Whamcloud - gitweb
LU-2901 ldlm: fix resource/fid check, use DLDLMRES
[fs/lustre-release.git] / lustre / mdt / mdt_handler.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2010, 2013, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * 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 #define DEBUG_SUBSYSTEM S_MDS
50
51 #include <linux/module.h>
52 /*
53  * struct OBD_{ALLOC,FREE}*()
54  */
55 #include <obd_support.h>
56 /* struct ptlrpc_request */
57 #include <lustre_net.h>
58 /* struct obd_export */
59 #include <lustre_export.h>
60 /* struct obd_device */
61 #include <obd.h>
62 /* lu2dt_dev() */
63 #include <dt_object.h>
64 #include <lustre_mds.h>
65 #include <lustre_mdt.h>
66 #include <lustre_log.h>
67 #include "mdt_internal.h"
68 #include <lustre_acl.h>
69 #include <lustre_param.h>
70 #include <lustre_quota.h>
71 #include <lustre_linkea.h>
72 #include <lustre_lfsck.h>
73
74 mdl_mode_t mdt_mdl_lock_modes[] = {
75         [LCK_MINMODE] = MDL_MINMODE,
76         [LCK_EX]      = MDL_EX,
77         [LCK_PW]      = MDL_PW,
78         [LCK_PR]      = MDL_PR,
79         [LCK_CW]      = MDL_CW,
80         [LCK_CR]      = MDL_CR,
81         [LCK_NL]      = MDL_NL,
82         [LCK_GROUP]   = MDL_GROUP
83 };
84
85 ldlm_mode_t mdt_dlm_lock_modes[] = {
86         [MDL_MINMODE] = LCK_MINMODE,
87         [MDL_EX]      = LCK_EX,
88         [MDL_PW]      = LCK_PW,
89         [MDL_PR]      = LCK_PR,
90         [MDL_CW]      = LCK_CW,
91         [MDL_CR]      = LCK_CR,
92         [MDL_NL]      = LCK_NL,
93         [MDL_GROUP]   = LCK_GROUP
94 };
95
96 static struct mdt_device *mdt_dev(struct lu_device *d);
97 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
98
99 static const struct lu_object_operations mdt_obj_ops;
100
101 /* Slab for MDT object allocation */
102 static struct kmem_cache *mdt_object_kmem;
103
104 static struct lu_kmem_descr mdt_caches[] = {
105         {
106                 .ckd_cache = &mdt_object_kmem,
107                 .ckd_name  = "mdt_obj",
108                 .ckd_size  = sizeof(struct mdt_object)
109         },
110         {
111                 .ckd_cache = NULL
112         }
113 };
114
115 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
116 {
117         if (!rep)
118                 return 0;
119         return (rep->lock_policy_res1 & flag);
120 }
121
122 void mdt_clear_disposition(struct mdt_thread_info *info,
123                            struct ldlm_reply *rep, int flag)
124 {
125         if (info)
126                 info->mti_opdata &= ~flag;
127         if (rep)
128                 rep->lock_policy_res1 &= ~flag;
129 }
130
131 void mdt_set_disposition(struct mdt_thread_info *info,
132                          struct ldlm_reply *rep, int flag)
133 {
134         if (info)
135                 info->mti_opdata |= flag;
136         if (rep)
137                 rep->lock_policy_res1 |= flag;
138 }
139
140 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
141 {
142         lh->mlh_pdo_hash = 0;
143         lh->mlh_reg_mode = lm;
144         lh->mlh_rreg_mode = lm;
145         lh->mlh_type = MDT_REG_LOCK;
146 }
147
148 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
149                        const char *name, int namelen)
150 {
151         lh->mlh_reg_mode = lm;
152         lh->mlh_rreg_mode = lm;
153         lh->mlh_type = MDT_PDO_LOCK;
154
155         if (name != NULL && (name[0] != '\0')) {
156                 LASSERT(namelen > 0);
157                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
158                 /* XXX Workaround for LU-2856
159                  * Zero is a valid return value of full_name_hash, but several
160                  * users of mlh_pdo_hash assume a non-zero hash value. We
161                  * therefore map zero onto an arbitrary, but consistent
162                  * value (1) to avoid problems further down the road. */
163                 if (unlikely(!lh->mlh_pdo_hash))
164                         lh->mlh_pdo_hash = 1;
165         } else {
166                 LASSERT(namelen == 0);
167                 lh->mlh_pdo_hash = 0ull;
168         }
169 }
170
171 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
172                               struct mdt_lock_handle *lh)
173 {
174         mdl_mode_t mode;
175         ENTRY;
176
177         /*
178          * Any dir access needs couple of locks:
179          *
180          * 1) on part of dir we gonna take lookup/modify;
181          *
182          * 2) on whole dir to protect it from concurrent splitting and/or to
183          * flush client's cache for readdir().
184          *
185          * so, for a given mode and object this routine decides what lock mode
186          * to use for lock #2:
187          *
188          * 1) if caller's gonna lookup in dir then we need to protect dir from
189          * being splitted only - LCK_CR
190          *
191          * 2) if caller's gonna modify dir then we need to protect dir from
192          * being splitted and to flush cache - LCK_CW
193          *
194          * 3) if caller's gonna modify dir and that dir seems ready for
195          * splitting then we need to protect it from any type of access
196          * (lookup/modify/split) - LCK_EX --bzzz
197          */
198
199         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
200         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
201
202         /*
203          * Ask underlaying level its opinion about preferable PDO lock mode
204          * having access type passed as regular lock mode:
205          *
206          * - MDL_MINMODE means that lower layer does not want to specify lock
207          * mode;
208          *
209          * - MDL_NL means that no PDO lock should be taken. This is used in some
210          * cases. Say, for non-splittable directories no need to use PDO locks
211          * at all.
212          */
213         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
214                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
215
216         if (mode != MDL_MINMODE) {
217                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
218         } else {
219                 /*
220                  * Lower layer does not want to specify locking mode. We do it
221                  * our selves. No special protection is needed, just flush
222                  * client's cache on modification and allow concurrent
223                  * mondification.
224                  */
225                 switch (lh->mlh_reg_mode) {
226                 case LCK_EX:
227                         lh->mlh_pdo_mode = LCK_EX;
228                         break;
229                 case LCK_PR:
230                         lh->mlh_pdo_mode = LCK_CR;
231                         break;
232                 case LCK_PW:
233                         lh->mlh_pdo_mode = LCK_CW;
234                         break;
235                 default:
236                         CERROR("Not expected lock type (0x%x)\n",
237                                (int)lh->mlh_reg_mode);
238                         LBUG();
239                 }
240         }
241
242         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
243         EXIT;
244 }
245
246 int mdt_getstatus(struct mdt_thread_info *info)
247 {
248         struct mdt_device       *mdt  = info->mti_mdt;
249         struct mdt_body         *repbody;
250         int                     rc;
251         ENTRY;
252
253         rc = mdt_check_ucred(info);
254         if (rc)
255                 RETURN(err_serious(rc));
256
257         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
258                 RETURN(err_serious(-ENOMEM));
259
260         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
261         repbody->fid1 = mdt->mdt_md_root_fid;
262         repbody->valid |= OBD_MD_FLID;
263
264         if (mdt->mdt_opts.mo_mds_capa &&
265             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
266                 struct mdt_object  *root;
267                 struct lustre_capa *capa;
268
269                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
270                 if (IS_ERR(root))
271                         RETURN(PTR_ERR(root));
272
273                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
274                 LASSERT(capa);
275                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
276                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
277                                  0);
278                 mdt_object_put(info->mti_env, root);
279                 if (rc == 0)
280                         repbody->valid |= OBD_MD_FLMDSCAPA;
281         }
282
283         RETURN(rc);
284 }
285
286 int mdt_statfs(struct mdt_thread_info *info)
287 {
288         struct ptlrpc_request           *req = mdt_info_req(info);
289         struct md_device                *next = info->mti_mdt->mdt_child;
290         struct ptlrpc_service_part      *svcpt;
291         struct obd_statfs               *osfs;
292         int                             rc;
293
294         ENTRY;
295
296         svcpt = info->mti_pill->rc_req->rq_rqbd->rqbd_svcpt;
297
298         /* This will trigger a watchdog timeout */
299         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
300                          (MDT_SERVICE_WATCHDOG_FACTOR *
301                           at_get(&svcpt->scp_at_estimate)) + 1);
302
303         rc = mdt_check_ucred(info);
304         if (rc)
305                 RETURN(err_serious(rc));
306
307         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
308                 RETURN(err_serious(-ENOMEM));
309
310         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
311         if (!osfs)
312                 RETURN(-EPROTO);
313
314         /** statfs information are cached in the mdt_device */
315         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
316                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
317                 /** statfs data is too old, get up-to-date one */
318                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
319                 if (rc)
320                         RETURN(rc);
321                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
322                 info->mti_mdt->mdt_osfs = *osfs;
323                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
324                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
325         } else {
326                 /** use cached statfs data */
327                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
328                 *osfs = info->mti_mdt->mdt_osfs;
329                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
330         }
331
332         if (rc == 0)
333                 mdt_counter_incr(req, LPROC_MDT_STATFS);
334
335         RETURN(rc);
336 }
337
338 /**
339  * Pack SOM attributes into the reply.
340  * Call under a DLM UPDATE lock.
341  */
342 static void mdt_pack_size2body(struct mdt_thread_info *info,
343                                struct mdt_object *mo)
344 {
345         struct mdt_body *b;
346         struct md_attr *ma = &info->mti_attr;
347
348         LASSERT(ma->ma_attr.la_valid & LA_MODE);
349         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
350
351         /* Check if Size-on-MDS is supported, if this is a regular file,
352          * if SOM is enabled on the object and if SOM cache exists and valid.
353          * Otherwise do not pack Size-on-MDS attributes to the reply. */
354         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
355             !S_ISREG(ma->ma_attr.la_mode) ||
356             !mdt_object_is_som_enabled(mo) ||
357             !(ma->ma_valid & MA_SOM))
358                 return;
359
360         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
361         b->size = ma->ma_som->msd_size;
362         b->blocks = ma->ma_som->msd_blocks;
363 }
364
365 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
366                         const struct lu_attr *attr, const struct lu_fid *fid)
367 {
368         struct md_attr *ma = &info->mti_attr;
369
370         LASSERT(ma->ma_valid & MA_INODE);
371
372         b->atime      = attr->la_atime;
373         b->mtime      = attr->la_mtime;
374         b->ctime      = attr->la_ctime;
375         b->mode       = attr->la_mode;
376         b->size       = attr->la_size;
377         b->blocks     = attr->la_blocks;
378         b->uid        = attr->la_uid;
379         b->gid        = attr->la_gid;
380         b->flags      = attr->la_flags;
381         b->nlink      = attr->la_nlink;
382         b->rdev       = attr->la_rdev;
383
384         /*XXX should pack the reply body according to lu_valid*/
385         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
386                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
387                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
388                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
389
390         if (!S_ISREG(attr->la_mode)) {
391                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
392         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
393                 /* means no objects are allocated on osts. */
394                 LASSERT(!(ma->ma_valid & MA_LOV));
395                 /* just ignore blocks occupied by extend attributes on MDS */
396                 b->blocks = 0;
397                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
398                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
399         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm &&
400                    (ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
401                 /* A released file stores its size on MDS. */
402                 b->blocks = 0;
403                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
404         }
405
406         if (fid) {
407                 b->fid1 = *fid;
408                 b->valid |= OBD_MD_FLID;
409                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
410                                 PFID(fid), b->nlink, b->mode, b->size);
411         }
412
413         if (info)
414                 mdt_body_reverse_idmap(info, b);
415
416         if (b->valid & OBD_MD_FLSIZE)
417                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
418                        PFID(fid), (unsigned long long)b->size);
419 }
420
421 static inline int mdt_body_has_lov(const struct lu_attr *la,
422                                    const struct mdt_body *body)
423 {
424         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
425                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
426 }
427
428 void mdt_client_compatibility(struct mdt_thread_info *info)
429 {
430         struct mdt_body       *body;
431         struct ptlrpc_request *req = mdt_info_req(info);
432         struct obd_export     *exp = req->rq_export;
433         struct md_attr        *ma = &info->mti_attr;
434         struct lu_attr        *la = &ma->ma_attr;
435         ENTRY;
436
437         if (exp_connect_layout(exp))
438                 /* the client can deal with 16-bit lmm_stripe_count */
439                 RETURN_EXIT;
440
441         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
442
443         if (!mdt_body_has_lov(la, body))
444                 RETURN_EXIT;
445
446         /* now we have a reply with a lov for a client not compatible with the
447          * layout lock so we have to clean the layout generation number */
448         if (S_ISREG(la->la_mode))
449                 ma->ma_lmm->lmm_layout_gen = 0;
450         EXIT;
451 }
452
453 static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
454                              char *name)
455 {
456         const struct lu_env *env = info->mti_env;
457         int rc;
458         ENTRY;
459
460         LASSERT(info->mti_big_lmm_used == 0);
461         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
462         if (rc < 0)
463                 RETURN(rc);
464
465         /* big_lmm may need to be grown */
466         if (info->mti_big_lmmsize < rc) {
467                 int size = size_roundup_power2(rc);
468
469                 if (info->mti_big_lmmsize > 0) {
470                         /* free old buffer */
471                         LASSERT(info->mti_big_lmm);
472                         OBD_FREE_LARGE(info->mti_big_lmm,
473                                        info->mti_big_lmmsize);
474                         info->mti_big_lmm = NULL;
475                         info->mti_big_lmmsize = 0;
476                 }
477
478                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
479                 if (info->mti_big_lmm == NULL)
480                         RETURN(-ENOMEM);
481                 info->mti_big_lmmsize = size;
482         }
483         LASSERT(info->mti_big_lmmsize >= rc);
484
485         info->mti_buf.lb_buf = info->mti_big_lmm;
486         info->mti_buf.lb_len = info->mti_big_lmmsize;
487         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
488
489         RETURN(rc);
490 }
491
492 int mdt_attr_get_lov(struct mdt_thread_info *info,
493                      struct mdt_object *o, struct md_attr *ma)
494 {
495         struct md_object *next = mdt_object_child(o);
496         struct lu_buf    *buf = &info->mti_buf;
497         int rc;
498
499         buf->lb_buf = ma->ma_lmm;
500         buf->lb_len = ma->ma_lmm_size;
501         rc = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_LOV);
502         if (rc > 0) {
503                 ma->ma_lmm_size = rc;
504                 ma->ma_valid |= MA_LOV;
505                 rc = 0;
506         } else if (rc == -ENODATA) {
507                 /* no LOV EA */
508                 rc = 0;
509         } else if (rc == -ERANGE) {
510                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LOV);
511                 if (rc > 0) {
512                         info->mti_big_lmm_used = 1;
513                         ma->ma_valid |= MA_LOV;
514                         ma->ma_lmm = info->mti_big_lmm;
515                         ma->ma_lmm_size = rc;
516                         /* update mdt_max_mdsize so all clients
517                          * will be aware about that */
518                         if (info->mti_mdt->mdt_max_mdsize < rc)
519                                 info->mti_mdt->mdt_max_mdsize = rc;
520                         rc = 0;
521                 }
522         }
523
524         return rc;
525 }
526
527 int mdt_attr_get_pfid(struct mdt_thread_info *info,
528                       struct mdt_object *o, struct lu_fid *pfid)
529 {
530         struct lu_buf           *buf = &info->mti_buf;
531         struct link_ea_header   *leh;
532         struct link_ea_entry    *lee;
533         int                      rc;
534         ENTRY;
535
536         buf->lb_buf = info->mti_big_lmm;
537         buf->lb_len = info->mti_big_lmmsize;
538         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
539                           buf, XATTR_NAME_LINK);
540         /* ignore errors, MA_PFID won't be set and it is
541          * up to the caller to treat this as an error */
542         if (rc == -ERANGE || buf->lb_len == 0) {
543                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
544                 buf->lb_buf = info->mti_big_lmm;
545                 buf->lb_len = info->mti_big_lmmsize;
546         }
547
548         if (rc < 0)
549                 RETURN(rc);
550         if (rc < sizeof(*leh)) {
551                 CERROR("short LinkEA on "DFID": rc = %d\n",
552                        PFID(mdt_object_fid(o)), rc);
553                 RETURN(-ENODATA);
554         }
555
556         leh = (struct link_ea_header *) buf->lb_buf;
557         lee = (struct link_ea_entry *)(leh + 1);
558         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
559                 leh->leh_magic = LINK_EA_MAGIC;
560                 leh->leh_reccount = __swab32(leh->leh_reccount);
561                 leh->leh_len = __swab64(leh->leh_len);
562         }
563         if (leh->leh_magic != LINK_EA_MAGIC)
564                 RETURN(-EINVAL);
565         if (leh->leh_reccount == 0)
566                 RETURN(-ENODATA);
567
568         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
569         fid_be_to_cpu(pfid, pfid);
570
571         RETURN(0);
572 }
573
574 int mdt_attr_get_complex(struct mdt_thread_info *info,
575                          struct mdt_object *o, struct md_attr *ma)
576 {
577         const struct lu_env *env = info->mti_env;
578         struct md_object    *next = mdt_object_child(o);
579         struct lu_buf       *buf = &info->mti_buf;
580         u32                  mode = lu_object_attr(&next->mo_lu);
581         int                  need = ma->ma_need;
582         int                  rc = 0, rc2;
583         ENTRY;
584
585         ma->ma_valid = 0;
586
587         if (need & MA_INODE) {
588                 ma->ma_need = MA_INODE;
589                 rc = mo_attr_get(env, next, ma);
590                 if (rc)
591                         GOTO(out, rc);
592                 ma->ma_valid |= MA_INODE;
593         }
594
595         if (need & MA_PFID) {
596                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
597                 if (rc == 0)
598                         ma->ma_valid |= MA_PFID;
599                 /* ignore this error, parent fid is not mandatory */
600                 rc = 0;
601         }
602
603         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
604                 rc = mdt_attr_get_lov(info, o, ma);
605                 if (rc)
606                         GOTO(out, rc);
607         }
608
609         if (need & MA_LMV && S_ISDIR(mode)) {
610                 buf->lb_buf = ma->ma_lmv;
611                 buf->lb_len = ma->ma_lmv_size;
612                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LMV);
613                 if (rc2 > 0) {
614                         ma->ma_lmv_size = rc2;
615                         ma->ma_valid |= MA_LMV;
616                 } else if (rc2 == -ENODATA) {
617                         /* no LMV EA */
618                         ma->ma_lmv_size = 0;
619                 } else
620                         GOTO(out, rc = rc2);
621         }
622
623         if (need & MA_SOM && S_ISREG(mode)) {
624                 buf->lb_buf = info->mti_xattr_buf;
625                 buf->lb_len = sizeof(info->mti_xattr_buf);
626                 CLASSERT(sizeof(struct som_attrs) <=
627                          sizeof(info->mti_xattr_buf));
628                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
629                 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
630                 if (rc2 == 0)
631                         ma->ma_valid |= MA_SOM;
632                 else if (rc2 < 0 && rc2 != -ENODATA)
633                         GOTO(out, rc = rc2);
634         }
635
636         if (need & MA_HSM && S_ISREG(mode)) {
637                 buf->lb_buf = info->mti_xattr_buf;
638                 buf->lb_len = sizeof(info->mti_xattr_buf);
639                 CLASSERT(sizeof(struct hsm_attrs) <=
640                          sizeof(info->mti_xattr_buf));
641                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
642                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
643                 if (rc2 == 0)
644                         ma->ma_valid |= MA_HSM;
645                 else if (rc2 < 0 && rc2 != -ENODATA)
646                         GOTO(out, rc = rc2);
647         }
648
649 #ifdef CONFIG_FS_POSIX_ACL
650         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
651                 buf->lb_buf = ma->ma_acl;
652                 buf->lb_len = ma->ma_acl_size;
653                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
654                 if (rc2 > 0) {
655                         ma->ma_acl_size = rc2;
656                         ma->ma_valid |= MA_ACL_DEF;
657                 } else if (rc2 == -ENODATA) {
658                         /* no ACLs */
659                         ma->ma_acl_size = 0;
660                 } else
661                         GOTO(out, rc = rc2);
662         }
663 #endif
664 out:
665         ma->ma_need = need;
666         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
667                rc, ma->ma_valid, ma->ma_lmm);
668         RETURN(rc);
669 }
670
671 static int mdt_getattr_internal(struct mdt_thread_info *info,
672                                 struct mdt_object *o, int ma_need)
673 {
674         struct md_object        *next = mdt_object_child(o);
675         const struct mdt_body   *reqbody = info->mti_body;
676         struct ptlrpc_request   *req = mdt_info_req(info);
677         struct md_attr          *ma = &info->mti_attr;
678         struct lu_attr          *la = &ma->ma_attr;
679         struct req_capsule      *pill = info->mti_pill;
680         const struct lu_env     *env = info->mti_env;
681         struct mdt_body         *repbody;
682         struct lu_buf           *buffer = &info->mti_buf;
683         int                     rc;
684         int                     is_root;
685         ENTRY;
686
687         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
688                 RETURN(err_serious(-ENOMEM));
689
690         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
691
692         ma->ma_valid = 0;
693
694         if (mdt_object_remote(o)) {
695                 /* This object is located on remote node.*/
696                 /* Return -EIO for old client */
697                 if (!mdt_is_dne_client(req->rq_export))
698                         GOTO(out, rc = -EIO);
699
700                 repbody->fid1 = *mdt_object_fid(o);
701                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
702                 GOTO(out, rc = 0);
703         }
704
705         buffer->lb_len = reqbody->eadatasize;
706         if (buffer->lb_len > 0)
707                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
708         else
709                 buffer->lb_buf = NULL;
710
711         /* If it is dir object and client require MEA, then we got MEA */
712         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
713             reqbody->valid & OBD_MD_MEA) {
714                 /* Assumption: MDT_MD size is enough for lmv size. */
715                 ma->ma_lmv = buffer->lb_buf;
716                 ma->ma_lmv_size = buffer->lb_len;
717                 ma->ma_need = MA_LMV | MA_INODE;
718         } else {
719                 ma->ma_lmm = buffer->lb_buf;
720                 ma->ma_lmm_size = buffer->lb_len;
721                 ma->ma_need = MA_LOV | MA_INODE | MA_HSM;
722         }
723
724         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
725             reqbody->valid & OBD_MD_FLDIREA  &&
726             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
727                 /* get default stripe info for this dir. */
728                 ma->ma_need |= MA_LOV_DEF;
729         }
730         ma->ma_need |= ma_need;
731         if (ma->ma_need & MA_SOM)
732                 ma->ma_som = &info->mti_u.som.data;
733
734         rc = mdt_attr_get_complex(info, o, ma);
735         if (unlikely(rc)) {
736                 CERROR("%s: getattr error for "DFID": rc = %d\n",
737                        mdt_obd_name(info->mti_mdt),
738                        PFID(mdt_object_fid(o)), rc);
739                 RETURN(rc);
740         }
741
742         /* if file is released, check if a restore is running */
743         if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
744             mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
745                 repbody->t_state = MS_RESTORE;
746                 repbody->valid |= OBD_MD_TSTATE;
747         }
748
749         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
750
751         /* the Lustre protocol supposes to return default striping
752          * on the user-visible root if explicitly requested */
753         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
754             (ma->ma_need & MA_LOV_DEF && is_root) && (ma->ma_need & MA_LOV)) {
755                 struct lu_fid      rootfid;
756                 struct mdt_object *root;
757                 struct mdt_device *mdt = info->mti_mdt;
758
759                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
760                 if (rc)
761                         RETURN(rc);
762                 root = mdt_object_find(env, mdt, &rootfid);
763                 if (IS_ERR(root))
764                         RETURN(PTR_ERR(root));
765                 rc = mdt_attr_get_lov(info, root, ma);
766                 mdt_object_put(info->mti_env, root);
767                 if (unlikely(rc)) {
768                         CERROR("%s: getattr error for "DFID": rc = %d\n",
769                                mdt_obd_name(info->mti_mdt),
770                                PFID(mdt_object_fid(o)), rc);
771                         RETURN(rc);
772                 }
773         }
774
775         if (likely(ma->ma_valid & MA_INODE))
776                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
777         else
778                 RETURN(-EFAULT);
779
780         if (mdt_body_has_lov(la, reqbody)) {
781                 if (ma->ma_valid & MA_LOV) {
782                         LASSERT(ma->ma_lmm_size);
783                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
784                         repbody->eadatasize = ma->ma_lmm_size;
785                         if (S_ISDIR(la->la_mode))
786                                 repbody->valid |= OBD_MD_FLDIREA;
787                         else
788                                 repbody->valid |= OBD_MD_FLEASIZE;
789                 }
790                 if (ma->ma_valid & MA_LMV) {
791                         LASSERT(S_ISDIR(la->la_mode));
792                         repbody->eadatasize = ma->ma_lmv_size;
793                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
794                 }
795         } else if (S_ISLNK(la->la_mode) &&
796                    reqbody->valid & OBD_MD_LINKNAME) {
797                 buffer->lb_buf = ma->ma_lmm;
798                 /* eadatasize from client includes NULL-terminator, so
799                  * there is no need to read it */
800                 buffer->lb_len = reqbody->eadatasize - 1;
801                 rc = mo_readlink(env, next, buffer);
802                 if (unlikely(rc <= 0)) {
803                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
804                                mdt_obd_name(info->mti_mdt),
805                                PFID(mdt_object_fid(o)), rc);
806                         rc = -EFAULT;
807                 } else {
808                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
809
810                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
811                                 rc -= 2;
812                         repbody->valid |= OBD_MD_LINKNAME;
813                         /* we need to report back size with NULL-terminator
814                          * because client expects that */
815                         repbody->eadatasize = rc + 1;
816                         if (repbody->eadatasize != reqbody->eadatasize)
817                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
818                                        "on "DFID ", expected %d\n",
819                                        mdt_obd_name(info->mti_mdt),
820                                        rc, PFID(mdt_object_fid(o)),
821                                        reqbody->eadatasize - 1);
822                         /* NULL terminate */
823                         ((char *)ma->ma_lmm)[rc] = 0;
824
825                         /* If the total CDEBUG() size is larger than a page, it
826                          * will print a warning to the console, avoid this by
827                          * printing just the last part of the symlink. */
828                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
829                                print_limit < rc ? "..." : "", print_limit,
830                                (char *)ma->ma_lmm + rc - print_limit, rc);
831                         rc = 0;
832                 }
833         }
834
835         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
836                 repbody->max_cookiesize = 0;
837                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
838                 repbody->valid |= OBD_MD_FLMODEASIZE;
839                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
840                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
841                        repbody->max_cookiesize);
842         }
843
844         if (exp_connect_rmtclient(info->mti_exp) &&
845             reqbody->valid & OBD_MD_FLRMTPERM) {
846                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
847
848                 /* mdt_getattr_lock only */
849                 rc = mdt_pack_remote_perm(info, o, buf);
850                 if (rc) {
851                         repbody->valid &= ~OBD_MD_FLRMTPERM;
852                         repbody->aclsize = 0;
853                         RETURN(rc);
854                 } else {
855                         repbody->valid |= OBD_MD_FLRMTPERM;
856                         repbody->aclsize = sizeof(struct mdt_remote_perm);
857                 }
858         }
859 #ifdef CONFIG_FS_POSIX_ACL
860         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
861                  (reqbody->valid & OBD_MD_FLACL)) {
862                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
863                 buffer->lb_len = req_capsule_get_size(pill,
864                                                       &RMF_ACL, RCL_SERVER);
865                 if (buffer->lb_len > 0) {
866                         rc = mo_xattr_get(env, next, buffer,
867                                           XATTR_NAME_ACL_ACCESS);
868                         if (rc < 0) {
869                                 if (rc == -ENODATA) {
870                                         repbody->aclsize = 0;
871                                         repbody->valid |= OBD_MD_FLACL;
872                                         rc = 0;
873                                 } else if (rc == -EOPNOTSUPP) {
874                                         rc = 0;
875                                 } else {
876                                         CERROR("%s: unable to read "DFID
877                                                " ACL: rc = %d\n",
878                                                mdt_obd_name(info->mti_mdt),
879                                                PFID(mdt_object_fid(o)), rc);
880                                 }
881                         } else {
882                                 repbody->aclsize = rc;
883                                 repbody->valid |= OBD_MD_FLACL;
884                                 rc = 0;
885                         }
886                 }
887         }
888 #endif
889
890         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
891             info->mti_mdt->mdt_opts.mo_mds_capa &&
892             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
893                 struct lustre_capa *capa;
894
895                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
896                 LASSERT(capa);
897                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
898                 rc = mo_capa_get(env, next, capa, 0);
899                 if (rc)
900                         RETURN(rc);
901                 repbody->valid |= OBD_MD_FLMDSCAPA;
902         }
903
904 out:
905         if (rc == 0)
906                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
907
908         RETURN(rc);
909 }
910
911 static int mdt_renew_capa(struct mdt_thread_info *info)
912 {
913         struct mdt_object  *obj = info->mti_object;
914         struct mdt_body    *body;
915         struct lustre_capa *capa, *c;
916         int rc;
917         ENTRY;
918
919         /* if object doesn't exist, or server has disabled capability,
920          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
921          * flag not set.
922          */
923         if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
924             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
925                 RETURN(0);
926
927         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
928         LASSERT(body != NULL);
929
930         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
931         LASSERT(c);
932
933         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
934         LASSERT(capa);
935
936         *capa = *c;
937         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
938         if (rc == 0)
939                 body->valid |= OBD_MD_FLOSSCAPA;
940         RETURN(rc);
941 }
942
943 int mdt_getattr(struct mdt_thread_info *info)
944 {
945         struct mdt_object       *obj = info->mti_object;
946         struct req_capsule      *pill = info->mti_pill;
947         struct mdt_body         *reqbody;
948         struct mdt_body         *repbody;
949         mode_t                   mode;
950         int rc, rc2;
951         ENTRY;
952
953         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
954         LASSERT(reqbody);
955
956         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
957                 rc = req_capsule_server_pack(pill);
958                 if (unlikely(rc))
959                         RETURN(err_serious(rc));
960                 rc = mdt_renew_capa(info);
961                 GOTO(out_shrink, rc);
962         }
963
964         LASSERT(obj != NULL);
965         LASSERT(lu_object_assert_exists(&obj->mot_obj));
966
967         mode = lu_object_attr(&obj->mot_obj);
968
969         /* old clients may not report needed easize, use max value then */
970         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
971                              reqbody->eadatasize == 0 ?
972                              info->mti_mdt->mdt_max_mdsize :
973                              reqbody->eadatasize);
974
975         rc = req_capsule_server_pack(pill);
976         if (unlikely(rc != 0))
977                 RETURN(err_serious(rc));
978
979         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
980         LASSERT(repbody != NULL);
981         repbody->eadatasize = 0;
982         repbody->aclsize = 0;
983
984         if (reqbody->valid & OBD_MD_FLRMTPERM)
985                 rc = mdt_init_ucred(info, reqbody);
986         else
987                 rc = mdt_check_ucred(info);
988         if (unlikely(rc))
989                 GOTO(out_shrink, rc);
990
991         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
992
993         /*
994          * Don't check capability at all, because rename might getattr for
995          * remote obj, and at that time no capability is available.
996          */
997         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
998         rc = mdt_getattr_internal(info, obj, 0);
999         if (reqbody->valid & OBD_MD_FLRMTPERM)
1000                 mdt_exit_ucred(info);
1001         EXIT;
1002 out_shrink:
1003         mdt_client_compatibility(info);
1004         rc2 = mdt_fix_reply(info);
1005         if (rc == 0)
1006                 rc = rc2;
1007         return rc;
1008 }
1009
1010 int mdt_is_subdir(struct mdt_thread_info *info)
1011 {
1012         struct mdt_object     *o = info->mti_object;
1013         struct req_capsule    *pill = info->mti_pill;
1014         const struct mdt_body *body = info->mti_body;
1015         struct mdt_body       *repbody;
1016         int                    rc;
1017         ENTRY;
1018
1019         LASSERT(o != NULL);
1020
1021         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1022
1023         /*
1024          * We save last checked parent fid to @repbody->fid1 for remote
1025          * directory case.
1026          */
1027         LASSERT(fid_is_sane(&body->fid2));
1028         LASSERT(mdt_object_exists(o) && !mdt_object_remote(o));
1029         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
1030                            &body->fid2, &repbody->fid1);
1031         if (rc == 0 || rc == -EREMOTE)
1032                 repbody->valid |= OBD_MD_FLID;
1033
1034         RETURN(rc);
1035 }
1036
1037 int mdt_swap_layouts(struct mdt_thread_info *info)
1038 {
1039         struct ptlrpc_request   *req = mdt_info_req(info);
1040         struct obd_export       *exp = req->rq_export;
1041         struct mdt_object       *o1, *o2, *o;
1042         struct mdt_lock_handle  *lh1, *lh2;
1043         struct mdc_swap_layouts *msl;
1044         int                      rc;
1045         ENTRY;
1046
1047         /* client does not support layout lock, so layout swaping
1048          * is disabled.
1049          * FIXME: there is a problem for old clients which don't support
1050          * layout lock yet. If those clients have already opened the file
1051          * they won't be notified at all so that old layout may still be
1052          * used to do IO. This can be fixed after file release is landed by
1053          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1054         if (!exp_connect_layout(exp))
1055                 RETURN(-EOPNOTSUPP);
1056
1057         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1058                 mdt_set_capainfo(info, 0, &info->mti_body->fid1,
1059                                  req_capsule_client_get(info->mti_pill,
1060                                                         &RMF_CAPA1));
1061
1062         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1063                 mdt_set_capainfo(info, 1, &info->mti_body->fid2,
1064                                  req_capsule_client_get(info->mti_pill,
1065                                                         &RMF_CAPA2));
1066
1067         o1 = info->mti_object;
1068         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1069                                 &info->mti_body->fid2);
1070         if (IS_ERR(o))
1071                 GOTO(out, rc = PTR_ERR(o));
1072
1073         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1074                 GOTO(put, rc = -ENOENT);
1075
1076         rc = lu_fid_cmp(&info->mti_body->fid1, &info->mti_body->fid2);
1077         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1078                 GOTO(put, rc);
1079
1080         if (rc < 0)
1081                 swap(o1, o2);
1082
1083         /* permission check. Make sure the calling process having permission
1084          * to write both files. */
1085         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1086                                 MAY_WRITE);
1087         if (rc < 0)
1088                 GOTO(put, rc);
1089
1090         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1091                                 MAY_WRITE);
1092         if (rc < 0)
1093                 GOTO(put, rc);
1094
1095         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1096         if (msl == NULL)
1097                 GOTO(put, rc = -EPROTO);
1098
1099         lh1 = &info->mti_lh[MDT_LH_NEW];
1100         mdt_lock_reg_init(lh1, LCK_EX);
1101         lh2 = &info->mti_lh[MDT_LH_OLD];
1102         mdt_lock_reg_init(lh2, LCK_EX);
1103
1104         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT,
1105                              MDT_LOCAL_LOCK);
1106         if (rc < 0)
1107                 GOTO(put, rc);
1108
1109         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT,
1110                              MDT_LOCAL_LOCK);
1111         if (rc < 0)
1112                 GOTO(unlock1, rc);
1113
1114         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1115                              mdt_object_child(o2), msl->msl_flags);
1116         GOTO(unlock2, rc);
1117 unlock2:
1118         mdt_object_unlock(info, o2, lh2, rc);
1119 unlock1:
1120         mdt_object_unlock(info, o1, lh1, rc);
1121 put:
1122         mdt_object_put(info->mti_env, o);
1123 out:
1124         RETURN(rc);
1125 }
1126
1127 static int mdt_raw_lookup(struct mdt_thread_info *info,
1128                           struct mdt_object *parent,
1129                           const struct lu_name *lname,
1130                           struct ldlm_reply *ldlm_rep)
1131 {
1132         struct md_object *next = mdt_object_child(info->mti_object);
1133         const struct mdt_body *reqbody = info->mti_body;
1134         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1135         struct mdt_body *repbody;
1136         int rc;
1137         ENTRY;
1138
1139         if (reqbody->valid != OBD_MD_FLID)
1140                 RETURN(0);
1141
1142         LASSERT(!info->mti_cross_ref);
1143
1144         /* Only got the fid of this obj by name */
1145         fid_zero(child_fid);
1146         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1147                         &info->mti_spec);
1148 #if 0
1149         /* XXX is raw_lookup possible as intent operation? */
1150         if (rc != 0) {
1151                 if (rc == -ENOENT)
1152                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1153                 RETURN(rc);
1154         } else
1155                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1156
1157         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1158 #endif
1159         if (rc == 0) {
1160                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1161                 repbody->fid1 = *child_fid;
1162                 repbody->valid = OBD_MD_FLID;
1163         }
1164         RETURN(1);
1165 }
1166
1167 /*
1168  * UPDATE lock should be taken against parent, and be release before exit;
1169  * child_bits lock should be taken against child, and be returned back:
1170  *            (1)normal request should release the child lock;
1171  *            (2)intent request will grant the lock to client.
1172  */
1173 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1174                                  struct mdt_lock_handle *lhc,
1175                                  __u64 child_bits,
1176                                  struct ldlm_reply *ldlm_rep)
1177 {
1178         struct ptlrpc_request  *req       = mdt_info_req(info);
1179         struct mdt_body        *reqbody   = NULL;
1180         struct mdt_object      *parent    = info->mti_object;
1181         struct mdt_object      *child;
1182         struct md_object       *next      = mdt_object_child(parent);
1183         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1184         struct lu_name         *lname     = NULL;
1185         const char             *name      = NULL;
1186         int                     namelen   = 0;
1187         struct mdt_lock_handle *lhp       = NULL;
1188         struct ldlm_lock       *lock;
1189         struct ldlm_res_id     *res_id;
1190         int                     is_resent;
1191         int                     ma_need = 0;
1192         int                     rc;
1193
1194         ENTRY;
1195
1196         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1197         LASSERT(ergo(is_resent,
1198                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1199
1200         LASSERT(parent != NULL);
1201         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
1202         if (name == NULL)
1203                 RETURN(err_serious(-EFAULT));
1204
1205         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
1206                                        RCL_CLIENT) - 1;
1207         if (!info->mti_cross_ref) {
1208                 /*
1209                  * XXX: Check for "namelen == 0" is for getattr by fid
1210                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
1211                  * that is the name must contain at least one character and
1212                  * the terminating '\0'
1213                  */
1214                 if (namelen == 0) {
1215                         reqbody = req_capsule_client_get(info->mti_pill,
1216                                                          &RMF_MDT_BODY);
1217                         if (unlikely(reqbody == NULL))
1218                                 RETURN(err_serious(-EFAULT));
1219
1220                         if (unlikely(!fid_is_sane(&reqbody->fid2)))
1221                                 RETURN(err_serious(-EINVAL));
1222
1223                         name = NULL;
1224                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1225                                "ldlm_rep = %p\n",
1226                                PFID(mdt_object_fid(parent)),
1227                                PFID(&reqbody->fid2), ldlm_rep);
1228                 } else {
1229                         lname = mdt_name(info->mti_env, (char *)name, namelen);
1230                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
1231                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1232                                name, ldlm_rep);
1233                 }
1234         }
1235         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1236
1237         if (unlikely(!mdt_object_exists(parent))) {
1238                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1239                                 &parent->mot_obj,
1240                                 "Parent doesn't exist!\n");
1241                 RETURN(-ESTALE);
1242         } else if (!info->mti_cross_ref) {
1243                 LASSERTF(!mdt_object_remote(parent),
1244                          "Parent "DFID" is on remote server\n",
1245                          PFID(mdt_object_fid(parent)));
1246         }
1247         if (lname) {
1248                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1249                 if (rc != 0) {
1250                         if (rc > 0)
1251                                 rc = 0;
1252                         RETURN(rc);
1253                 }
1254         }
1255
1256         if (info->mti_cross_ref) {
1257                 /* Only getattr on the child. Parent is on another node. */
1258                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1259                 child = parent;
1260                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1261                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
1262
1263                 if (is_resent) {
1264                         /* Do not take lock for resent case. */
1265                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1266                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1267                                  lhc->mlh_reg_lh.cookie);
1268                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
1269                                                 &lock->l_resource->lr_name));
1270                         LDLM_LOCK_PUT(lock);
1271                         rc = 0;
1272                 } else {
1273                         mdt_lock_handle_init(lhc);
1274                         mdt_lock_reg_init(lhc, LCK_PR);
1275
1276                         /*
1277                          * Object's name is on another MDS, no lookup lock is
1278                          * needed here but update is.
1279                          */
1280                         child_bits &= ~MDS_INODELOCK_LOOKUP;
1281                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1282
1283                         rc = mdt_object_lock(info, child, lhc, child_bits,
1284                                              MDT_LOCAL_LOCK);
1285                 }
1286                 if (rc == 0) {
1287                         /* Finally, we can get attr for child. */
1288                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
1289                                          BYPASS_CAPA);
1290                         rc = mdt_getattr_internal(info, child, 0);
1291                         if (unlikely(rc != 0))
1292                                 mdt_object_unlock(info, child, lhc, 1);
1293                 }
1294                 RETURN(rc);
1295         }
1296
1297         if (lname) {
1298                 /* step 1: lock parent only if parent is a directory */
1299                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1300                         lhp = &info->mti_lh[MDT_LH_PARENT];
1301                         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
1302                         rc = mdt_object_lock(info, parent, lhp,
1303                                              MDS_INODELOCK_UPDATE,
1304                                              MDT_LOCAL_LOCK);
1305                         if (unlikely(rc != 0))
1306                                 RETURN(rc);
1307                 }
1308
1309                 /* step 2: lookup child's fid by name */
1310                 fid_zero(child_fid);
1311                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1312                                 &info->mti_spec);
1313
1314                 if (rc != 0) {
1315                         if (rc == -ENOENT)
1316                                 mdt_set_disposition(info, ldlm_rep,
1317                                                     DISP_LOOKUP_NEG);
1318                         GOTO(out_parent, rc);
1319                 } else
1320                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1321         } else {
1322                 *child_fid = reqbody->fid2;
1323                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1324         }
1325
1326         /*
1327          *step 3: find the child object by fid & lock it.
1328          *        regardless if it is local or remote.
1329          */
1330         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
1331
1332         if (unlikely(IS_ERR(child)))
1333                 GOTO(out_parent, rc = PTR_ERR(child));
1334         if (is_resent) {
1335                 /* Do not take lock for resent case. */
1336                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1337                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1338                          lhc->mlh_reg_lh.cookie);
1339
1340                 res_id = &lock->l_resource->lr_name;
1341                 if (!fid_res_name_eq(mdt_object_fid(child),
1342                                      &lock->l_resource->lr_name)) {
1343                         LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1344                                                  &lock->l_resource->lr_name),
1345                                  "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1346                                  PLDLMRES(lock->l_resource),
1347                                  PFID(mdt_object_fid(parent)));
1348                         CWARN("Although resent, but still not get child lock"
1349                               "parent:"DFID" child:"DFID"\n",
1350                               PFID(mdt_object_fid(parent)),
1351                               PFID(mdt_object_fid(child)));
1352                         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1353                         LDLM_LOCK_PUT(lock);
1354                         GOTO(relock, 0);
1355                 }
1356                 LDLM_LOCK_PUT(lock);
1357                 rc = 0;
1358         } else {
1359                 bool try_layout = false;
1360
1361 relock:
1362                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1363                 mdt_lock_handle_init(lhc);
1364                 mdt_lock_reg_init(lhc, LCK_PR);
1365
1366                 if (!mdt_object_exists(child)) {
1367                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1368                                         &child->mot_obj,
1369                                         "Object doesn't exist!\n");
1370                         GOTO(out_child, rc = -ENOENT);
1371                 }
1372
1373                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1374                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1375                         struct md_attr *ma = &info->mti_attr;
1376
1377                         ma->ma_valid = 0;
1378                         ma->ma_need = MA_INODE;
1379                         rc = mdt_attr_get_complex(info, child, ma);
1380                         if (unlikely(rc != 0))
1381                                 GOTO(out_child, rc);
1382
1383                         /* If the file has not been changed for some time, we
1384                          * return not only a LOOKUP lock, but also an UPDATE
1385                          * lock and this might save us RPC on later STAT. For
1386                          * directories, it also let negative dentry starts
1387                          * working for this dir. */
1388                         if (ma->ma_valid & MA_INODE &&
1389                             ma->ma_attr.la_valid & LA_CTIME &&
1390                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1391                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1392                                 child_bits |= MDS_INODELOCK_UPDATE;
1393                 }
1394
1395                 /* layout lock must be granted in a best-effort way
1396                  * for IT operations */
1397                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1398                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1399                     exp_connect_layout(info->mti_exp) &&
1400                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1401                     ldlm_rep != NULL) {
1402                         /* try to grant layout lock for regular file. */
1403                         try_layout = true;
1404                 }
1405
1406                 rc = 0;
1407                 if (try_layout) {
1408                         child_bits |= MDS_INODELOCK_LAYOUT;
1409                         /* try layout lock, it may fail to be granted due to
1410                          * contention at LOOKUP or UPDATE */
1411                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1412                                                  MDT_CROSS_LOCK)) {
1413                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1414                                 LASSERT(child_bits != 0);
1415                                 rc = mdt_object_lock(info, child, lhc,
1416                                                 child_bits, MDT_CROSS_LOCK);
1417                         } else {
1418                                 ma_need |= MA_LOV;
1419                         }
1420                 } else {
1421                         rc = mdt_object_lock(info, child, lhc, child_bits,
1422                                                 MDT_CROSS_LOCK);
1423                 }
1424                 if (unlikely(rc != 0))
1425                         GOTO(out_child, rc);
1426         }
1427
1428         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1429         /* Get MA_SOM attributes if update lock is given. */
1430         if (lock &&
1431             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1432             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1433                 ma_need |= MA_SOM;
1434
1435         /* finally, we can get attr for child. */
1436         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1437         rc = mdt_getattr_internal(info, child, ma_need);
1438         if (unlikely(rc != 0)) {
1439                 mdt_object_unlock(info, child, lhc, 1);
1440         } else if (lock) {
1441                 /* Debugging code. */
1442                 res_id = &lock->l_resource->lr_name;
1443                 LDLM_DEBUG(lock, "Returning lock to client");
1444                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1445                                          &lock->l_resource->lr_name),
1446                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1447                          PLDLMRES(lock->l_resource),
1448                          PFID(mdt_object_fid(child)));
1449                 if (mdt_object_exists(child) && !mdt_object_remote(child))
1450                         mdt_pack_size2body(info, child);
1451         }
1452         if (lock)
1453                 LDLM_LOCK_PUT(lock);
1454
1455         EXIT;
1456 out_child:
1457         mdt_object_put(info->mti_env, child);
1458 out_parent:
1459         if (lhp)
1460                 mdt_object_unlock(info, parent, lhp, 1);
1461         return rc;
1462 }
1463
1464 /* normal handler: should release the child lock */
1465 int mdt_getattr_name(struct mdt_thread_info *info)
1466 {
1467         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1468         struct mdt_body        *reqbody;
1469         struct mdt_body        *repbody;
1470         int rc, rc2;
1471         ENTRY;
1472
1473         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1474         LASSERT(reqbody != NULL);
1475         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1476         LASSERT(repbody != NULL);
1477
1478         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1479         repbody->eadatasize = 0;
1480         repbody->aclsize = 0;
1481
1482         rc = mdt_init_ucred(info, reqbody);
1483         if (unlikely(rc))
1484                 GOTO(out_shrink, rc);
1485
1486         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1487         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1488                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1489                 lhc->mlh_reg_lh.cookie = 0;
1490         }
1491         mdt_exit_ucred(info);
1492         EXIT;
1493 out_shrink:
1494         mdt_client_compatibility(info);
1495         rc2 = mdt_fix_reply(info);
1496         if (rc == 0)
1497                 rc = rc2;
1498         return rc;
1499 }
1500
1501 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1502                          void *karg, void *uarg);
1503
1504 int mdt_set_info(struct mdt_thread_info *info)
1505 {
1506         struct ptlrpc_request *req = mdt_info_req(info);
1507         char *key;
1508         void *val;
1509         int keylen, vallen, rc = 0;
1510         ENTRY;
1511
1512         rc = req_capsule_server_pack(info->mti_pill);
1513         if (rc)
1514                 RETURN(rc);
1515
1516         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1517         if (key == NULL) {
1518                 DEBUG_REQ(D_HA, req, "no set_info key");
1519                 RETURN(-EFAULT);
1520         }
1521
1522         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1523                                       RCL_CLIENT);
1524
1525         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1526         if (val == NULL) {
1527                 DEBUG_REQ(D_HA, req, "no set_info val");
1528                 RETURN(-EFAULT);
1529         }
1530
1531         vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1532                                       RCL_CLIENT);
1533
1534         /* Swab any part of val you need to here */
1535         if (KEY_IS(KEY_READ_ONLY)) {
1536                 req->rq_status = 0;
1537                 lustre_msg_set_status(req->rq_repmsg, 0);
1538
1539                 spin_lock(&req->rq_export->exp_lock);
1540                 if (*(__u32 *)val)
1541                         *exp_connect_flags_ptr(req->rq_export) |=
1542                                 OBD_CONNECT_RDONLY;
1543                 else
1544                         *exp_connect_flags_ptr(req->rq_export) &=
1545                                 ~OBD_CONNECT_RDONLY;
1546                 spin_unlock(&req->rq_export->exp_lock);
1547
1548         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1549                 struct changelog_setinfo *cs =
1550                         (struct changelog_setinfo *)val;
1551                 if (vallen != sizeof(*cs)) {
1552                         CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1553                         RETURN(-EINVAL);
1554                 }
1555                 if (ptlrpc_req_need_swab(req)) {
1556                         __swab64s(&cs->cs_recno);
1557                         __swab32s(&cs->cs_id);
1558                 }
1559
1560                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1561                                    vallen, val, NULL);
1562                 lustre_msg_set_status(req->rq_repmsg, rc);
1563
1564         } else {
1565                 RETURN(-EINVAL);
1566         }
1567         RETURN(0);
1568 }
1569
1570 /**
1571  * Top-level handler for MDT connection requests.
1572  */
1573 int mdt_connect(struct mdt_thread_info *info)
1574 {
1575         int rc;
1576         struct obd_connect_data *reply;
1577         struct obd_export *exp;
1578         struct ptlrpc_request *req = mdt_info_req(info);
1579
1580         rc = target_handle_connect(req);
1581         if (rc != 0)
1582                 return err_serious(rc);
1583
1584         LASSERT(req->rq_export != NULL);
1585         info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1586         rc = mdt_init_sec_level(info);
1587         if (rc != 0) {
1588                 obd_disconnect(class_export_get(req->rq_export));
1589                 return rc;
1590         }
1591
1592         /* To avoid exposing partially initialized connection flags, changes up
1593          * to this point have been staged in reply->ocd_connect_flags. Now that
1594          * connection handling has completed successfully, atomically update
1595          * the connect flags in the shared export data structure. LU-1623 */
1596         reply = req_capsule_server_get(info->mti_pill, &RMF_CONNECT_DATA);
1597         exp = req->rq_export;
1598         spin_lock(&exp->exp_lock);
1599         *exp_connect_flags_ptr(exp) = reply->ocd_connect_flags;
1600         spin_unlock(&exp->exp_lock);
1601
1602         rc = mdt_init_idmap(info);
1603         if (rc != 0)
1604                 obd_disconnect(class_export_get(req->rq_export));
1605
1606         return rc;
1607 }
1608
1609 int mdt_disconnect(struct mdt_thread_info *info)
1610 {
1611         int rc;
1612         ENTRY;
1613
1614         rc = target_handle_disconnect(mdt_info_req(info));
1615         if (rc)
1616                 rc = err_serious(rc);
1617         RETURN(rc);
1618 }
1619
1620 static int mdt_sendpage(struct mdt_thread_info *info,
1621                         struct lu_rdpg *rdpg, int nob)
1622 {
1623         struct ptlrpc_request   *req = mdt_info_req(info);
1624         struct obd_export       *exp = req->rq_export;
1625         struct ptlrpc_bulk_desc *desc;
1626         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1627         int                      tmpcount;
1628         int                      tmpsize;
1629         int                      i;
1630         int                      rc;
1631         ENTRY;
1632
1633         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1634                                     MDS_BULK_PORTAL);
1635         if (desc == NULL)
1636                 RETURN(-ENOMEM);
1637
1638         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1639                 /* old client requires reply size in it's PAGE_SIZE,
1640                  * which is rdpg->rp_count */
1641                 nob = rdpg->rp_count;
1642
1643         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1644              i++, tmpcount -= tmpsize) {
1645                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1646                 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1647         }
1648
1649         LASSERT(desc->bd_nob == nob);
1650         rc = target_bulk_io(exp, desc, lwi);
1651         ptlrpc_free_bulk_pin(desc);
1652         RETURN(rc);
1653 }
1654
1655 int mdt_readpage(struct mdt_thread_info *info)
1656 {
1657         struct mdt_object *object = info->mti_object;
1658         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1659         struct mdt_body   *reqbody;
1660         struct mdt_body   *repbody;
1661         int                rc;
1662         int                i;
1663         ENTRY;
1664
1665         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1666                 RETURN(err_serious(-ENOMEM));
1667
1668         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1669         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1670         if (reqbody == NULL || repbody == NULL)
1671                 RETURN(err_serious(-EFAULT));
1672
1673         /*
1674          * prepare @rdpg before calling lower layers and transfer itself. Here
1675          * reqbody->size contains offset of where to start to read and
1676          * reqbody->nlink contains number bytes to read.
1677          */
1678         rdpg->rp_hash = reqbody->size;
1679         if (rdpg->rp_hash != reqbody->size) {
1680                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1681                        rdpg->rp_hash, reqbody->size);
1682                 RETURN(-EFAULT);
1683         }
1684
1685         rdpg->rp_attrs = reqbody->mode;
1686         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_64BITHASH)
1687                 rdpg->rp_attrs |= LUDA_64BITHASH;
1688         rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
1689                                 exp_max_brw_size(info->mti_exp));
1690         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1691                           PAGE_CACHE_SHIFT;
1692         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1693         if (rdpg->rp_pages == NULL)
1694                 RETURN(-ENOMEM);
1695
1696         for (i = 0; i < rdpg->rp_npages; ++i) {
1697                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1698                 if (rdpg->rp_pages[i] == NULL)
1699                         GOTO(free_rdpg, rc = -ENOMEM);
1700         }
1701
1702         /* call lower layers to fill allocated pages with directory data */
1703         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1704         if (rc < 0)
1705                 GOTO(free_rdpg, rc);
1706
1707         /* send pages to client */
1708         rc = mdt_sendpage(info, rdpg, rc);
1709
1710         EXIT;
1711 free_rdpg:
1712
1713         for (i = 0; i < rdpg->rp_npages; i++)
1714                 if (rdpg->rp_pages[i] != NULL)
1715                         __free_page(rdpg->rp_pages[i]);
1716         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1717
1718         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1719                 RETURN(0);
1720
1721         return rc;
1722 }
1723
1724 static int mdt_reint_internal(struct mdt_thread_info *info,
1725                               struct mdt_lock_handle *lhc,
1726                               __u32 op)
1727 {
1728         struct req_capsule      *pill = info->mti_pill;
1729         struct mdt_body         *repbody;
1730         int                      rc = 0, rc2;
1731         ENTRY;
1732
1733
1734         rc = mdt_reint_unpack(info, op);
1735         if (rc != 0) {
1736                 CERROR("Can't unpack reint, rc %d\n", rc);
1737                 RETURN(err_serious(rc));
1738         }
1739
1740         /* for replay (no_create) lmm is not needed, client has it already */
1741         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1742                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1743                                      info->mti_rr.rr_eadatalen);
1744
1745         /* llog cookies are always 0, the field is kept for compatibility */
1746         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1747                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1748
1749         rc = req_capsule_server_pack(pill);
1750         if (rc != 0) {
1751                 CERROR("Can't pack response, rc %d\n", rc);
1752                 RETURN(err_serious(rc));
1753         }
1754
1755         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1756                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1757                 LASSERT(repbody);
1758                 repbody->eadatasize = 0;
1759                 repbody->aclsize = 0;
1760         }
1761
1762         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1763
1764         /* for replay no cookkie / lmm need, because client have this already */
1765         if (info->mti_spec.no_create)
1766                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1767                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1768
1769         rc = mdt_init_ucred_reint(info);
1770         if (rc)
1771                 GOTO(out_shrink, rc);
1772
1773         rc = mdt_fix_attr_ucred(info, op);
1774         if (rc != 0)
1775                 GOTO(out_ucred, rc = err_serious(rc));
1776
1777         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1778                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1779                 GOTO(out_ucred, rc);
1780         }
1781         rc = mdt_reint_rec(info, lhc);
1782         EXIT;
1783 out_ucred:
1784         mdt_exit_ucred(info);
1785 out_shrink:
1786         mdt_client_compatibility(info);
1787         rc2 = mdt_fix_reply(info);
1788         if (rc == 0)
1789                 rc = rc2;
1790         return rc;
1791 }
1792
1793 static long mdt_reint_opcode(struct mdt_thread_info *info,
1794                              const struct req_format **fmt)
1795 {
1796         struct mdt_rec_reint *rec;
1797         long opc;
1798
1799         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1800         if (rec != NULL) {
1801                 opc = rec->rr_opcode;
1802                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1803                 if (opc < REINT_MAX && fmt[opc] != NULL)
1804                         req_capsule_extend(info->mti_pill, fmt[opc]);
1805                 else {
1806                         CERROR("%s: Unsupported opcode '%ld' from client '%s': "
1807                                "rc = %d\n", mdt_obd_name(info->mti_mdt), opc,
1808                                info->mti_mdt->mdt_ldlm_client->cli_name,
1809                                -EFAULT);
1810                         opc = err_serious(-EFAULT);
1811                 }
1812         } else {
1813                 opc = err_serious(-EFAULT);
1814         }
1815         return opc;
1816 }
1817
1818 int mdt_reint(struct mdt_thread_info *info)
1819 {
1820         long opc;
1821         int  rc;
1822
1823         static const struct req_format *reint_fmts[REINT_MAX] = {
1824                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1825                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1826                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1827                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1828                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1829                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1830                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1831                 [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK
1832         };
1833
1834         ENTRY;
1835
1836         opc = mdt_reint_opcode(info, reint_fmts);
1837         if (opc >= 0) {
1838                 /*
1839                  * No lock possible here from client to pass it to reint code
1840                  * path.
1841                  */
1842                 rc = mdt_reint_internal(info, NULL, opc);
1843         } else {
1844                 rc = opc;
1845         }
1846
1847         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1848         RETURN(rc);
1849 }
1850
1851 /* this should sync the whole device */
1852 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1853 {
1854         struct dt_device *dt = mdt->mdt_bottom;
1855         int rc;
1856         ENTRY;
1857
1858         rc = dt->dd_ops->dt_sync(env, dt);
1859         RETURN(rc);
1860 }
1861
1862 /* this should sync this object */
1863 static int mdt_object_sync(struct mdt_thread_info *info)
1864 {
1865         struct md_object *next;
1866         int rc;
1867         ENTRY;
1868
1869         if (!mdt_object_exists(info->mti_object)) {
1870                 CWARN("Non existing object  "DFID"!\n",
1871                       PFID(mdt_object_fid(info->mti_object)));
1872                 RETURN(-ESTALE);
1873         }
1874         next = mdt_object_child(info->mti_object);
1875         rc = mo_object_sync(info->mti_env, next);
1876
1877         RETURN(rc);
1878 }
1879
1880 int mdt_sync(struct mdt_thread_info *info)
1881 {
1882         struct ptlrpc_request *req = mdt_info_req(info);
1883         struct req_capsule *pill = info->mti_pill;
1884         struct mdt_body *body;
1885         int rc;
1886         ENTRY;
1887
1888         /* The fid may be zero, so we req_capsule_set manually */
1889         req_capsule_set(pill, &RQF_MDS_SYNC);
1890
1891         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1892         if (body == NULL)
1893                 RETURN(err_serious(-EINVAL));
1894
1895         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1896                 RETURN(err_serious(-ENOMEM));
1897
1898         if (fid_seq(&body->fid1) == 0) {
1899                 /* sync the whole device */
1900                 rc = req_capsule_server_pack(pill);
1901                 if (rc == 0)
1902                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1903                 else
1904                         rc = err_serious(rc);
1905         } else {
1906                 /* sync an object */
1907                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1908                 if (rc == 0) {
1909                         rc = mdt_object_sync(info);
1910                         if (rc == 0) {
1911                                 const struct lu_fid *fid;
1912                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1913
1914                                 info->mti_attr.ma_need = MA_INODE;
1915                                 info->mti_attr.ma_valid = 0;
1916                                 rc = mdt_attr_get_complex(info, info->mti_object,
1917                                                           &info->mti_attr);
1918                                 if (rc == 0) {
1919                                         body = req_capsule_server_get(pill,
1920                                                                 &RMF_MDT_BODY);
1921                                         fid = mdt_object_fid(info->mti_object);
1922                                         mdt_pack_attr2body(info, body, la, fid);
1923                                 }
1924                         }
1925                 } else
1926                         rc = err_serious(rc);
1927         }
1928         if (rc == 0)
1929                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1930
1931         RETURN(rc);
1932 }
1933
1934 /*
1935  * Quotacheck handler.
1936  * in-kernel quotacheck isn't supported any more.
1937  */
1938 int mdt_quotacheck(struct mdt_thread_info *info)
1939 {
1940         struct obd_quotactl     *oqctl;
1941         int                      rc;
1942         ENTRY;
1943
1944         oqctl = req_capsule_client_get(info->mti_pill, &RMF_OBD_QUOTACTL);
1945         if (oqctl == NULL)
1946                 RETURN(err_serious(-EPROTO));
1947
1948         rc = req_capsule_server_pack(info->mti_pill);
1949         if (rc)
1950                 RETURN(err_serious(rc));
1951
1952         /* deprecated, not used any more */
1953         RETURN(-EOPNOTSUPP);
1954 }
1955
1956 /*
1957  * Handle quota control requests to consult current usage/limit, but also
1958  * to configure quota enforcement
1959  */
1960 int mdt_quotactl(struct mdt_thread_info *info)
1961 {
1962         struct obd_export       *exp  = info->mti_exp;
1963         struct req_capsule      *pill = info->mti_pill;
1964         struct obd_quotactl     *oqctl, *repoqc;
1965         int                      id, rc;
1966         struct lu_device        *qmt = info->mti_mdt->mdt_qmt_dev;
1967         ENTRY;
1968
1969         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1970         if (oqctl == NULL)
1971                 RETURN(err_serious(-EPROTO));
1972
1973         rc = req_capsule_server_pack(pill);
1974         if (rc)
1975                 RETURN(err_serious(rc));
1976
1977         switch (oqctl->qc_cmd) {
1978         case Q_QUOTACHECK:
1979         case LUSTRE_Q_INVALIDATE:
1980         case LUSTRE_Q_FINVALIDATE:
1981         case Q_QUOTAON:
1982         case Q_QUOTAOFF:
1983         case Q_INITQUOTA:
1984                 /* deprecated, not used any more */
1985                 RETURN(-EOPNOTSUPP);
1986                 /* master quotactl */
1987         case Q_GETINFO:
1988         case Q_SETINFO:
1989         case Q_SETQUOTA:
1990         case Q_GETQUOTA:
1991                 if (qmt == NULL)
1992                         RETURN(-EOPNOTSUPP);
1993                 /* slave quotactl */
1994         case Q_GETOINFO:
1995         case Q_GETOQUOTA:
1996                 break;
1997         default:
1998                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1999                 RETURN(-EFAULT);
2000         }
2001
2002         /* map uid/gid for remote client */
2003         id = oqctl->qc_id;
2004         if (exp_connect_rmtclient(exp)) {
2005                 struct lustre_idmap_table *idmap;
2006
2007                 idmap = mdt_req2med(mdt_info_req(info))->med_idmap;
2008
2009                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2010                              oqctl->qc_cmd != Q_GETINFO))
2011                         RETURN(-EPERM);
2012
2013                 if (oqctl->qc_type == USRQUOTA)
2014                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2015                                                      oqctl->qc_id);
2016                 else if (oqctl->qc_type == GRPQUOTA)
2017                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2018                                                      oqctl->qc_id);
2019                 else
2020                         RETURN(-EINVAL);
2021
2022                 if (id == CFS_IDMAP_NOTFOUND) {
2023                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2024                         RETURN(-EACCES);
2025                 }
2026         }
2027
2028         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2029         if (repoqc == NULL)
2030                 RETURN(err_serious(-EFAULT));
2031
2032         if (oqctl->qc_id != id)
2033                 swap(oqctl->qc_id, id);
2034
2035         switch (oqctl->qc_cmd) {
2036
2037         case Q_GETINFO:
2038         case Q_SETINFO:
2039         case Q_SETQUOTA:
2040         case Q_GETQUOTA:
2041                 /* forward quotactl request to QMT */
2042                 rc = qmt_hdls.qmth_quotactl(info->mti_env, qmt, oqctl);
2043                 break;
2044
2045         case Q_GETOINFO:
2046         case Q_GETOQUOTA:
2047                 /* slave quotactl */
2048                 rc = lquotactl_slv(info->mti_env, info->mti_mdt->mdt_bottom,
2049                                    oqctl);
2050                 break;
2051
2052         default:
2053                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2054                 RETURN(-EFAULT);
2055         }
2056
2057         if (oqctl->qc_id != id)
2058                 swap(oqctl->qc_id, id);
2059
2060         *repoqc = *oqctl;
2061         RETURN(rc);
2062 }
2063
2064 /*
2065  * OBD PING and other handlers.
2066  */
2067 int mdt_obd_ping(struct mdt_thread_info *info)
2068 {
2069         int rc;
2070         ENTRY;
2071
2072         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
2073
2074         rc = target_handle_ping(mdt_info_req(info));
2075         if (rc < 0)
2076                 rc = err_serious(rc);
2077         RETURN(rc);
2078 }
2079
2080 /*
2081  * OBD_IDX_READ handler
2082  */
2083 int mdt_obd_idx_read(struct mdt_thread_info *info)
2084 {
2085         struct mdt_device       *mdt = info->mti_mdt;
2086         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2087         struct idx_info         *req_ii, *rep_ii;
2088         int                      rc, i;
2089         ENTRY;
2090
2091         memset(rdpg, 0, sizeof(*rdpg));
2092         req_capsule_set(info->mti_pill, &RQF_OBD_IDX_READ);
2093
2094         /* extract idx_info buffer from request & reply */
2095         req_ii = req_capsule_client_get(info->mti_pill, &RMF_IDX_INFO);
2096         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
2097                 RETURN(err_serious(-EPROTO));
2098
2099         rc = req_capsule_server_pack(info->mti_pill);
2100         if (rc)
2101                 RETURN(err_serious(rc));
2102
2103         rep_ii = req_capsule_server_get(info->mti_pill, &RMF_IDX_INFO);
2104         if (rep_ii == NULL)
2105                 RETURN(err_serious(-EFAULT));
2106         rep_ii->ii_magic = IDX_INFO_MAGIC;
2107
2108         /* extract hash to start with */
2109         rdpg->rp_hash = req_ii->ii_hash_start;
2110
2111         /* extract requested attributes */
2112         rdpg->rp_attrs = req_ii->ii_attrs;
2113
2114         /* check that fid packed in request is valid and supported */
2115         if (!fid_is_sane(&req_ii->ii_fid))
2116                 RETURN(-EINVAL);
2117         rep_ii->ii_fid = req_ii->ii_fid;
2118
2119         /* copy flags */
2120         rep_ii->ii_flags = req_ii->ii_flags;
2121
2122         /* compute number of pages to allocate, ii_count is the number of 4KB
2123          * containers */
2124         if (req_ii->ii_count <= 0)
2125                 GOTO(out, rc = -EFAULT);
2126         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
2127                                exp_max_brw_size(info->mti_exp));
2128         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
2129                                 PAGE_CACHE_SHIFT;
2130
2131         /* allocate pages to store the containers */
2132         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
2133         if (rdpg->rp_pages == NULL)
2134                 GOTO(out, rc = -ENOMEM);
2135         for (i = 0; i < rdpg->rp_npages; i++) {
2136                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
2137                 if (rdpg->rp_pages[i] == NULL)
2138                         GOTO(out, rc = -ENOMEM);
2139         }
2140
2141         /* populate pages with key/record pairs */
2142         rc = dt_index_read(info->mti_env, mdt->mdt_bottom, rep_ii, rdpg);
2143         if (rc < 0)
2144                 GOTO(out, rc);
2145
2146         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
2147                  "asked %d > %d\n", rc, rdpg->rp_count);
2148
2149         /* send pages to client */
2150         rc = mdt_sendpage(info, rdpg, rc);
2151
2152         GOTO(out, rc);
2153 out:
2154         if (rdpg->rp_pages) {
2155                 for (i = 0; i < rdpg->rp_npages; i++)
2156                         if (rdpg->rp_pages[i])
2157                                 __free_page(rdpg->rp_pages[i]);
2158                 OBD_FREE(rdpg->rp_pages,
2159                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
2160         }
2161         return rc;
2162 }
2163
2164 int mdt_obd_log_cancel(struct mdt_thread_info *info)
2165 {
2166         return err_serious(-EOPNOTSUPP);
2167 }
2168
2169 int mdt_obd_qc_callback(struct mdt_thread_info *info)
2170 {
2171         return err_serious(-EOPNOTSUPP);
2172 }
2173
2174 /*
2175  * LLOG handlers.
2176  */
2177
2178 /** clone llog ctxt from child (mdd)
2179  * This allows remote llog (replicator) access.
2180  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2181  * context was originally set up, or we can handle them directly.
2182  * I choose the latter, but that means I need any llog
2183  * contexts set up by child to be accessable by the mdt.  So we clone the
2184  * context into our context list here.
2185  */
2186 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2187                                int idx)
2188 {
2189         struct md_device  *next = mdt->mdt_child;
2190         struct llog_ctxt *ctxt;
2191         int rc;
2192
2193         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2194                 return 0;
2195
2196         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2197         if (rc || ctxt == NULL) {
2198                 return 0;
2199         }
2200
2201         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2202         if (rc)
2203                 CERROR("Can't set mdt ctxt %d\n", rc);
2204
2205         return rc;
2206 }
2207
2208 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2209                                  struct mdt_device *mdt, int idx)
2210 {
2211         struct llog_ctxt *ctxt;
2212
2213         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2214         if (ctxt == NULL)
2215                 return 0;
2216         /* Put once for the get we just did, and once for the clone */
2217         llog_ctxt_put(ctxt);
2218         llog_ctxt_put(ctxt);
2219         return 0;
2220 }
2221
2222 int mdt_llog_create(struct mdt_thread_info *info)
2223 {
2224         int rc;
2225
2226         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
2227         rc = llog_origin_handle_open(mdt_info_req(info));
2228         return (rc < 0 ? err_serious(rc) : rc);
2229 }
2230
2231 int mdt_llog_destroy(struct mdt_thread_info *info)
2232 {
2233         int rc;
2234
2235         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
2236         rc = llog_origin_handle_destroy(mdt_info_req(info));
2237         return (rc < 0 ? err_serious(rc) : rc);
2238 }
2239
2240 int mdt_llog_read_header(struct mdt_thread_info *info)
2241 {
2242         int rc;
2243
2244         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
2245         rc = llog_origin_handle_read_header(mdt_info_req(info));
2246         return (rc < 0 ? err_serious(rc) : rc);
2247 }
2248
2249 int mdt_llog_next_block(struct mdt_thread_info *info)
2250 {
2251         int rc;
2252
2253         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
2254         rc = llog_origin_handle_next_block(mdt_info_req(info));
2255         return (rc < 0 ? err_serious(rc) : rc);
2256 }
2257
2258 int mdt_llog_prev_block(struct mdt_thread_info *info)
2259 {
2260         int rc;
2261
2262         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
2263         rc = llog_origin_handle_prev_block(mdt_info_req(info));
2264         return (rc < 0 ? err_serious(rc) : rc);
2265 }
2266
2267
2268 /*
2269  * DLM handlers.
2270  */
2271
2272 static struct ldlm_callback_suite cbs = {
2273         .lcs_completion = ldlm_server_completion_ast,
2274         .lcs_blocking   = ldlm_server_blocking_ast,
2275         .lcs_glimpse    = ldlm_server_glimpse_ast
2276 };
2277
2278 int mdt_enqueue(struct mdt_thread_info *info)
2279 {
2280         struct ptlrpc_request *req;
2281         int rc;
2282
2283         /*
2284          * info->mti_dlm_req already contains swapped and (if necessary)
2285          * converted dlm request.
2286          */
2287         LASSERT(info->mti_dlm_req != NULL);
2288
2289         req = mdt_info_req(info);
2290         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
2291                                   req, info->mti_dlm_req, &cbs);
2292         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
2293         return rc ? err_serious(rc) : req->rq_status;
2294 }
2295
2296 int mdt_convert(struct mdt_thread_info *info)
2297 {
2298         int rc;
2299         struct ptlrpc_request *req;
2300
2301         LASSERT(info->mti_dlm_req);
2302         req = mdt_info_req(info);
2303         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2304         return rc ? err_serious(rc) : req->rq_status;
2305 }
2306
2307 int mdt_bl_callback(struct mdt_thread_info *info)
2308 {
2309         CERROR("bl callbacks should not happen on MDS\n");
2310         LBUG();
2311         return err_serious(-EOPNOTSUPP);
2312 }
2313
2314 int mdt_cp_callback(struct mdt_thread_info *info)
2315 {
2316         CERROR("cp callbacks should not happen on MDS\n");
2317         LBUG();
2318         return err_serious(-EOPNOTSUPP);
2319 }
2320
2321 /*
2322  * sec context handlers
2323  */
2324 int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2325 {
2326         int rc;
2327
2328         rc = mdt_handle_idmap(info);
2329
2330         if (unlikely(rc)) {
2331                 struct ptlrpc_request *req = mdt_info_req(info);
2332                 __u32                  opc;
2333
2334                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2335                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2336                         sptlrpc_svc_ctx_invalidate(req);
2337         }
2338
2339         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2340
2341         return rc;
2342 }
2343
2344 /*
2345  * quota request handlers
2346  */
2347 int mdt_quota_dqacq(struct mdt_thread_info *info)
2348 {
2349         struct lu_device        *qmt = info->mti_mdt->mdt_qmt_dev;
2350         int                      rc;
2351         ENTRY;
2352
2353         if (qmt == NULL)
2354                 RETURN(err_serious(-EOPNOTSUPP));
2355
2356         rc = qmt_hdls.qmth_dqacq(info->mti_env, qmt, mdt_info_req(info));
2357         RETURN(rc);
2358 }
2359
2360 static struct mdt_object *mdt_obj(struct lu_object *o)
2361 {
2362         LASSERT(lu_device_is_mdt(o->lo_dev));
2363         return container_of0(o, struct mdt_object, mot_obj);
2364 }
2365
2366 struct mdt_object *mdt_object_new(const struct lu_env *env,
2367                                   struct mdt_device *d,
2368                                   const struct lu_fid *f)
2369 {
2370         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2371         struct lu_object *o;
2372         struct mdt_object *m;
2373         ENTRY;
2374
2375         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2376         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2377         if (unlikely(IS_ERR(o)))
2378                 m = (struct mdt_object *)o;
2379         else
2380                 m = mdt_obj(o);
2381         RETURN(m);
2382 }
2383
2384 struct mdt_object *mdt_object_find(const struct lu_env *env,
2385                                    struct mdt_device *d,
2386                                    const struct lu_fid *f)
2387 {
2388         struct lu_object *o;
2389         struct mdt_object *m;
2390         ENTRY;
2391
2392         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2393         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2394         if (unlikely(IS_ERR(o)))
2395                 m = (struct mdt_object *)o;
2396         else
2397                 m = mdt_obj(o);
2398
2399         RETURN(m);
2400 }
2401
2402 /**
2403  * Asyncronous commit for mdt device.
2404  *
2405  * Pass asynchonous commit call down the MDS stack.
2406  *
2407  * \param env environment
2408  * \param mdt the mdt device
2409  */
2410 static void mdt_device_commit_async(const struct lu_env *env,
2411                                     struct mdt_device *mdt)
2412 {
2413         struct dt_device *dt = mdt->mdt_bottom;
2414         int rc;
2415
2416         rc = dt->dd_ops->dt_commit_async(env, dt);
2417         if (unlikely(rc != 0))
2418                 CWARN("async commit start failed with rc = %d", rc);
2419 }
2420
2421 /**
2422  * Mark the lock as "synchonous".
2423  *
2424  * Mark the lock to deffer transaction commit to the unlock time.
2425  *
2426  * \param lock the lock to mark as "synchonous"
2427  *
2428  * \see mdt_is_lock_sync
2429  * \see mdt_save_lock
2430  */
2431 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2432 {
2433         lock->l_ast_data = (void*)1;
2434 }
2435
2436 /**
2437  * Check whehter the lock "synchonous" or not.
2438  *
2439  * \param lock the lock to check
2440  * \retval 1 the lock is "synchonous"
2441  * \retval 0 the lock isn't "synchronous"
2442  *
2443  * \see mdt_set_lock_sync
2444  * \see mdt_save_lock
2445  */
2446 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2447 {
2448         return lock->l_ast_data != NULL;
2449 }
2450
2451 /**
2452  * Blocking AST for mdt locks.
2453  *
2454  * Starts transaction commit if in case of COS lock conflict or
2455  * deffers such a commit to the mdt_save_lock.
2456  *
2457  * \param lock the lock which blocks a request or cancelling lock
2458  * \param desc unused
2459  * \param data unused
2460  * \param flag indicates whether this cancelling or blocking callback
2461  * \retval 0
2462  * \see ldlm_blocking_ast_nocheck
2463  */
2464 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2465                      void *data, int flag)
2466 {
2467         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2468         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2469         int rc;
2470         ENTRY;
2471
2472         if (flag == LDLM_CB_CANCELING)
2473                 RETURN(0);
2474         lock_res_and_lock(lock);
2475         if (lock->l_blocking_ast != mdt_blocking_ast) {
2476                 unlock_res_and_lock(lock);
2477                 RETURN(0);
2478         }
2479         if (mdt_cos_is_enabled(mdt) &&
2480             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2481             lock->l_blocking_lock != NULL &&
2482             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2483                 mdt_set_lock_sync(lock);
2484         }
2485         rc = ldlm_blocking_ast_nocheck(lock);
2486
2487         /* There is no lock conflict if l_blocking_lock == NULL,
2488          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2489          * when the last reference to a local lock was released */
2490         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2491                 struct lu_env env;
2492
2493                 rc = lu_env_init(&env, LCT_LOCAL);
2494                 if (unlikely(rc != 0))
2495                         CWARN("lu_env initialization failed with rc = %d,"
2496                               "cannot start asynchronous commit\n", rc);
2497                 else
2498                         mdt_device_commit_async(&env, mdt);
2499                 lu_env_fini(&env);
2500         }
2501         RETURN(rc);
2502 }
2503
2504 int mdt_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2505                         void *data, int flag)
2506 {
2507         struct lustre_handle lockh;
2508         int               rc;
2509
2510         switch (flag) {
2511         case LDLM_CB_BLOCKING:
2512                 ldlm_lock2handle(lock, &lockh);
2513                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2514                 if (rc < 0) {
2515                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2516                         RETURN(rc);
2517                 }
2518                 break;
2519         case LDLM_CB_CANCELING:
2520                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2521                 break;
2522         default:
2523                 LBUG();
2524         }
2525         RETURN(0);
2526 }
2527
2528 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2529                            struct mdt_object *o, struct lustre_handle *lh,
2530                            ldlm_mode_t mode, __u64 ibits)
2531 {
2532         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2533         ldlm_policy_data_t *policy = &mti->mti_policy;
2534         int rc = 0;
2535         ENTRY;
2536
2537         LASSERT(mdt_object_remote(o));
2538
2539         LASSERT((ibits & MDS_INODELOCK_UPDATE));
2540
2541         memset(einfo, 0, sizeof(*einfo));
2542         einfo->ei_type = LDLM_IBITS;
2543         einfo->ei_mode = mode;
2544         einfo->ei_cb_bl = mdt_md_blocking_ast;
2545         einfo->ei_cb_cp = ldlm_completion_ast;
2546
2547         memset(policy, 0, sizeof(*policy));
2548         policy->l_inodebits.bits = ibits;
2549
2550         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2551                             policy);
2552         RETURN(rc);
2553 }
2554
2555 static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o,
2556                             struct mdt_lock_handle *lh, __u64 ibits,
2557                             bool nonblock, int locality)
2558 {
2559         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2560         ldlm_policy_data_t *policy = &info->mti_policy;
2561         struct ldlm_res_id *res_id = &info->mti_res_id;
2562         __u64 dlmflags;
2563         int rc;
2564         ENTRY;
2565
2566         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2567         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2568         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2569         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2570
2571         if (mdt_object_remote(o)) {
2572                 if (locality == MDT_CROSS_LOCK) {
2573                         ibits &= ~(MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM);
2574                         ibits |= MDS_INODELOCK_LOOKUP;
2575                 } else {
2576                         LASSERTF(!(ibits &
2577                                   (MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM)),
2578                                 "%s: wrong bit "LPX64" for remote obj "DFID"\n",
2579                                 mdt_obd_name(info->mti_mdt), ibits,
2580                                 PFID(mdt_object_fid(o)));
2581                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2582                 }
2583                 /* No PDO lock on remote object */
2584                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2585         }
2586
2587         if (lh->mlh_type == MDT_PDO_LOCK) {
2588                 /* check for exists after object is locked */
2589                 if (mdt_object_exists(o) == 0) {
2590                         /* Non-existent object shouldn't have PDO lock */
2591                         RETURN(-ESTALE);
2592                 } else {
2593                         /* Non-dir object shouldn't have PDO lock */
2594                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2595                                 RETURN(-ENOTDIR);
2596                 }
2597         }
2598
2599         memset(policy, 0, sizeof(*policy));
2600         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2601
2602         dlmflags = LDLM_FL_ATOMIC_CB;
2603         if (nonblock)
2604                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2605
2606         /*
2607          * Take PDO lock on whole directory and build correct @res_id for lock
2608          * on part of directory.
2609          */
2610         if (lh->mlh_pdo_hash != 0) {
2611                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2612                 mdt_lock_pdo_mode(info, o, lh);
2613                 if (lh->mlh_pdo_mode != LCK_NL) {
2614                         /*
2615                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2616                          * is never going to be sent to client and we do not
2617                          * want it slowed down due to possible cancels.
2618                          */
2619                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2620                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2621                                           policy, res_id, dlmflags,
2622                                           &info->mti_exp->exp_handle.h_cookie);
2623                         if (unlikely(rc))
2624                                 RETURN(rc);
2625                 }
2626
2627                 /*
2628                  * Finish res_id initializing by name hash marking part of
2629                  * directory which is taking modification.
2630                  */
2631                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2632         }
2633
2634         policy->l_inodebits.bits = ibits;
2635
2636         /*
2637          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2638          * going to be sent to client. If it is - mdt_intent_policy() path will
2639          * fix it up and turn FL_LOCAL flag off.
2640          */
2641         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2642                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2643                           &info->mti_exp->exp_handle.h_cookie);
2644         if (rc)
2645                 mdt_object_unlock(info, o, lh, 1);
2646         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2647                  lh->mlh_pdo_hash != 0 &&
2648                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2649                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2650         }
2651
2652         RETURN(rc);
2653 }
2654
2655 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2656                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2657 {
2658         return mdt_object_lock0(info, o, lh, ibits, false, locality);
2659 }
2660
2661 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2662                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2663 {
2664         struct mdt_lock_handle tmp = *lh;
2665         int rc;
2666
2667         rc = mdt_object_lock0(info, o, &tmp, ibits, true, locality);
2668         if (rc == 0)
2669                 *lh = tmp;
2670
2671         return rc == 0;
2672 }
2673
2674 /**
2675  * Save a lock within request object.
2676  *
2677  * Keep the lock referenced until whether client ACK or transaction
2678  * commit happens or release the lock immediately depending on input
2679  * parameters. If COS is ON, a write lock is converted to COS lock
2680  * before saving.
2681  *
2682  * \param info thead info object
2683  * \param h lock handle
2684  * \param mode lock mode
2685  * \param decref force immediate lock releasing
2686  */
2687 static
2688 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2689                    ldlm_mode_t mode, int decref)
2690 {
2691         ENTRY;
2692
2693         if (lustre_handle_is_used(h)) {
2694                 if (decref || !info->mti_has_trans ||
2695                     !(mode & (LCK_PW | LCK_EX))){
2696                         mdt_fid_unlock(h, mode);
2697                 } else {
2698                         struct mdt_device *mdt = info->mti_mdt;
2699                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2700                         struct ptlrpc_request *req = mdt_info_req(info);
2701                         int no_ack = 0;
2702
2703                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2704                                  h->cookie);
2705                         CDEBUG(D_HA, "request = %p reply state = %p"
2706                                " transno = "LPD64"\n",
2707                                req, req->rq_reply_state, req->rq_transno);
2708                         if (mdt_cos_is_enabled(mdt)) {
2709                                 no_ack = 1;
2710                                 ldlm_lock_downgrade(lock, LCK_COS);
2711                                 mode = LCK_COS;
2712                         }
2713                         ptlrpc_save_lock(req, h, mode, no_ack);
2714                         if (mdt_is_lock_sync(lock)) {
2715                                 CDEBUG(D_HA, "found sync-lock,"
2716                                        " async commit started\n");
2717                                 mdt_device_commit_async(info->mti_env,
2718                                                         mdt);
2719                         }
2720                         LDLM_LOCK_PUT(lock);
2721                 }
2722                 h->cookie = 0ull;
2723         }
2724
2725         EXIT;
2726 }
2727
2728 /**
2729  * Unlock mdt object.
2730  *
2731  * Immeditely release the regular lock and the PDO lock or save the
2732  * lock in reqeuest and keep them referenced until client ACK or
2733  * transaction commit.
2734  *
2735  * \param info thread info object
2736  * \param o mdt object
2737  * \param lh mdt lock handle referencing regular and PDO locks
2738  * \param decref force immediate lock releasing
2739  */
2740 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2741                        struct mdt_lock_handle *lh, int decref)
2742 {
2743         ENTRY;
2744
2745         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2746         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2747
2748         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2749                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2750
2751         EXIT;
2752 }
2753
2754 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2755                                         const struct lu_fid *f,
2756                                         struct mdt_lock_handle *lh,
2757                                         __u64 ibits)
2758 {
2759         struct mdt_object *o;
2760
2761         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2762         if (!IS_ERR(o)) {
2763                 int rc;
2764
2765                 rc = mdt_object_lock(info, o, lh, ibits,
2766                                      MDT_LOCAL_LOCK);
2767                 if (rc != 0) {
2768                         mdt_object_put(info->mti_env, o);
2769                         o = ERR_PTR(rc);
2770                 }
2771         }
2772         return o;
2773 }
2774
2775 void mdt_object_unlock_put(struct mdt_thread_info * info,
2776                            struct mdt_object * o,
2777                            struct mdt_lock_handle *lh,
2778                            int decref)
2779 {
2780         mdt_object_unlock(info, o, lh, decref);
2781         mdt_object_put(info->mti_env, o);
2782 }
2783
2784 struct mdt_handler *mdt_handler_find(__u32 opc, struct mdt_opc_slice *supported)
2785 {
2786         struct mdt_opc_slice *s;
2787         struct mdt_handler   *h;
2788
2789         h = NULL;
2790         for (s = supported; s->mos_hs != NULL; s++) {
2791                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2792                         h = s->mos_hs + (opc - s->mos_opc_start);
2793                         if (likely(h->mh_opc != 0))
2794                                 LASSERTF(h->mh_opc == opc,
2795                                          "opcode mismatch %d != %d\n",
2796                                          h->mh_opc, opc);
2797                         else
2798                                 h = NULL; /* unsupported opc */
2799                         break;
2800                 }
2801         }
2802         return h;
2803 }
2804
2805 static int mdt_lock_resname_compat(struct mdt_device *m,
2806                                    struct ldlm_request *req)
2807 {
2808         /* XXX something... later. */
2809         return 0;
2810 }
2811
2812 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2813 {
2814         /* XXX something... later. */
2815         return 0;
2816 }
2817
2818 /*
2819  * Generic code handling requests that have struct mdt_body passed in:
2820  *
2821  *  - extract mdt_body from request and save it in @info, if present;
2822  *
2823  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2824  *  @info;
2825  *
2826  *  - if HABEO_CORPUS flag is set for this request type check whether object
2827  *  actually exists on storage (lu_object_exists()).
2828  *
2829  */
2830 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2831 {
2832         const struct mdt_body    *body;
2833         struct mdt_object        *obj;
2834         const struct lu_env      *env;
2835         struct req_capsule       *pill;
2836         int                       rc;
2837         ENTRY;
2838
2839         env = info->mti_env;
2840         pill = info->mti_pill;
2841
2842         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2843         if (body == NULL)
2844                 RETURN(-EFAULT);
2845
2846         if (!(body->valid & OBD_MD_FLID))
2847                 RETURN(0);
2848
2849         if (!fid_is_sane(&body->fid1)) {
2850                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2851                 RETURN(-EINVAL);
2852         }
2853
2854         /*
2855          * Do not get size or any capa fields before we check that request
2856          * contains capa actually. There are some requests which do not, for
2857          * instance MDS_IS_SUBDIR.
2858          */
2859         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2860             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2861                 mdt_set_capainfo(info, 0, &body->fid1,
2862                                  req_capsule_client_get(pill, &RMF_CAPA1));
2863
2864         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2865         if (!IS_ERR(obj)) {
2866                 if ((flags & HABEO_CORPUS) &&
2867                     !mdt_object_exists(obj)) {
2868                         mdt_object_put(env, obj);
2869                         /* for capability renew ENOENT will be handled in
2870                          * mdt_renew_capa */
2871                         if (body->valid & OBD_MD_FLOSSCAPA)
2872                                 rc = 0;
2873                         else
2874                                 rc = -ENOENT;
2875                 } else {
2876                         info->mti_object = obj;
2877                         rc = 0;
2878                 }
2879         } else
2880                 rc = PTR_ERR(obj);
2881
2882         RETURN(rc);
2883 }
2884
2885 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2886 {
2887         struct req_capsule *pill = info->mti_pill;
2888         int rc;
2889         ENTRY;
2890
2891         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2892                 rc = mdt_body_unpack(info, flags);
2893         else
2894                 rc = 0;
2895
2896         if (rc == 0 && (flags & HABEO_REFERO)) {
2897                 /* Pack reply. */
2898                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2899                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2900                                              info->mti_body->eadatasize);
2901                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2902                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2903                                              RCL_SERVER, 0);
2904
2905                 rc = req_capsule_server_pack(pill);
2906         }
2907         RETURN(rc);
2908 }
2909
2910 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2911 {
2912         struct md_device *next = m->mdt_child;
2913
2914         return next->md_ops->mdo_init_capa_ctxt(env, next,
2915                                                 m->mdt_opts.mo_mds_capa,
2916                                                 m->mdt_capa_timeout,
2917                                                 m->mdt_capa_alg,
2918                                                 m->mdt_capa_keys);
2919 }
2920
2921 /*
2922  * Invoke handler for this request opc. Also do necessary preprocessing
2923  * (according to handler ->mh_flags), and post-processing (setting of
2924  * ->last_{xid,committed}).
2925  */
2926 static int mdt_req_handle(struct mdt_thread_info *info,
2927                           struct mdt_handler *h, struct ptlrpc_request *req)
2928 {
2929         int   rc, serious = 0;
2930         __u32 flags;
2931
2932         ENTRY;
2933
2934         LASSERT(h->mh_act != NULL);
2935         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2936         LASSERT(current->journal_info == NULL);
2937
2938         /*
2939          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2940          * to put same checks into handlers like mdt_close(), mdt_reint(),
2941          * etc., without talking to mdt authors first. Checking same thing
2942          * there again is useless and returning 0 error without packing reply
2943          * is buggy! Handlers either pack reply or return error.
2944          *
2945          * We return 0 here and do not send any reply in order to emulate
2946          * network failure. Do not send any reply in case any of NET related
2947          * fail_id has occured.
2948          */
2949         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2950                 RETURN(0);
2951
2952         rc = 0;
2953         flags = h->mh_flags;
2954         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2955
2956         if (h->mh_fmt != NULL) {
2957                 req_capsule_set(info->mti_pill, h->mh_fmt);
2958                 rc = mdt_unpack_req_pack_rep(info, flags);
2959         }
2960
2961         if (rc == 0 && flags & MUTABOR &&
2962             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
2963                 /* should it be rq_status? */
2964                 rc = -EROFS;
2965
2966         if (rc == 0 && flags & HABEO_CLAVIS) {
2967                 struct ldlm_request *dlm_req;
2968
2969                 LASSERT(h->mh_fmt != NULL);
2970
2971                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2972                 if (dlm_req != NULL) {
2973                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2974                                         LDLM_IBITS &&
2975                                      dlm_req->lock_desc.l_policy_data.\
2976                                         l_inodebits.bits == 0)) {
2977                                 /*
2978                                  * Lock without inodebits makes no sense and
2979                                  * will oops later in ldlm. If client miss to
2980                                  * set such bits, do not trigger ASSERTION.
2981                                  *
2982                                  * For liblustre flock case, it maybe zero.
2983                                  */
2984                                 rc = -EPROTO;
2985                         } else {
2986                                 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2987                                         rc = mdt_lock_resname_compat(
2988                                                                 info->mti_mdt,
2989                                                                 dlm_req);
2990                                 info->mti_dlm_req = dlm_req;
2991                         }
2992                 } else {
2993                         rc = -EFAULT;
2994                 }
2995         }
2996
2997         /* capability setting changed via /proc, needs reinitialize ctxt */
2998         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2999                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
3000                 info->mti_mdt->mdt_capa_conf = 0;
3001         }
3002
3003         if (likely(rc == 0)) {
3004                 /*
3005                  * Process request, there can be two types of rc:
3006                  * 1) errors with msg unpack/pack, other failures outside the
3007                  * operation itself. This is counted as serious errors;
3008                  * 2) errors during fs operation, should be placed in rq_status
3009                  * only
3010                  */
3011                 rc = h->mh_act(info);
3012                 if (rc == 0 &&
3013                     !req->rq_no_reply && req->rq_reply_state == NULL) {
3014                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
3015                                   "pack reply and returned 0 error\n",
3016                                   h->mh_name);
3017                         LBUG();
3018                 }
3019                 serious = is_serious(rc);
3020                 rc = clear_serious(rc);
3021         } else
3022                 serious = 1;
3023
3024         req->rq_status = rc;
3025
3026         /*
3027          * ELDLM_* codes which > 0 should be in rq_status only as well as
3028          * all non-serious errors.
3029          */
3030         if (rc > 0 || !serious)
3031                 rc = 0;
3032
3033         LASSERT(current->journal_info == NULL);
3034
3035         if (rc == 0 && (flags & HABEO_CLAVIS) &&
3036             info->mti_mdt->mdt_opts.mo_compat_resname) {
3037                 struct ldlm_reply *dlmrep;
3038
3039                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3040                 if (dlmrep != NULL)
3041                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
3042         }
3043
3044         /* If we're DISCONNECTing, the mdt_export_data is already freed */
3045         if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
3046                 target_committed_to_req(req);
3047
3048         if (unlikely(req_is_replay(req) &&
3049                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
3050                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
3051                 LBUG();
3052         }
3053
3054         target_send_reply(req, rc, info->mti_fail_id);
3055         RETURN(0);
3056 }
3057
3058 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3059 {
3060         lh->mlh_type = MDT_NUL_LOCK;
3061         lh->mlh_reg_lh.cookie = 0ull;
3062         lh->mlh_reg_mode = LCK_MINMODE;
3063         lh->mlh_pdo_lh.cookie = 0ull;
3064         lh->mlh_pdo_mode = LCK_MINMODE;
3065         lh->mlh_rreg_lh.cookie = 0ull;
3066         lh->mlh_rreg_mode = LCK_MINMODE;
3067 }
3068
3069 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3070 {
3071         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3072         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3073 }
3074
3075 /*
3076  * Initialize fields of struct mdt_thread_info. Other fields are left in
3077  * uninitialized state, because it's too expensive to zero out whole
3078  * mdt_thread_info (> 1K) on each request arrival.
3079  */
3080 static void mdt_thread_info_init(struct ptlrpc_request *req,
3081                                  struct mdt_thread_info *info)
3082 {
3083         int i;
3084
3085         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
3086         info->mti_pill = &req->rq_pill;
3087
3088         /* lock handle */
3089         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3090                 mdt_lock_handle_init(&info->mti_lh[i]);
3091
3092         /* mdt device: it can be NULL while CONNECT */
3093         if (req->rq_export) {
3094                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3095                 info->mti_exp = req->rq_export;
3096         } else
3097                 info->mti_mdt = NULL;
3098         info->mti_env = req->rq_svc_thread->t_env;
3099         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
3100         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3101         info->mti_mos = NULL;
3102
3103         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3104         info->mti_big_buf = LU_BUF_NULL;
3105         info->mti_body = NULL;
3106         info->mti_object = NULL;
3107         info->mti_dlm_req = NULL;
3108         info->mti_has_trans = 0;
3109         info->mti_cross_ref = 0;
3110         info->mti_opdata = 0;
3111         info->mti_big_lmm_used = 0;
3112
3113         /* To not check for split by default. */
3114         info->mti_spec.no_create = 0;
3115         info->mti_spec.sp_rm_entry = 0;
3116 }
3117
3118 static void mdt_thread_info_fini(struct mdt_thread_info *info)
3119 {
3120         int i;
3121
3122         req_capsule_fini(info->mti_pill);
3123         if (info->mti_object != NULL) {
3124                 mdt_object_put(info->mti_env, info->mti_object);
3125                 info->mti_object = NULL;
3126         }
3127
3128         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3129                 mdt_lock_handle_fini(&info->mti_lh[i]);
3130         info->mti_env = NULL;
3131
3132         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3133                 lu_buf_free(&info->mti_big_buf);
3134 }
3135
3136 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
3137                                        struct obd_device *obd, int *process)
3138 {
3139         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3140         case MDS_CONNECT: /* This will never get here, but for completeness. */
3141         case OST_CONNECT: /* This will never get here, but for completeness. */
3142         case MDS_DISCONNECT:
3143         case OST_DISCONNECT:
3144         case OBD_IDX_READ:
3145                *process = 1;
3146                RETURN(0);
3147
3148         case MDS_CLOSE:
3149         case MDS_DONE_WRITING:
3150         case MDS_SYNC: /* used in unmounting */
3151         case OBD_PING:
3152         case MDS_REINT:
3153         case UPDATE_OBJ:
3154         case SEQ_QUERY:
3155         case FLD_QUERY:
3156         case LDLM_ENQUEUE:
3157                 *process = target_queue_recovery_request(req, obd);
3158                 RETURN(0);
3159
3160         default:
3161                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
3162                 *process = -EAGAIN;
3163                 RETURN(0);
3164         }
3165 }
3166
3167 /*
3168  * Handle recovery. Return:
3169  *        +1: continue request processing;
3170  *       -ve: abort immediately with the given error code;
3171  *         0: send reply with error code in req->rq_status;
3172  */
3173 static int mdt_recovery(struct mdt_thread_info *info)
3174 {
3175         struct ptlrpc_request *req = mdt_info_req(info);
3176         struct obd_device *obd;
3177
3178         ENTRY;
3179
3180         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3181         case MDS_CONNECT:
3182         case SEC_CTX_INIT:
3183         case SEC_CTX_INIT_CONT:
3184         case SEC_CTX_FINI:
3185                 {
3186 #if 0
3187                         int rc;
3188
3189                         rc = mdt_handle_idmap(info);
3190                         if (rc)
3191                                 RETURN(rc);
3192                         else
3193 #endif
3194                                 RETURN(+1);
3195                 }
3196         }
3197
3198         if (unlikely(!class_connected_export(req->rq_export))) {
3199                 CDEBUG(D_HA, "operation %d on unconnected MDS from %s\n",
3200                        lustre_msg_get_opc(req->rq_reqmsg),
3201                        libcfs_id2str(req->rq_peer));
3202                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
3203                  * mds_A will get -ENOTCONN(especially for ping req),
3204                  * which will cause that mds_A deactive timeout, then when
3205                  * mds_A cleanup, the cleanup process will be suspended since
3206                  * deactive timeout is not zero.
3207                  */
3208                 req->rq_status = -ENOTCONN;
3209                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
3210                 RETURN(0);
3211         }
3212
3213         /* sanity check: if the xid matches, the request must be marked as a
3214          * resent or replayed */
3215         if (req_xid_is_last(req)) {
3216                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
3217                       (MSG_RESENT | MSG_REPLAY))) {
3218                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
3219                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
3220                   &nbs