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