2789                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2790                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2791                                   lustre_msg_get_flags(req->rq_reqmsg));
2792                         LBUG();
2793                         req->rq_status = -ENOTCONN;
2794                         RETURN(-ENOTCONN);
2795                 }
2796         }
2797
2798         /* else: note the opposite is not always true; a RESENT req after a
2799          * failover will usually not match the last_xid, since it was likely
2800          * never committed. A REPLAYed request will almost never match the
2801          * last xid, however it could for a committed, but still retained,
2802          * open. */
2803
2804         obd = req->rq_export->exp_obd;
2805
2806         /* Check for aborted recovery... */
2807         spin_lock_bh(&obd->obd_processing_task_lock);
2808         recovering = obd->obd_recovering;
2809         spin_unlock_bh(&obd->obd_processing_task_lock);
2810         if (unlikely(recovering)) {
2811                 int rc;
2812                 int should_process;
2813                 DEBUG_REQ(D_INFO, req, "Got new replay");
2814                 rc = mdt_filter_recovery_request(req, obd, &should_process);
2815                 if (rc != 0 || !should_process)
2816                         RETURN(rc);
2817                 else if (should_process < 0) {
2818                         req->rq_status = should_process;
2819                         rc = ptlrpc_error(req);
2820                         RETURN(rc);
2821                 }
2822         }
2823         RETURN(+1);
2824 }
2825
2826 static int mdt_msg_check_version(struct lustre_msg *msg)
2827 {
2828         int rc;
2829
2830         switch (lustre_msg_get_opc(msg)) {
2831         case MDS_CONNECT:
2832         case MDS_DISCONNECT:
2833         case OBD_PING:
2834         case SEC_CTX_INIT:
2835         case SEC_CTX_INIT_CONT:
2836         case SEC_CTX_FINI:
2837                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2838                 if (rc)
2839                         CERROR("bad opc %u version %08x, expecting %08x\n",
2840                                lustre_msg_get_opc(msg),
2841                                lustre_msg_get_version(msg),
2842                                LUSTRE_OBD_VERSION);
2843                 break;
2844         case MDS_GETSTATUS:
2845         case MDS_GETATTR:
2846         case MDS_GETATTR_NAME:
2847         case MDS_STATFS:
2848         case MDS_READPAGE:
2849         case MDS_WRITEPAGE:
2850         case MDS_IS_SUBDIR:
2851         case MDS_REINT:
2852         case MDS_CLOSE:
2853         case MDS_DONE_WRITING:
2854         case MDS_PIN:
2855         case MDS_SYNC:
2856         case MDS_GETXATTR:
2857         case MDS_SETXATTR:
2858         case MDS_SET_INFO:
2859         case MDS_GET_INFO:
2860         case MDS_QUOTACHECK:
2861         case MDS_QUOTACTL:
2862         case QUOTA_DQACQ:
2863         case QUOTA_DQREL:
2864         case SEQ_QUERY:
2865         case FLD_QUERY:
2866                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2867                 if (rc)
2868                         CERROR("bad opc %u version %08x, expecting %08x\n",
2869                                lustre_msg_get_opc(msg),
2870                                lustre_msg_get_version(msg),
2871                                LUSTRE_MDS_VERSION);
2872                 break;
2873         case LDLM_ENQUEUE:
2874         case LDLM_CONVERT:
2875         case LDLM_BL_CALLBACK:
2876         case LDLM_CP_CALLBACK:
2877                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2878                 if (rc)
2879                         CERROR("bad opc %u version %08x, expecting %08x\n",
2880                                lustre_msg_get_opc(msg),
2881                                lustre_msg_get_version(msg),
2882                                LUSTRE_DLM_VERSION);
2883                 break;
2884         case OBD_LOG_CANCEL:
2885         case LLOG_ORIGIN_HANDLE_CREATE:
2886         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2887         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2888         case LLOG_ORIGIN_HANDLE_CLOSE:
2889         case LLOG_ORIGIN_HANDLE_DESTROY:
2890         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2891         case LLOG_CATINFO:
2892                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2893                 if (rc)
2894                         CERROR("bad opc %u version %08x, expecting %08x\n",
2895                                lustre_msg_get_opc(msg),
2896                                lustre_msg_get_version(msg),
2897                                LUSTRE_LOG_VERSION);
2898                 break;
2899         default:
2900                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2901                 rc = -ENOTSUPP;
2902         }
2903         return rc;
2904 }
2905
2906 static int mdt_handle0(struct ptlrpc_request *req,
2907                        struct mdt_thread_info *info,
2908                        struct mdt_opc_slice *supported)
2909 {
2910         struct mdt_handler *h;
2911         struct lustre_msg  *msg;
2912         int                 rc;
2913
2914         ENTRY;
2915
2916         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2917                 RETURN(0);
2918
2919         LASSERT(current->journal_info == NULL);
2920
2921         msg = req->rq_reqmsg;
2922         rc = mdt_msg_check_version(msg);
2923         if (likely(rc == 0)) {
2924                 rc = mdt_recovery(info);
2925                 if (likely(rc == +1)) {
2926                         switch (lustre_msg_get_opc(msg)) {
2927                         case MDS_READPAGE:
2928                                 req->rq_bulk_read = 1;
2929                                 break;
2930                         case MDS_WRITEPAGE:
2931                                 req->rq_bulk_write = 1;
2932                                 break;
2933                         }
2934
2935                         h = mdt_handler_find(lustre_msg_get_opc(msg),
2936                                              supported);
2937                         if (likely(h != NULL)) {
2938                                 rc = mdt_req_handle(info, h, req);
2939                         } else {
2940                                 CERROR("The unsupported opc: 0x%x\n",
2941                                        lustre_msg_get_opc(msg) );
2942                                 req->rq_status = -ENOTSUPP;
2943                                 rc = ptlrpc_error(req);
2944                                 RETURN(rc);
2945                         }
2946                 }
2947         } else
2948                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2949         RETURN(rc);
2950 }
2951
2952 /*
2953  * MDT handler function called by ptlrpc service thread when request comes.
2954  *
2955  * XXX common "target" functionality should be factored into separate module
2956  * shared by mdt, ost and stand-alone services like fld.
2957  */
2958 static int mdt_handle_common(struct ptlrpc_request *req,
2959                              struct mdt_opc_slice *supported)
2960 {
2961         struct lu_env          *env;
2962         struct mdt_thread_info *info;
2963         int                     rc;
2964         ENTRY;
2965
2966         env = req->rq_svc_thread->t_env;
2967         LASSERT(env != NULL);
2968         LASSERT(env->le_ses != NULL);
2969         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2970         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2971         LASSERT(info != NULL);
2972
2973         mdt_thread_info_init(req, info);
2974
2975         rc = mdt_handle0(req, info, supported);
2976
2977         mdt_thread_info_fini(info);
2978         RETURN(rc);
2979 }
2980
2981 /*
2982  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2983  * as well.
2984  */
2985 int mdt_recovery_handle(struct ptlrpc_request *req)
2986 {
2987         int rc;
2988         ENTRY;
2989
2990         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2991         case FLD_QUERY:
2992                 rc = mdt_handle_common(req, mdt_fld_handlers);
2993                 break;
2994         case SEQ_QUERY:
2995                 rc = mdt_handle_common(req, mdt_seq_handlers);
2996                 break;
2997         default:
2998                 rc = mdt_handle_common(req, mdt_regular_handlers);
2999                 break;
3000         }
3001
3002         RETURN(rc);
3003 }
3004
3005 static int mdt_regular_handle(struct ptlrpc_request *req)
3006 {
3007         return mdt_handle_common(req, mdt_regular_handlers);
3008 }
3009
3010 static int mdt_readpage_handle(struct ptlrpc_request *req)
3011 {
3012         return mdt_handle_common(req, mdt_readpage_handlers);
3013 }
3014
3015 static int mdt_xmds_handle(struct ptlrpc_request *req)
3016 {
3017         return mdt_handle_common(req, mdt_xmds_handlers);
3018 }
3019
3020 static int mdt_mdsc_handle(struct ptlrpc_request *req)
3021 {
3022         return mdt_handle_common(req, mdt_seq_handlers);
3023 }
3024
3025 static int mdt_mdss_handle(struct ptlrpc_request *req)
3026 {
3027         return mdt_handle_common(req, mdt_seq_handlers);
3028 }
3029
3030 static int mdt_dtss_handle(struct ptlrpc_request *req)
3031 {
3032         return mdt_handle_common(req, mdt_seq_handlers);
3033 }
3034
3035 static int mdt_fld_handle(struct ptlrpc_request *req)
3036 {
3037         return mdt_handle_common(req, mdt_fld_handlers);
3038 }
3039
3040 enum mdt_it_code {
3041         MDT_IT_OPEN,
3042         MDT_IT_OCREAT,
3043         MDT_IT_CREATE,
3044         MDT_IT_GETATTR,
3045         MDT_IT_READDIR,
3046         MDT_IT_LOOKUP,
3047         MDT_IT_UNLINK,
3048         MDT_IT_TRUNC,
3049         MDT_IT_GETXATTR,
3050         MDT_IT_NR
3051 };
3052
3053 static int mdt_intent_getattr(enum mdt_it_code opcode,
3054                               struct mdt_thread_info *info,
3055                               struct ldlm_lock **,
3056                               int);
3057 static int mdt_intent_reint(enum mdt_it_code opcode,
3058                             struct mdt_thread_info *info,
3059                             struct ldlm_lock **,
3060                             int);
3061
3062 static struct mdt_it_flavor {
3063         const struct req_format *it_fmt;
3064         __u32                    it_flags;
3065         int                    (*it_act)(enum mdt_it_code ,
3066                                          struct mdt_thread_info *,
3067                                          struct ldlm_lock **,
3068                                          int);
3069         long                     it_reint;
3070 } mdt_it_flavor[] = {
3071         [MDT_IT_OPEN]     = {
3072                 .it_fmt   = &RQF_LDLM_INTENT,
3073                 /*.it_flags = HABEO_REFERO,*/
3074                 .it_flags = 0,
3075                 .it_act   = mdt_intent_reint,
3076                 .it_reint = REINT_OPEN
3077         },
3078         [MDT_IT_OCREAT]   = {
3079                 .it_fmt   = &RQF_LDLM_INTENT,
3080                 .it_flags = MUTABOR,
3081                 .it_act   = mdt_intent_reint,
3082                 .it_reint = REINT_OPEN
3083         },
3084         [MDT_IT_CREATE]   = {
3085                 .it_fmt   = &RQF_LDLM_INTENT,
3086                 .it_flags = MUTABOR,
3087                 .it_act   = mdt_intent_reint,
3088                 .it_reint = REINT_CREATE
3089         },
3090         [MDT_IT_GETATTR]  = {
3091                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3092                 .it_flags = HABEO_REFERO,
3093                 .it_act   = mdt_intent_getattr
3094         },
3095         [MDT_IT_READDIR]  = {
3096                 .it_fmt   = NULL,
3097                 .it_flags = 0,
3098                 .it_act   = NULL
3099         },
3100         [MDT_IT_LOOKUP]   = {
3101                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3102                 .it_flags = HABEO_REFERO,
3103                 .it_act   = mdt_intent_getattr
3104         },
3105         [MDT_IT_UNLINK]   = {
3106                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3107                 .it_flags = MUTABOR,
3108                 .it_act   = NULL,
3109                 .it_reint = REINT_UNLINK
3110         },
3111         [MDT_IT_TRUNC]    = {
3112                 .it_fmt   = NULL,
3113                 .it_flags = MUTABOR,
3114                 .it_act   = NULL
3115         },
3116         [MDT_IT_GETXATTR] = {
3117                 .it_fmt   = NULL,
3118                 .it_flags = 0,
3119                 .it_act   = NULL
3120         }
3121 };
3122
3123 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3124                             struct ldlm_lock **lockp,
3125                             struct ldlm_lock *new_lock,
3126                             struct mdt_lock_handle *lh,
3127                             int flags)
3128 {
3129         struct ptlrpc_request  *req = mdt_info_req(info);
3130         struct ldlm_lock       *lock = *lockp;
3131
3132         /*
3133          * Get new lock only for cases when possible resent did not find any
3134          * lock.
3135          */
3136         if (new_lock == NULL)
3137                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3138
3139         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3140                 lh->mlh_reg_lh.cookie = 0;
3141                 RETURN(0);
3142         }
3143
3144         LASSERTF(new_lock != NULL,
3145                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3146
3147         /*
3148          * If we've already given this lock to a client once, then we should
3149          * have no readers or writers.  Otherwise, we should have one reader
3150          * _or_ writer ref (which will be zeroed below) before returning the
3151          * lock to a client.
3152          */
3153         if (new_lock->l_export == req->rq_export) {
3154                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3155         } else {
3156                 LASSERT(new_lock->l_export == NULL);
3157                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3158         }
3159
3160         *lockp = new_lock;
3161
3162         if (new_lock->l_export == req->rq_export) {
3163                 /*
3164                  * Already gave this to the client, which means that we
3165                  * reconstructed a reply.
3166                  */
3167                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3168                         MSG_RESENT);
3169                 lh->mlh_reg_lh.cookie = 0;
3170                 RETURN(ELDLM_LOCK_REPLACED);
3171         }
3172
3173         /*
3174          * Fixup the lock to be given to the client.
3175          */
3176         lock_res_and_lock(new_lock);
3177         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3178          * possible blocking AST. */
3179         while (new_lock->l_readers > 0) {
3180                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3181                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3182                 new_lock->l_readers--;
3183         }
3184         while (new_lock->l_writers > 0) {
3185                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3186                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3187                 new_lock->l_writers--;
3188         }
3189
3190         new_lock->l_export = class_export_get(req->rq_export);
3191         new_lock->l_blocking_ast = lock->l_blocking_ast;
3192         new_lock->l_completion_ast = lock->l_completion_ast;
3193         new_lock->l_remote_handle = lock->l_remote_handle;
3194         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3195
3196         unlock_res_and_lock(new_lock);
3197
3198         lustre_hash_add(new_lock->l_export->exp_lock_hash,
3199                         &new_lock->l_remote_handle,
3200                         &new_lock->l_exp_hash);
3201
3202         LDLM_LOCK_RELEASE(new_lock);
3203         lh->mlh_reg_lh.cookie = 0;
3204
3205         RETURN(ELDLM_LOCK_REPLACED);
3206 }
3207
3208 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3209                                     struct ldlm_lock *new_lock,
3210                                     struct ldlm_lock **old_lock,
3211                                     struct mdt_lock_handle *lh)
3212 {
3213         struct ptlrpc_request  *req = mdt_info_req(info);
3214         struct obd_export      *exp = req->rq_export;
3215         struct lustre_handle    remote_hdl;
3216         struct ldlm_request    *dlmreq;
3217         struct ldlm_lock       *lock;
3218
3219         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3220                 return;
3221
3222         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3223         remote_hdl = dlmreq->lock_handle[0];
3224
3225         lock = lustre_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3226         if (lock) {
3227                 if (lock != new_lock) {
3228                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3229                         lh->mlh_reg_mode = lock->l_granted_mode;
3230
3231                         LDLM_DEBUG(lock, "Restoring lock cookie");
3232                         DEBUG_REQ(D_DLMTRACE, req,
3233                                   "restoring lock cookie "LPX64,
3234                                   lh->mlh_reg_lh.cookie);
3235                         if (old_lock)
3236                                 *old_lock = LDLM_LOCK_GET(lock);
3237                         lh_put(exp->exp_lock_hash, &lock->l_exp_hash);
3238                         return;
3239                 }
3240
3241                 lh_put(exp->exp_lock_hash, &lock->l_exp_hash);
3242         }
3243
3244         /*
3245          * If the xid matches, then we know this is a resent request, and allow
3246          * it. (It's probably an OPEN, for which we don't send a lock.
3247          */
3248         if (req_xid_is_last(req))
3249                 return;
3250
3251         /*
3252          * This remote handle isn't enqueued, so we never received or processed
3253          * this request.  Clear MSG_RESENT, because it can be handled like any
3254          * normal request now.
3255          */
3256         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3257
3258         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3259                   remote_hdl.cookie);
3260 }
3261
3262 static int mdt_intent_getattr(enum mdt_it_code opcode,
3263                               struct mdt_thread_info *info,
3264                               struct ldlm_lock **lockp,
3265                               int flags)
3266 {
3267         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3268         struct ldlm_lock       *new_lock = NULL;
3269         __u64                   child_bits;
3270         struct ldlm_reply      *ldlm_rep;
3271         struct ptlrpc_request  *req;
3272         struct mdt_body        *reqbody;
3273         struct mdt_body        *repbody;
3274         int                     rc;
3275         ENTRY;
3276
3277         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3278         LASSERT(reqbody);
3279
3280         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3281         LASSERT(repbody);
3282
3283         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
3284         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3285         repbody->eadatasize = 0;
3286         repbody->aclsize = 0;
3287
3288         switch (opcode) {
3289         case MDT_IT_LOOKUP:
3290                 child_bits = MDS_INODELOCK_LOOKUP;
3291                 break;
3292         case MDT_IT_GETATTR:
3293                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3294                 break;
3295         default:
3296                 CERROR("Unhandled till now");
3297                 GOTO(out_shrink, rc = -EINVAL);
3298         }
3299
3300         rc = mdt_init_ucred(info, reqbody);
3301         if (rc)
3302                 GOTO(out_shrink, rc);
3303
3304         req = info->mti_pill->rc_req;
3305         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3306         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3307
3308         /* Get lock from request for possible resent case. */
3309         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3310
3311         ldlm_rep->lock_policy_res2 =
3312                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3313
3314         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3315                 ldlm_rep->lock_policy_res2 = 0;
3316         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3317             ldlm_rep->lock_policy_res2) {
3318                 lhc->mlh_reg_lh.cookie = 0ull;
3319                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3320         }
3321
3322         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3323         EXIT;
3324 out_ucred:
3325         mdt_exit_ucred(info);
3326 out_shrink:
3327         mdt_shrink_reply(info);
3328         return rc;
3329 }
3330
3331 static int mdt_intent_reint(enum mdt_it_code opcode,
3332                             struct mdt_thread_info *info,
3333                             struct ldlm_lock **lockp,
3334                             int flags)
3335 {
3336         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3337         struct ldlm_reply      *rep = NULL;
3338         long                    opc;
3339         int                     rc;
3340
3341         static const struct req_format *intent_fmts[REINT_MAX] = {
3342                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3343                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3344         };
3345
3346         ENTRY;
3347
3348         opc = mdt_reint_opcode(info, intent_fmts);
3349         if (opc < 0)
3350                 RETURN(opc);
3351
3352         if (mdt_it_flavor[opcode].it_reint != opc) {
3353                 CERROR("Reint code %ld doesn't match intent: %d\n",
3354                        opc, opcode);
3355                 RETURN(err_serious(-EPROTO));
3356         }
3357
3358         /* Get lock from request for possible resent case. */
3359         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3360
3361         rc = mdt_reint_internal(info, lhc, opc);
3362
3363         /* Check whether the reply has been packed successfully. */
3364         if (mdt_info_req(info)->rq_repmsg != NULL)
3365                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3366         if (rep == NULL)
3367                 RETURN(err_serious(-EFAULT));
3368
3369         /* MDC expects this in any case */
3370         if (rc != 0)
3371                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3372
3373         /* Cross-ref case, the lock should be returned to the client */
3374         if (rc == -EREMOTE) {
3375                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3376                 rep->lock_policy_res2 = 0;
3377                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3378                 RETURN(rc);
3379         }
3380         rep->lock_policy_res2 = clear_serious(rc);
3381
3382         if (rc == -ENOTCONN || rc == -ENODEV ||
3383             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3384                 /*
3385                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3386                  * will be returned by rq_status, and client at ptlrpc layer
3387                  * will detect this, then disconnect, reconnect the import
3388                  * immediately, instead of impacting the following the rpc.
3389                  */
3390                 lhc->mlh_reg_lh.cookie = 0ull;
3391                 RETURN(rc);
3392         } else {
3393                 /*
3394                  * For other cases, the error will be returned by intent.
3395                  * and client will retrieve the result from intent.
3396                  */
3397                  /*
3398                   * FIXME: when open lock is finished, that should be
3399                   * checked here.
3400                   */
3401                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3402                         rep->lock_policy_res2 = 0;
3403                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3404                         RETURN(rc);
3405                 } else {
3406                         lhc->mlh_reg_lh.cookie = 0ull;
3407                         RETURN(ELDLM_LOCK_ABORTED);
3408                 }
3409         }
3410 }
3411
3412 static int mdt_intent_code(long itcode)
3413 {
3414         int rc;
3415
3416         switch(itcode) {
3417         case IT_OPEN:
3418                 rc = MDT_IT_OPEN;
3419                 break;
3420         case IT_OPEN|IT_CREAT:
3421                 rc = MDT_IT_OCREAT;
3422                 break;
3423         case IT_CREAT:
3424                 rc = MDT_IT_CREATE;
3425                 break;
3426         case IT_READDIR:
3427                 rc = MDT_IT_READDIR;
3428                 break;
3429         case IT_GETATTR:
3430                 rc = MDT_IT_GETATTR;
3431                 break;
3432         case IT_LOOKUP:
3433                 rc = MDT_IT_LOOKUP;
3434                 break;
3435         case IT_UNLINK:
3436                 rc = MDT_IT_UNLINK;
3437                 break;
3438         case IT_TRUNC:
3439                 rc = MDT_IT_TRUNC;
3440                 break;
3441         case IT_GETXATTR:
3442                 rc = MDT_IT_GETXATTR;
3443                 break;
3444         default:
3445                 CERROR("Unknown intent opcode: %ld\n", itcode);
3446                 rc = -EINVAL;
3447                 break;
3448         }
3449         return rc;
3450 }
3451
3452 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3453                           struct ldlm_lock **lockp, int flags)
3454 {
3455         struct req_capsule   *pill;
3456         struct mdt_it_flavor *flv;
3457         int opc;
3458         int rc;
3459         ENTRY;
3460
3461         opc = mdt_intent_code(itopc);
3462         if (opc < 0)
3463                 RETURN(-EINVAL);
3464
3465         pill = info->mti_pill;
3466         flv  = &mdt_it_flavor[opc];
3467
3468         if (flv->it_fmt != NULL)
3469                 req_capsule_extend(pill, flv->it_fmt);
3470
3471         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3472         if (rc == 0) {
3473                 struct ptlrpc_request *req = mdt_info_req(info);
3474                 if (flv->it_flags & MUTABOR &&
3475                     req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3476                         RETURN(-EROFS);
3477         }
3478         if (rc == 0 && flv->it_act != NULL) {
3479                 /* execute policy */
3480                 rc = flv->it_act(opc, info, lockp, flags);
3481         } else {
3482                 rc = -EOPNOTSUPP;
3483         }
3484         RETURN(rc);
3485 }
3486
3487 static int mdt_intent_policy(struct ldlm_namespace *ns,
3488                              struct ldlm_lock **lockp, void *req_cookie,
3489                              ldlm_mode_t mode, int flags, void *data)
3490 {
3491         struct mdt_thread_info *info;
3492         struct ptlrpc_request  *req  =  req_cookie;
3493         struct ldlm_intent     *it;
3494         struct req_capsule     *pill;
3495         int rc;
3496
3497         ENTRY;
3498
3499         LASSERT(req != NULL);
3500
3501         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3502                                   &mdt_thread_key);
3503         LASSERT(info != NULL);
3504         pill = info->mti_pill;
3505         LASSERT(pill->rc_req == req);
3506
3507         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3508                 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3509                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3510                 if (it != NULL) {
3511                         const struct ldlm_request *dlmreq;
3512                         __u64 req_bits;
3513
3514                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3515                         if (rc == 0)
3516                                 rc = ELDLM_OK;
3517
3518                         /*
3519                          * Lock without inodebits makes no sense and will oops
3520                          * later in ldlm. Let's check it now to see if we have
3521                          * wrong lock from client or bits get corrupted
3522                          * somewhere in mdt_intent_opc().
3523                          */
3524                         dlmreq = info->mti_dlm_req;
3525                         req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
3526                         LASSERT(req_bits != 0);
3527
3528                 } else
3529                         rc = err_serious(-EFAULT);
3530         } else {
3531                 /* No intent was provided */
3532                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3533                 rc = req_capsule_server_pack(pill);
3534                 if (rc)
3535                         rc = err_serious(rc);
3536         }
3537         RETURN(rc);
3538 }
3539
3540 /*
3541  * Seq wrappers
3542  */
3543 static void mdt_seq_adjust(const struct lu_env *env,
3544                           struct mdt_device *m, int lost)
3545 {
3546         struct md_site *ms = mdt_md_site(m);
3547         struct lu_seq_range out;
3548         ENTRY;
3549
3550         LASSERT(ms && ms->ms_server_seq);
3551         LASSERT(lost >= 0);
3552         /* get extra seq from seq_server, moving it's range up */
3553         while (lost-- > 0) {
3554                 seq_server_alloc_meta(ms->ms_server_seq, NULL, &out, env);
3555         }
3556         EXIT;
3557 }
3558
3559 static int mdt_seq_fini(const struct lu_env *env,
3560                         struct mdt_device *m)
3561 {
3562         struct md_site *ms = mdt_md_site(m);
3563         ENTRY;
3564
3565         if (ms != NULL) {
3566                 if (ms->ms_server_seq) {
3567                         seq_server_fini(ms->ms_server_seq, env);
3568                         OBD_FREE_PTR(ms->ms_server_seq);
3569                         ms->ms_server_seq = NULL;
3570         }
3571
3572                 if (ms->ms_control_seq) {
3573                         seq_server_fini(ms->ms_control_seq, env);
3574                         OBD_FREE_PTR(ms->ms_control_seq);
3575                         ms->ms_control_seq = NULL;
3576         }
3577
3578                 if (ms->ms_client_seq) {
3579                         seq_client_fini(ms->ms_client_seq);
3580                         OBD_FREE_PTR(ms->ms_client_seq);
3581                         ms->ms_client_seq = NULL;
3582                 }
3583         }
3584
3585         RETURN(0);
3586 }
3587
3588 static int mdt_seq_init(const struct lu_env *env,
3589                         const char *uuid,
3590                         struct mdt_device *m)
3591 {
3592         struct md_site *ms;
3593         char *prefix;
3594         int rc;
3595         ENTRY;
3596
3597         ms = mdt_md_site(m);
3598
3599         /*
3600          * This is sequence-controller node. Init seq-controller server on local
3601          * MDT.
3602          */
3603         if (ms->ms_node_id == 0) {
3604                 LASSERT(ms->ms_control_seq == NULL);
3605
3606                 OBD_ALLOC_PTR(ms->ms_control_seq);
3607                 if (ms->ms_control_seq == NULL)
3608                         RETURN(-ENOMEM);
3609
3610                 rc = seq_server_init(ms->ms_control_seq,
3611                                      m->mdt_bottom, uuid,
3612                                      LUSTRE_SEQ_CONTROLLER,
3613                                      ms,
3614                                      env);
3615
3616                 if (rc)
3617                         GOTO(out_seq_fini, rc);
3618
3619                 OBD_ALLOC_PTR(ms->ms_client_seq);
3620                 if (ms->ms_client_seq == NULL)
3621                         GOTO(out_seq_fini, rc = -ENOMEM);
3622
3623                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3624                 if (prefix == NULL) {
3625                         OBD_FREE_PTR(ms->ms_client_seq);
3626                         GOTO(out_seq_fini, rc = -ENOMEM);
3627                 }
3628
3629                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3630                          uuid);
3631
3632                 /*
3633                  * Init seq-controller client after seq-controller server is
3634                  * ready. Pass ms->ms_control_seq to it for direct talking.
3635                  */
3636                 rc = seq_client_init(ms->ms_client_seq, NULL,
3637                                      LUSTRE_SEQ_METADATA, prefix,
3638                                      ms->ms_control_seq);
3639                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3640
3641                 if (rc)
3642                         GOTO(out_seq_fini, rc);
3643         }
3644
3645         /* Init seq-server on local MDT */
3646         LASSERT(ms->ms_server_seq == NULL);
3647
3648         OBD_ALLOC_PTR(ms->ms_server_seq);
3649         if (ms->ms_server_seq == NULL)
3650                 GOTO(out_seq_fini, rc = -ENOMEM);
3651
3652         rc = seq_server_init(ms->ms_server_seq,
3653                              m->mdt_bottom, uuid,
3654                              LUSTRE_SEQ_SERVER,
3655                              ms,
3656                              env);
3657         if (rc)
3658                 GOTO(out_seq_fini, rc = -ENOMEM);
3659
3660         /* Assign seq-controller client to local seq-server. */
3661         if (ms->ms_node_id == 0) {
3662                 LASSERT(ms->ms_client_seq != NULL);
3663
3664                 rc = seq_server_set_cli(ms->ms_server_seq,
3665                                         ms->ms_client_seq,
3666                                         env);
3667         }
3668
3669         EXIT;
3670 out_seq_fini:
3671         if (rc)
3672                 mdt_seq_fini(env, m);
3673
3674         return rc;
3675 }
3676 /*
3677  * Init client sequence manager which is used by local MDS to talk to sequence
3678  * controller on remote node.
3679  */
3680 static int mdt_seq_init_cli(const struct lu_env *env,
3681                             struct mdt_device *m,
3682                             struct lustre_cfg *cfg)
3683 {
3684         struct md_site    *ms = mdt_md_site(m);
3685         struct obd_device *mdc;
3686         struct obd_uuid   *uuidp, *mdcuuidp;
3687         char              *uuid_str, *mdc_uuid_str;
3688         int                rc;
3689         int                index;
3690         struct mdt_thread_info *info;
3691         char *p, *index_string = lustre_cfg_string(cfg, 2);
3692         ENTRY;
3693
3694         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3695         uuidp = &info->mti_u.uuid[0];
3696         mdcuuidp = &info->mti_u.uuid[1];
3697
3698         LASSERT(index_string);
3699
3700         index = simple_strtol(index_string, &p, 10);
3701         if (*p) {
3702                 CERROR("Invalid index in lustre_cgf, offset 2\n");
3703                 RETURN(-EINVAL);
3704         }
3705
3706         /* check if this is adding the first MDC and controller is not yet
3707          * initialized. */
3708         if (index != 0 || ms->ms_client_seq)
3709                 RETURN(0);
3710
3711         uuid_str = lustre_cfg_string(cfg, 1);
3712         mdc_uuid_str = lustre_cfg_string(cfg, 4);
3713         obd_str2uuid(uuidp, uuid_str);
3714         obd_str2uuid(mdcuuidp, mdc_uuid_str);
3715
3716         mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3717         if (!mdc) {
3718                 CERROR("can't find controller MDC by uuid %s\n",
3719                        uuid_str);
3720                 rc = -ENOENT;
3721         } else if (!mdc->obd_set_up) {
3722                 CERROR("target %s not set up\n", mdc->obd_name);
3723                 rc = -EINVAL;
3724         } else {
3725                 LASSERT(ms->ms_control_exp);
3726                 OBD_ALLOC_PTR(ms->ms_client_seq);
3727                 if (ms->ms_client_seq != NULL) {
3728                         char *prefix;
3729
3730                         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3731                         if (!prefix)
3732                                 RETURN(-ENOMEM);
3733
3734                         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3735                                  mdc->obd_name);
3736
3737                         rc = seq_client_init(ms->ms_client_seq,
3738                                              ms->ms_control_exp,
3739                                              LUSTRE_SEQ_METADATA,
3740                                              prefix, NULL);
3741                         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3742                 } else
3743                         rc = -ENOMEM;
3744
3745                 if (rc)
3746                         RETURN(rc);
3747
3748                 LASSERT(ms->ms_server_seq != NULL);
3749                 rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
3750                                         env);
3751         }
3752
3753         RETURN(rc);
3754 }
3755
3756 static void mdt_seq_fini_cli(struct mdt_device *m)
3757 {
3758         struct md_site *ms;
3759
3760         ENTRY;
3761
3762         ms = mdt_md_site(m);
3763