Whamcloud - gitweb
f2fc8f00ba3281c161d80104d263444c1c9ca104
[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
72 mdl_mode_t mdt_mdl_lock_modes[] = {
73         [LCK_MINMODE] = MDL_MINMODE,
74         [LCK_EX]      = MDL_EX,
75         [LCK_PW]      = MDL_PW,
76         [LCK_PR]      = MDL_PR,
77         [LCK_CW]      = MDL_CW,
78         [LCK_CR]      = MDL_CR,
79         [LCK_NL]      = MDL_NL,
80         [LCK_GROUP]   = MDL_GROUP
81 };
82
83 ldlm_mode_t mdt_dlm_lock_modes[] = {
84         [MDL_MINMODE] = LCK_MINMODE,
85         [MDL_EX]      = LCK_EX,
86         [MDL_PW]      = LCK_PW,
87         [MDL_PR]      = LCK_PR,
88         [MDL_CW]      = LCK_CW,
89         [MDL_CR]      = LCK_CR,
90         [MDL_NL]      = LCK_NL,
91         [MDL_GROUP]   = LCK_GROUP
92 };
93
94
95 static struct mdt_device *mdt_dev(struct lu_device *d);
96 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
97 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
98                         struct getinfo_fid2path *fp);
99
100 static const struct lu_object_operations mdt_obj_ops;
101
102 /* Slab for MDT object allocation */
103 static cfs_mem_cache_t *mdt_object_kmem;
104
105 static struct lu_kmem_descr mdt_caches[] = {
106         {
107                 .ckd_cache = &mdt_object_kmem,
108                 .ckd_name  = "mdt_obj",
109                 .ckd_size  = sizeof(struct mdt_object)
110         },
111         {
112                 .ckd_cache = NULL
113         }
114 };
115
116 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
117 {
118         if (!rep)
119                 return 0;
120         return (rep->lock_policy_res1 & flag);
121 }
122
123 void mdt_clear_disposition(struct mdt_thread_info *info,
124                            struct ldlm_reply *rep, int flag)
125 {
126         if (info)
127                 info->mti_opdata &= ~flag;
128         if (rep)
129                 rep->lock_policy_res1 &= ~flag;
130 }
131
132 void mdt_set_disposition(struct mdt_thread_info *info,
133                          struct ldlm_reply *rep, int flag)
134 {
135         if (info)
136                 info->mti_opdata |= flag;
137         if (rep)
138                 rep->lock_policy_res1 |= flag;
139 }
140
141 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
142 {
143         lh->mlh_pdo_hash = 0;
144         lh->mlh_reg_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_type = MDT_PDO_LOCK;
153
154         if (name != NULL && (name[0] != '\0')) {
155                 LASSERT(namelen > 0);
156                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
157         } else {
158                 LASSERT(namelen == 0);
159                 lh->mlh_pdo_hash = 0ull;
160         }
161 }
162
163 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
164                               struct mdt_lock_handle *lh)
165 {
166         mdl_mode_t mode;
167         ENTRY;
168
169         /*
170          * Any dir access needs couple of locks:
171          *
172          * 1) on part of dir we gonna take lookup/modify;
173          *
174          * 2) on whole dir to protect it from concurrent splitting and/or to
175          * flush client's cache for readdir().
176          *
177          * so, for a given mode and object this routine decides what lock mode
178          * to use for lock #2:
179          *
180          * 1) if caller's gonna lookup in dir then we need to protect dir from
181          * being splitted only - LCK_CR
182          *
183          * 2) if caller's gonna modify dir then we need to protect dir from
184          * being splitted and to flush cache - LCK_CW
185          *
186          * 3) if caller's gonna modify dir and that dir seems ready for
187          * splitting then we need to protect it from any type of access
188          * (lookup/modify/split) - LCK_EX --bzzz
189          */
190
191         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
192         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
193
194         /*
195          * Ask underlaying level its opinion about preferable PDO lock mode
196          * having access type passed as regular lock mode:
197          *
198          * - MDL_MINMODE means that lower layer does not want to specify lock
199          * mode;
200          *
201          * - MDL_NL means that no PDO lock should be taken. This is used in some
202          * cases. Say, for non-splittable directories no need to use PDO locks
203          * at all.
204          */
205         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
206                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
207
208         if (mode != MDL_MINMODE) {
209                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
210         } else {
211                 /*
212                  * Lower layer does not want to specify locking mode. We do it
213                  * our selves. No special protection is needed, just flush
214                  * client's cache on modification and allow concurrent
215                  * mondification.
216                  */
217                 switch (lh->mlh_reg_mode) {
218                 case LCK_EX:
219                         lh->mlh_pdo_mode = LCK_EX;
220                         break;
221                 case LCK_PR:
222                         lh->mlh_pdo_mode = LCK_CR;
223                         break;
224                 case LCK_PW:
225                         lh->mlh_pdo_mode = LCK_CW;
226                         break;
227                 default:
228                         CERROR("Not expected lock type (0x%x)\n",
229                                (int)lh->mlh_reg_mode);
230                         LBUG();
231                 }
232         }
233
234         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
235         EXIT;
236 }
237
238 int mdt_getstatus(struct mdt_thread_info *info)
239 {
240         struct mdt_device *mdt  = info->mti_mdt;
241         struct md_device  *next = mdt->mdt_child;
242         struct mdt_body   *repbody;
243         int                rc;
244
245         ENTRY;
246
247         rc = mdt_check_ucred(info);
248         if (rc)
249                 RETURN(err_serious(rc));
250
251         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
252                 RETURN(err_serious(-ENOMEM));
253
254         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
255         rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
256         if (rc != 0)
257                 RETURN(rc);
258
259         repbody->valid |= OBD_MD_FLID;
260
261         if (mdt->mdt_opts.mo_mds_capa &&
262             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
263                 struct mdt_object  *root;
264                 struct lustre_capa *capa;
265
266                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
267                 if (IS_ERR(root))
268                         RETURN(PTR_ERR(root));
269
270                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
271                 LASSERT(capa);
272                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
273                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
274                                  0);
275                 mdt_object_put(info->mti_env, root);
276                 if (rc == 0)
277                         repbody->valid |= OBD_MD_FLMDSCAPA;
278         }
279
280         RETURN(rc);
281 }
282
283 int mdt_statfs(struct mdt_thread_info *info)
284 {
285         struct ptlrpc_request           *req = mdt_info_req(info);
286         struct md_device                *next = info->mti_mdt->mdt_child;
287         struct ptlrpc_service_part      *svcpt;
288         struct obd_statfs               *osfs;
289         int                             rc;
290
291         ENTRY;
292
293         svcpt = info->mti_pill->rc_req->rq_rqbd->rqbd_svcpt;
294
295         /* This will trigger a watchdog timeout */
296         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
297                          (MDT_SERVICE_WATCHDOG_FACTOR *
298                           at_get(&svcpt->scp_at_estimate)) + 1);
299
300         rc = mdt_check_ucred(info);
301         if (rc)
302                 RETURN(err_serious(rc));
303
304         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
305                 RETURN(err_serious(-ENOMEM));
306
307         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
308         if (!osfs)
309                 RETURN(-EPROTO);
310
311         /** statfs information are cached in the mdt_device */
312         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
313                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
314                 /** statfs data is too old, get up-to-date one */
315                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
316                 if (rc)
317                         RETURN(rc);
318                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
319                 info->mti_mdt->mdt_osfs = *osfs;
320                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
321                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
322         } else {
323                 /** use cached statfs data */
324                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
325                 *osfs = info->mti_mdt->mdt_osfs;
326                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
327         }
328
329         if (rc == 0)
330                 mdt_counter_incr(req, LPROC_MDT_STATFS);
331
332         RETURN(rc);
333 }
334
335 /**
336  * Pack SOM attributes into the reply.
337  * Call under a DLM UPDATE lock.
338  */
339 static void mdt_pack_size2body(struct mdt_thread_info *info,
340                                struct mdt_object *mo)
341 {
342         struct mdt_body *b;
343         struct md_attr *ma = &info->mti_attr;
344
345         LASSERT(ma->ma_attr.la_valid & LA_MODE);
346         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
347
348         /* Check if Size-on-MDS is supported, if this is a regular file,
349          * if SOM is enabled on the object and if SOM cache exists and valid.
350          * Otherwise do not pack Size-on-MDS attributes to the reply. */
351         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
352             !S_ISREG(ma->ma_attr.la_mode) ||
353             !mdt_object_is_som_enabled(mo) ||
354             !(ma->ma_valid & MA_SOM))
355                 return;
356
357         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
358         b->size = ma->ma_som->msd_size;
359         b->blocks = ma->ma_som->msd_blocks;
360 }
361
362 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
363                         const struct lu_attr *attr, const struct lu_fid *fid)
364 {
365         struct md_attr *ma = &info->mti_attr;
366
367         LASSERT(ma->ma_valid & MA_INODE);
368
369         b->atime      = attr->la_atime;
370         b->mtime      = attr->la_mtime;
371         b->ctime      = attr->la_ctime;
372         b->mode       = attr->la_mode;
373         b->size       = attr->la_size;
374         b->blocks     = attr->la_blocks;
375         b->uid        = attr->la_uid;
376         b->gid        = attr->la_gid;
377         b->flags      = attr->la_flags;
378         b->nlink      = attr->la_nlink;
379         b->rdev       = attr->la_rdev;
380
381         /*XXX should pack the reply body according to lu_valid*/
382         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
383                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
384                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
385                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
386
387         if (!S_ISREG(attr->la_mode)) {
388                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
389         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
390                 /* means no objects are allocated on osts. */
391                 LASSERT(!(ma->ma_valid & MA_LOV));
392                 /* just ignore blocks occupied by extend attributes on MDS */
393                 b->blocks = 0;
394                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
395                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
396         }
397
398         if (fid) {
399                 b->fid1 = *fid;
400                 b->valid |= OBD_MD_FLID;
401
402                 /* FIXME: these should be fixed when new igif ready.*/
403                 b->ino  =  fid_oid(fid);       /* 1.6 compatibility */
404                 b->generation = fid_ver(fid);  /* 1.6 compatibility */
405                 b->valid |= OBD_MD_FLGENER;    /* 1.6 compatibility */
406
407                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
408                                 PFID(fid), b->nlink, b->mode, b->size);
409         }
410
411         if (info)
412                 mdt_body_reverse_idmap(info, b);
413
414         if (b->valid & OBD_MD_FLSIZE)
415                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
416                        PFID(fid), (unsigned long long)b->size);
417 }
418
419 static inline int mdt_body_has_lov(const struct lu_attr *la,
420                                    const struct mdt_body *body)
421 {
422         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
423                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
424 }
425
426 void mdt_client_compatibility(struct mdt_thread_info *info)
427 {
428         struct mdt_body       *body;
429         struct ptlrpc_request *req = mdt_info_req(info);
430         struct obd_export     *exp = req->rq_export;
431         struct md_attr        *ma = &info->mti_attr;
432         struct lu_attr        *la = &ma->ma_attr;
433         ENTRY;
434
435         if (exp->exp_connect_flags & OBD_CONNECT_LAYOUTLOCK)
436                 /* the client can deal with 16-bit lmm_stripe_count */
437                 RETURN_EXIT;
438
439         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
440
441         if (!mdt_body_has_lov(la, body))
442                 RETURN_EXIT;
443
444         /* now we have a reply with a lov for a client not compatible with the
445          * layout lock so we have to clean the layout generation number */
446         if (S_ISREG(la->la_mode))
447                 ma->ma_lmm->lmm_layout_gen = 0;
448         EXIT;
449 }
450
451 static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
452                              char *name)
453 {
454         const struct lu_env *env = info->mti_env;
455         int rc;
456         ENTRY;
457
458         LASSERT(info->mti_big_lmm_used == 0);
459         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
460         if (rc < 0)
461                 RETURN(rc);
462
463         /* big_lmm may need to be grown */
464         if (info->mti_big_lmmsize < rc) {
465                 int size = size_roundup_power2(rc);
466
467                 if (info->mti_big_lmmsize > 0) {
468                         /* free old buffer */
469                         LASSERT(info->mti_big_lmm);
470                         OBD_FREE_LARGE(info->mti_big_lmm,
471                                        info->mti_big_lmmsize);
472                         info->mti_big_lmm = NULL;
473                         info->mti_big_lmmsize = 0;
474                 }
475
476                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
477                 if (info->mti_big_lmm == NULL)
478                         RETURN(-ENOMEM);
479                 info->mti_big_lmmsize = size;
480         }
481         LASSERT(info->mti_big_lmmsize >= rc);
482
483         info->mti_buf.lb_buf = info->mti_big_lmm;
484         info->mti_buf.lb_len = info->mti_big_lmmsize;
485         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
486
487         RETURN(rc);
488 }
489
490 int mdt_attr_get_lov(struct mdt_thread_info *info,
491                      struct mdt_object *o, struct md_attr *ma)
492 {
493         struct md_object *next = mdt_object_child(o);
494         struct lu_buf    *buf = &info->mti_buf;
495         int rc;
496
497         buf->lb_buf = ma->ma_lmm;
498         buf->lb_len = ma->ma_lmm_size;
499         rc = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_LOV);
500         if (rc > 0) {
501                 ma->ma_lmm_size = rc;
502                 ma->ma_valid |= MA_LOV;
503                 rc = 0;
504         } else if (rc == -ENODATA) {
505                 /* no LOV EA */
506                 rc = 0;
507         } else if (rc == -ERANGE) {
508                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LOV);
509                 if (rc > 0) {
510                         info->mti_big_lmm_used = 1;
511                         ma->ma_valid |= MA_LOV;
512                         ma->ma_lmm = info->mti_big_lmm;
513                         ma->ma_lmm_size = rc;
514                         /* update mdt_max_mdsize so all clients
515                          * will be aware about that */
516                         if (info->mti_mdt->mdt_max_mdsize < rc)
517                                 info->mti_mdt->mdt_max_mdsize = rc;
518                         rc = 0;
519                 }
520         }
521
522         return rc;
523 }
524
525 int mdt_attr_get_pfid(struct mdt_thread_info *info,
526                       struct mdt_object *o, struct lu_fid *pfid)
527 {
528         struct lu_buf           *buf = &info->mti_buf;
529         struct link_ea_header   *leh;
530         struct link_ea_entry    *lee;
531         int                      rc;
532         ENTRY;
533
534         buf->lb_buf = info->mti_big_lmm;
535         buf->lb_len = info->mti_big_lmmsize;
536         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
537                           buf, XATTR_NAME_LINK);
538         /* ignore errors, MA_PFID won't be set and it is
539          * up to the caller to treat this as an error */
540         if (rc == -ERANGE || buf->lb_len == 0) {
541                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
542                 buf->lb_buf = info->mti_big_lmm;
543                 buf->lb_len = info->mti_big_lmmsize;
544         }
545
546         if (rc < 0)
547                 RETURN(rc);
548         if (rc < sizeof(*leh)) {
549                 CERROR("short LinkEA on "DFID": rc = %d\n",
550                        PFID(mdt_object_fid(o)), rc);
551                 RETURN(-ENODATA);
552         }
553
554         leh = (struct link_ea_header *) buf->lb_buf;
555         lee = (struct link_ea_entry *)(leh + 1);
556         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
557                 leh->leh_magic = LINK_EA_MAGIC;
558                 leh->leh_reccount = __swab32(leh->leh_reccount);
559                 leh->leh_len = __swab64(leh->leh_len);
560         }
561         if (leh->leh_magic != LINK_EA_MAGIC)
562                 RETURN(-EINVAL);
563         if (leh->leh_reccount == 0)
564                 RETURN(-ENODATA);
565
566         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
567         fid_be_to_cpu(pfid, pfid);
568
569         RETURN(0);
570 }
571
572 int mdt_attr_get_complex(struct mdt_thread_info *info,
573                          struct mdt_object *o, struct md_attr *ma)
574 {
575         const struct lu_env *env = info->mti_env;
576         struct md_object    *next = mdt_object_child(o);
577         struct lu_buf       *buf = &info->mti_buf;
578         u32                  mode = lu_object_attr(&next->mo_lu);
579         int                  need = ma->ma_need;
580         int                  rc = 0, rc2;
581         ENTRY;
582
583         ma->ma_valid = 0;
584
585         if (need & MA_INODE) {
586                 ma->ma_need = MA_INODE;
587                 rc = mo_attr_get(env, next, ma);
588                 if (rc)
589                         GOTO(out, rc);
590                 ma->ma_valid |= MA_INODE;
591         }
592
593         if (need & MA_PFID) {
594                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
595                 if (rc == 0)
596                         ma->ma_valid |= MA_PFID;
597                 /* ignore this error, parent fid is not mandatory */
598                 rc = 0;
599         }
600
601         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
602                 rc = mdt_attr_get_lov(info, o, ma);
603                 if (rc)
604                         GOTO(out, rc);
605         }
606
607         if (need & MA_LMV && S_ISDIR(mode)) {
608                 buf->lb_buf = ma->ma_lmv;
609                 buf->lb_len = ma->ma_lmv_size;
610                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LMV);
611                 if (rc2 > 0) {
612                         ma->ma_lmv_size = rc2;
613                         ma->ma_valid |= MA_LMV;
614                 } else if (rc2 == -ENODATA) {
615                         /* no LMV EA */
616                         ma->ma_lmv_size = 0;
617                 } else
618                         GOTO(out, rc = rc2);
619         }
620
621         if (need & MA_SOM && S_ISREG(mode)) {
622                 buf->lb_buf = info->mti_xattr_buf;
623                 buf->lb_len = sizeof(info->mti_xattr_buf);
624                 CLASSERT(sizeof(struct som_attrs) <=
625                          sizeof(info->mti_xattr_buf));
626                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
627                 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
628                 if (rc2 == 0)
629                         ma->ma_valid |= MA_SOM;
630                 else if (rc2 < 0 && rc2 != -ENODATA)
631                         GOTO(out, rc = rc2);
632         }
633
634         if (need & MA_HSM && S_ISREG(mode)) {
635                 buf->lb_buf = info->mti_xattr_buf;
636                 buf->lb_len = sizeof(info->mti_xattr_buf);
637                 CLASSERT(sizeof(struct hsm_attrs) <=
638                          sizeof(info->mti_xattr_buf));
639                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
640                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
641                 if (rc2 == 0)
642                         ma->ma_valid |= MA_HSM;
643                 else if (rc2 < 0 && rc2 != -ENODATA)
644                         GOTO(out, rc = rc2);
645         }
646
647 #ifdef CONFIG_FS_POSIX_ACL
648         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
649                 buf->lb_buf = ma->ma_acl;
650                 buf->lb_len = ma->ma_acl_size;
651                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
652                 if (rc2 > 0) {
653                         ma->ma_acl_size = rc2;
654                         ma->ma_valid |= MA_ACL_DEF;
655                 } else if (rc2 == -ENODATA) {
656                         /* no ACLs */
657                         ma->ma_acl_size = 0;
658                 } else
659                         GOTO(out, rc = rc2);
660         }
661 #endif
662 out:
663         ma->ma_need = need;
664         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
665                rc, ma->ma_valid, ma->ma_lmm);
666         RETURN(rc);
667 }
668
669 static int mdt_getattr_internal(struct mdt_thread_info *info,
670                                 struct mdt_object *o, int ma_need)
671 {
672         struct md_object        *next = mdt_object_child(o);
673         const struct mdt_body   *reqbody = info->mti_body;
674         struct ptlrpc_request   *req = mdt_info_req(info);
675         struct md_attr          *ma = &info->mti_attr;
676         struct lu_attr          *la = &ma->ma_attr;
677         struct req_capsule      *pill = info->mti_pill;
678         const struct lu_env     *env = info->mti_env;
679         struct mdt_body         *repbody;
680         struct lu_buf           *buffer = &info->mti_buf;
681         int                     rc;
682         int                     is_root;
683         ENTRY;
684
685         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
686                 RETURN(err_serious(-ENOMEM));
687
688         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
689
690         ma->ma_valid = 0;
691
692         rc = mdt_object_exists(o);
693         if (rc < 0) {
694                 /* This object is located on remote node.*/
695                 repbody->fid1 = *mdt_object_fid(o);
696                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
697                 GOTO(out, rc = 0);
698         }
699
700         buffer->lb_len = reqbody->eadatasize;
701         if (buffer->lb_len > 0)
702                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
703         else
704                 buffer->lb_buf = NULL;
705
706         /* If it is dir object and client require MEA, then we got MEA */
707         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
708             reqbody->valid & OBD_MD_MEA) {
709                 /* Assumption: MDT_MD size is enough for lmv size. */
710                 ma->ma_lmv = buffer->lb_buf;
711                 ma->ma_lmv_size = buffer->lb_len;
712                 ma->ma_need = MA_LMV | MA_INODE;
713         } else {
714                 ma->ma_lmm = buffer->lb_buf;
715                 ma->ma_lmm_size = buffer->lb_len;
716                 ma->ma_need = MA_LOV | MA_INODE | MA_HSM;
717         }
718
719         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
720             reqbody->valid & OBD_MD_FLDIREA  &&
721             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
722                 /* get default stripe info for this dir. */
723                 ma->ma_need |= MA_LOV_DEF;
724         }
725         ma->ma_need |= ma_need;
726         if (ma->ma_need & MA_SOM)
727                 ma->ma_som = &info->mti_u.som.data;
728
729         rc = mdt_attr_get_complex(info, o, ma);
730         if (unlikely(rc)) {
731                 CERROR("getattr error for "DFID": %d\n",
732                         PFID(mdt_object_fid(o)), rc);
733                 RETURN(rc);
734         }
735
736         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
737
738         /* the Lustre protocol supposes to return default striping
739          * on the user-visible root if explicitly requested */
740         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
741             (ma->ma_need & MA_LOV_DEF && is_root) && (ma->ma_need & MA_LOV)) {
742                 struct lu_fid      rootfid;
743                 struct mdt_object *root;
744                 struct mdt_device *mdt = info->mti_mdt;
745
746                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
747                 if (rc)
748                         RETURN(rc);
749                 root = mdt_object_find(env, mdt, &rootfid);
750                 if (IS_ERR(root))
751                         RETURN(PTR_ERR(root));
752                 rc = mdt_attr_get_lov(info, root, ma);
753                 mdt_object_put(info->mti_env, root);
754                 if (unlikely(rc)) {
755                         CERROR("getattr error for "DFID": %d\n",
756                                         PFID(mdt_object_fid(o)), rc);
757                         RETURN(rc);
758                 }
759         }
760
761         if (likely(ma->ma_valid & MA_INODE))
762                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
763         else
764                 RETURN(-EFAULT);
765
766         if (mdt_body_has_lov(la, reqbody)) {
767                 if (ma->ma_valid & MA_LOV) {
768                         LASSERT(ma->ma_lmm_size);
769                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
770                         repbody->eadatasize = ma->ma_lmm_size;
771                         if (S_ISDIR(la->la_mode))
772                                 repbody->valid |= OBD_MD_FLDIREA;
773                         else
774                                 repbody->valid |= OBD_MD_FLEASIZE;
775                 }
776                 if (ma->ma_valid & MA_LMV) {
777                         LASSERT(S_ISDIR(la->la_mode));
778                         repbody->eadatasize = ma->ma_lmv_size;
779                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
780                 }
781         } else if (S_ISLNK(la->la_mode) &&
782                    reqbody->valid & OBD_MD_LINKNAME) {
783                 buffer->lb_buf = ma->ma_lmm;
784                 /* eadatasize from client includes NULL-terminator, so
785                  * there is no need to read it */
786                 buffer->lb_len = reqbody->eadatasize - 1;
787                 rc = mo_readlink(env, next, buffer);
788                 if (unlikely(rc <= 0)) {
789                         CERROR("readlink failed: %d\n", rc);
790                         rc = -EFAULT;
791                 } else {
792                         int print_limit = min_t(int, CFS_PAGE_SIZE - 128, rc);
793
794                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
795                                 rc -= 2;
796                         repbody->valid |= OBD_MD_LINKNAME;
797                         /* we need to report back size with NULL-terminator
798                          * because client expects that */
799                         repbody->eadatasize = rc + 1;
800                         if (repbody->eadatasize != reqbody->eadatasize)
801                                 CERROR("Read shorter symlink %d, expected %d\n",
802                                        rc, reqbody->eadatasize - 1);
803                         /* NULL terminate */
804                         ((char *)ma->ma_lmm)[rc] = 0;
805
806                         /* If the total CDEBUG() size is larger than a page, it
807                          * will print a warning to the console, avoid this by
808                          * printing just the last part of the symlink. */
809                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
810                                print_limit < rc ? "..." : "", print_limit,
811                                (char *)ma->ma_lmm + rc - print_limit, rc);
812                         rc = 0;
813                 }
814         }
815
816         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
817                 repbody->max_cookiesize = 0;
818                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
819                 repbody->valid |= OBD_MD_FLMODEASIZE;
820                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
821                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
822                        repbody->max_cookiesize);
823         }
824
825         if (exp_connect_rmtclient(info->mti_exp) &&
826             reqbody->valid & OBD_MD_FLRMTPERM) {
827                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
828
829                 /* mdt_getattr_lock only */
830                 rc = mdt_pack_remote_perm(info, o, buf);
831                 if (rc) {
832                         repbody->valid &= ~OBD_MD_FLRMTPERM;
833                         repbody->aclsize = 0;
834                         RETURN(rc);
835                 } else {
836                         repbody->valid |= OBD_MD_FLRMTPERM;
837                         repbody->aclsize = sizeof(struct mdt_remote_perm);
838                 }
839         }
840 #ifdef CONFIG_FS_POSIX_ACL
841         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
842                  (reqbody->valid & OBD_MD_FLACL)) {
843                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
844                 buffer->lb_len = req_capsule_get_size(pill,
845                                                       &RMF_ACL, RCL_SERVER);
846                 if (buffer->lb_len > 0) {
847                         rc = mo_xattr_get(env, next, buffer,
848                                           XATTR_NAME_ACL_ACCESS);
849                         if (rc < 0) {
850                                 if (rc == -ENODATA) {
851                                         repbody->aclsize = 0;
852                                         repbody->valid |= OBD_MD_FLACL;
853                                         rc = 0;
854                                 } else if (rc == -EOPNOTSUPP) {
855                                         rc = 0;
856                                 } else {
857                                         CERROR("got acl size: %d\n", rc);
858                                 }
859                         } else {
860                                 repbody->aclsize = rc;
861                                 repbody->valid |= OBD_MD_FLACL;
862                                 rc = 0;
863                         }
864                 }
865         }
866 #endif
867
868         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
869             info->mti_mdt->mdt_opts.mo_mds_capa &&
870             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
871                 struct lustre_capa *capa;
872
873                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
874                 LASSERT(capa);
875                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
876                 rc = mo_capa_get(env, next, capa, 0);
877                 if (rc)
878                         RETURN(rc);
879                 repbody->valid |= OBD_MD_FLMDSCAPA;
880         }
881
882 out:
883         if (rc == 0)
884                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
885
886         RETURN(rc);
887 }
888
889 static int mdt_renew_capa(struct mdt_thread_info *info)
890 {
891         struct mdt_object  *obj = info->mti_object;
892         struct mdt_body    *body;
893         struct lustre_capa *capa, *c;
894         int rc;
895         ENTRY;
896
897         /* if object doesn't exist, or server has disabled capability,
898          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
899          * flag not set.
900          */
901         if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
902             !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
903                 RETURN(0);
904
905         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
906         LASSERT(body != NULL);
907
908         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
909         LASSERT(c);
910
911         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
912         LASSERT(capa);
913
914         *capa = *c;
915         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
916         if (rc == 0)
917                 body->valid |= OBD_MD_FLOSSCAPA;
918         RETURN(rc);
919 }
920
921 int mdt_getattr(struct mdt_thread_info *info)
922 {
923         struct mdt_object       *obj = info->mti_object;
924         struct req_capsule      *pill = info->mti_pill;
925         struct mdt_body         *reqbody;
926         struct mdt_body         *repbody;
927         mode_t                   mode;
928         int rc, rc2;
929         ENTRY;
930
931         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
932         LASSERT(reqbody);
933
934         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
935                 rc = req_capsule_server_pack(pill);
936                 if (unlikely(rc))
937                         RETURN(err_serious(rc));
938                 rc = mdt_renew_capa(info);
939                 GOTO(out_shrink, rc);
940         }
941
942         LASSERT(obj != NULL);
943         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
944
945         mode = lu_object_attr(&obj->mot_obj.mo_lu);
946
947         /* old clients may not report needed easize, use max value then */
948         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
949                              reqbody->eadatasize == 0 ?
950                              info->mti_mdt->mdt_max_mdsize :
951                              reqbody->eadatasize);
952
953         rc = req_capsule_server_pack(pill);
954         if (unlikely(rc != 0))
955                 RETURN(err_serious(rc));
956
957         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
958         LASSERT(repbody != NULL);
959         repbody->eadatasize = 0;
960         repbody->aclsize = 0;
961
962         if (reqbody->valid & OBD_MD_FLRMTPERM)
963                 rc = mdt_init_ucred(info, reqbody);
964         else
965                 rc = mdt_check_ucred(info);
966         if (unlikely(rc))
967                 GOTO(out_shrink, rc);
968
969         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
970
971         /*
972          * Don't check capability at all, because rename might getattr for
973          * remote obj, and at that time no capability is available.
974          */
975         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
976         rc = mdt_getattr_internal(info, obj, 0);
977         if (reqbody->valid & OBD_MD_FLRMTPERM)
978                 mdt_exit_ucred(info);
979         EXIT;
980 out_shrink:
981         mdt_client_compatibility(info);
982         rc2 = mdt_fix_reply(info);
983         if (rc == 0)
984                 rc = rc2;
985         return rc;
986 }
987
988 int mdt_is_subdir(struct mdt_thread_info *info)
989 {
990         struct mdt_object     *o = info->mti_object;
991         struct req_capsule    *pill = info->mti_pill;
992         const struct mdt_body *body = info->mti_body;
993         struct mdt_body       *repbody;
994         int                    rc;
995         ENTRY;
996
997         LASSERT(o != NULL);
998
999         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1000
1001         /*
1002          * We save last checked parent fid to @repbody->fid1 for remote
1003          * directory case.
1004          */
1005         LASSERT(fid_is_sane(&body->fid2));
1006         LASSERT(mdt_object_exists(o) > 0);
1007         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
1008                            &body->fid2, &repbody->fid1);
1009         if (rc == 0 || rc == -EREMOTE)
1010                 repbody->valid |= OBD_MD_FLID;
1011
1012         RETURN(rc);
1013 }
1014
1015 static int mdt_raw_lookup(struct mdt_thread_info *info,
1016                           struct mdt_object *parent,
1017                           const struct lu_name *lname,
1018                           struct ldlm_reply *ldlm_rep)
1019 {
1020         struct md_object *next = mdt_object_child(info->mti_object);
1021         const struct mdt_body *reqbody = info->mti_body;
1022         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1023         struct mdt_body *repbody;
1024         int rc;
1025         ENTRY;
1026
1027         if (reqbody->valid != OBD_MD_FLID)
1028                 RETURN(0);
1029
1030         LASSERT(!info->mti_cross_ref);
1031
1032         /* Only got the fid of this obj by name */
1033         fid_zero(child_fid);
1034         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1035                         &info->mti_spec);
1036 #if 0
1037         /* XXX is raw_lookup possible as intent operation? */
1038         if (rc != 0) {
1039                 if (rc == -ENOENT)
1040                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1041                 RETURN(rc);
1042         } else
1043                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1044
1045         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1046 #endif
1047         if (rc == 0) {
1048                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1049                 repbody->fid1 = *child_fid;
1050                 repbody->valid = OBD_MD_FLID;
1051         }
1052         RETURN(1);
1053 }
1054
1055 /*
1056  * UPDATE lock should be taken against parent, and be release before exit;
1057  * child_bits lock should be taken against child, and be returned back:
1058  *            (1)normal request should release the child lock;
1059  *            (2)intent request will grant the lock to client.
1060  */
1061 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1062                                  struct mdt_lock_handle *lhc,
1063                                  __u64 child_bits,
1064                                  struct ldlm_reply *ldlm_rep)
1065 {
1066         struct ptlrpc_request  *req       = mdt_info_req(info);
1067         struct mdt_body        *reqbody   = NULL;
1068         struct mdt_object      *parent    = info->mti_object;
1069         struct mdt_object      *child;
1070         struct md_object       *next      = mdt_object_child(parent);
1071         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1072         struct lu_name         *lname     = NULL;
1073         const char             *name      = NULL;
1074         int                     namelen   = 0;
1075         struct mdt_lock_handle *lhp       = NULL;
1076         struct ldlm_lock       *lock;
1077         struct ldlm_res_id     *res_id;
1078         int                     is_resent;
1079         int                     ma_need = 0;
1080         int                     rc;
1081
1082         ENTRY;
1083
1084         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1085         LASSERT(ergo(is_resent,
1086                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1087
1088         LASSERT(parent != NULL);
1089         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
1090         if (name == NULL)
1091                 RETURN(err_serious(-EFAULT));
1092
1093         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
1094                                        RCL_CLIENT) - 1;
1095         if (!info->mti_cross_ref) {
1096                 /*
1097                  * XXX: Check for "namelen == 0" is for getattr by fid
1098                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
1099                  * that is the name must contain at least one character and
1100                  * the terminating '\0'
1101                  */
1102                 if (namelen == 0) {
1103                         reqbody = req_capsule_client_get(info->mti_pill,
1104                                                          &RMF_MDT_BODY);
1105                         if (unlikely(reqbody == NULL))
1106                                 RETURN(err_serious(-EFAULT));
1107
1108                         if (unlikely(!fid_is_sane(&reqbody->fid2)))
1109                                 RETURN(err_serious(-EINVAL));
1110
1111                         name = NULL;
1112                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1113                                "ldlm_rep = %p\n",
1114                                PFID(mdt_object_fid(parent)),
1115                                PFID(&reqbody->fid2), ldlm_rep);
1116                 } else {
1117                         lname = mdt_name(info->mti_env, (char *)name, namelen);
1118                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
1119                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1120                                name, ldlm_rep);
1121                 }
1122         }
1123         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1124
1125         rc = mdt_object_exists(parent);
1126         if (unlikely(rc == 0)) {
1127                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1128                                 &parent->mot_obj.mo_lu,
1129                                 "Parent doesn't exist!\n");
1130                 RETURN(-ESTALE);
1131         } else if (!info->mti_cross_ref) {
1132                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
1133                          PFID(mdt_object_fid(parent)));
1134         }
1135         if (lname) {
1136                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1137                 if (rc != 0) {
1138                         if (rc > 0)
1139                                 rc = 0;
1140                         RETURN(rc);
1141                 }
1142         }
1143
1144         if (info->mti_cross_ref) {
1145                 /* Only getattr on the child. Parent is on another node. */
1146                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1147                 child = parent;
1148                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1149                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
1150
1151                 if (is_resent) {
1152                         /* Do not take lock for resent case. */
1153                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1154                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1155                                  lhc->mlh_reg_lh.cookie);
1156                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
1157                                                 &lock->l_resource->lr_name));
1158                         LDLM_LOCK_PUT(lock);
1159                         rc = 0;
1160                 } else {
1161                         mdt_lock_handle_init(lhc);
1162                         mdt_lock_reg_init(lhc, LCK_PR);
1163
1164                         /*
1165                          * Object's name is on another MDS, no lookup lock is
1166                          * needed here but update is.
1167                          */
1168                         child_bits &= ~MDS_INODELOCK_LOOKUP;
1169                         child_bits |= MDS_INODELOCK_UPDATE;
1170
1171                         rc = mdt_object_lock(info, child, lhc, child_bits,
1172                                              MDT_LOCAL_LOCK);
1173                 }
1174                 if (rc == 0) {
1175                         /* Finally, we can get attr for child. */
1176                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
1177                                          BYPASS_CAPA);
1178                         rc = mdt_getattr_internal(info, child, 0);
1179                         if (unlikely(rc != 0))
1180                                 mdt_object_unlock(info, child, lhc, 1);
1181                 }
1182                 RETURN(rc);
1183         }
1184
1185         if (lname) {
1186                 /* step 1: lock parent only if parent is a directory */
1187                 if (S_ISDIR(lu_object_attr(&parent->mot_obj.mo_lu))) {
1188                         lhp = &info->mti_lh[MDT_LH_PARENT];
1189                         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
1190                         rc = mdt_object_lock(info, parent, lhp,
1191                                              MDS_INODELOCK_UPDATE,
1192                                              MDT_LOCAL_LOCK);
1193                         if (unlikely(rc != 0))
1194                                 RETURN(rc);
1195                 }
1196
1197                 /* step 2: lookup child's fid by name */
1198                 fid_zero(child_fid);
1199                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1200                                 &info->mti_spec);
1201
1202                 if (rc != 0) {
1203                         if (rc == -ENOENT)
1204                                 mdt_set_disposition(info, ldlm_rep,
1205                                                     DISP_LOOKUP_NEG);
1206                         GOTO(out_parent, rc);
1207                 } else
1208                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1209         } else {
1210                 *child_fid = reqbody->fid2;
1211                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1212         }
1213
1214         /*
1215          *step 3: find the child object by fid & lock it.
1216          *        regardless if it is local or remote.
1217          */
1218         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
1219
1220         if (unlikely(IS_ERR(child)))
1221                 GOTO(out_parent, rc = PTR_ERR(child));
1222         if (is_resent) {
1223                 /* Do not take lock for resent case. */
1224                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1225                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1226                          lhc->mlh_reg_lh.cookie);
1227
1228                 res_id = &lock->l_resource->lr_name;
1229                 if (!fid_res_name_eq(mdt_object_fid(child),
1230                                     &lock->l_resource->lr_name)) {
1231                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1232                                                  &lock->l_resource->lr_name),
1233                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1234                                  (unsigned long)res_id->name[0],
1235                                  (unsigned long)res_id->name[1],
1236                                  (unsigned long)res_id->name[2],
1237                                  PFID(mdt_object_fid(parent)));
1238                           CWARN("Although resent, but still not get child lock"
1239                                 "parent:"DFID" child:"DFID"\n",
1240                                 PFID(mdt_object_fid(parent)),
1241                                 PFID(mdt_object_fid(child)));
1242                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1243                           LDLM_LOCK_PUT(lock);
1244                           GOTO(relock, 0);
1245                 }
1246                 LDLM_LOCK_PUT(lock);
1247                 rc = 0;
1248         } else {
1249                 bool try_layout = false;
1250
1251 relock:
1252                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1253                 mdt_lock_handle_init(lhc);
1254                 mdt_lock_reg_init(lhc, LCK_PR);
1255
1256                 if (mdt_object_exists(child) == 0) {
1257                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1258                                         &child->mot_obj.mo_lu,
1259                                         "Object doesn't exist!\n");
1260                         GOTO(out_child, rc = -ENOENT);
1261                 }
1262
1263                 if (!(child_bits & MDS_INODELOCK_UPDATE)) {
1264                         struct md_attr *ma = &info->mti_attr;
1265
1266                         ma->ma_valid = 0;
1267                         ma->ma_need = MA_INODE;
1268                         rc = mdt_attr_get_complex(info, child, ma);
1269                         if (unlikely(rc != 0))
1270                                 GOTO(out_child, rc);
1271
1272                         /* If the file has not been changed for some time, we
1273                          * return not only a LOOKUP lock, but also an UPDATE
1274                          * lock and this might save us RPC on later STAT. For
1275                          * directories, it also let negative dentry starts
1276                          * working for this dir. */
1277                         if (ma->ma_valid & MA_INODE &&
1278                             ma->ma_attr.la_valid & LA_CTIME &&
1279                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1280                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1281                                 child_bits |= MDS_INODELOCK_UPDATE;
1282                 }
1283
1284                 /* layout lock must be granted in a best-effort way
1285                  * for IT operations */
1286                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1287                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1288                     exp_connect_layout(info->mti_exp) &&
1289                     S_ISREG(lu_object_attr(&child->mot_obj.mo_lu)) &&
1290                     ldlm_rep != NULL) {
1291                         /* try to grant layout lock for regular file. */
1292                         try_layout = true;
1293                 }
1294
1295                 rc = 0;
1296                 if (try_layout) {
1297                         child_bits |= MDS_INODELOCK_LAYOUT;
1298                         /* try layout lock, it may fail to be granted due to
1299                          * contention at LOOKUP or UPDATE */
1300                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1301                                                  MDT_CROSS_LOCK)) {
1302                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1303                                 LASSERT(child_bits != 0);
1304                                 rc = mdt_object_lock(info, child, lhc,
1305                                                 child_bits, MDT_CROSS_LOCK);
1306                         } else {
1307                                 ma_need |= MA_LOV;
1308                         }
1309                 } else {
1310                         rc = mdt_object_lock(info, child, lhc, child_bits,
1311                                                 MDT_CROSS_LOCK);
1312                 }
1313                 if (unlikely(rc != 0))
1314                         GOTO(out_child, rc);
1315         }
1316
1317         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1318         /* Get MA_SOM attributes if update lock is given. */
1319         if (lock &&
1320             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1321             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1322                 ma_need |= MA_SOM;
1323
1324         /* finally, we can get attr for child. */
1325         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1326         rc = mdt_getattr_internal(info, child, ma_need);
1327         if (unlikely(rc != 0)) {
1328                 mdt_object_unlock(info, child, lhc, 1);
1329         } else if (lock) {
1330                 /* Debugging code. */
1331                 res_id = &lock->l_resource->lr_name;
1332                 LDLM_DEBUG(lock, "Returning lock to client");
1333                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1334                                          &lock->l_resource->lr_name),
1335                          "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1336                          (unsigned long)res_id->name[0],
1337                          (unsigned long)res_id->name[1],
1338                          (unsigned long)res_id->name[2],
1339                          PFID(mdt_object_fid(child)));
1340                 mdt_pack_size2body(info, child);
1341         }
1342         if (lock)
1343                 LDLM_LOCK_PUT(lock);
1344
1345         EXIT;
1346 out_child:
1347         mdt_object_put(info->mti_env, child);
1348 out_parent:
1349         if (lhp)
1350                 mdt_object_unlock(info, parent, lhp, 1);
1351         return rc;
1352 }
1353
1354 /* normal handler: should release the child lock */
1355 int mdt_getattr_name(struct mdt_thread_info *info)
1356 {
1357         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1358         struct mdt_body        *reqbody;
1359         struct mdt_body        *repbody;
1360         int rc, rc2;
1361         ENTRY;
1362
1363         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1364         LASSERT(reqbody != NULL);
1365         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1366         LASSERT(repbody != NULL);
1367
1368         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1369         repbody->eadatasize = 0;
1370         repbody->aclsize = 0;
1371
1372         rc = mdt_init_ucred(info, reqbody);
1373         if (unlikely(rc))
1374                 GOTO(out_shrink, rc);
1375
1376         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1377         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1378                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1379                 lhc->mlh_reg_lh.cookie = 0;
1380         }
1381         mdt_exit_ucred(info);
1382         EXIT;
1383 out_shrink:
1384         mdt_client_compatibility(info);
1385         rc2 = mdt_fix_reply(info);
1386         if (rc == 0)
1387                 rc = rc2;
1388         return rc;
1389 }
1390
1391 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1392                          void *karg, void *uarg);
1393
1394 int mdt_set_info(struct mdt_thread_info *info)
1395 {
1396         struct ptlrpc_request *req = mdt_info_req(info);
1397         char *key;
1398         void *val;
1399         int keylen, vallen, rc = 0;
1400         ENTRY;
1401
1402         rc = req_capsule_server_pack(info->mti_pill);
1403         if (rc)
1404                 RETURN(rc);
1405
1406         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1407         if (key == NULL) {
1408                 DEBUG_REQ(D_HA, req, "no set_info key");
1409                 RETURN(-EFAULT);
1410         }
1411
1412         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1413                                       RCL_CLIENT);
1414
1415         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1416         if (val == NULL) {
1417                 DEBUG_REQ(D_HA, req, "no set_info val");
1418                 RETURN(-EFAULT);
1419         }
1420
1421         vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1422                                       RCL_CLIENT);
1423
1424         /* Swab any part of val you need to here */
1425         if (KEY_IS(KEY_READ_ONLY)) {
1426                 req->rq_status = 0;
1427                 lustre_msg_set_status(req->rq_repmsg, 0);
1428
1429                 spin_lock(&req->rq_export->exp_lock);
1430                 if (*(__u32 *)val)
1431                         req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1432                 else
1433                         req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
1434                 spin_unlock(&req->rq_export->exp_lock);
1435
1436         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1437                 struct changelog_setinfo *cs =
1438                         (struct changelog_setinfo *)val;
1439                 if (vallen != sizeof(*cs)) {
1440                         CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1441                         RETURN(-EINVAL);
1442                 }
1443                 if (ptlrpc_req_need_swab(req)) {
1444                         __swab64s(&cs->cs_recno);
1445                         __swab32s(&cs->cs_id);
1446                 }
1447
1448                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1449                                    vallen, val, NULL);
1450                 lustre_msg_set_status(req->rq_repmsg, rc);
1451
1452         } else {
1453                 RETURN(-EINVAL);
1454         }
1455         RETURN(0);
1456 }
1457
1458 /**
1459  * Top-level handler for MDT connection requests.
1460  */
1461 int mdt_connect(struct mdt_thread_info *info)
1462 {
1463         int rc;
1464         struct obd_connect_data *reply;
1465         struct obd_export *exp;
1466         struct ptlrpc_request *req = mdt_info_req(info);
1467
1468         rc = target_handle_connect(req);
1469         if (rc != 0)
1470                 return err_serious(rc);
1471
1472         LASSERT(req->rq_export != NULL);
1473         info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1474         rc = mdt_init_sec_level(info);
1475         if (rc != 0) {
1476                 obd_disconnect(class_export_get(req->rq_export));
1477                 return rc;
1478         }
1479
1480         /* To avoid exposing partially initialized connection flags, changes up
1481          * to this point have been staged in reply->ocd_connect_flags. Now that
1482          * connection handling has completed successfully, atomically update
1483          * the connect flags in the shared export data structure. LU-1623 */
1484         reply = req_capsule_server_get(info->mti_pill, &RMF_CONNECT_DATA);
1485         exp = req->rq_export;
1486         spin_lock(&exp->exp_lock);
1487         exp->exp_connect_flags = reply->ocd_connect_flags;
1488         spin_unlock(&exp->exp_lock);
1489
1490         rc = mdt_init_idmap(info);
1491         if (rc != 0)
1492                 obd_disconnect(class_export_get(req->rq_export));
1493
1494         return rc;
1495 }
1496
1497 int mdt_disconnect(struct mdt_thread_info *info)
1498 {
1499         int rc;
1500         ENTRY;
1501
1502         rc = target_handle_disconnect(mdt_info_req(info));
1503         if (rc)
1504                 rc = err_serious(rc);
1505         RETURN(rc);
1506 }
1507
1508 static int mdt_sendpage(struct mdt_thread_info *info,
1509                         struct lu_rdpg *rdpg, int nob)
1510 {
1511         struct ptlrpc_request   *req = mdt_info_req(info);
1512         struct obd_export       *exp = req->rq_export;
1513         struct ptlrpc_bulk_desc *desc;
1514         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1515         int                      tmpcount;
1516         int                      tmpsize;
1517         int                      i;
1518         int                      rc;
1519         ENTRY;
1520
1521         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1522                                     MDS_BULK_PORTAL);
1523         if (desc == NULL)
1524                 RETURN(-ENOMEM);
1525
1526         if (!(exp->exp_connect_flags & OBD_CONNECT_BRW_SIZE))
1527                 /* old client requires reply size in it's PAGE_SIZE,
1528                  * which is rdpg->rp_count */
1529                 nob = rdpg->rp_count;
1530
1531         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1532              i++, tmpcount -= tmpsize) {
1533                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1534                 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1535         }
1536
1537         LASSERT(desc->bd_nob == nob);
1538         rc = target_bulk_io(exp, desc, lwi);
1539         ptlrpc_free_bulk_pin(desc);
1540         RETURN(rc);
1541 }
1542
1543 int mdt_readpage(struct mdt_thread_info *info)
1544 {
1545         struct mdt_object *object = info->mti_object;
1546         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1547         struct mdt_body   *reqbody;
1548         struct mdt_body   *repbody;
1549         int                rc;
1550         int                i;
1551         ENTRY;
1552
1553         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1554                 RETURN(err_serious(-ENOMEM));
1555
1556         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1557         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1558         if (reqbody == NULL || repbody == NULL)
1559                 RETURN(err_serious(-EFAULT));
1560
1561         /*
1562          * prepare @rdpg before calling lower layers and transfer itself. Here
1563          * reqbody->size contains offset of where to start to read and
1564          * reqbody->nlink contains number bytes to read.
1565          */
1566         rdpg->rp_hash = reqbody->size;
1567         if (rdpg->rp_hash != reqbody->size) {
1568                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1569                        rdpg->rp_hash, reqbody->size);
1570                 RETURN(-EFAULT);
1571         }
1572
1573         rdpg->rp_attrs = reqbody->mode;
1574         if (info->mti_exp->exp_connect_flags & OBD_CONNECT_64BITHASH)
1575                 rdpg->rp_attrs |= LUDA_64BITHASH;
1576         rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
1577                                 PTLRPC_MAX_BRW_SIZE);
1578         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1) >>
1579                           CFS_PAGE_SHIFT;
1580         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1581         if (rdpg->rp_pages == NULL)
1582                 RETURN(-ENOMEM);
1583
1584         for (i = 0; i < rdpg->rp_npages; ++i) {
1585                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1586                 if (rdpg->rp_pages[i] == NULL)
1587                         GOTO(free_rdpg, rc = -ENOMEM);
1588         }
1589
1590         /* call lower layers to fill allocated pages with directory data */
1591         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1592         if (rc < 0)
1593                 GOTO(free_rdpg, rc);
1594
1595         /* send pages to client */
1596         rc = mdt_sendpage(info, rdpg, rc);
1597
1598         EXIT;
1599 free_rdpg:
1600
1601         for (i = 0; i < rdpg->rp_npages; i++)
1602                 if (rdpg->rp_pages[i] != NULL)
1603                         cfs_free_page(rdpg->rp_pages[i]);
1604         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1605
1606         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1607                 RETURN(0);
1608
1609         return rc;
1610 }
1611
1612 static int mdt_reint_internal(struct mdt_thread_info *info,
1613                               struct mdt_lock_handle *lhc,
1614                               __u32 op)
1615 {
1616         struct req_capsule      *pill = info->mti_pill;
1617         struct mdt_body         *repbody;
1618         int                      rc = 0, rc2;
1619         ENTRY;
1620
1621
1622         rc = mdt_reint_unpack(info, op);
1623         if (rc != 0) {
1624                 CERROR("Can't unpack reint, rc %d\n", rc);
1625                 RETURN(err_serious(rc));
1626         }
1627
1628         /* for replay (no_create) lmm is not needed, client has it already */
1629         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1630                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1631                                      info->mti_rr.rr_eadatalen);
1632
1633         /* llog cookies are always 0, the field is kept for compatibility */
1634         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1635                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1636
1637         rc = req_capsule_server_pack(pill);
1638         if (rc != 0) {
1639                 CERROR("Can't pack response, rc %d\n", rc);
1640                 RETURN(err_serious(rc));
1641         }
1642
1643         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1644                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1645                 LASSERT(repbody);
1646                 repbody->eadatasize = 0;
1647                 repbody->aclsize = 0;
1648         }
1649
1650         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1651
1652         /* for replay no cookkie / lmm need, because client have this already */
1653         if (info->mti_spec.no_create)
1654                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1655                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1656
1657         rc = mdt_init_ucred_reint(info);
1658         if (rc)
1659                 GOTO(out_shrink, rc);
1660
1661         rc = mdt_fix_attr_ucred(info, op);
1662         if (rc != 0)
1663                 GOTO(out_ucred, rc = err_serious(rc));
1664
1665         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1666                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1667                 GOTO(out_ucred, rc);
1668         }
1669         rc = mdt_reint_rec(info, lhc);
1670         EXIT;
1671 out_ucred:
1672         mdt_exit_ucred(info);
1673 out_shrink:
1674         mdt_client_compatibility(info);
1675         rc2 = mdt_fix_reply(info);
1676         if (rc == 0)
1677                 rc = rc2;
1678         return rc;
1679 }
1680
1681 static long mdt_reint_opcode(struct mdt_thread_info *info,
1682                              const struct req_format **fmt)
1683 {
1684         struct mdt_rec_reint *rec;
1685         long opc;
1686
1687         opc = err_serious(-EFAULT);
1688         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1689         if (rec != NULL) {
1690                 opc = rec->rr_opcode;
1691                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1692                 if (opc < REINT_MAX && fmt[opc] != NULL)
1693                         req_capsule_extend(info->mti_pill, fmt[opc]);
1694                 else {
1695                         CERROR("Unsupported opc: %ld\n", opc);
1696                         opc = err_serious(opc);
1697                 }
1698         }
1699         return opc;
1700 }
1701
1702 int mdt_reint(struct mdt_thread_info *info)
1703 {
1704         long opc;
1705         int  rc;
1706
1707         static const struct req_format *reint_fmts[REINT_MAX] = {
1708                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1709                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1710                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1711                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1712                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1713                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1714                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1715         };
1716
1717         ENTRY;
1718
1719         opc = mdt_reint_opcode(info, reint_fmts);
1720         if (opc >= 0) {
1721                 /*
1722                  * No lock possible here from client to pass it to reint code
1723                  * path.
1724                  */
1725                 rc = mdt_reint_internal(info, NULL, opc);
1726         } else {
1727                 rc = opc;
1728         }
1729
1730         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1731         RETURN(rc);
1732 }
1733
1734 /* this should sync the whole device */
1735 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1736 {
1737         struct dt_device *dt = mdt->mdt_bottom;
1738         int rc;
1739         ENTRY;
1740
1741         rc = dt->dd_ops->dt_sync(env, dt);
1742         RETURN(rc);
1743 }
1744
1745 /* this should sync this object */
1746 static int mdt_object_sync(struct mdt_thread_info *info)
1747 {
1748         struct md_object *next;
1749         int rc;
1750         ENTRY;
1751
1752         if (!mdt_object_exists(info->mti_object)) {
1753                 CWARN("Non existing object  "DFID"!\n",
1754                       PFID(mdt_object_fid(info->mti_object)));
1755                 RETURN(-ESTALE);
1756         }
1757         next = mdt_object_child(info->mti_object);
1758         rc = mo_object_sync(info->mti_env, next);
1759
1760         RETURN(rc);
1761 }
1762
1763 int mdt_sync(struct mdt_thread_info *info)
1764 {
1765         struct ptlrpc_request *req = mdt_info_req(info);
1766         struct req_capsule *pill = info->mti_pill;
1767         struct mdt_body *body;
1768         int rc;
1769         ENTRY;
1770
1771         /* The fid may be zero, so we req_capsule_set manually */
1772         req_capsule_set(pill, &RQF_MDS_SYNC);
1773
1774         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1775         if (body == NULL)
1776                 RETURN(err_serious(-EINVAL));
1777
1778         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1779                 RETURN(err_serious(-ENOMEM));
1780
1781         if (fid_seq(&body->fid1) == 0) {
1782                 /* sync the whole device */
1783                 rc = req_capsule_server_pack(pill);
1784                 if (rc == 0)
1785                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1786                 else
1787                         rc = err_serious(rc);
1788         } else {
1789                 /* sync an object */
1790                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1791                 if (rc == 0) {
1792                         rc = mdt_object_sync(info);
1793                         if (rc == 0) {
1794                                 const struct lu_fid *fid;
1795                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1796
1797                                 info->mti_attr.ma_need = MA_INODE;
1798                                 info->mti_attr.ma_valid = 0;
1799                                 rc = mdt_attr_get_complex(info, info->mti_object,
1800                                                           &info->mti_attr);
1801                                 if (rc == 0) {
1802                                         body = req_capsule_server_get(pill,
1803                                                                 &RMF_MDT_BODY);
1804                                         fid = mdt_object_fid(info->mti_object);
1805                                         mdt_pack_attr2body(info, body, la, fid);
1806                                 }
1807                         }
1808                 } else
1809                         rc = err_serious(rc);
1810         }
1811         if (rc == 0)
1812                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1813
1814         RETURN(rc);
1815 }
1816
1817 /*
1818  * Quotacheck handler.
1819  * in-kernel quotacheck isn't supported any more.
1820  */
1821 int mdt_quotacheck(struct mdt_thread_info *info)
1822 {
1823         struct obd_quotactl     *oqctl;
1824         int                      rc;
1825         ENTRY;
1826
1827         oqctl = req_capsule_client_get(info->mti_pill, &RMF_OBD_QUOTACTL);
1828         if (oqctl == NULL)
1829                 RETURN(err_serious(-EPROTO));
1830
1831         rc = req_capsule_server_pack(info->mti_pill);
1832         if (rc)
1833                 RETURN(err_serious(rc));
1834
1835         /* deprecated, not used any more */
1836         RETURN(-EOPNOTSUPP);
1837 }
1838
1839 /*
1840  * Handle quota control requests to consult current usage/limit, but also
1841  * to configure quota enforcement
1842  */
1843 int mdt_quotactl(struct mdt_thread_info *info)
1844 {
1845         struct obd_export       *exp  = info->mti_exp;
1846         struct req_capsule      *pill = info->mti_pill;
1847         struct obd_quotactl     *oqctl, *repoqc;
1848         int                      id, rc;
1849         struct lu_device        *qmt = info->mti_mdt->mdt_qmt_dev;
1850         ENTRY;
1851
1852         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1853         if (oqctl == NULL)
1854                 RETURN(err_serious(-EPROTO));
1855
1856         rc = req_capsule_server_pack(pill);
1857         if (rc)
1858                 RETURN(err_serious(rc));
1859
1860         switch (oqctl->qc_cmd) {
1861         case Q_QUOTACHECK:
1862         case LUSTRE_Q_INVALIDATE:
1863         case LUSTRE_Q_FINVALIDATE:
1864         case Q_QUOTAON:
1865         case Q_QUOTAOFF:
1866         case Q_INITQUOTA:
1867                 /* deprecated, not used any more */
1868                 RETURN(-EOPNOTSUPP);
1869                 /* master quotactl */
1870         case Q_GETINFO:
1871         case Q_SETINFO:
1872         case Q_SETQUOTA:
1873         case Q_GETQUOTA:
1874                 if (qmt == NULL)
1875                         RETURN(-EOPNOTSUPP);
1876                 /* slave quotactl */
1877         case Q_GETOINFO:
1878         case Q_GETOQUOTA:
1879                 break;
1880         default:
1881                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1882                 RETURN(-EFAULT);
1883         }
1884
1885         /* map uid/gid for remote client */
1886         id = oqctl->qc_id;
1887         if (exp_connect_rmtclient(exp)) {
1888                 struct lustre_idmap_table *idmap;
1889
1890                 idmap = mdt_req2med(mdt_info_req(info))->med_idmap;
1891
1892                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1893                              oqctl->qc_cmd != Q_GETINFO))
1894                         RETURN(-EPERM);
1895
1896                 if (oqctl->qc_type == USRQUOTA)
1897                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1898                                                      oqctl->qc_id);
1899                 else if (oqctl->qc_type == GRPQUOTA)
1900                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1901                                                      oqctl->qc_id);
1902                 else
1903                         RETURN(-EINVAL);
1904
1905                 if (id == CFS_IDMAP_NOTFOUND) {
1906                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
1907                         RETURN(-EACCES);
1908                 }
1909         }
1910
1911         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1912         if (repoqc == NULL)
1913                 RETURN(err_serious(-EFAULT));
1914
1915         if (oqctl->qc_id != id)
1916                 swap(oqctl->qc_id, id);
1917
1918         switch (oqctl->qc_cmd) {
1919
1920         case Q_GETINFO:
1921         case Q_SETINFO:
1922         case Q_SETQUOTA:
1923         case Q_GETQUOTA:
1924                 /* forward quotactl request to QMT */
1925                 rc = qmt_hdls.qmth_quotactl(info->mti_env, qmt, oqctl);
1926                 break;
1927
1928         case Q_GETOINFO:
1929         case Q_GETOQUOTA:
1930                 /* slave quotactl */
1931                 rc = lquotactl_slv(info->mti_env, info->mti_mdt->mdt_bottom,
1932                                    oqctl);
1933                 break;
1934
1935         default:
1936                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1937                 RETURN(-EFAULT);
1938         }
1939
1940         if (oqctl->qc_id != id)
1941                 swap(oqctl->qc_id, id);
1942
1943         *repoqc = *oqctl;
1944         RETURN(rc);
1945 }
1946
1947 /*
1948  * OBD PING and other handlers.
1949  */
1950 int mdt_obd_ping(struct mdt_thread_info *info)
1951 {
1952         int rc;
1953         ENTRY;
1954
1955         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1956
1957         rc = target_handle_ping(mdt_info_req(info));
1958         if (rc < 0)
1959                 rc = err_serious(rc);
1960         RETURN(rc);
1961 }
1962
1963 /*
1964  * OBD_IDX_READ handler
1965  */
1966 int mdt_obd_idx_read(struct mdt_thread_info *info)
1967 {
1968         struct mdt_device       *mdt = info->mti_mdt;
1969         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1970         struct idx_info         *req_ii, *rep_ii;
1971         int                      rc, i;
1972         ENTRY;
1973
1974         memset(rdpg, 0, sizeof(*rdpg));
1975         req_capsule_set(info->mti_pill, &RQF_OBD_IDX_READ);
1976
1977         /* extract idx_info buffer from request & reply */
1978         req_ii = req_capsule_client_get(info->mti_pill, &RMF_IDX_INFO);
1979         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1980                 RETURN(err_serious(-EPROTO));
1981
1982         rc = req_capsule_server_pack(info->mti_pill);
1983         if (rc)
1984                 RETURN(err_serious(rc));
1985
1986         rep_ii = req_capsule_server_get(info->mti_pill, &RMF_IDX_INFO);
1987         if (rep_ii == NULL)
1988                 RETURN(err_serious(-EFAULT));
1989         rep_ii->ii_magic = IDX_INFO_MAGIC;
1990
1991         /* extract hash to start with */
1992         rdpg->rp_hash = req_ii->ii_hash_start;
1993
1994         /* extract requested attributes */
1995         rdpg->rp_attrs = req_ii->ii_attrs;
1996
1997         /* check that fid packed in request is valid and supported */
1998         if (!fid_is_sane(&req_ii->ii_fid))
1999                 RETURN(-EINVAL);
2000         rep_ii->ii_fid = req_ii->ii_fid;
2001
2002         /* copy flags */
2003         rep_ii->ii_flags = req_ii->ii_flags;
2004
2005         /* compute number of pages to allocate, ii_count is the number of 4KB
2006          * containers */
2007         if (req_ii->ii_count <= 0)
2008                 GOTO(out, rc = -EFAULT);
2009         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
2010                                PTLRPC_MAX_BRW_SIZE);
2011         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE -1) >> CFS_PAGE_SHIFT;
2012
2013         /* allocate pages to store the containers */
2014         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
2015         if (rdpg->rp_pages == NULL)
2016                 GOTO(out, rc = -ENOMEM);
2017         for (i = 0; i < rdpg->rp_npages; i++) {
2018                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
2019                 if (rdpg->rp_pages[i] == NULL)
2020                         GOTO(out, rc = -ENOMEM);
2021         }
2022
2023         /* populate pages with key/record pairs */
2024         rc = dt_index_read(info->mti_env, mdt->mdt_bottom, rep_ii, rdpg);
2025         if (rc < 0)
2026                 GOTO(out, rc);
2027
2028         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
2029                  "asked %d > %d\n", rc, rdpg->rp_count);
2030
2031         /* send pages to client */
2032         rc = mdt_sendpage(info, rdpg, rc);
2033
2034         GOTO(out, rc);
2035 out:
2036         if (rdpg->rp_pages) {
2037                 for (i = 0; i < rdpg->rp_npages; i++)
2038                         if (rdpg->rp_pages[i])
2039                                 cfs_free_page(rdpg->rp_pages[i]);
2040                 OBD_FREE(rdpg->rp_pages,
2041                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
2042         }
2043         return rc;
2044 }
2045
2046 int mdt_obd_log_cancel(struct mdt_thread_info *info)
2047 {
2048         return err_serious(-EOPNOTSUPP);
2049 }
2050
2051 int mdt_obd_qc_callback(struct mdt_thread_info *info)
2052 {
2053         return err_serious(-EOPNOTSUPP);
2054 }
2055
2056 /*
2057  * LLOG handlers.
2058  */
2059
2060 /** clone llog ctxt from child (mdd)
2061  * This allows remote llog (replicator) access.
2062  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2063  * context was originally set up, or we can handle them directly.
2064  * I choose the latter, but that means I need any llog
2065  * contexts set up by child to be accessable by the mdt.  So we clone the
2066  * context into our context list here.
2067  */
2068 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2069                                int idx)
2070 {
2071         struct md_device  *next = mdt->mdt_child;
2072         struct llog_ctxt *ctxt;
2073         int rc;
2074
2075         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2076                 return 0;
2077
2078         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2079         if (rc || ctxt == NULL) {
2080                 return 0;
2081         }
2082
2083         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2084         if (rc)
2085                 CERROR("Can't set mdt ctxt %d\n", rc);
2086
2087         return rc;
2088 }
2089
2090 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2091                                  struct mdt_device *mdt, int idx)
2092 {
2093         struct llog_ctxt *ctxt;
2094
2095         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2096         if (ctxt == NULL)
2097                 return 0;
2098         /* Put once for the get we just did, and once for the clone */
2099         llog_ctxt_put(ctxt);
2100         llog_ctxt_put(ctxt);
2101         return 0;
2102 }
2103
2104 int mdt_llog_create(struct mdt_thread_info *info)
2105 {
2106         int rc;
2107
2108         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
2109         rc = llog_origin_handle_open(mdt_info_req(info));
2110         return (rc < 0 ? err_serious(rc) : rc);
2111 }
2112
2113 int mdt_llog_destroy(struct mdt_thread_info *info)
2114 {
2115         int rc;
2116
2117         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
2118         rc = llog_origin_handle_destroy(mdt_info_req(info));
2119         return (rc < 0 ? err_serious(rc) : rc);
2120 }
2121
2122 int mdt_llog_read_header(struct mdt_thread_info *info)
2123 {
2124         int rc;
2125
2126         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
2127         rc = llog_origin_handle_read_header(mdt_info_req(info));
2128         return (rc < 0 ? err_serious(rc) : rc);
2129 }
2130
2131 int mdt_llog_next_block(struct mdt_thread_info *info)
2132 {
2133         int rc;
2134
2135         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
2136         rc = llog_origin_handle_next_block(mdt_info_req(info));
2137         return (rc < 0 ? err_serious(rc) : rc);
2138 }
2139
2140 int mdt_llog_prev_block(struct mdt_thread_info *info)
2141 {
2142         int rc;
2143
2144         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
2145         rc = llog_origin_handle_prev_block(mdt_info_req(info));
2146         return (rc < 0 ? err_serious(rc) : rc);
2147 }
2148
2149
2150 /*
2151  * DLM handlers.
2152  */
2153
2154 static struct ldlm_callback_suite cbs = {
2155         .lcs_completion = ldlm_server_completion_ast,
2156         .lcs_blocking   = ldlm_server_blocking_ast,
2157         .lcs_glimpse    = ldlm_server_glimpse_ast
2158 };
2159
2160 int mdt_enqueue(struct mdt_thread_info *info)
2161 {
2162         struct ptlrpc_request *req;
2163         int rc;
2164
2165         /*
2166          * info->mti_dlm_req already contains swapped and (if necessary)
2167          * converted dlm request.
2168          */
2169         LASSERT(info->mti_dlm_req != NULL);
2170
2171         req = mdt_info_req(info);
2172         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
2173                                   req, info->mti_dlm_req, &cbs);
2174         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
2175         return rc ? err_serious(rc) : req->rq_status;
2176 }
2177
2178 int mdt_convert(struct mdt_thread_info *info)
2179 {
2180         int rc;
2181         struct ptlrpc_request *req;
2182
2183         LASSERT(info->mti_dlm_req);
2184         req = mdt_info_req(info);
2185         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2186         return rc ? err_serious(rc) : req->rq_status;
2187 }
2188
2189 int mdt_bl_callback(struct mdt_thread_info *info)
2190 {
2191         CERROR("bl callbacks should not happen on MDS\n");
2192         LBUG();
2193         return err_serious(-EOPNOTSUPP);
2194 }
2195
2196 int mdt_cp_callback(struct mdt_thread_info *info)
2197 {
2198         CERROR("cp callbacks should not happen on MDS\n");
2199         LBUG();
2200         return err_serious(-EOPNOTSUPP);
2201 }
2202
2203 /*
2204  * sec context handlers
2205  */
2206 int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2207 {
2208         int rc;
2209
2210         rc = mdt_handle_idmap(info);
2211
2212         if (unlikely(rc)) {
2213                 struct ptlrpc_request *req = mdt_info_req(info);
2214                 __u32                  opc;
2215
2216                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2217                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2218                         sptlrpc_svc_ctx_invalidate(req);
2219         }
2220
2221         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2222
2223         return rc;
2224 }
2225
2226 /*
2227  * quota request handlers
2228  */
2229 int mdt_quota_dqacq(struct mdt_thread_info *info)
2230 {
2231         struct lu_device        *qmt = info->mti_mdt->mdt_qmt_dev;
2232         int                      rc;
2233         ENTRY;
2234
2235         if (qmt == NULL)
2236                 RETURN(err_serious(-EOPNOTSUPP));
2237
2238         rc = qmt_hdls.qmth_dqacq(info->mti_env, qmt, mdt_info_req(info));
2239         RETURN(rc);
2240 }
2241
2242 static struct mdt_object *mdt_obj(struct lu_object *o)
2243 {
2244         LASSERT(lu_device_is_mdt(o->lo_dev));
2245         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2246 }
2247
2248 struct mdt_object *mdt_object_new(const struct lu_env *env,
2249                                   struct mdt_device *d,
2250                                   const struct lu_fid *f)
2251 {
2252         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2253         struct lu_object *o;
2254         struct mdt_object *m;
2255         ENTRY;
2256
2257         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2258         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, &conf);
2259         if (unlikely(IS_ERR(o)))
2260                 m = (struct mdt_object *)o;
2261         else
2262                 m = mdt_obj(o);
2263         RETURN(m);
2264 }
2265
2266 struct mdt_object *mdt_object_find(const struct lu_env *env,
2267                                    struct mdt_device *d,
2268                                    const struct lu_fid *f)
2269 {
2270         struct lu_object *o;
2271         struct mdt_object *m;
2272         ENTRY;
2273
2274         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2275         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2276         if (unlikely(IS_ERR(o)))
2277                 m = (struct mdt_object *)o;
2278         else
2279                 m = mdt_obj(o);
2280         RETURN(m);
2281 }
2282
2283 /**
2284  * Asyncronous commit for mdt device.
2285  *
2286  * Pass asynchonous commit call down the MDS stack.
2287  *
2288  * \param env environment
2289  * \param mdt the mdt device
2290  */
2291 static void mdt_device_commit_async(const struct lu_env *env,
2292                                     struct mdt_device *mdt)
2293 {
2294         struct dt_device *dt = mdt->mdt_bottom;
2295         int rc;
2296
2297         rc = dt->dd_ops->dt_commit_async(env, dt);
2298         if (unlikely(rc != 0))
2299                 CWARN("async commit start failed with rc = %d", rc);
2300 }
2301
2302 /**
2303  * Mark the lock as "synchonous".
2304  *
2305  * Mark the lock to deffer transaction commit to the unlock time.
2306  *
2307  * \param lock the lock to mark as "synchonous"
2308  *
2309  * \see mdt_is_lock_sync
2310  * \see mdt_save_lock
2311  */
2312 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2313 {
2314         lock->l_ast_data = (void*)1;
2315 }
2316
2317 /**
2318  * Check whehter the lock "synchonous" or not.
2319  *
2320  * \param lock the lock to check
2321  * \retval 1 the lock is "synchonous"
2322  * \retval 0 the lock isn't "synchronous"
2323  *
2324  * \see mdt_set_lock_sync
2325  * \see mdt_save_lock
2326  */
2327 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2328 {
2329         return lock->l_ast_data != NULL;
2330 }
2331
2332 /**
2333  * Blocking AST for mdt locks.
2334  *
2335  * Starts transaction commit if in case of COS lock conflict or
2336  * deffers such a commit to the mdt_save_lock.
2337  *
2338  * \param lock the lock which blocks a request or cancelling lock
2339  * \param desc unused
2340  * \param data unused
2341  * \param flag indicates whether this cancelling or blocking callback
2342  * \retval 0
2343  * \see ldlm_blocking_ast_nocheck
2344  */
2345 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2346                      void *data, int flag)
2347 {
2348         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2349         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2350         int rc;
2351         ENTRY;
2352
2353         if (flag == LDLM_CB_CANCELING)
2354                 RETURN(0);
2355         lock_res_and_lock(lock);
2356         if (lock->l_blocking_ast != mdt_blocking_ast) {
2357                 unlock_res_and_lock(lock);
2358                 RETURN(0);
2359         }
2360         if (mdt_cos_is_enabled(mdt) &&
2361             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2362             lock->l_blocking_lock != NULL &&
2363             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2364                 mdt_set_lock_sync(lock);
2365         }
2366         rc = ldlm_blocking_ast_nocheck(lock);
2367
2368         /* There is no lock conflict if l_blocking_lock == NULL,
2369          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2370          * when the last reference to a local lock was released */
2371         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2372                 struct lu_env env;
2373
2374                 rc = lu_env_init(&env, LCT_LOCAL);
2375                 if (unlikely(rc != 0))
2376                         CWARN("lu_env initialization failed with rc = %d,"
2377                               "cannot start asynchronous commit\n", rc);
2378                 else
2379                         mdt_device_commit_async(&env, mdt);
2380                 lu_env_fini(&env);
2381         }
2382         RETURN(rc);
2383 }
2384
2385 static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o,
2386                             struct mdt_lock_handle *lh, __u64 ibits,
2387                             bool nonblock, int locality)
2388 {
2389         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2390         ldlm_policy_data_t *policy = &info->mti_policy;
2391         struct ldlm_res_id *res_id = &info->mti_res_id;
2392         __u64 dlmflags;
2393         int rc;
2394         ENTRY;
2395
2396         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2397         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2398         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2399         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2400
2401         if (mdt_object_exists(o) < 0) {
2402                 if (locality == MDT_CROSS_LOCK) {
2403                         /* cross-ref object fix */
2404                         ibits &= ~MDS_INODELOCK_UPDATE;
2405                         ibits |= MDS_INODELOCK_LOOKUP;
2406                 } else {
2407                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2408                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2409                 }
2410                 /* No PDO lock on remote object */
2411                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2412         }
2413
2414         if (lh->mlh_type == MDT_PDO_LOCK) {
2415                 /* check for exists after object is locked */
2416                 if (mdt_object_exists(o) == 0) {
2417                         /* Non-existent object shouldn't have PDO lock */
2418                         RETURN(-ESTALE);
2419                 } else {
2420                         /* Non-dir object shouldn't have PDO lock */
2421                         if (!S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)))
2422                                 RETURN(-ENOTDIR);
2423                 }
2424         }
2425
2426         memset(policy, 0, sizeof(*policy));
2427         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2428
2429         dlmflags = LDLM_FL_ATOMIC_CB;
2430         if (nonblock)
2431                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2432
2433         /*
2434          * Take PDO lock on whole directory and build correct @res_id for lock
2435          * on part of directory.
2436          */
2437         if (lh->mlh_pdo_hash != 0) {
2438                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2439                 mdt_lock_pdo_mode(info, o, lh);
2440                 if (lh->mlh_pdo_mode != LCK_NL) {
2441                         /*
2442                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2443                          * is never going to be sent to client and we do not
2444                          * want it slowed down due to possible cancels.
2445                          */
2446                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2447                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2448                                           policy, res_id, dlmflags,
2449                                           &info->mti_exp->exp_handle.h_cookie);
2450                         if (unlikely(rc))
2451                                 RETURN(rc);
2452                 }
2453
2454                 /*
2455                  * Finish res_id initializing by name hash marking part of
2456                  * directory which is taking modification.
2457                  */
2458                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2459         }
2460
2461         policy->l_inodebits.bits = ibits;
2462
2463         /*
2464          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2465          * going to be sent to client. If it is - mdt_intent_policy() path will
2466          * fix it up and turn FL_LOCAL flag off.
2467          */
2468         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2469                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2470                           &info->mti_exp->exp_handle.h_cookie);
2471         if (rc)
2472                 mdt_object_unlock(info, o, lh, 1);
2473         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2474                  lh->mlh_pdo_hash != 0 &&
2475                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2476                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2477         }
2478
2479         RETURN(rc);
2480 }
2481
2482 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2483                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2484 {
2485         return mdt_object_lock0(info, o, lh, ibits, false, locality);
2486 }
2487
2488 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2489                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2490 {
2491         struct mdt_lock_handle tmp = *lh;
2492         int rc;
2493
2494         rc = mdt_object_lock0(info, o, &tmp, ibits, true, locality);
2495         if (rc == 0)
2496                 *lh = tmp;
2497
2498         return rc == 0;
2499 }
2500
2501 /**
2502  * Save a lock within request object.
2503  *
2504  * Keep the lock referenced until whether client ACK or transaction
2505  * commit happens or release the lock immediately depending on input
2506  * parameters. If COS is ON, a write lock is converted to COS lock
2507  * before saving.
2508  *
2509  * \param info thead info object
2510  * \param h lock handle
2511  * \param mode lock mode
2512  * \param decref force immediate lock releasing
2513  */
2514 static
2515 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2516                    ldlm_mode_t mode, int decref)
2517 {
2518         ENTRY;
2519
2520         if (lustre_handle_is_used(h)) {
2521                 if (decref || !info->mti_has_trans ||
2522                     !(mode & (LCK_PW | LCK_EX))){
2523                         mdt_fid_unlock(h, mode);
2524                 } else {
2525                         struct mdt_device *mdt = info->mti_mdt;
2526                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2527                         struct ptlrpc_request *req = mdt_info_req(info);
2528                         int no_ack = 0;
2529
2530                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2531                                  h->cookie);
2532                         CDEBUG(D_HA, "request = %p reply state = %p"
2533                                " transno = "LPD64"\n",
2534                                req, req->rq_reply_state, req->rq_transno);
2535                         if (mdt_cos_is_enabled(mdt)) {
2536                                 no_ack = 1;
2537                                 ldlm_lock_downgrade(lock, LCK_COS);
2538                                 mode = LCK_COS;
2539                         }
2540                         ptlrpc_save_lock(req, h, mode, no_ack);
2541                         if (mdt_is_lock_sync(lock)) {
2542                                 CDEBUG(D_HA, "found sync-lock,"
2543                                        " async commit started\n");
2544                                 mdt_device_commit_async(info->mti_env,
2545                                                         mdt);
2546                         }
2547                         LDLM_LOCK_PUT(lock);
2548                 }
2549                 h->cookie = 0ull;
2550         }
2551
2552         EXIT;
2553 }
2554
2555 /**
2556  * Unlock mdt object.
2557  *
2558  * Immeditely release the regular lock and the PDO lock or save the
2559  * lock in reqeuest and keep them referenced until client ACK or
2560  * transaction commit.
2561  *
2562  * \param info thread info object
2563  * \param o mdt object
2564  * \param lh mdt lock handle referencing regular and PDO locks
2565  * \param decref force immediate lock releasing
2566  */
2567 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2568                        struct mdt_lock_handle *lh, int decref)
2569 {
2570         ENTRY;
2571
2572         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2573         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2574
2575         EXIT;
2576 }
2577
2578 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2579                                         const struct lu_fid *f,
2580                                         struct mdt_lock_handle *lh,
2581                                         __u64 ibits)
2582 {
2583         struct mdt_object *o;
2584
2585         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2586         if (!IS_ERR(o)) {
2587                 int rc;
2588
2589                 rc = mdt_object_lock(info, o, lh, ibits,
2590                                      MDT_LOCAL_LOCK);
2591                 if (rc != 0) {
2592                         mdt_object_put(info->mti_env, o);
2593                         o = ERR_PTR(rc);
2594                 }
2595         }
2596         return o;
2597 }
2598
2599 void mdt_object_unlock_put(struct mdt_thread_info * info,
2600                            struct mdt_object * o,
2601                            struct mdt_lock_handle *lh,
2602                            int decref)
2603 {
2604         mdt_object_unlock(info, o, lh, decref);
2605         mdt_object_put(info->mti_env, o);
2606 }
2607
2608 struct mdt_handler *mdt_handler_find(__u32 opc, struct mdt_opc_slice *supported)
2609 {
2610         struct mdt_opc_slice *s;
2611         struct mdt_handler   *h;
2612
2613         h = NULL;
2614         for (s = supported; s->mos_hs != NULL; s++) {
2615                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2616                         h = s->mos_hs + (opc - s->mos_opc_start);
2617                         if (likely(h->mh_opc != 0))
2618                                 LASSERTF(h->mh_opc == opc,
2619                                          "opcode mismatch %d != %d\n",
2620                                          h->mh_opc, opc);
2621                         else
2622                                 h = NULL; /* unsupported opc */
2623                         break;
2624                 }
2625         }
2626         return h;
2627 }
2628
2629 static int mdt_lock_resname_compat(struct mdt_device *m,
2630                                    struct ldlm_request *req)
2631 {
2632         /* XXX something... later. */
2633         return 0;
2634 }
2635
2636 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2637 {
2638         /* XXX something... later. */
2639         return 0;
2640 }
2641
2642 /*
2643  * Generic code handling requests that have struct mdt_body passed in:
2644  *
2645  *  - extract mdt_body from request and save it in @info, if present;
2646  *
2647  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2648  *  @info;
2649  *
2650  *  - if HABEO_CORPUS flag is set for this request type check whether object
2651  *  actually exists on storage (lu_object_exists()).
2652  *
2653  */
2654 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2655 {
2656         const struct mdt_body    *body;
2657         struct mdt_object        *obj;
2658         const struct lu_env      *env;
2659         struct req_capsule       *pill;
2660         int                       rc;
2661         ENTRY;
2662
2663         env = info->mti_env;
2664         pill = info->mti_pill;
2665
2666         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2667         if (body == NULL)
2668                 RETURN(-EFAULT);
2669
2670         if (!(body->valid & OBD_MD_FLID))
2671                 RETURN(0);
2672
2673         if (!fid_is_sane(&body->fid1)) {
2674                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2675                 RETURN(-EINVAL);
2676         }
2677
2678         /*
2679          * Do not get size or any capa fields before we check that request
2680          * contains capa actually. There are some requests which do not, for
2681          * instance MDS_IS_SUBDIR.
2682          */
2683         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2684             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2685                 mdt_set_capainfo(info, 0, &body->fid1,
2686                                  req_capsule_client_get(pill, &RMF_CAPA1));
2687
2688         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2689         if (!IS_ERR(obj)) {
2690                 if ((flags & HABEO_CORPUS) &&
2691                     !mdt_object_exists(obj)) {
2692                         mdt_object_put(env, obj);
2693                         /* for capability renew ENOENT will be handled in
2694                          * mdt_renew_capa */
2695                         if (body->valid & OBD_MD_FLOSSCAPA)
2696                                 rc = 0;
2697                         else
2698                                 rc = -ENOENT;
2699                 } else {
2700                         info->mti_object = obj;
2701                         rc = 0;
2702                 }
2703         } else
2704                 rc = PTR_ERR(obj);
2705
2706         RETURN(rc);
2707 }
2708
2709 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2710 {
2711         struct req_capsule *pill = info->mti_pill;
2712         int rc;
2713         ENTRY;
2714
2715         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2716                 rc = mdt_body_unpack(info, flags);
2717         else
2718                 rc = 0;
2719
2720         if (rc == 0 && (flags & HABEO_REFERO)) {
2721                 /* Pack reply. */
2722                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2723                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2724                                              info->mti_body->eadatasize);
2725                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2726                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2727                                              RCL_SERVER, 0);
2728
2729                 rc = req_capsule_server_pack(pill);
2730         }
2731         RETURN(rc);
2732 }
2733
2734 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2735 {
2736         struct md_device *next = m->mdt_child;
2737
2738         return next->md_ops->mdo_init_capa_ctxt(env, next,
2739                                                 m->mdt_opts.mo_mds_capa,
2740                                                 m->mdt_capa_timeout,
2741                                                 m->mdt_capa_alg,
2742                                                 m->mdt_capa_keys);
2743 }
2744
2745 /*
2746  * Invoke handler for this request opc. Also do necessary preprocessing
2747  * (according to handler ->mh_flags), and post-processing (setting of
2748  * ->last_{xid,committed}).
2749  */
2750 static int mdt_req_handle(struct mdt_thread_info *info,
2751                           struct mdt_handler *h, struct ptlrpc_request *req)
2752 {
2753         int   rc, serious = 0;
2754         __u32 flags;
2755
2756         ENTRY;
2757
2758         LASSERT(h->mh_act != NULL);
2759         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2760         LASSERT(current->journal_info == NULL);
2761
2762         /*
2763          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2764          * to put same checks into handlers like mdt_close(), mdt_reint(),
2765          * etc., without talking to mdt authors first. Checking same thing
2766          * there again is useless and returning 0 error without packing reply
2767          * is buggy! Handlers either pack reply or return error.
2768          *
2769          * We return 0 here and do not send any reply in order to emulate
2770          * network failure. Do not send any reply in case any of NET related
2771          * fail_id has occured.
2772          */
2773         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2774                 RETURN(0);
2775
2776         rc = 0;
2777         flags = h->mh_flags;
2778         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2779
2780         if (h->mh_fmt != NULL) {
2781                 req_capsule_set(info->mti_pill, h->mh_fmt);
2782                 rc = mdt_unpack_req_pack_rep(info, flags);
2783         }
2784
2785         if (rc == 0 && flags & MUTABOR &&
2786             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2787                 /* should it be rq_status? */
2788                 rc = -EROFS;
2789
2790         if (rc == 0 && flags & HABEO_CLAVIS) {
2791                 struct ldlm_request *dlm_req;
2792
2793                 LASSERT(h->mh_fmt != NULL);
2794
2795                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2796                 if (dlm_req != NULL) {
2797                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2798                                         LDLM_IBITS &&
2799                                      dlm_req->lock_desc.l_policy_data.\
2800                                         l_inodebits.bits == 0)) {
2801                                 /*
2802                                  * Lock without inodebits makes no sense and
2803                                  * will oops later in ldlm. If client miss to
2804                                  * set such bits, do not trigger ASSERTION.
2805                                  *
2806                                  * For liblustre flock case, it maybe zero.
2807                                  */
2808                                 rc = -EPROTO;
2809                         } else {
2810                                 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2811                                         rc = mdt_lock_resname_compat(
2812                                                                 info->mti_mdt,
2813                                                                 dlm_req);
2814                                 info->mti_dlm_req = dlm_req;
2815                         }
2816                 } else {
2817                         rc = -EFAULT;
2818                 }
2819         }
2820
2821         /* capability setting changed via /proc, needs reinitialize ctxt */
2822         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2823                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2824                 info->mti_mdt->mdt_capa_conf = 0;
2825         }
2826
2827         if (likely(rc == 0)) {
2828                 /*
2829                  * Process request, there can be two types of rc:
2830                  * 1) errors with msg unpack/pack, other failures outside the
2831                  * operation itself. This is counted as serious errors;
2832                  * 2) errors during fs operation, should be placed in rq_status
2833                  * only
2834                  */
2835                 rc = h->mh_act(info);
2836                 if (rc == 0 &&
2837                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2838                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2839                                   "pack reply and returned 0 error\n",
2840                                   h->mh_name);
2841                         LBUG();
2842                 }
2843                 serious = is_serious(rc);
2844                 rc = clear_serious(rc);
2845         } else
2846                 serious = 1;
2847
2848         req->rq_status = rc;
2849
2850         /*
2851          * ELDLM_* codes which > 0 should be in rq_status only as well as
2852          * all non-serious errors.
2853          */
2854         if (rc > 0 || !serious)
2855                 rc = 0;
2856
2857         LASSERT(current->journal_info == NULL);
2858
2859         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2860             info->mti_mdt->mdt_opts.mo_compat_resname) {
2861                 struct ldlm_reply *dlmrep;
2862
2863                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2864                 if (dlmrep != NULL)
2865                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2866         }
2867
2868         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2869         if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2870                 target_committed_to_req(req);
2871
2872         if (unlikely(req_is_replay(req) &&
2873                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2874                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2875                 LBUG();
2876         }
2877
2878         target_send_reply(req, rc, info->mti_fail_id);
2879         RETURN(0);
2880 }
2881
2882 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2883 {
2884         lh->mlh_type = MDT_NUL_LOCK;
2885         lh->mlh_reg_lh.cookie = 0ull;
2886         lh->mlh_reg_mode = LCK_MINMODE;
2887         lh->mlh_pdo_lh.cookie = 0ull;
2888         lh->mlh_pdo_mode = LCK_MINMODE;
2889 }
2890
2891 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2892 {
2893         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2894         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2895 }
2896
2897 /*
2898  * Initialize fields of struct mdt_thread_info. Other fields are left in
2899  * uninitialized state, because it's too expensive to zero out whole
2900  * mdt_thread_info (> 1K) on each request arrival.
2901  */
2902 static void mdt_thread_info_init(struct ptlrpc_request *req,
2903                                  struct mdt_thread_info *info)
2904 {
2905         int i;
2906
2907         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2908         info->mti_pill = &req->rq_pill;
2909
2910         /* lock handle */
2911         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2912                 mdt_lock_handle_init(&info->mti_lh[i]);
2913
2914         /* mdt device: it can be NULL while CONNECT */
2915         if (req->rq_export) {
2916                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2917                 info->mti_exp = req->rq_export;
2918         } else
2919                 info->mti_mdt = NULL;
2920         info->mti_env = req->rq_svc_thread->t_env;
2921         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2922         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2923         info->mti_mos = NULL;
2924
2925         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2926         info->mti_body = NULL;
2927         info->mti_object = NULL;
2928         info->mti_dlm_req = NULL;
2929         info->mti_has_trans = 0;
2930         info->mti_cross_ref = 0;
2931         info->mti_opdata = 0;
2932         info->mti_big_lmm_used = 0;
2933
2934         /* To not check for split by default. */
2935         info->mti_spec.no_create = 0;
2936 }
2937
2938 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2939 {
2940         int i;
2941
2942         req_capsule_fini(info->mti_pill);
2943         if (info->mti_object != NULL) {
2944                 mdt_object_put(info->mti_env, info->mti_object);
2945                 info->mti_object = NULL;
2946         }
2947         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2948                 mdt_lock_handle_fini(&info->mti_lh[i]);
2949         info->mti_env = NULL;
2950 }
2951
2952 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2953                                        struct obd_device *obd, int *process)
2954 {
2955         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2956         case MDS_CONNECT: /* This will never get here, but for completeness. */
2957         case OST_CONNECT: /* This will never get here, but for completeness. */
2958         case MDS_DISCONNECT:
2959         case OST_DISCONNECT:
2960         case OBD_IDX_READ:
2961                *process = 1;
2962                RETURN(0);
2963
2964         case MDS_CLOSE:
2965         case MDS_DONE_WRITING:
2966         case MDS_SYNC: /* used in unmounting */
2967         case OBD_PING:
2968         case MDS_REINT:
2969         case SEQ_QUERY:
2970         case FLD_QUERY:
2971         case LDLM_ENQUEUE:
2972                 *process = target_queue_recovery_request(req, obd);
2973                 RETURN(0);
2974
2975         default:
2976                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2977                 *process = -EAGAIN;
2978                 RETURN(0);
2979         }
2980 }
2981
2982 /*
2983  * Handle recovery. Return:
2984  *        +1: continue request processing;
2985  *       -ve: abort immediately with the given error code;
2986  *         0: send reply with error code in req->rq_status;
2987  */
2988 static int mdt_recovery(struct mdt_thread_info *info)
2989 {
2990         struct ptlrpc_request *req = mdt_info_req(info);
2991         struct obd_device *obd;
2992
2993         ENTRY;
2994
2995         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2996         case MDS_CONNECT:
2997         case SEC_CTX_INIT:
2998         case SEC_CTX_INIT_CONT:
2999         case SEC_CTX_FINI:
3000                 {
3001 #if 0
3002                         int rc;
3003
3004                         rc = mdt_handle_idmap(info);
3005                         if (rc)
3006                                 RETURN(rc);
3007                         else
3008 #endif
3009                                 RETURN(+1);
3010                 }
3011         }
3012
3013         if (unlikely(!class_connected_export(req->rq_export))) {
3014                 CDEBUG(D_HA, "operation %d on unconnected MDS from %s\n",
3015                        lustre_msg_get_opc(req->rq_reqmsg),
3016                        libcfs_id2str(req->rq_peer));
3017                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
3018                  * mds_A will get -ENOTCONN(especially for ping req),
3019                  * which will cause that mds_A deactive timeout, then when
3020                  * mds_A cleanup, the cleanup process will be suspended since
3021                  * deactive timeout is not zero.
3022                  */
3023                 req->rq_status = -ENOTCONN;
3024                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
3025                 RETURN(0);
3026         }
3027
3028         /* sanity check: if the xid matches, the request must be marked as a
3029          * resent or replayed */
3030         if (req_xid_is_last(req)) {
3031                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
3032                       (MSG_RESENT | MSG_REPLAY))) {
3033                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
3034                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
3035                                   lustre_msg_get_flags(req->rq_reqmsg));
3036                         LBUG();
3037                         req->rq_status = -ENOTCONN;
3038                         RETURN(-ENOTCONN);
3039                 }
3040         }
3041
3042         /* else: note the opposite is not always true; a RESENT req after a
3043          * failover will usually not match the last_xid, since it was likely
3044          * never committed. A REPLAYed request will almost never match the
3045          * last xid, however it could for a committed, but still retained,
3046          * open. */
3047
3048         obd = req->rq_export->exp_obd;
3049
3050         /* Check for aborted recovery... */
3051         if (unlikely(obd->obd_recovering)) {
3052                 int rc;
3053                 int should_process;
3054                 DEBUG_REQ(D_INFO, req, "Got new replay");
3055                 rc = mdt_filter_recovery_request(req, obd, &should_process);
3056                 if (rc != 0 || !should_process)
3057                         RETURN(rc);
3058                 else if (should_process < 0) {
3059                         req->rq_status = should_process;
3060                         rc = ptlrpc_error(req);
3061                         RETURN(rc);
3062                 }
3063         }
3064         RETURN(+1);
3065 }
3066
3067 static int mdt_msg_check_version(struct lustre_msg *msg)
3068 {
3069         int rc;
3070
3071         switch (lustre_msg_get_opc(msg)) {
3072         case MDS_CONNECT:
3073         case MDS_DISCONNECT:
3074         case OBD_PING:
3075         case SEC_CTX_INIT:
3076         case SEC_CTX_INIT_CONT:
3077         case SEC_CTX_FINI:
3078         case OBD_IDX_READ:
3079                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
3080                 if (rc)
3081                         CERROR("bad opc %u version %08x, expecting %08x\n",
3082                                lustre_msg_get_opc(msg),
3083                                lustre_msg_get_version(msg),
3084                                LUSTRE_OBD_VERSION);
3085                 break;
3086         case MDS_GETSTATUS:
3087         case MDS_GETATTR:
3088         case MDS_GETATTR_NAME:
3089         case MDS_STATFS:
3090         case MDS_READPAGE:
3091         case MDS_WRITEPAGE:
3092         case MDS_IS_SUBDIR:
3093         case MDS_REINT:
3094         case MDS_CLOSE:
3095         case MDS_DONE_WRITING:
3096         case MDS_PIN:
3097         case MDS_SYNC:
3098         case MDS_GETXATTR:
3099         case MDS_SETXATTR:
3100         case MDS_SET_INFO:
3101         case MDS_GET_INFO:
3102         case MDS_HSM_PROGRESS:
3103         case MDS_HSM_REQUEST:
3104         case MDS_HSM_CT_REGISTER:
3105         case MDS_HSM_CT_UNREGISTER:
3106         case MDS_HSM_STATE_GET:
3107         case MDS_HSM_STATE_SET:
3108         case MDS_HSM_ACTION:
3109         case MDS_QUOTACHECK:
3110         case MDS_QUOTACTL:
3111         case UPDATE_OBJ:
3112         case QUOTA_DQACQ:
3113         case QUOTA_DQREL:
3114         case SEQ_QUERY:
3115         case FLD_QUERY:
3116                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
3117                 if (rc)
3118                         CERROR("bad opc %u version %08x, expecting %08x\n",
3119                                lustre_msg_get_opc(msg),
3120                                lustre_msg_get_version(msg),
3121                                LUSTRE_MDS_VERSION);
3122                 break;
3123         case LDLM_ENQUEUE:
3124         case LDLM_CONVERT:
3125         case LDLM_BL_CALLBACK:
3126         case LDLM_CP_CALLBACK:
3127                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
3128                 if (rc)
3129                         CERROR("bad opc %u version %08x, expecting %08x\n",
3130                                lustre_msg_get_opc(msg),
3131                                lustre_msg_get_version(msg),
3132                                LUSTRE_DLM_VERSION);
3133                 break;
3134         case OBD_LOG_CANCEL:
3135         case LLOG_ORIGIN_HANDLE_CREATE:
3136         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
3137         case LLOG_ORIGIN_HANDLE_READ_HEADER:
3138         case LLOG_ORIGIN_HANDLE_CLOSE:
3139         case LLOG_ORIGIN_HANDLE_DESTROY:
3140         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
3141         case LLOG_CATINFO:
3142                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
3143                 if (rc)
3144                         CERROR("bad opc %u version %08x, expecting %08x\n",
3145                                lustre_msg_get_opc(msg),
3146                                lustre_msg_get_version(msg),
3147                                LUSTRE_LOG_VERSION);
3148                 break;
3149         default:
3150                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
3151                 rc = -ENOTSUPP;
3152         }
3153         return rc;
3154 }
3155
3156 static int mdt_handle0(struct ptlrpc_request *req,
3157                        struct mdt_thread_info *info,
3158                        struct mdt_opc_slice *supported)
3159 {
3160         struct mdt_handler *h;
3161         struct lustre_msg  *msg;
3162         int                 rc;
3163
3164         ENTRY;
3165
3166         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
3167                 RETURN(0);
3168
3169         LASSERT(current->journal_info == NULL);
3170
3171         msg = req->rq_reqmsg;
3172         rc = mdt_msg_check_version(msg);
3173         if (likely(rc == 0)) {
3174                 rc = mdt_recovery(info);
3175                 if (likely(rc == +1)) {
3176                         h = mdt_handler_find(lustre_msg_get_opc(msg),
3177                                              supported);
3178                         if (likely(h != NULL)) {
3179                                 rc = mdt_req_handle(info, h, req);
3180                         } else {
3181                                 CERROR("The unsupported opc: 0x%x\n",
3182                                        lustre_msg_get_opc(msg) );
3183                                 req->rq_status = -ENOTSUPP;
3184                                 rc = ptlrpc_error(req);
3185                                 RETURN(rc);
3186                         }
3187                 }
3188         } else
3189                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
3190         RETURN(rc);
3191 }
3192
3193 /*
3194  * MDT handler function called by ptlrpc service thread when request comes.
3195  *
3196  * XXX common "target" functionality should be factored into separate module
3197  * shared by mdt, ost and stand-alone services like fld.
3198  */
3199 int mdt_handle_common(struct ptlrpc_request *req,
3200                       struct mdt_opc_slice *supported)
3201 {
3202         struct lu_env          *env;
3203         struct mdt_thread_info *info;
3204         int                     rc;
3205         ENTRY;
3206
3207         env = req->rq_svc_thread->t_env;
3208         /* Refill(initilize) the context(mdt_thread_info), in case it is
3209          * not initialized yet. Usually it happens during start up, after
3210          * MDS(ptlrpc threads) is start up, it gets the first CONNECT request,
3211          * before MDT_thread_info is initialized */
3212         lu_env_refill(env);
3213         LASSERT(env != NULL);
3214         LASSERT(env->le_ses != NULL);
3215         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
3216         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3217         LASSERT(info != NULL);
3218
3219         mdt_thread_info_init(req, info);
3220
3221         rc = mdt_handle0(req, info, supported);
3222
3223         mdt_thread_info_fini(info);
3224         RETURN(rc);
3225 }
3226
3227 /*
3228  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
3229  * as well.
3230  */
3231 int mdt_recovery_handle(struct ptlrpc_request *req)
3232 {
3233         int rc;
3234         ENTRY;
3235
3236         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3237         case FLD_QUERY:
3238                 rc = mdt_handle_common(req, mdt_fld_handlers);
3239                 break;
3240         case SEQ_QUERY:
3241                 rc = mdt_handle_common(req, mdt_seq_handlers);
3242                 break;
3243         default:
3244                 rc = mdt_handle_common(req, mdt_regular_handlers);
3245                 break;
3246         }
3247
3248         RETURN(rc);
3249 }
3250
3251 enum mdt_it_code {
3252         MDT_IT_OPEN,
3253         MDT_IT_OCREAT,
3254         MDT_IT_CREATE,
3255         MDT_IT_GETATTR,
3256         MDT_IT_READDIR,
3257         MDT_IT_LOOKUP,
3258         MDT_IT_UNLINK,
3259         MDT_IT_TRUNC,
3260         MDT_IT_GETXATTR,
3261         MDT_IT_LAYOUT,
3262         MDT_IT_QUOTA,
3263         MDT_IT_NR
3264 };
3265
3266 static int mdt_intent_getattr(enum mdt_it_code opcode,
3267                               struct mdt_thread_info *info,
3268                               struct ldlm_lock **,
3269                               __u64);
3270 static int mdt_intent_layout(enum mdt_it_code opcode,
3271                              struct mdt_thread_info *info,
3272                              struct ldlm_lock **,
3273                              __u64);
3274 static int mdt_intent_reint(enum mdt_it_code opcode,
3275                             struct mdt_thread_info *info,
3276                             struct ldlm_lock **,
3277                             __u64);
3278
3279 static struct mdt_it_flavor {
3280         const struct req_format *it_fmt;
3281         __u32                    it_flags;
3282         int                    (*it_act)(enum mdt_it_code ,
3283                                          struct mdt_thread_info *,
3284                                          struct ldlm_lock **,
3285                                          __u64);
3286         long                     it_reint;
3287 } mdt_it_flavor[] = {
3288         [MDT_IT_OPEN]     = {
3289                 .it_fmt   = &RQF_LDLM_INTENT,
3290                 /*.it_flags = HABEO_REFERO,*/
3291                 .it_flags = 0,
3292                 .it_act   = mdt_intent_reint,
3293                 .it_reint = REINT_OPEN
3294         },
3295         [MDT_IT_OCREAT]   = {
3296                 .it_fmt   = &RQF_LDLM_INTENT,
3297                 .it_flags = MUTABOR,
3298                 .it_act   = mdt_intent_reint,
3299                 .it_reint = REINT_OPEN
3300         },
3301         [MDT_IT_CREATE]   = {
3302                 .it_fmt   = &RQF_LDLM_INTENT,
3303                 .it_flags = MUTABOR,
3304                 .it_act   = mdt_intent_reint,
3305                 .it_reint = REINT_CREATE
3306         },
3307         [MDT_IT_GETATTR]  = {
3308                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3309                 .it_flags = HABEO_REFERO,
3310                 .it_act   = mdt_intent_getattr
3311         },
3312         [MDT_IT_READDIR]  = {
3313                 .it_fmt   = NULL,
3314                 .it_flags = 0,
3315                 .it_act   = NULL
3316         },
3317         [MDT_IT_LOOKUP]   = {
3318                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3319                 .it_flags = HABEO_REFERO,
3320                 .it_act   = mdt_intent_getattr
3321         },
3322         [MDT_IT_UNLINK]   = {
3323                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3324                 .it_flags = MUTABOR,
3325                 .it_act   = NULL,
3326                 .it_reint = REINT_UNLINK
3327         },
3328         [MDT_IT_TRUNC]    = {
3329                 .it_fmt   = NULL,
3330                 .it_flags = MUTABOR,
3331                 .it_act   = NULL
3332         },
3333         [MDT_IT_GETXATTR] = {
3334                 .it_fmt   = NULL,
3335                 .it_flags = 0,
3336                 .it_act   = NULL
3337         },
3338         [MDT_IT_LAYOUT] = {
3339                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3340                 .it_flags = 0,
3341                 .it_act   = mdt_intent_layout
3342         }
3343 };
3344
3345 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3346                             struct ldlm_lock **lockp,
3347                             struct ldlm_lock *new_lock,
3348                             struct mdt_lock_handle *lh,
3349                             __u64 flags)
3350 {
3351         struct ptlrpc_request  *req = mdt_info_req(info);
3352         struct ldlm_lock       *lock = *lockp;
3353
3354         /*
3355          * Get new lock only for cases when possible resent did not find any
3356          * lock.
3357          */
3358         if (new_lock == NULL)
3359                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3360
3361         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3362                 lh->mlh_reg_lh.cookie = 0;
3363                 RETURN(0);
3364         }
3365
3366         LASSERTF(new_lock != NULL,
3367                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3368
3369         /*
3370          * If we've already given this lock to a client once, then we should
3371          * have no readers or writers.  Otherwise, we should have one reader
3372          * _or_ writer ref (which will be zeroed below) before returning the
3373          * lock to a client.
3374          */
3375         if (new_lock->l_export == req->rq_export) {
3376                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3377         } else {
3378                 LASSERT(new_lock->l_export == NULL);
3379                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3380         }
3381
3382         *lockp = new_lock;
3383
3384         if (new_lock->l_export == req->rq_export) {
3385                 /*
3386                  * Already gave this to the client, which means that we
3387                  * reconstructed a reply.
3388                  */
3389                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3390                         MSG_RESENT);
3391                 lh->mlh_reg_lh.cookie = 0;
3392                 RETURN(ELDLM_LOCK_REPLACED);
3393         }
3394
3395         /*
3396          * Fixup the lock to be given to the client.
3397          */
3398         lock_res_and_lock(new_lock);
3399         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3400          * possible blocking AST. */
3401         while (new_lock->l_readers > 0) {
3402                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3403                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3404                 new_lock->l_readers--;
3405         }
3406         while (new_lock->l_writers > 0) {
3407                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3408                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3409                 new_lock->l_writers--;
3410         }
3411
3412         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3413         new_lock->l_blocking_ast = lock->l_blocking_ast;
3414         new_lock->l_completion_ast = lock->l_completion_ast;
3415         new_lock->l_remote_handle = lock->l_remote_handle;
3416         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3417
3418         unlock_res_and_lock(new_lock);
3419
3420         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3421                      &new_lock->l_remote_handle,
3422                      &new_lock->l_exp_hash);
3423
3424         LDLM_LOCK_RELEASE(new_lock);
3425         lh->mlh_reg_lh.cookie = 0;
3426
3427         RETURN(ELDLM_LOCK_REPLACED);
3428 }
3429
3430 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3431                                     struct ldlm_lock *new_lock,
3432                                     struct ldlm_lock **old_lock,
3433                                     struct mdt_lock_handle *lh)
3434 {
3435         struct ptlrpc_request  *req = mdt_info_req(info);
3436         struct obd_export      *exp = req->rq_export;
3437         struct lustre_handle    remote_hdl;
3438         struct ldlm_request    *dlmreq;
3439         struct ldlm_lock       *lock;
3440
3441         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3442                 return;
3443
3444         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3445         remote_hdl = dlmreq->lock_handle[0];
3446
3447         /* In the function below, .hs_keycmp resolves to
3448          * ldlm_export_lock_keycmp() */
3449         /* coverity[overrun-buffer-val] */
3450         lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3451         if (lock) {
3452                 if (lock != new_lock) {
3453                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3454                         lh->mlh_reg_mode = lock->l_granted_mode;
3455
3456                         LDLM_DEBUG(lock, "Restoring lock cookie");
3457                         DEBUG_REQ(D_DLMTRACE, req,
3458                                   "restoring lock cookie "LPX64,
3459                                   lh->mlh_reg_lh.cookie);
3460                         if (old_lock)
3461                                 *old_lock = LDLM_LOCK_GET(lock);
3462                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3463                         return;
3464                 }
3465
3466                 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3467         }
3468
3469         /*
3470          * If the xid matches, then we know this is a resent request, and allow
3471          * it. (It's probably an OPEN, for which we don't send a lock.
3472          */
3473         if (req_xid_is_last(req))
3474                 return;
3475
3476         /*
3477          * This remote handle isn't enqueued, so we never received or processed
3478          * this request.  Clear MSG_RESENT, because it can be handled like any
3479          * normal request now.
3480          */
3481         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3482
3483         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3484                   remote_hdl.cookie);
3485 }
3486
3487 static int mdt_intent_getattr(enum mdt_it_code opcode,
3488                               struct mdt_thread_info *info,
3489                               struct ldlm_lock **lockp,
3490                               __u64 flags)
3491 {
3492         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3493         struct ldlm_lock       *new_lock = NULL;
3494         __u64                   child_bits;
3495         struct ldlm_reply      *ldlm_rep;
3496         struct ptlrpc_request  *req;
3497         struct mdt_body        *reqbody;
3498         struct mdt_body        *repbody;
3499         int                     rc, rc2;
3500         ENTRY;
3501
3502         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3503         LASSERT(reqbody);
3504
3505         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3506         LASSERT(repbody);
3507
3508         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3509         repbody->eadatasize = 0;
3510         repbody->aclsize = 0;
3511
3512         switch (opcode) {
3513         case MDT_IT_LOOKUP:
3514                 child_bits = MDS_INODELOCK_LOOKUP;
3515                 break;
3516         case MDT_IT_GETATTR:
3517                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3518                 break;
3519         default:
3520                 CERROR("Unsupported intent (%d)\n", opcode);
3521                 GOTO(out_shrink, rc = -EINVAL);
3522         }
3523
3524         rc = mdt_init_ucred(info, reqbody);
3525         if (rc)
3526                 GOTO(out_shrink, rc);
3527
3528         req = info->mti_pill->rc_req;
3529         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3530         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3531
3532         /* Get lock from request for possible resent case. */
3533         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3534
3535         ldlm_rep->lock_policy_res2 =
3536                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3537
3538         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3539                 ldlm_rep->lock_policy_res2 = 0;
3540         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3541             ldlm_rep->lock_policy_res2) {
3542                 lhc->mlh_reg_lh.cookie = 0ull;
3543                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3544         }
3545
3546         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3547         EXIT;
3548 out_ucred:
3549         mdt_exit_ucred(info);
3550 out_shrink:
3551         mdt_client_compatibility(info);
3552         rc2 = mdt_fix_reply(info);
3553         if (rc == 0)
3554                 rc = rc2;
3555         return rc;
3556 }
3557
3558 static int mdt_intent_layout(enum mdt_it_code opcode,
3559                              struct mdt_thread_info *info,
3560                              struct ldlm_lock **lockp,
3561                              __u64 flags)
3562 {
3563         struct layout_intent *layout;
3564         int rc;
3565         ENTRY;
3566
3567         if (opcode != MDT_IT_LAYOUT) {
3568                 CERROR("%s: Unknown intent (%d)\n",
3569                         info->mti_exp->exp_obd->obd_name, opcode);
3570                 RETURN(-EINVAL);
3571         }
3572
3573         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3574         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3575                         ldlm_lvbo_size(*lockp));
3576         rc = req_capsule_server_pack(info->mti_pill);
3577         if (rc != 0)
3578                 RETURN(-EINVAL);
3579
3580         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3581         LASSERT(layout != NULL);
3582         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3583                 /* return to normal ldlm handling */
3584                 RETURN(0);
3585
3586         CERROR("%s: Unsupported layout intent (%d)\n",
3587                 info->mti_exp->exp_obd->obd_name, layout->li_opc);
3588         RETURN(-EINVAL);
3589 }
3590
3591 static int mdt_intent_reint(enum mdt_it_code opcode,
3592                             struct mdt_thread_info *info,
3593                             struct ldlm_lock **lockp,
3594                             __u64 flags)
3595 {
3596         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3597         struct ldlm_reply      *rep = NULL;
3598         long                    opc;
3599         int                     rc;
3600
3601         static const struct req_format *intent_fmts[REINT_MAX] = {
3602                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3603                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3604         };
3605
3606         ENTRY;
3607
3608         opc = mdt_reint_opcode(info, intent_fmts);
3609         if (opc < 0)
3610                 RETURN(opc);
3611
3612         if (mdt_it_flavor[opcode].it_reint != opc) {
3613                 CERROR("Reint code %ld doesn't match intent: %d\n",
3614                        opc, opcode);
3615                 RETURN(err_serious(-EPROTO));
3616         }
3617
3618         /* Get lock from request for possible resent case. */
3619         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3620
3621         rc = mdt_reint_internal(info, lhc, opc);
3622
3623         /* Check whether the reply has been packed successfully. */
3624         if (mdt_info_req(info)->rq_repmsg != NULL)
3625                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3626         if (rep == NULL)
3627                 RETURN(err_serious(-EFAULT));
3628
3629         /* MDC expects this in any case */
3630         if (rc != 0)
3631                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3632
3633         /* Cross-ref case, the lock should be returned to the client */
3634         if (rc == -EREMOTE) {
3635                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3636                 rep->lock_policy_res2 = 0;
3637                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3638                 RETURN(rc);
3639         }
3640         rep->lock_policy_res2 = clear_serious(rc);
3641
3642         if (rep->lock_policy_res2 == -ENOENT &&
3643             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3644                 rep->lock_policy_res2 = 0;
3645
3646         if (rc == -ENOTCONN || rc == -ENODEV ||
3647             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3648                 /*
3649                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3650                  * will be returned by rq_status, and client at ptlrpc layer
3651                  * will detect this, then disconnect, reconnect the import
3652                  * immediately, instead of impacting the following the rpc.
3653                  */
3654                 lhc->mlh_reg_lh.cookie = 0ull;
3655                 RETURN(rc);
3656         } else {
3657                 /*
3658                  * For other cases, the error will be returned by intent.
3659                  * and client will retrieve the result from intent.
3660                  */
3661                  /*
3662                   * FIXME: when open lock is finished, that should be
3663                   * checked here.
3664                   */
3665                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3666                         LASSERTF(rc == 0, "Error occurred but lock handle "
3667                                  "is still in use, rc = %d\n", rc);
3668                         rep->lock_policy_res2 = 0;
3669                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3670                         RETURN(rc);
3671                 } else {
3672                         lhc->mlh_reg_lh.cookie = 0ull;
3673                         RETURN(ELDLM_LOCK_ABORTED);
3674                 }
3675         }
3676 }
3677
3678 static int mdt_intent_code(long itcode)
3679 {
3680         int rc;
3681
3682         switch(itcode) {
3683         case IT_OPEN:
3684                 rc = MDT_IT_OPEN;
3685                 break;
3686         case IT_OPEN|IT_CREAT:
3687                 rc = MDT_IT_OCREAT;
3688                 break;
3689         case IT_CREAT:
3690                 rc = MDT_IT_CREATE;
3691                 break;
3692         case IT_READDIR:
3693                 rc = MDT_IT_READDIR;
3694                 break;
3695         case IT_GETATTR:
3696                 rc = MDT_IT_GETATTR;
3697                 break;
3698         case IT_LOOKUP:
3699                 rc = MDT_IT_LOOKUP;
3700                 break;
3701         case IT_UNLINK:
3702                 rc = MDT_IT_UNLINK;
3703                 break;
3704         case IT_TRUNC:
3705                 rc = MDT_IT_TRUNC;
3706                 break;
3707         case IT_GETXATTR:
3708                 rc = MDT_IT_GETXATTR;
3709                 break;
3710         case IT_LAYOUT:
3711                 rc = MDT_IT_LAYOUT;
3712                 break;
3713         case IT_QUOTA_DQACQ:
3714         case IT_QUOTA_CONN:
3715                 rc = MDT_IT_QUOTA;
3716                 break;
3717         default:
3718                 CERROR("Unknown intent opcode: %ld\n", itcode);
3719                 rc = -EINVAL;
3720                 break;
3721         }
3722         return rc;
3723 }
3724
3725 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3726                           struct ldlm_lock **lockp, __u64 flags)
3727 {
3728         struct req_capsule   *pill;
3729         struct mdt_it_flavor *flv;
3730         int opc;
3731         int rc;
3732         ENTRY;
3733
3734         opc = mdt_intent_code(itopc);
3735         if (opc < 0)
3736                 RETURN(-EINVAL);
3737
3738         pill = info->mti_pill;
3739
3740         if (opc == MD