Whamcloud - gitweb
fa705177ae2e827a2aab00b7a6c95cce9bd9bd65
[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_log.h>
66 #include "mdt_internal.h"
67 #include <lustre_acl.h>
68 #include <lustre_param.h>
69 #include <lustre_quota.h>
70 #include <lustre_lfsck.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 static struct mdt_device *mdt_dev(struct lu_device *d);
95 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
96
97 static const struct lu_object_operations mdt_obj_ops;
98
99 /* Slab for MDT object allocation */
100 static struct kmem_cache *mdt_object_kmem;
101
102 /* For HSM restore handles */
103 struct kmem_cache *mdt_hsm_cdt_kmem;
104
105 /* For HSM request handles */
106 struct kmem_cache *mdt_hsm_car_kmem;
107
108 static struct lu_kmem_descr mdt_caches[] = {
109         {
110                 .ckd_cache = &mdt_object_kmem,
111                 .ckd_name  = "mdt_obj",
112                 .ckd_size  = sizeof(struct mdt_object)
113         },
114         {
115                 .ckd_cache      = &mdt_hsm_cdt_kmem,
116                 .ckd_name       = "mdt_cdt_restore_handle",
117                 .ckd_size       = sizeof(struct cdt_restore_handle)
118         },
119         {
120                 .ckd_cache      = &mdt_hsm_car_kmem,
121                 .ckd_name       = "mdt_cdt_agent_req",
122                 .ckd_size       = sizeof(struct cdt_agent_req)
123         },
124         {
125                 .ckd_cache = NULL
126         }
127 };
128
129 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
130 {
131         if (!rep)
132                 return 0;
133         return (rep->lock_policy_res1 & flag);
134 }
135
136 void mdt_clear_disposition(struct mdt_thread_info *info,
137                            struct ldlm_reply *rep, int flag)
138 {
139         if (info) {
140                 info->mti_opdata &= ~flag;
141                 tgt_opdata_clear(info->mti_env, flag);
142         }
143         if (rep)
144                 rep->lock_policy_res1 &= ~flag;
145 }
146
147 void mdt_set_disposition(struct mdt_thread_info *info,
148                          struct ldlm_reply *rep, int flag)
149 {
150         if (info) {
151                 info->mti_opdata |= flag;
152                 tgt_opdata_set(info->mti_env, flag);
153         }
154         if (rep)
155                 rep->lock_policy_res1 |= flag;
156 }
157
158 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
159 {
160         lh->mlh_pdo_hash = 0;
161         lh->mlh_reg_mode = lm;
162         lh->mlh_rreg_mode = lm;
163         lh->mlh_type = MDT_REG_LOCK;
164 }
165
166 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
167                        const struct lu_name *lname)
168 {
169         lh->mlh_reg_mode = lock_mode;
170         lh->mlh_rreg_mode = lock_mode;
171         lh->mlh_type = MDT_PDO_LOCK;
172
173         if (lu_name_is_valid(lname)) {
174                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
175                                                   lname->ln_namelen);
176                 /* XXX Workaround for LU-2856
177                  *
178                  * Zero is a valid return value of full_name_hash, but
179                  * several users of mlh_pdo_hash assume a non-zero
180                  * hash value. We therefore map zero onto an
181                  * arbitrary, but consistent value (1) to avoid
182                  * problems further down the road. */
183                 if (unlikely(lh->mlh_pdo_hash == 0))
184                         lh->mlh_pdo_hash = 1;
185         } else {
186                 lh->mlh_pdo_hash = 0;
187         }
188 }
189
190 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
191                               struct mdt_lock_handle *lh)
192 {
193         mdl_mode_t mode;
194         ENTRY;
195
196         /*
197          * Any dir access needs couple of locks:
198          *
199          * 1) on part of dir we gonna take lookup/modify;
200          *
201          * 2) on whole dir to protect it from concurrent splitting and/or to
202          * flush client's cache for readdir().
203          *
204          * so, for a given mode and object this routine decides what lock mode
205          * to use for lock #2:
206          *
207          * 1) if caller's gonna lookup in dir then we need to protect dir from
208          * being splitted only - LCK_CR
209          *
210          * 2) if caller's gonna modify dir then we need to protect dir from
211          * being splitted and to flush cache - LCK_CW
212          *
213          * 3) if caller's gonna modify dir and that dir seems ready for
214          * splitting then we need to protect it from any type of access
215          * (lookup/modify/split) - LCK_EX --bzzz
216          */
217
218         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
219         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
220
221         /*
222          * Ask underlaying level its opinion about preferable PDO lock mode
223          * having access type passed as regular lock mode:
224          *
225          * - MDL_MINMODE means that lower layer does not want to specify lock
226          * mode;
227          *
228          * - MDL_NL means that no PDO lock should be taken. This is used in some
229          * cases. Say, for non-splittable directories no need to use PDO locks
230          * at all.
231          */
232         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
233                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
234
235         if (mode != MDL_MINMODE) {
236                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
237         } else {
238                 /*
239                  * Lower layer does not want to specify locking mode. We do it
240                  * our selves. No special protection is needed, just flush
241                  * client's cache on modification and allow concurrent
242                  * mondification.
243                  */
244                 switch (lh->mlh_reg_mode) {
245                 case LCK_EX:
246                         lh->mlh_pdo_mode = LCK_EX;
247                         break;
248                 case LCK_PR:
249                         lh->mlh_pdo_mode = LCK_CR;
250                         break;
251                 case LCK_PW:
252                         lh->mlh_pdo_mode = LCK_CW;
253                         break;
254                 default:
255                         CERROR("Not expected lock type (0x%x)\n",
256                                (int)lh->mlh_reg_mode);
257                         LBUG();
258                 }
259         }
260
261         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
262         EXIT;
263 }
264
265 int mdt_getstatus(struct tgt_session_info *tsi)
266 {
267         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
268         struct mdt_device       *mdt = info->mti_mdt;
269         struct mdt_body         *repbody;
270         int                      rc;
271
272         ENTRY;
273
274         rc = mdt_check_ucred(info);
275         if (rc)
276                 GOTO(out, rc = err_serious(rc));
277
278         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
279                 GOTO(out, rc = err_serious(-ENOMEM));
280
281         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
282         repbody->fid1 = mdt->mdt_md_root_fid;
283         repbody->valid |= OBD_MD_FLID;
284
285         if (tsi->tsi_tgt->lut_mds_capa &&
286             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
287                 struct mdt_object       *root;
288                 struct lustre_capa      *capa;
289
290                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
291                 if (IS_ERR(root))
292                         GOTO(out, rc = PTR_ERR(root));
293
294                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
295                 LASSERT(capa);
296                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
297                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
298                                  0);
299                 mdt_object_put(info->mti_env, root);
300                 if (rc == 0)
301                         repbody->valid |= OBD_MD_FLMDSCAPA;
302         }
303         EXIT;
304 out:
305         mdt_thread_info_fini(info);
306         return rc;
307 }
308
309 int mdt_statfs(struct tgt_session_info *tsi)
310 {
311         struct ptlrpc_request           *req = tgt_ses_req(tsi);
312         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
313         struct md_device                *next = info->mti_mdt->mdt_child;
314         struct ptlrpc_service_part      *svcpt;
315         struct obd_statfs               *osfs;
316         int                             rc;
317
318         ENTRY;
319
320         svcpt = req->rq_rqbd->rqbd_svcpt;
321
322         /* This will trigger a watchdog timeout */
323         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
324                          (MDT_SERVICE_WATCHDOG_FACTOR *
325                           at_get(&svcpt->scp_at_estimate)) + 1);
326
327         rc = mdt_check_ucred(info);
328         if (rc)
329                 GOTO(out, rc = err_serious(rc));
330
331         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
332                 GOTO(out, rc = err_serious(-ENOMEM));
333
334         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
335         if (!osfs)
336                 GOTO(out, rc = -EPROTO);
337
338         /** statfs information are cached in the mdt_device */
339         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
340                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
341                 /** statfs data is too old, get up-to-date one */
342                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
343                 if (rc)
344                         GOTO(out, rc);
345                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
346                 info->mti_mdt->mdt_osfs = *osfs;
347                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
348                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
349         } else {
350                 /** use cached statfs data */
351                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
352                 *osfs = info->mti_mdt->mdt_osfs;
353                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
354         }
355
356         if (rc == 0)
357                 mdt_counter_incr(req, LPROC_MDT_STATFS);
358 out:
359         mdt_thread_info_fini(info);
360         RETURN(rc);
361 }
362
363 /**
364  * Pack SOM attributes into the reply.
365  * Call under a DLM UPDATE lock.
366  */
367 static void mdt_pack_size2body(struct mdt_thread_info *info,
368                                struct mdt_object *mo)
369 {
370         struct mdt_body *b;
371         struct md_attr *ma = &info->mti_attr;
372
373         LASSERT(ma->ma_attr.la_valid & LA_MODE);
374         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
375
376         /* Check if Size-on-MDS is supported, if this is a regular file,
377          * if SOM is enabled on the object and if SOM cache exists and valid.
378          * Otherwise do not pack Size-on-MDS attributes to the reply. */
379         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
380             !S_ISREG(ma->ma_attr.la_mode) ||
381             !mdt_object_is_som_enabled(mo) ||
382             !(ma->ma_valid & MA_SOM))
383                 return;
384
385         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
386         b->size = ma->ma_som->msd_size;
387         b->blocks = ma->ma_som->msd_blocks;
388 }
389
390 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
391                         const struct lu_attr *attr, const struct lu_fid *fid)
392 {
393         struct md_attr *ma = &info->mti_attr;
394
395         LASSERT(ma->ma_valid & MA_INODE);
396
397         b->atime      = attr->la_atime;
398         b->mtime      = attr->la_mtime;
399         b->ctime      = attr->la_ctime;
400         b->mode       = attr->la_mode;
401         b->size       = attr->la_size;
402         b->blocks     = attr->la_blocks;
403         b->uid        = attr->la_uid;
404         b->gid        = attr->la_gid;
405         b->flags      = attr->la_flags;
406         b->nlink      = attr->la_nlink;
407         b->rdev       = attr->la_rdev;
408
409         /*XXX should pack the reply body according to lu_valid*/
410         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
411                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
412                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
413                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
414
415         if (!S_ISREG(attr->la_mode)) {
416                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
417         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
418                 /* means no objects are allocated on osts. */
419                 LASSERT(!(ma->ma_valid & MA_LOV));
420                 /* just ignore blocks occupied by extend attributes on MDS */
421                 b->blocks = 0;
422                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
423                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
424         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
425                    ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
426                 /* A released file stores its size on MDS. */
427                 /* But return 1 block for released file, unless tools like tar
428                  * will consider it fully sparse. (LU-3864)
429                  */
430                 if (unlikely(b->size == 0))
431                         b->blocks = 0;
432                 else
433                         b->blocks = 1;
434                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
435         }
436
437         if (fid) {
438                 b->fid1 = *fid;
439                 b->valid |= OBD_MD_FLID;
440                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
441                                 PFID(fid), b->nlink, b->mode, b->size);
442         }
443
444         if (info)
445                 mdt_body_reverse_idmap(info, b);
446
447         if (fid != NULL && (b->valid & OBD_MD_FLSIZE))
448                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
449                        PFID(fid), (unsigned long long)b->size);
450 }
451
452 static inline int mdt_body_has_lov(const struct lu_attr *la,
453                                    const struct mdt_body *body)
454 {
455         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
456                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
457 }
458
459 void mdt_client_compatibility(struct mdt_thread_info *info)
460 {
461         struct mdt_body       *body;
462         struct ptlrpc_request *req = mdt_info_req(info);
463         struct obd_export     *exp = req->rq_export;
464         struct md_attr        *ma = &info->mti_attr;
465         struct lu_attr        *la = &ma->ma_attr;
466         ENTRY;
467
468         if (exp_connect_layout(exp))
469                 /* the client can deal with 16-bit lmm_stripe_count */
470                 RETURN_EXIT;
471
472         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
473
474         if (!mdt_body_has_lov(la, body))
475                 RETURN_EXIT;
476
477         /* now we have a reply with a lov for a client not compatible with the
478          * layout lock so we have to clean the layout generation number */
479         if (S_ISREG(la->la_mode))
480                 ma->ma_lmm->lmm_layout_gen = 0;
481         EXIT;
482 }
483
484 int mdt_attr_get_eabuf_size(struct mdt_thread_info *info, struct mdt_object *o)
485 {
486         const struct lu_env *env = info->mti_env;
487         int rc, rc2;
488
489         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
490                           XATTR_NAME_LOV);
491
492         if (rc == -ENODATA)
493                 rc = 0;
494
495         if (rc < 0)
496                 goto out;
497
498         /* Is it a directory? Let's check for the LMV as well */
499         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
500                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
501                                    XATTR_NAME_LMV);
502                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
503                         rc = rc2;
504         }
505
506 out:
507         return rc;
508 }
509
510 static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
511                              const char *name)
512 {
513         const struct lu_env *env = info->mti_env;
514         int rc;
515         ENTRY;
516
517         LASSERT(info->mti_big_lmm_used == 0);
518         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
519         if (rc < 0)
520                 RETURN(rc);
521
522         /* big_lmm may need to be grown */
523         if (info->mti_big_lmmsize < rc) {
524                 int size = size_roundup_power2(rc);
525
526                 if (info->mti_big_lmmsize > 0) {
527                         /* free old buffer */
528                         LASSERT(info->mti_big_lmm);
529                         OBD_FREE_LARGE(info->mti_big_lmm,
530                                        info->mti_big_lmmsize);
531                         info->mti_big_lmm = NULL;
532                         info->mti_big_lmmsize = 0;
533                 }
534
535                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
536                 if (info->mti_big_lmm == NULL)
537                         RETURN(-ENOMEM);
538                 info->mti_big_lmmsize = size;
539         }
540         LASSERT(info->mti_big_lmmsize >= rc);
541
542         info->mti_buf.lb_buf = info->mti_big_lmm;
543         info->mti_buf.lb_len = info->mti_big_lmmsize;
544         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
545
546         RETURN(rc);
547 }
548
549 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
550                    struct md_attr *ma, const char *name)
551 {
552         struct md_object *next = mdt_object_child(o);
553         struct lu_buf    *buf = &info->mti_buf;
554         int rc;
555
556         if (strcmp(name, XATTR_NAME_LOV) == 0) {
557                 buf->lb_buf = ma->ma_lmm;
558                 buf->lb_len = ma->ma_lmm_size;
559                 LASSERT(!(ma->ma_valid & MA_LOV));
560         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
561                 buf->lb_buf = ma->ma_lmv;
562                 buf->lb_len = ma->ma_lmv_size;
563                 LASSERT(!(ma->ma_valid & MA_LMV));
564         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
565                 buf->lb_buf = ma->ma_lmv;
566                 buf->lb_len = ma->ma_lmv_size;
567                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
568         } else {
569                 return -EINVAL;
570         }
571
572         rc = mo_xattr_get(info->mti_env, next, buf, name);
573         if (rc > 0) {
574                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
575                         ma->ma_lmm_size = rc;
576                         ma->ma_valid |= MA_LOV;
577                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
578                         ma->ma_lmv_size = rc;
579                         ma->ma_valid |= MA_LMV;
580                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
581                         ma->ma_lmv_size = rc;
582                         ma->ma_valid |= MA_LMV_DEF;
583                 }
584
585                 rc = 0;
586         } else if (rc == -ENODATA) {
587                 /* no LOV EA */
588                 rc = 0;
589         } else if (rc == -ERANGE) {
590                 /* Default LMV has fixed size, so it must be able to fit
591                  * in the original buffer */
592                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
593                         return rc;
594                 rc = mdt_big_xattr_get(info, o, name);
595                 if (rc > 0) {
596                         info->mti_big_lmm_used = 1;
597                         if (!strcmp(name, XATTR_NAME_LOV)) {
598                                 ma->ma_valid |= MA_LOV;
599                                 ma->ma_lmm = info->mti_big_lmm;
600                                 ma->ma_lmm_size = rc;
601                         } else if (!strcmp(name, XATTR_NAME_LMV)) {
602                                 ma->ma_valid |= MA_LMV;
603                                 ma->ma_lmv = info->mti_big_lmm;
604                                 ma->ma_lmv_size = rc;
605                         } else {
606                                 return -EINVAL;
607                         }
608
609                         /* update mdt_max_mdsize so all clients
610                          * will be aware about that */
611                         if (info->mti_mdt->mdt_max_mdsize < rc)
612                                 info->mti_mdt->mdt_max_mdsize = rc;
613                         rc = 0;
614                 }
615         }
616
617         return rc;
618 }
619
620 int mdt_attr_get_pfid(struct mdt_thread_info *info,
621                       struct mdt_object *o, struct lu_fid *pfid)
622 {
623         struct lu_buf           *buf = &info->mti_buf;
624         struct link_ea_header   *leh;
625         struct link_ea_entry    *lee;
626         int                      rc;
627         ENTRY;
628
629         buf->lb_buf = info->mti_big_lmm;
630         buf->lb_len = info->mti_big_lmmsize;
631         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
632                           buf, XATTR_NAME_LINK);
633         /* ignore errors, MA_PFID won't be set and it is
634          * up to the caller to treat this as an error */
635         if (rc == -ERANGE || buf->lb_len == 0) {
636                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
637                 buf->lb_buf = info->mti_big_lmm;
638                 buf->lb_len = info->mti_big_lmmsize;
639         }
640
641         if (rc < 0)
642                 RETURN(rc);
643         if (rc < sizeof(*leh)) {
644                 CERROR("short LinkEA on "DFID": rc = %d\n",
645                        PFID(mdt_object_fid(o)), rc);
646                 RETURN(-ENODATA);
647         }
648
649         leh = (struct link_ea_header *) buf->lb_buf;
650         lee = (struct link_ea_entry *)(leh + 1);
651         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
652                 leh->leh_magic = LINK_EA_MAGIC;
653                 leh->leh_reccount = __swab32(leh->leh_reccount);
654                 leh->leh_len = __swab64(leh->leh_len);
655         }
656         if (leh->leh_magic != LINK_EA_MAGIC)
657                 RETURN(-EINVAL);
658         if (leh->leh_reccount == 0)
659                 RETURN(-ENODATA);
660
661         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
662         fid_be_to_cpu(pfid, pfid);
663
664         RETURN(0);
665 }
666
667 int mdt_attr_get_complex(struct mdt_thread_info *info,
668                          struct mdt_object *o, struct md_attr *ma)
669 {
670         const struct lu_env *env = info->mti_env;
671         struct md_object    *next = mdt_object_child(o);
672         struct lu_buf       *buf = &info->mti_buf;
673         u32                  mode = lu_object_attr(&next->mo_lu);
674         int                  need = ma->ma_need;
675         int                  rc = 0, rc2;
676         ENTRY;
677
678         ma->ma_valid = 0;
679
680         if (need & MA_INODE) {
681                 ma->ma_need = MA_INODE;
682                 rc = mo_attr_get(env, next, ma);
683                 if (rc)
684                         GOTO(out, rc);
685                 ma->ma_valid |= MA_INODE;
686         }
687
688         if (need & MA_PFID) {
689                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
690                 if (rc == 0)
691                         ma->ma_valid |= MA_PFID;
692                 /* ignore this error, parent fid is not mandatory */
693                 rc = 0;
694         }
695
696         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
697                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
698                 if (rc)
699                         GOTO(out, rc);
700         }
701
702         if (need & MA_LMV && S_ISDIR(mode)) {
703                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
704                 if (rc != 0)
705                         GOTO(out, rc);
706         }
707
708         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
709                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
710                 if (rc != 0)
711                         GOTO(out, rc);
712         }
713
714         if (need & MA_SOM && S_ISREG(mode)) {
715                 buf->lb_buf = info->mti_xattr_buf;
716                 buf->lb_len = sizeof(info->mti_xattr_buf);
717                 CLASSERT(sizeof(struct som_attrs) <=
718                          sizeof(info->mti_xattr_buf));
719                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
720                 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
721                 if (rc2 == 0)
722                         ma->ma_valid |= MA_SOM;
723                 else if (rc2 < 0 && rc2 != -ENODATA)
724                         GOTO(out, rc = rc2);
725         }
726
727         if (need & MA_HSM && S_ISREG(mode)) {
728                 buf->lb_buf = info->mti_xattr_buf;
729                 buf->lb_len = sizeof(info->mti_xattr_buf);
730                 CLASSERT(sizeof(struct hsm_attrs) <=
731                          sizeof(info->mti_xattr_buf));
732                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
733                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
734                 if (rc2 == 0)
735                         ma->ma_valid |= MA_HSM;
736                 else if (rc2 < 0 && rc2 != -ENODATA)
737                         GOTO(out, rc = rc2);
738         }
739
740 #ifdef CONFIG_FS_POSIX_ACL
741         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
742                 buf->lb_buf = ma->ma_acl;
743                 buf->lb_len = ma->ma_acl_size;
744                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
745                 if (rc2 > 0) {
746                         ma->ma_acl_size = rc2;
747                         ma->ma_valid |= MA_ACL_DEF;
748                 } else if (rc2 == -ENODATA) {
749                         /* no ACLs */
750                         ma->ma_acl_size = 0;
751                 } else
752                         GOTO(out, rc = rc2);
753         }
754 #endif
755 out:
756         ma->ma_need = need;
757         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
758                rc, ma->ma_valid, ma->ma_lmm);
759         RETURN(rc);
760 }
761
762 static int mdt_getattr_internal(struct mdt_thread_info *info,
763                                 struct mdt_object *o, int ma_need)
764 {
765         struct md_object        *next = mdt_object_child(o);
766         const struct mdt_body   *reqbody = info->mti_body;
767         struct ptlrpc_request   *req = mdt_info_req(info);
768         struct md_attr          *ma = &info->mti_attr;
769         struct lu_attr          *la = &ma->ma_attr;
770         struct req_capsule      *pill = info->mti_pill;
771         const struct lu_env     *env = info->mti_env;
772         struct mdt_body         *repbody;
773         struct lu_buf           *buffer = &info->mti_buf;
774         int                     rc;
775         int                     is_root;
776         ENTRY;
777
778         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
779                 RETURN(err_serious(-ENOMEM));
780
781         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
782
783         ma->ma_valid = 0;
784
785         if (mdt_object_remote(o)) {
786                 /* This object is located on remote node.*/
787                 /* Return -EIO for old client */
788                 if (!mdt_is_dne_client(req->rq_export))
789                         GOTO(out, rc = -EIO);
790
791                 repbody->fid1 = *mdt_object_fid(o);
792                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
793                 GOTO(out, rc = 0);
794         }
795
796         if (reqbody->eadatasize > 0) {
797                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
798                 if (buffer->lb_buf == NULL)
799                         GOTO(out, rc = -EPROTO);
800                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
801                                                       RCL_SERVER);
802         } else {
803                 buffer->lb_buf = NULL;
804                 buffer->lb_len = 0;
805                 ma_need &= ~(MA_LOV | MA_LMV);
806                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
807                        mdt_obd_name(info->mti_mdt),
808                        req->rq_export->exp_client_uuid.uuid);
809         }
810
811         /* If it is dir object and client require MEA, then we got MEA */
812         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
813             (reqbody->valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
814                 /* Assumption: MDT_MD size is enough for lmv size. */
815                 ma->ma_lmv = buffer->lb_buf;
816                 ma->ma_lmv_size = buffer->lb_len;
817                 ma->ma_need = MA_INODE;
818                 if (ma->ma_lmv_size > 0) {
819                         if (reqbody->valid & OBD_MD_MEA)
820                                 ma->ma_need |= MA_LMV;
821                         else if (reqbody->valid & OBD_MD_DEFAULT_MEA)
822                                 ma->ma_need |= MA_LMV_DEF;
823                 }
824         } else {
825                 ma->ma_lmm = buffer->lb_buf;
826                 ma->ma_lmm_size = buffer->lb_len;
827                 ma->ma_need = MA_INODE | MA_HSM;
828                 if (ma->ma_lmm_size > 0)
829                         ma->ma_need |= MA_LOV;
830         }
831
832         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
833             reqbody->valid & OBD_MD_FLDIREA  &&
834             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
835                 /* get default stripe info for this dir. */
836                 ma->ma_need |= MA_LOV_DEF;
837         }
838         ma->ma_need |= ma_need;
839         if (ma->ma_need & MA_SOM)
840                 ma->ma_som = &info->mti_u.som.data;
841
842         rc = mdt_attr_get_complex(info, o, ma);
843         if (unlikely(rc)) {
844                 CERROR("%s: getattr error for "DFID": rc = %d\n",
845                        mdt_obd_name(info->mti_mdt),
846                        PFID(mdt_object_fid(o)), rc);
847                 RETURN(rc);
848         }
849
850         /* if file is released, check if a restore is running */
851         if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
852             mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
853                 repbody->t_state = MS_RESTORE;
854                 repbody->valid |= OBD_MD_TSTATE;
855         }
856
857         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
858
859         /* the Lustre protocol supposes to return default striping
860          * on the user-visible root if explicitly requested */
861         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
862             (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
863                 struct lu_fid      rootfid;
864                 struct mdt_object *root;
865                 struct mdt_device *mdt = info->mti_mdt;
866
867                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
868                 if (rc)
869                         RETURN(rc);
870                 root = mdt_object_find(env, mdt, &rootfid);
871                 if (IS_ERR(root))
872                         RETURN(PTR_ERR(root));
873                 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
874                 mdt_object_put(info->mti_env, root);
875                 if (unlikely(rc)) {
876                         CERROR("%s: getattr error for "DFID": rc = %d\n",
877                                mdt_obd_name(info->mti_mdt),
878                                PFID(mdt_object_fid(o)), rc);
879                         RETURN(rc);
880                 }
881         }
882
883         if (likely(ma->ma_valid & MA_INODE))
884                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
885         else
886                 RETURN(-EFAULT);
887
888         if (mdt_body_has_lov(la, reqbody)) {
889                 if (ma->ma_valid & MA_LOV) {
890                         LASSERT(ma->ma_lmm_size);
891                         repbody->eadatasize = ma->ma_lmm_size;
892                         if (S_ISDIR(la->la_mode))
893                                 repbody->valid |= OBD_MD_FLDIREA;
894                         else
895                                 repbody->valid |= OBD_MD_FLEASIZE;
896                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->valid);
897                 }
898                 if (ma->ma_valid & MA_LMV) {
899                         LASSERT(S_ISDIR(la->la_mode));
900                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
901                         repbody->eadatasize = ma->ma_lmv_size;
902                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
903                 }
904                 if (ma->ma_valid & MA_LMV_DEF) {
905                         LASSERT(S_ISDIR(la->la_mode));
906                         repbody->eadatasize = ma->ma_lmv_size;
907                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_DEFAULT_MEA);
908                 }
909         } else if (S_ISLNK(la->la_mode) &&
910                    reqbody->valid & OBD_MD_LINKNAME) {
911                 buffer->lb_buf = ma->ma_lmm;
912                 /* eadatasize from client includes NULL-terminator, so
913                  * there is no need to read it */
914                 buffer->lb_len = reqbody->eadatasize - 1;
915                 rc = mo_readlink(env, next, buffer);
916                 if (unlikely(rc <= 0)) {
917                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
918                                mdt_obd_name(info->mti_mdt),
919                                PFID(mdt_object_fid(o)), rc);
920                         rc = -EFAULT;
921                 } else {
922                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
923
924                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
925                                 rc -= 2;
926                         repbody->valid |= OBD_MD_LINKNAME;
927                         /* we need to report back size with NULL-terminator
928                          * because client expects that */
929                         repbody->eadatasize = rc + 1;
930                         if (repbody->eadatasize != reqbody->eadatasize)
931                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
932                                        "on "DFID ", expected %d\n",
933                                        mdt_obd_name(info->mti_mdt),
934                                        rc, PFID(mdt_object_fid(o)),
935                                        reqbody->eadatasize - 1);
936                         /* NULL terminate */
937                         ((char *)ma->ma_lmm)[rc] = 0;
938
939                         /* If the total CDEBUG() size is larger than a page, it
940                          * will print a warning to the console, avoid this by
941                          * printing just the last part of the symlink. */
942                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
943                                print_limit < rc ? "..." : "", print_limit,
944                                (char *)ma->ma_lmm + rc - print_limit, rc);
945                         rc = 0;
946                 }
947         }
948
949         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
950                 repbody->max_cookiesize = 0;
951                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
952                 repbody->valid |= OBD_MD_FLMODEASIZE;
953                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
954                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
955                        repbody->max_cookiesize);
956         }
957
958         if (exp_connect_rmtclient(info->mti_exp) &&
959             reqbody->valid & OBD_MD_FLRMTPERM) {
960                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
961
962                 /* mdt_getattr_lock only */
963                 rc = mdt_pack_remote_perm(info, o, buf);
964                 if (rc) {
965                         repbody->valid &= ~OBD_MD_FLRMTPERM;
966                         repbody->aclsize = 0;
967                         RETURN(rc);
968                 } else {
969                         repbody->valid |= OBD_MD_FLRMTPERM;
970                         repbody->aclsize = sizeof(struct mdt_remote_perm);
971                 }
972         }
973 #ifdef CONFIG_FS_POSIX_ACL
974         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
975                  (reqbody->valid & OBD_MD_FLACL)) {
976                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
977                 buffer->lb_len = req_capsule_get_size(pill,
978                                                       &RMF_ACL, RCL_SERVER);
979                 if (buffer->lb_len > 0) {
980                         rc = mo_xattr_get(env, next, buffer,
981                                           XATTR_NAME_ACL_ACCESS);
982                         if (rc < 0) {
983                                 if (rc == -ENODATA) {
984                                         repbody->aclsize = 0;
985                                         repbody->valid |= OBD_MD_FLACL;
986                                         rc = 0;
987                                 } else if (rc == -EOPNOTSUPP) {
988                                         rc = 0;
989                                 } else {
990                                         CERROR("%s: unable to read "DFID
991                                                " ACL: rc = %d\n",
992                                                mdt_obd_name(info->mti_mdt),
993                                                PFID(mdt_object_fid(o)), rc);
994                                 }
995                         } else {
996                                 repbody->aclsize = rc;
997                                 repbody->valid |= OBD_MD_FLACL;
998                                 rc = 0;
999                         }
1000                 }
1001         }
1002 #endif
1003
1004         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
1005             info->mti_mdt->mdt_lut.lut_mds_capa &&
1006             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
1007                 struct lustre_capa *capa;
1008
1009                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
1010                 LASSERT(capa);
1011                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
1012                 rc = mo_capa_get(env, next, capa, 0);
1013                 if (rc)
1014                         RETURN(rc);
1015                 repbody->valid |= OBD_MD_FLMDSCAPA;
1016         }
1017
1018 out:
1019         if (rc == 0)
1020                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1021
1022         RETURN(rc);
1023 }
1024
1025 static int mdt_renew_capa(struct mdt_thread_info *info)
1026 {
1027         struct mdt_object  *obj = info->mti_object;
1028         struct mdt_body    *body;
1029         struct lustre_capa *capa, *c;
1030         int rc;
1031         ENTRY;
1032
1033         /* if object doesn't exist, or server has disabled capability,
1034          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
1035          * flag not set.
1036          */
1037         if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa ||
1038             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
1039                 RETURN(0);
1040
1041         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1042         LASSERT(body != NULL);
1043
1044         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
1045         LASSERT(c);
1046
1047         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
1048         LASSERT(capa);
1049
1050         *capa = *c;
1051         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
1052         if (rc == 0)
1053                 body->valid |= OBD_MD_FLOSSCAPA;
1054         RETURN(rc);
1055 }
1056
1057 int mdt_getattr(struct tgt_session_info *tsi)
1058 {
1059         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1060         struct mdt_object       *obj = info->mti_object;
1061         struct req_capsule      *pill = info->mti_pill;
1062         struct mdt_body         *reqbody;
1063         struct mdt_body         *repbody;
1064         mode_t                   mode;
1065         int rc, rc2;
1066         ENTRY;
1067
1068         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1069         LASSERT(reqbody);
1070
1071         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
1072                 rc = req_capsule_server_pack(pill);
1073                 if (unlikely(rc))
1074                         RETURN(err_serious(rc));
1075                 rc = mdt_renew_capa(info);
1076                 GOTO(out_shrink, rc);
1077         }
1078
1079         LASSERT(obj != NULL);
1080         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1081
1082         mode = lu_object_attr(&obj->mot_obj);
1083
1084         /* Readlink */
1085         if (reqbody->valid & OBD_MD_LINKNAME) {
1086                 /* No easy way to know how long is the symlink, but it cannot
1087                  * be more than PATH_MAX, so we allocate +1 */
1088                 rc = PATH_MAX + 1;
1089
1090         /* A special case for fs ROOT: getattr there might fetch
1091          * default EA for entire fs, not just for this dir!
1092          */
1093         } else if (lu_fid_eq(mdt_object_fid(obj),
1094                              &info->mti_mdt->mdt_md_root_fid) &&
1095                    (reqbody->valid & OBD_MD_FLDIREA) &&
1096                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1097                                                                  MDS_GETATTR)) {
1098                 /* Should the default strping be bigger, mdt_fix_reply
1099                  * will reallocate */
1100                 rc = DEF_REP_MD_SIZE;
1101         } else {
1102                 /* Hopefully no race in EA change for either file or directory?
1103                  */
1104                 rc = mdt_attr_get_eabuf_size(info, obj);
1105         }
1106
1107         if (rc < 0)
1108                 GOTO(out_shrink, rc);
1109
1110         /* old clients may not report needed easize, use max value then */
1111         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1112
1113         rc = req_capsule_server_pack(pill);
1114         if (unlikely(rc != 0))
1115                 GOTO(out, rc = err_serious(rc));
1116
1117         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1118         LASSERT(repbody != NULL);
1119         repbody->eadatasize = 0;
1120         repbody->aclsize = 0;
1121
1122         if (reqbody->valid & OBD_MD_FLRMTPERM)
1123                 rc = mdt_init_ucred(info, reqbody);
1124         else
1125                 rc = mdt_check_ucred(info);
1126         if (unlikely(rc))
1127                 GOTO(out_shrink, rc);
1128
1129         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1130
1131         /*
1132          * Don't check capability at all, because rename might getattr for
1133          * remote obj, and at that time no capability is available.
1134          */
1135         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
1136         rc = mdt_getattr_internal(info, obj, 0);
1137         if (reqbody->valid & OBD_MD_FLRMTPERM)
1138                 mdt_exit_ucred(info);
1139         EXIT;
1140 out_shrink:
1141         mdt_client_compatibility(info);
1142         rc2 = mdt_fix_reply(info);
1143         if (rc == 0)
1144                 rc = rc2;
1145 out:
1146         mdt_thread_info_fini(info);
1147         return rc;
1148 }
1149
1150 int mdt_is_subdir(struct tgt_session_info *tsi)
1151 {
1152         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1153         struct mdt_object     *o = info->mti_object;
1154         struct req_capsule    *pill = info->mti_pill;
1155         const struct mdt_body *body = info->mti_body;
1156         struct mdt_body       *repbody;
1157         int                    rc;
1158         ENTRY;
1159
1160         LASSERT(o != NULL);
1161
1162         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1163
1164         /*
1165          * We save last checked parent fid to @repbody->fid1 for remote
1166          * directory case.
1167          */
1168         LASSERT(fid_is_sane(&body->fid2));
1169         LASSERT(mdt_object_exists(o) && !mdt_object_remote(o));
1170         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
1171                            &body->fid2, &repbody->fid1);
1172         if (rc == 0 || rc == -EREMOTE)
1173                 repbody->valid |= OBD_MD_FLID;
1174
1175         mdt_thread_info_fini(info);
1176         RETURN(rc);
1177 }
1178
1179 int mdt_swap_layouts(struct tgt_session_info *tsi)
1180 {
1181         struct mdt_thread_info  *info;
1182         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1183         struct obd_export       *exp = req->rq_export;
1184         struct mdt_object       *o1, *o2, *o;
1185         struct mdt_lock_handle  *lh1, *lh2;
1186         struct mdc_swap_layouts *msl;
1187         int                      rc;
1188         ENTRY;
1189
1190         /* client does not support layout lock, so layout swaping
1191          * is disabled.
1192          * FIXME: there is a problem for old clients which don't support
1193          * layout lock yet. If those clients have already opened the file
1194          * they won't be notified at all so that old layout may still be
1195          * used to do IO. This can be fixed after file release is landed by
1196          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1197         if (!exp_connect_layout(exp))
1198                 RETURN(-EOPNOTSUPP);
1199
1200         info = tsi2mdt_info(tsi);
1201
1202         if (info->mti_dlm_req != NULL)
1203                 ldlm_request_cancel(req, info->mti_dlm_req, 0);
1204
1205         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1206                 mdt_set_capainfo(info, 0, &info->mti_body->fid1,
1207                                  req_capsule_client_get(info->mti_pill,
1208                                                         &RMF_CAPA1));
1209
1210         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1211                 mdt_set_capainfo(info, 1, &info->mti_body->fid2,
1212                                  req_capsule_client_get(info->mti_pill,
1213                                                         &RMF_CAPA2));
1214
1215         o1 = info->mti_object;
1216         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1217                                 &info->mti_body->fid2);
1218         if (IS_ERR(o))
1219                 GOTO(out, rc = PTR_ERR(o));
1220
1221         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1222                 GOTO(put, rc = -ENOENT);
1223
1224         rc = lu_fid_cmp(&info->mti_body->fid1, &info->mti_body->fid2);
1225         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1226                 GOTO(put, rc);
1227
1228         if (rc < 0)
1229                 swap(o1, o2);
1230
1231         /* permission check. Make sure the calling process having permission
1232          * to write both files. */
1233         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1234                                 MAY_WRITE);
1235         if (rc < 0)
1236                 GOTO(put, rc);
1237
1238         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1239                                 MAY_WRITE);
1240         if (rc < 0)
1241                 GOTO(put, rc);
1242
1243         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1244         if (msl == NULL)
1245                 GOTO(put, rc = -EPROTO);
1246
1247         lh1 = &info->mti_lh[MDT_LH_NEW];
1248         mdt_lock_reg_init(lh1, LCK_EX);
1249         lh2 = &info->mti_lh[MDT_LH_OLD];
1250         mdt_lock_reg_init(lh2, LCK_EX);
1251
1252         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1253                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1254         if (rc < 0)
1255                 GOTO(put, rc);
1256
1257         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1258                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1259         if (rc < 0)
1260                 GOTO(unlock1, rc);
1261
1262         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1263                              mdt_object_child(o2), msl->msl_flags);
1264         GOTO(unlock2, rc);
1265 unlock2:
1266         mdt_object_unlock(info, o2, lh2, rc);
1267 unlock1:
1268         mdt_object_unlock(info, o1, lh1, rc);
1269 put:
1270         mdt_object_put(info->mti_env, o);
1271 out:
1272         mdt_thread_info_fini(info);
1273         RETURN(rc);
1274 }
1275
1276 static int mdt_raw_lookup(struct mdt_thread_info *info,
1277                           struct mdt_object *parent,
1278                           const struct lu_name *lname,
1279                           struct ldlm_reply *ldlm_rep)
1280 {
1281         struct md_object *next = mdt_object_child(info->mti_object);
1282         const struct mdt_body *reqbody = info->mti_body;
1283         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1284         struct mdt_body *repbody;
1285         int rc;
1286         ENTRY;
1287
1288         if (reqbody->valid != OBD_MD_FLID)
1289                 RETURN(0);
1290
1291         LASSERT(!info->mti_cross_ref);
1292
1293         /* Only got the fid of this obj by name */
1294         fid_zero(child_fid);
1295         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1296                         &info->mti_spec);
1297 #if 0
1298         /* XXX is raw_lookup possible as intent operation? */
1299         if (rc != 0) {
1300                 if (rc == -ENOENT)
1301                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1302                 RETURN(rc);
1303         } else
1304                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1305
1306         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1307 #endif
1308         if (rc == 0) {
1309                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1310                 repbody->fid1 = *child_fid;
1311                 repbody->valid = OBD_MD_FLID;
1312         }
1313         RETURN(1);
1314 }
1315
1316 /*
1317  * UPDATE lock should be taken against parent, and be release before exit;
1318  * child_bits lock should be taken against child, and be returned back:
1319  *            (1)normal request should release the child lock;
1320  *            (2)intent request will grant the lock to client.
1321  */
1322 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1323                                  struct mdt_lock_handle *lhc,
1324                                  __u64 child_bits,
1325                                  struct ldlm_reply *ldlm_rep)
1326 {
1327         struct ptlrpc_request  *req       = mdt_info_req(info);
1328         struct mdt_body        *reqbody   = NULL;
1329         struct mdt_object      *parent    = info->mti_object;
1330         struct mdt_object      *child;
1331         struct md_object       *next      = mdt_object_child(parent);
1332         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1333         struct lu_name         *lname     = NULL;
1334         struct mdt_lock_handle *lhp       = NULL;
1335         struct ldlm_lock       *lock;
1336         struct ldlm_res_id     *res_id;
1337         int                     is_resent;
1338         int                     ma_need = 0;
1339         int                     rc;
1340
1341         ENTRY;
1342
1343         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1344         LASSERT(ergo(is_resent,
1345                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1346
1347         LASSERT(parent != NULL);
1348
1349         lname = &info->mti_name;
1350         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1351
1352         if (!info->mti_cross_ref) {
1353                 /*
1354                  * XXX: Check for anonymous name is for getattr by fid
1355                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
1356                  * that is the name must contain at least one character and
1357                  * the terminating '\0'.
1358                  */
1359                 if (!lu_name_is_valid(lname)) {
1360                         reqbody = req_capsule_client_get(info->mti_pill,
1361                                                          &RMF_MDT_BODY);
1362                         if (unlikely(reqbody == NULL))
1363                                 RETURN(err_serious(-EFAULT));
1364
1365                         if (unlikely(!fid_is_sane(&reqbody->fid2)))
1366                                 RETURN(err_serious(-EINVAL));
1367
1368                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1369                                "ldlm_rep = %p\n",
1370                                PFID(mdt_object_fid(parent)),
1371                                PFID(&reqbody->fid2), ldlm_rep);
1372                 } else {
1373                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1374                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1375                                PNAME(lname), ldlm_rep);
1376                 }
1377         }
1378
1379         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1380
1381         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1382                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1383                                 &parent->mot_obj,
1384                                 "Parent doesn't exist!\n");
1385                 RETURN(-ESTALE);
1386         } else if (!info->mti_cross_ref) {
1387                 LASSERTF(!mdt_object_remote(parent),
1388                          "Parent "DFID" is on remote server\n",
1389                          PFID(mdt_object_fid(parent)));
1390         }
1391
1392         if (lu_name_is_valid(lname)) {
1393                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1394                 if (rc != 0) {
1395                         if (rc > 0)
1396                                 rc = 0;
1397                         RETURN(rc);
1398                 }
1399         }
1400
1401         if (info->mti_cross_ref) {
1402                 /* Only getattr on the child. Parent is on another node. */
1403                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1404                 child = parent;
1405                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1406                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
1407
1408                 if (is_resent) {
1409                         /* Do not take lock for resent case. */
1410                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1411                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1412                                  lhc->mlh_reg_lh.cookie);
1413                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
1414                                                 &lock->l_resource->lr_name));
1415                         LDLM_LOCK_PUT(lock);
1416                         rc = 0;
1417                 } else {
1418                         mdt_lock_handle_init(lhc);
1419                         mdt_lock_reg_init(lhc, LCK_PR);
1420
1421                         /*
1422                          * Object's name is on another MDS, no lookup or layout
1423                          * lock is needed here but update lock is.
1424                          */
1425                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1426                                         MDS_INODELOCK_LAYOUT);
1427                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1428
1429                         rc = mdt_object_lock(info, child, lhc, child_bits,
1430                                              MDT_LOCAL_LOCK);
1431                 }
1432                 if (rc == 0) {
1433                         /* Finally, we can get attr for child. */
1434                         if (!mdt_object_exists(child)) {
1435                                 LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1436                                                 &child->mot_obj,
1437                                              "remote object doesn't exist.\n");
1438                                 mdt_object_unlock(info, child, lhc, 1);
1439                                 RETURN(-ENOENT);
1440                         }
1441
1442                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
1443                                          BYPASS_CAPA);
1444                         rc = mdt_getattr_internal(info, child, 0);
1445                         if (unlikely(rc != 0))
1446                                 mdt_object_unlock(info, child, lhc, 1);
1447                 }
1448                 RETURN(rc);
1449         }
1450
1451         if (lu_name_is_valid(lname)) {
1452                 /* step 1: lock parent only if parent is a directory */
1453                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1454                         lhp = &info->mti_lh[MDT_LH_PARENT];
1455                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1456                         rc = mdt_object_lock(info, parent, lhp,
1457                                              MDS_INODELOCK_UPDATE,
1458                                              MDT_LOCAL_LOCK);
1459                         if (unlikely(rc != 0))
1460                                 RETURN(rc);
1461                 }
1462
1463                 /* step 2: lookup child's fid by name */
1464                 fid_zero(child_fid);
1465                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1466                                 &info->mti_spec);
1467
1468                 if (rc != 0) {
1469                         if (rc == -ENOENT)
1470                                 mdt_set_disposition(info, ldlm_rep,
1471                                                     DISP_LOOKUP_NEG);
1472                         GOTO(out_parent, rc);
1473                 } else
1474                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1475         } else {
1476                 *child_fid = reqbody->fid2;
1477                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1478         }
1479
1480         /*
1481          *step 3: find the child object by fid & lock it.
1482          *        regardless if it is local or remote.
1483          *
1484          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1485          *      set parent dir fid the same as child fid in getattr by fid case
1486          *      we should not lu_object_find() the object again, could lead
1487          *      to hung if there is a concurrent unlink destroyed the object.
1488          */
1489         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1490                 mdt_object_get(info->mti_env, parent);
1491                 child = parent;
1492         } else {
1493                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1494                                         child_fid);
1495         }
1496
1497         if (unlikely(IS_ERR(child)))
1498                 GOTO(out_parent, rc = PTR_ERR(child));
1499         if (is_resent) {
1500                 /* Do not take lock for resent case. */
1501                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1502                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1503                          lhc->mlh_reg_lh.cookie);
1504
1505                 res_id = &lock->l_resource->lr_name;
1506                 if (!fid_res_name_eq(mdt_object_fid(child),
1507                                      &lock->l_resource->lr_name)) {
1508                         LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1509                                                  &lock->l_resource->lr_name),
1510                                  "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1511                                  PLDLMRES(lock->l_resource),
1512                                  PFID(mdt_object_fid(parent)));
1513                         CWARN("Although resent, but still not get child lock"
1514                               "parent:"DFID" child:"DFID"\n",
1515                               PFID(mdt_object_fid(parent)),
1516                               PFID(mdt_object_fid(child)));
1517                         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1518                         LDLM_LOCK_PUT(lock);
1519                         GOTO(relock, 0);
1520                 }
1521                 LDLM_LOCK_PUT(lock);
1522                 rc = 0;
1523         } else {
1524                 bool try_layout = false;
1525
1526 relock:
1527                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1528                 mdt_lock_handle_init(lhc);
1529                 mdt_lock_reg_init(lhc, LCK_PR);
1530
1531                 if (!mdt_object_exists(child)) {
1532                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1533                                         &child->mot_obj,
1534                                         "Object doesn't exist!\n");
1535                         GOTO(out_child, rc = -ENOENT);
1536                 }
1537
1538                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1539                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1540                         struct md_attr *ma = &info->mti_attr;
1541
1542                         ma->ma_valid = 0;
1543                         ma->ma_need = MA_INODE;
1544                         rc = mdt_attr_get_complex(info, child, ma);
1545                         if (unlikely(rc != 0))
1546                                 GOTO(out_child, rc);
1547
1548                         /* If the file has not been changed for some time, we
1549                          * return not only a LOOKUP lock, but also an UPDATE
1550                          * lock and this might save us RPC on later STAT. For
1551                          * directories, it also let negative dentry cache start
1552                          * working for this dir. */
1553                         if (ma->ma_valid & MA_INODE &&
1554                             ma->ma_attr.la_valid & LA_CTIME &&
1555                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1556                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1557                                 child_bits |= MDS_INODELOCK_UPDATE;
1558                 }
1559
1560                 /* layout lock must be granted in a best-effort way
1561                  * for IT operations */
1562                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1563                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1564                     exp_connect_layout(info->mti_exp) &&
1565                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1566                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1567                         /* try to grant layout lock for regular file. */
1568                         try_layout = true;
1569                 }
1570
1571                 rc = 0;
1572                 if (try_layout) {
1573                         child_bits |= MDS_INODELOCK_LAYOUT;
1574                         /* try layout lock, it may fail to be granted due to
1575                          * contention at LOOKUP or UPDATE */
1576                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1577                                                  MDT_CROSS_LOCK)) {
1578                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1579                                 LASSERT(child_bits != 0);
1580                                 rc = mdt_object_lock(info, child, lhc,
1581                                                 child_bits, MDT_CROSS_LOCK);
1582                         } else {
1583                                 ma_need |= MA_LOV;
1584                         }
1585                 } else {
1586                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1587                          * client will enqueue the lock to the remote MDT */
1588                         if (mdt_object_remote(child))
1589                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1590                         rc = mdt_object_lock(info, child, lhc, child_bits,
1591                                                 MDT_CROSS_LOCK);
1592                 }
1593                 if (unlikely(rc != 0))
1594                         GOTO(out_child, rc);
1595         }
1596
1597         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1598         /* Get MA_SOM attributes if update lock is given. */
1599         if (lock &&
1600             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1601             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1602                 ma_need |= MA_SOM;
1603
1604         /* finally, we can get attr for child. */
1605         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1606         rc = mdt_getattr_internal(info, child, ma_need);
1607         if (unlikely(rc != 0)) {
1608                 mdt_object_unlock(info, child, lhc, 1);
1609         } else if (lock) {
1610                 /* Debugging code. */
1611                 res_id = &lock->l_resource->lr_name;
1612                 LDLM_DEBUG(lock, "Returning lock to client");
1613                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1614                                          &lock->l_resource->lr_name),
1615                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1616                          PLDLMRES(lock->l_resource),
1617                          PFID(mdt_object_fid(child)));
1618                 if (mdt_object_exists(child) && !mdt_object_remote(child))
1619                         mdt_pack_size2body(info, child);
1620         }
1621         if (lock)
1622                 LDLM_LOCK_PUT(lock);
1623
1624         EXIT;
1625 out_child:
1626         mdt_object_put(info->mti_env, child);
1627 out_parent:
1628         if (lhp)
1629                 mdt_object_unlock(info, parent, lhp, 1);
1630         return rc;
1631 }
1632
1633 /* normal handler: should release the child lock */
1634 int mdt_getattr_name(struct tgt_session_info *tsi)
1635 {
1636         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1637         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1638         struct mdt_body        *reqbody;
1639         struct mdt_body        *repbody;
1640         int rc, rc2;
1641         ENTRY;
1642
1643         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1644         LASSERT(reqbody != NULL);
1645         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1646         LASSERT(repbody != NULL);
1647
1648         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1649         repbody->eadatasize = 0;
1650         repbody->aclsize = 0;
1651
1652         rc = mdt_init_ucred(info, reqbody);
1653         if (unlikely(rc))
1654                 GOTO(out_shrink, rc);
1655
1656         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1657         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1658                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1659                 lhc->mlh_reg_lh.cookie = 0;
1660         }
1661         mdt_exit_ucred(info);
1662         EXIT;
1663 out_shrink:
1664         mdt_client_compatibility(info);
1665         rc2 = mdt_fix_reply(info);
1666         if (rc == 0)
1667                 rc = rc2;
1668         mdt_thread_info_fini(info);
1669         return rc;
1670 }
1671
1672 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1673                          void *karg, void *uarg);
1674
1675 int mdt_set_info(struct tgt_session_info *tsi)
1676 {
1677         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1678         char                    *key;
1679         void                    *val;
1680         int                      keylen, vallen, rc = 0;
1681
1682         ENTRY;
1683
1684         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1685         if (key == NULL) {
1686                 DEBUG_REQ(D_HA, req, "no set_info key");
1687                 RETURN(err_serious(-EFAULT));
1688         }
1689
1690         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1691                                       RCL_CLIENT);
1692
1693         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1694         if (val == NULL) {
1695                 DEBUG_REQ(D_HA, req, "no set_info val");
1696                 RETURN(err_serious(-EFAULT));
1697         }
1698
1699         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1700                                       RCL_CLIENT);
1701
1702         /* Swab any part of val you need to here */
1703         if (KEY_IS(KEY_READ_ONLY)) {
1704                 spin_lock(&req->rq_export->exp_lock);
1705                 if (*(__u32 *)val)
1706                         *exp_connect_flags_ptr(req->rq_export) |=
1707                                 OBD_CONNECT_RDONLY;
1708                 else
1709                         *exp_connect_flags_ptr(req->rq_export) &=
1710                                 ~OBD_CONNECT_RDONLY;
1711                 spin_unlock(&req->rq_export->exp_lock);
1712         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1713                 struct changelog_setinfo *cs = val;
1714
1715                 if (vallen != sizeof(*cs)) {
1716                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1717                                tgt_name(tsi->tsi_tgt), vallen);
1718                         RETURN(-EINVAL);
1719                 }
1720                 if (ptlrpc_req_need_swab(req)) {
1721                         __swab64s(&cs->cs_recno);
1722                         __swab32s(&cs->cs_id);
1723                 }
1724
1725                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1726                                    vallen, val, NULL);
1727         } else {
1728                 RETURN(-EINVAL);
1729         }
1730         RETURN(rc);
1731 }
1732
1733 int mdt_readpage(struct tgt_session_info *tsi)
1734 {
1735         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1736         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1737         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1738         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1739         struct mdt_body         *repbody;
1740         int                      rc;
1741         int                      i;
1742
1743         ENTRY;
1744
1745         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1746                 RETURN(err_serious(-ENOMEM));
1747
1748         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1749         if (repbody == NULL || reqbody == NULL)
1750                 RETURN(err_serious(-EFAULT));
1751
1752         /*
1753          * prepare @rdpg before calling lower layers and transfer itself. Here
1754          * reqbody->size contains offset of where to start to read and
1755          * reqbody->nlink contains number bytes to read.
1756          */
1757         rdpg->rp_hash = reqbody->size;
1758         if (rdpg->rp_hash != reqbody->size) {
1759                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1760                        rdpg->rp_hash, reqbody->size);
1761                 RETURN(-EFAULT);
1762         }
1763
1764         rdpg->rp_attrs = reqbody->mode;
1765         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1766                 rdpg->rp_attrs |= LUDA_64BITHASH;
1767         rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
1768                                 exp_max_brw_size(tsi->tsi_exp));
1769         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1770                           PAGE_CACHE_SHIFT;
1771         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1772         if (rdpg->rp_pages == NULL)
1773                 RETURN(-ENOMEM);
1774
1775         for (i = 0; i < rdpg->rp_npages; ++i) {
1776                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1777                 if (rdpg->rp_pages[i] == NULL)
1778                         GOTO(free_rdpg, rc = -ENOMEM);
1779         }
1780
1781         /* call lower layers to fill allocated pages with directory data */
1782         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1783         if (rc < 0)
1784                 GOTO(free_rdpg, rc);
1785
1786         /* send pages to client */
1787         rc = tgt_sendpage(tsi, rdpg, rc);
1788
1789         EXIT;
1790 free_rdpg:
1791
1792         for (i = 0; i < rdpg->rp_npages; i++)
1793                 if (rdpg->rp_pages[i] != NULL)
1794                         __free_page(rdpg->rp_pages[i]);
1795         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1796
1797         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1798                 RETURN(0);
1799
1800         return rc;
1801 }
1802
1803 static int mdt_reint_internal(struct mdt_thread_info *info,
1804                               struct mdt_lock_handle *lhc,
1805                               __u32 op)
1806 {
1807         struct req_capsule      *pill = info->mti_pill;
1808         struct mdt_body         *repbody;
1809         int                      rc = 0, rc2;
1810
1811         ENTRY;
1812
1813         rc = mdt_reint_unpack(info, op);
1814         if (rc != 0) {
1815                 CERROR("Can't unpack reint, rc %d\n", rc);
1816                 RETURN(err_serious(rc));
1817         }
1818
1819         /* for replay (no_create) lmm is not needed, client has it already */
1820         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1821                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1822                                      DEF_REP_MD_SIZE);
1823
1824         /* llog cookies are always 0, the field is kept for compatibility */
1825         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1826                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1827
1828         rc = req_capsule_server_pack(pill);
1829         if (rc != 0) {
1830                 CERROR("Can't pack response, rc %d\n", rc);
1831                 RETURN(err_serious(rc));
1832         }
1833
1834         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1835                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1836                 LASSERT(repbody);
1837                 repbody->eadatasize = 0;
1838                 repbody->aclsize = 0;
1839         }
1840
1841         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1842
1843         /* for replay no cookkie / lmm need, because client have this already */
1844         if (info->mti_spec.no_create)
1845                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1846                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1847
1848         rc = mdt_init_ucred_reint(info);
1849         if (rc)
1850                 GOTO(out_shrink, rc);
1851
1852         rc = mdt_fix_attr_ucred(info, op);
1853         if (rc != 0)
1854                 GOTO(out_ucred, rc = err_serious(rc));
1855
1856         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1857                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1858                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1859                 GOTO(out_ucred, rc);
1860         }
1861         rc = mdt_reint_rec(info, lhc);
1862         EXIT;
1863 out_ucred:
1864         mdt_exit_ucred(info);
1865 out_shrink:
1866         mdt_client_compatibility(info);
1867         rc2 = mdt_fix_reply(info);
1868         if (rc == 0)
1869                 rc = rc2;
1870         return rc;
1871 }
1872
1873 static long mdt_reint_opcode(struct ptlrpc_request *req,
1874                              const struct req_format **fmt)
1875 {
1876         struct mdt_device       *mdt;
1877         struct mdt_rec_reint    *rec;
1878         long                     opc;
1879
1880         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1881         if (rec != NULL) {
1882                 opc = rec->rr_opcode;
1883                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1884                 if (opc < REINT_MAX && fmt[opc] != NULL)
1885                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1886                 else {
1887                         mdt = mdt_exp2dev(req->rq_export);
1888                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1889                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1890                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1891                         opc = err_serious(-EFAULT);
1892                 }
1893         } else {
1894                 opc = err_serious(-EFAULT);
1895         }
1896         return opc;
1897 }
1898
1899 int mdt_reint(struct tgt_session_info *tsi)
1900 {
1901         long opc;
1902         int  rc;
1903         static const struct req_format *reint_fmts[REINT_MAX] = {
1904                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1905                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1906                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1907                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1908                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1909                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1910                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1911                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1912                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
1913         };
1914
1915         ENTRY;
1916
1917         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1918         if (opc >= 0) {
1919                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1920                 /*
1921                  * No lock possible here from client to pass it to reint code
1922                  * path.
1923                  */
1924                 rc = mdt_reint_internal(info, NULL, opc);
1925                 mdt_thread_info_fini(info);
1926         } else {
1927                 rc = opc;
1928         }
1929
1930         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1931         RETURN(rc);
1932 }
1933
1934 /* this should sync the whole device */
1935 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1936 {
1937         struct dt_device *dt = mdt->mdt_bottom;
1938         int rc;
1939         ENTRY;
1940
1941         rc = dt->dd_ops->dt_sync(env, dt);
1942         RETURN(rc);
1943 }
1944
1945 /* this should sync this object */
1946 static int mdt_object_sync(struct mdt_thread_info *info)
1947 {
1948         struct md_object *next;
1949         int rc;
1950         ENTRY;
1951
1952         if (!mdt_object_exists(info->mti_object)) {
1953                 CWARN("Non existing object  "DFID"!\n",
1954                       PFID(mdt_object_fid(info->mti_object)));
1955                 RETURN(-ESTALE);
1956         }
1957         next = mdt_object_child(info->mti_object);
1958         rc = mo_object_sync(info->mti_env, next);
1959
1960         RETURN(rc);
1961 }
1962
1963 int mdt_sync(struct tgt_session_info *tsi)
1964 {
1965         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1966         struct req_capsule      *pill = tsi->tsi_pill;
1967         struct mdt_body         *body;
1968         int                      rc;
1969
1970         ENTRY;
1971
1972         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1973                 RETURN(err_serious(-ENOMEM));
1974
1975         if (fid_seq(&tsi->tsi_mdt_body->fid1) == 0) {
1976                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1977         } else {
1978                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1979
1980                 /* sync an object */
1981                 rc = mdt_object_sync(info);
1982                 if (rc == 0) {
1983                         const struct lu_fid *fid;
1984                         struct lu_attr *la = &info->mti_attr.ma_attr;
1985
1986                         info->mti_attr.ma_need = MA_INODE;
1987                         info->mti_attr.ma_valid = 0;
1988                         rc = mdt_attr_get_complex(info, info->mti_object,
1989                                                   &info->mti_attr);
1990                         if (rc == 0) {
1991                                 body = req_capsule_server_get(pill,
1992                                                               &RMF_MDT_BODY);
1993                                 fid = mdt_object_fid(info->mti_object);
1994                                 mdt_pack_attr2body(info, body, la, fid);
1995                         }
1996                 }
1997                 mdt_thread_info_fini(info);
1998         }
1999         if (rc == 0)
2000                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2001
2002         RETURN(rc);
2003 }
2004
2005 /*
2006  * Handle quota control requests to consult current usage/limit, but also
2007  * to configure quota enforcement
2008  */
2009 int mdt_quotactl(struct tgt_session_info *tsi)
2010 {
2011         struct obd_export       *exp  = tsi->tsi_exp;
2012         struct req_capsule      *pill = tsi->tsi_pill;
2013         struct obd_quotactl     *oqctl, *repoqc;
2014         int                      id, rc;
2015         struct mdt_device       *mdt = mdt_exp2dev(exp);
2016         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2017         ENTRY;
2018
2019         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2020         if (oqctl == NULL)
2021                 RETURN(err_serious(-EPROTO));
2022
2023         rc = req_capsule_server_pack(pill);
2024         if (rc)
2025                 RETURN(err_serious(rc));
2026
2027         switch (oqctl->qc_cmd) {
2028         case Q_QUOTACHECK:
2029         case LUSTRE_Q_INVALIDATE:
2030         case LUSTRE_Q_FINVALIDATE:
2031         case Q_QUOTAON:
2032         case Q_QUOTAOFF:
2033         case Q_INITQUOTA:
2034                 /* deprecated, not used any more */
2035                 RETURN(-EOPNOTSUPP);
2036                 /* master quotactl */
2037         case Q_GETINFO:
2038         case Q_SETINFO:
2039         case Q_SETQUOTA:
2040         case Q_GETQUOTA:
2041                 if (qmt == NULL)
2042                         RETURN(-EOPNOTSUPP);
2043                 /* slave quotactl */
2044         case Q_GETOINFO:
2045         case Q_GETOQUOTA:
2046                 break;
2047         default:
2048                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2049                 RETURN(-EFAULT);
2050         }
2051
2052         /* map uid/gid for remote client */
2053         id = oqctl->qc_id;
2054         if (exp_connect_rmtclient(exp)) {
2055                 struct lustre_idmap_table *idmap;
2056
2057                 idmap = exp->exp_mdt_data.med_idmap;
2058
2059                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2060                              oqctl->qc_cmd != Q_GETINFO))
2061                         RETURN(-EPERM);
2062
2063                 if (oqctl->qc_type == USRQUOTA)
2064                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2065                                                      oqctl->qc_id);
2066                 else if (oqctl->qc_type == GRPQUOTA)
2067                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2068                                                      oqctl->qc_id);
2069                 else
2070                         RETURN(-EINVAL);
2071
2072                 if (id == CFS_IDMAP_NOTFOUND) {
2073                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2074                         RETURN(-EACCES);
2075                 }
2076         }
2077
2078         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2079         if (repoqc == NULL)
2080                 RETURN(err_serious(-EFAULT));
2081
2082         if (oqctl->qc_id != id)
2083                 swap(oqctl->qc_id, id);
2084
2085         switch (oqctl->qc_cmd) {
2086
2087         case Q_GETINFO:
2088         case Q_SETINFO:
2089         case Q_SETQUOTA:
2090         case Q_GETQUOTA:
2091                 /* forward quotactl request to QMT */
2092                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2093                 break;
2094
2095         case Q_GETOINFO:
2096         case Q_GETOQUOTA:
2097                 /* slave quotactl */
2098                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2099                                    oqctl);
2100                 break;
2101
2102         default:
2103                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2104                 RETURN(-EFAULT);
2105         }
2106
2107         if (oqctl->qc_id != id)
2108                 swap(oqctl->qc_id, id);
2109
2110         *repoqc = *oqctl;
2111         RETURN(rc);
2112 }
2113
2114 /** clone llog ctxt from child (mdd)
2115  * This allows remote llog (replicator) access.
2116  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2117  * context was originally set up, or we can handle them directly.
2118  * I choose the latter, but that means I need any llog
2119  * contexts set up by child to be accessable by the mdt.  So we clone the
2120  * context into our context list here.
2121  */
2122 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2123                                int idx)
2124 {
2125         struct md_device  *next = mdt->mdt_child;
2126         struct llog_ctxt *ctxt;
2127         int rc;
2128
2129         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2130                 return 0;
2131
2132         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2133         if (rc || ctxt == NULL) {
2134                 return 0;
2135         }
2136
2137         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2138         if (rc)
2139                 CERROR("Can't set mdt ctxt %d\n", rc);
2140
2141         return rc;
2142 }
2143
2144 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2145                                  struct mdt_device *mdt, int idx)
2146 {
2147         struct llog_ctxt *ctxt;
2148
2149         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2150         if (ctxt == NULL)
2151                 return 0;
2152         /* Put once for the get we just did, and once for the clone */
2153         llog_ctxt_put(ctxt);
2154         llog_ctxt_put(ctxt);
2155         return 0;
2156 }
2157
2158 /*
2159  * sec context handlers
2160  */
2161 int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2162 {
2163         int rc;
2164
2165         rc = mdt_handle_idmap(tsi);
2166         if (unlikely(rc)) {
2167                 struct ptlrpc_request   *req = tgt_ses_req(tsi);
2168                 __u32                    opc;
2169
2170                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2171                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2172                         sptlrpc_svc_ctx_invalidate(req);
2173         }
2174
2175         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2176
2177         return rc;
2178 }
2179
2180 /*
2181  * quota request handlers
2182  */
2183 int mdt_quota_dqacq(struct tgt_session_info *tsi)
2184 {
2185         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2186         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2187         int                      rc;
2188         ENTRY;
2189
2190         if (qmt == NULL)
2191                 RETURN(err_serious(-EOPNOTSUPP));
2192
2193         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2194         RETURN(rc);
2195 }
2196
2197 struct mdt_object *mdt_object_new(const struct lu_env *env,
2198                                   struct mdt_device *d,
2199                                   const struct lu_fid *f)
2200 {
2201         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2202         struct lu_object *o;
2203         struct mdt_object *m;
2204         ENTRY;
2205
2206         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2207         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2208         if (unlikely(IS_ERR(o)))
2209                 m = (struct mdt_object *)o;
2210         else
2211                 m = mdt_obj(o);
2212         RETURN(m);
2213 }
2214
2215 struct mdt_object *mdt_object_find(const struct lu_env *env,
2216                                    struct mdt_device *d,
2217                                    const struct lu_fid *f)
2218 {
2219         struct lu_object *o;
2220         struct mdt_object *m;
2221         ENTRY;
2222
2223         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2224         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2225         if (unlikely(IS_ERR(o)))
2226                 m = (struct mdt_object *)o;
2227         else
2228                 m = mdt_obj(o);
2229
2230         RETURN(m);
2231 }
2232
2233 /**
2234  * Asyncronous commit for mdt device.
2235  *
2236  * Pass asynchonous commit call down the MDS stack.
2237  *
2238  * \param env environment
2239  * \param mdt the mdt device
2240  */
2241 static void mdt_device_commit_async(const struct lu_env *env,
2242                                     struct mdt_device *mdt)
2243 {
2244         struct dt_device *dt = mdt->mdt_bottom;
2245         int rc;
2246
2247         rc = dt->dd_ops->dt_commit_async(env, dt);
2248         if (unlikely(rc != 0))
2249                 CWARN("async commit start failed with rc = %d", rc);
2250 }
2251
2252 /**
2253  * Mark the lock as "synchonous".
2254  *
2255  * Mark the lock to deffer transaction commit to the unlock time.
2256  *
2257  * \param lock the lock to mark as "synchonous"
2258  *
2259  * \see mdt_is_lock_sync
2260  * \see mdt_save_lock
2261  */
2262 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2263 {
2264         lock->l_ast_data = (void*)1;
2265 }
2266
2267 /**
2268  * Check whehter the lock "synchonous" or not.
2269  *
2270  * \param lock the lock to check
2271  * \retval 1 the lock is "synchonous"
2272  * \retval 0 the lock isn't "synchronous"
2273  *
2274  * \see mdt_set_lock_sync
2275  * \see mdt_save_lock
2276  */
2277 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2278 {
2279         return lock->l_ast_data != NULL;
2280 }
2281
2282 /**
2283  * Blocking AST for mdt locks.
2284  *
2285  * Starts transaction commit if in case of COS lock conflict or
2286  * deffers such a commit to the mdt_save_lock.
2287  *
2288  * \param lock the lock which blocks a request or cancelling lock
2289  * \param desc unused
2290  * \param data unused
2291  * \param flag indicates whether this cancelling or blocking callback
2292  * \retval 0
2293  * \see ldlm_blocking_ast_nocheck
2294  */
2295 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2296                      void *data, int flag)
2297 {
2298         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2299         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2300         int rc;
2301         ENTRY;
2302
2303         if (flag == LDLM_CB_CANCELING)
2304                 RETURN(0);
2305         lock_res_and_lock(lock);
2306         if (lock->l_blocking_ast != mdt_blocking_ast) {
2307                 unlock_res_and_lock(lock);
2308                 RETURN(0);
2309         }
2310         if (mdt_cos_is_enabled(mdt) &&
2311             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2312             lock->l_blocking_lock != NULL &&
2313             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2314                 mdt_set_lock_sync(lock);
2315         }
2316         rc = ldlm_blocking_ast_nocheck(lock);
2317
2318         /* There is no lock conflict if l_blocking_lock == NULL,
2319          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2320          * when the last reference to a local lock was released */
2321         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2322                 struct lu_env env;
2323
2324                 rc = lu_env_init(&env, LCT_LOCAL);
2325                 if (unlikely(rc != 0))
2326                         CWARN("lu_env initialization failed with rc = %d,"
2327                               "cannot start asynchronous commit\n", rc);
2328                 else
2329                         mdt_device_commit_async(&env, mdt);
2330                 lu_env_fini(&env);
2331         }
2332         RETURN(rc);
2333 }
2334
2335 /* Used for cross-MDT lock */
2336 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2337                             void *data, int flag)
2338 {
2339         struct lustre_handle lockh;
2340         int               rc;
2341
2342         switch (flag) {
2343         case LDLM_CB_BLOCKING:
2344                 ldlm_lock2handle(lock, &lockh);
2345                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2346                 if (rc < 0) {
2347                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2348                         RETURN(rc);
2349                 }
2350                 break;
2351         case LDLM_CB_CANCELING:
2352                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2353                 break;
2354         default:
2355                 LBUG();
2356         }
2357         RETURN(0);
2358 }
2359
2360 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2361                            struct mdt_object *o, const struct lu_fid *fid,
2362                            struct lustre_handle *lh, ldlm_mode_t mode,
2363                            __u64 ibits)
2364 {
2365         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2366         ldlm_policy_data_t *policy = &mti->mti_policy;
2367         struct ldlm_res_id *res_id = &mti->mti_res_id;
2368         int rc = 0;
2369         ENTRY;
2370
2371         LASSERT(mdt_object_remote(o));
2372
2373         LASSERT(ibits == MDS_INODELOCK_UPDATE);
2374
2375         fid_build_reg_res_name(fid, res_id);
2376
2377         memset(einfo, 0, sizeof(*einfo));
2378         einfo->ei_type = LDLM_IBITS;
2379         einfo->ei_mode = mode;
2380         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2381         einfo->ei_cb_cp = ldlm_completion_ast;
2382         einfo->ei_enq_slave = 0;
2383         einfo->ei_res_id = res_id;
2384
2385         memset(policy, 0, sizeof(*policy));
2386         policy->l_inodebits.bits = ibits;
2387
2388         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2389                             policy);
2390         RETURN(rc);
2391 }
2392
2393 static int mdt_object_local_lock(struct mdt_thread_info *info,
2394                                  struct mdt_object *o,
2395                                  struct mdt_lock_handle *lh, __u64 ibits,
2396                                  bool nonblock, int locality)
2397 {
2398         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2399         ldlm_policy_data_t *policy = &info->mti_policy;
2400         struct ldlm_res_id *res_id = &info->mti_res_id;
2401         __u64 dlmflags;
2402         int rc;
2403         ENTRY;
2404
2405         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2406         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2407         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2408         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2409
2410         /* Only enqueue LOOKUP lock for remote object */
2411         if (mdt_object_remote(o))
2412                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
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)))
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 == NULL ? NULL :
2450                                           &info->mti_exp->exp_handle.h_cookie);
2451                         if (unlikely(rc))
2452                                 RETURN(rc);
2453                 }
2454
2455                 /*
2456                  * Finish res_id initializing by name hash marking part of
2457                  * directory which is taking modification.
2458                  */
2459                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2460         }
2461
2462         policy->l_inodebits.bits = ibits;
2463
2464         /*
2465          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2466          * going to be sent to client. If it is - mdt_intent_policy() path will
2467          * fix it up and turn FL_LOCAL flag off.
2468          */
2469         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2470                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2471                           info->mti_exp == NULL ? NULL :
2472                           &info->mti_exp->exp_handle.h_cookie);
2473         if (rc)
2474                 mdt_object_unlock(info, o, lh, 1);
2475         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2476                  lh->mlh_pdo_hash != 0 &&
2477                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2478                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2479         }
2480
2481         RETURN(rc);
2482 }
2483
2484 int mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2485                              struct mdt_lock_handle *lh, __u64 ibits,
2486                              bool nonblock, int locality)
2487 {
2488         int rc;
2489         ENTRY;
2490
2491         if (!mdt_object_remote(o))
2492                 return mdt_object_local_lock(info, o, lh, ibits, nonblock,
2493                                              locality);
2494
2495         if (locality == MDT_LOCAL_LOCK) {
2496                 CERROR("%s: try to get local lock for remote object"
2497                        DFID".\n", mdt_obd_name(info->mti_mdt),
2498                        PFID(mdt_object_fid(o)));
2499                 RETURN(-EPROTO);
2500         }
2501
2502         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2503         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2504                    MDS_INODELOCK_XATTR);
2505         if (ibits & MDS_INODELOCK_UPDATE) {
2506                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2507                  * enqueue RPC can only request 1 lock, to avoid extra
2508                  * RPC, so it will instead enqueue EX lock for remote
2509                  * object anyway XXX*/
2510                 if (lh->mlh_type == MDT_PDO_LOCK &&
2511                     lh->mlh_pdo_hash != 0) {
2512                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2513                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2514                                PFID(mdt_object_fid(o)));
2515                         lh->mlh_pdo_hash = 0;
2516                         lh->mlh_rreg_mode = LCK_EX;
2517                         lh->mlh_type = MDT_REG_LOCK;
2518                 }
2519                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2520                                             &lh->mlh_rreg_lh,
2521                                             lh->mlh_rreg_mode,
2522                                             MDS_INODELOCK_UPDATE);
2523                 if (rc != ELDLM_OK)
2524                         RETURN(rc);
2525         }
2526
2527         /* Only enqueue LOOKUP lock for remote object */
2528         if (ibits & MDS_INODELOCK_LOOKUP) {
2529                 rc = mdt_object_local_lock(info, o, lh,
2530                                            MDS_INODELOCK_LOOKUP,
2531                                            nonblock, locality);
2532                 if (rc != ELDLM_OK)
2533                         RETURN(rc);
2534         }
2535
2536         RETURN(0);
2537 }
2538
2539 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2540                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2541 {
2542         return mdt_object_lock_internal(info, o, lh, ibits, false, locality);
2543 }
2544
2545 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2546                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2547 {
2548         struct mdt_lock_handle tmp = *lh;
2549         int rc;
2550
2551         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality);
2552         if (rc == 0)
2553                 *lh = tmp;
2554
2555         return rc == 0;
2556 }
2557
2558 /**
2559  * Save a lock within request object.
2560  *
2561  * Keep the lock referenced until whether client ACK or transaction
2562  * commit happens or release the lock immediately depending on input
2563  * parameters. If COS is ON, a write lock is converted to COS lock
2564  * before saving.
2565  *
2566  * \param info thead info object
2567  * \param h lock handle
2568  * \param mode lock mode
2569  * \param decref force immediate lock releasing
2570  */
2571 static
2572 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2573                    ldlm_mode_t mode, int decref)
2574 {
2575         ENTRY;
2576
2577         if (lustre_handle_is_used(h)) {
2578                 if (decref || !info->mti_has_trans ||
2579                     !(mode & (LCK_PW | LCK_EX))){
2580                         mdt_fid_unlock(h, mode);
2581                 } else {
2582                         struct mdt_device *mdt = info->mti_mdt;
2583                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2584                         struct ptlrpc_request *req = mdt_info_req(info);
2585                         int no_ack = 0;
2586
2587                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2588                                  h->cookie);
2589                         /* there is no request if mdt_object_unlock() is called
2590                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2591                         if (likely(req != NULL)) {
2592                                 CDEBUG(D_HA, "request = %p reply state = %p"
2593                                        " transno = "LPD64"\n", req,
2594                                        req->rq_reply_state, req->rq_transno);
2595                                 if (mdt_cos_is_enabled(mdt)) {
2596                                         no_ack = 1;
2597                                         ldlm_lock_downgrade(lock, LCK_COS);
2598                                         mode = LCK_COS;
2599                                 }
2600                                 ptlrpc_save_lock(req, h, mode, no_ack);
2601                         } else {
2602                                 ldlm_lock_decref(h, mode);
2603                         }
2604                         if (mdt_is_lock_sync(lock)) {
2605                                 CDEBUG(D_HA, "found sync-lock,"
2606                                        " async commit started\n");
2607                                 mdt_device_commit_async(info->mti_env,
2608                                                         mdt);
2609                         }
2610                         LDLM_LOCK_PUT(lock);
2611                 }
2612                 h->cookie = 0ull;
2613         }
2614
2615         EXIT;
2616 }
2617
2618 /**
2619  * Unlock mdt object.
2620  *
2621  * Immeditely release the regular lock and the PDO lock or save the
2622  * lock in reqeuest and keep them referenced until client ACK or
2623  * transaction commit.
2624  *
2625  * \param info thread info object
2626  * \param o mdt object
2627  * \param lh mdt lock handle referencing regular and PDO locks
2628  * \param decref force immediate lock releasing
2629  */
2630 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2631                        struct mdt_lock_handle *lh, int decref)
2632 {
2633         ENTRY;
2634
2635         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2636         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2637
2638         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2639                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2640
2641         EXIT;
2642 }
2643
2644 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2645                                         const struct lu_fid *f,
2646                                         struct mdt_lock_handle *lh,
2647                                         __u64 ibits)
2648 {
2649         struct mdt_object *o;
2650
2651         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2652         if (!IS_ERR(o)) {
2653                 int rc;
2654
2655                 rc = mdt_object_lock(info, o, lh, ibits,
2656                                      MDT_LOCAL_LOCK);
2657                 if (rc != 0) {
2658                         mdt_object_put(info->mti_env, o);
2659                         o = ERR_PTR(rc);
2660                 }
2661         }
2662         return o;
2663 }
2664
2665 void mdt_object_unlock_put(struct mdt_thread_info * info,
2666                            struct mdt_object * o,
2667                            struct mdt_lock_handle *lh,
2668                            int decref)
2669 {
2670         mdt_object_unlock(info, o, lh, decref);
2671         mdt_object_put(info->mti_env, o);
2672 }
2673
2674 /*
2675  * Generic code handling requests that have struct mdt_body passed in:
2676  *
2677  *  - extract mdt_body from request and save it in @info, if present;
2678  *
2679  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2680  *  @info;
2681  *
2682  *  - if HABEO_CORPUS flag is set for this request type check whether object
2683  *  actually exists on storage (lu_object_exists()).
2684  *
2685  */
2686 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2687 {
2688         const struct mdt_body    *body;
2689         struct mdt_object        *obj;
2690         const struct lu_env      *env;
2691         struct req_capsule       *pill;
2692         int                       rc;
2693         ENTRY;
2694
2695         env = info->mti_env;
2696         pill = info->mti_pill;
2697
2698         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2699         if (body == NULL)
2700                 RETURN(-EFAULT);
2701
2702         if (!(body->valid & OBD_MD_FLID))
2703                 RETURN(0);
2704
2705         if (!fid_is_sane(&body->fid1)) {
2706                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2707                 RETURN(-EINVAL);
2708         }
2709
2710         /*
2711          * Do not get size or any capa fields before we check that request
2712          * contains capa actually. There are some requests which do not, for
2713          * instance MDS_IS_SUBDIR.
2714          */
2715         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2716             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2717                 mdt_set_capainfo(info, 0, &body->fid1,
2718                                  req_capsule_client_get(pill, &RMF_CAPA1));
2719
2720         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2721         if (!IS_ERR(obj)) {
2722                 if ((flags & HABEO_CORPUS) &&
2723                     !mdt_object_exists(obj)) {
2724                         mdt_object_put(env, obj);
2725                         /* for capability renew ENOENT will be handled in
2726                          * mdt_renew_capa */
2727                         if (body->valid & OBD_MD_FLOSSCAPA)
2728                                 rc = 0;
2729                         else
2730                                 rc = -ENOENT;
2731                 } else {
2732                         info->mti_object = obj;
2733                         rc = 0;
2734                 }
2735         } else
2736                 rc = PTR_ERR(obj);
2737
2738         RETURN(rc);
2739 }
2740
2741 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2742 {
2743         struct req_capsule *pill = info->mti_pill;
2744         int rc;
2745         ENTRY;
2746
2747         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2748                 rc = mdt_body_unpack(info, flags);
2749         else
2750                 rc = 0;
2751
2752         if (rc == 0 && (flags & HABEO_REFERO)) {
2753                 /* Pack reply. */
2754                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2755                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2756                                              DEF_REP_MD_SIZE);
2757                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2758                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2759                                              RCL_SERVER, 0);
2760
2761                 rc = req_capsule_server_pack(pill);
2762         }
2763         RETURN(rc);
2764 }
2765
2766 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2767 {
2768         struct md_device *next = m->mdt_child;
2769
2770         return next->md_ops->mdo_init_capa_ctxt(env, next,
2771                                                 m->mdt_lut.lut_mds_capa,
2772                                                 m->mdt_capa_timeout,
2773                                                 m->mdt_capa_alg,
2774                                                 m->mdt_capa_keys);
2775 }
2776
2777 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2778 {
2779         lh->mlh_type = MDT_NUL_LOCK;
2780         lh->mlh_reg_lh.cookie = 0ull;
2781         lh->mlh_reg_mode = LCK_MINMODE;
2782         lh->mlh_pdo_lh.cookie = 0ull;
2783         lh->mlh_pdo_mode = LCK_MINMODE;
2784         lh->mlh_rreg_lh.cookie = 0ull;
2785         lh->mlh_rreg_mode = LCK_MINMODE;
2786 }
2787
2788 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2789 {
2790         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2791         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2792 }
2793
2794 /*
2795  * Initialize fields of struct mdt_thread_info. Other fields are left in
2796  * uninitialized state, because it's too expensive to zero out whole
2797  * mdt_thread_info (> 1K) on each request arrival.
2798  */
2799 void mdt_thread_info_init(struct ptlrpc_request *req,
2800                           struct mdt_thread_info *info)
2801 {
2802         int i;
2803
2804         info->mti_pill = &req->rq_pill;
2805
2806         /* lock handle */
2807         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2808                 mdt_lock_handle_init(&info->mti_lh[i]);
2809
2810         /* mdt device: it can be NULL while CONNECT */
2811         if (req->rq_export) {
2812                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2813                 info->mti_exp = req->rq_export;
2814         } else
2815                 info->mti_mdt = NULL;
2816         info->mti_env = req->rq_svc_thread->t_env;
2817         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2818
2819         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2820         info->mti_big_buf = LU_BUF_NULL;
2821         info->mti_body = NULL;
2822         info->mti_object = NULL;
2823         info->mti_dlm_req = NULL;
2824         info->mti_has_trans = 0;
2825         info->mti_cross_ref = 0;
2826         info->mti_opdata = 0;
2827         info->mti_big_lmm_used = 0;
2828
2829         info->mti_spec.no_create = 0;
2830         info->mti_spec.sp_rm_entry = 0;
2831
2832         info->mti_spec.u.sp_ea.eadata = NULL;
2833         info->mti_spec.u.sp_ea.eadatalen = 0;
2834 }
2835
2836 void mdt_thread_info_fini(struct mdt_thread_info *info)
2837 {
2838         int i;
2839
2840         if (info->mti_object != NULL) {
2841                 mdt_object_put(info->mti_env, info->mti_object);
2842                 info->mti_object = NULL;
2843         }
2844
2845         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2846                 mdt_lock_handle_fini(&info->mti_lh[i]);
2847         info->mti_env = NULL;
2848         info->mti_pill = NULL;
2849         info->mti_exp = NULL;
2850
2851         if (unlikely(info->mti_big_buf.lb_buf != NULL))
2852                 lu_buf_free(&info->mti_big_buf);
2853 }
2854
2855 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2856 {
2857         struct mdt_thread_info  *mti;
2858         struct lustre_capa      *lc;
2859
2860         mti = mdt_th_info(tsi->tsi_env);
2861         LASSERT(mti != NULL);
2862
2863         mdt_thread_info_init(tgt_ses_req(tsi), mti);
2864         if (tsi->tsi_corpus != NULL) {
2865                 struct req_capsule *pill = tsi->tsi_pill;
2866
2867                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2868                 lu_object_get(tsi->tsi_corpus);
2869
2870                 /*
2871                  * XXX: must be part of tgt_mdt_body_unpack but moved here
2872                  * due to mdt_set_capainfo().
2873                  */
2874                 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2875                     req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) {
2876                         lc = req_capsule_client_get(pill, &RMF_CAPA1);
2877                         mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->fid1, lc);
2878                 }
2879         }
2880         mti->mti_body = tsi->tsi_mdt_body;
2881         mti->mti_dlm_req = tsi->tsi_dlm_req;
2882
2883         return mti;
2884 }
2885
2886 int mdt_tgt_connect(struct tgt_session_info *tsi)
2887 {
2888         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2889         int                      rc;
2890
2891         ENTRY;
2892
2893         rc = tgt_connect(tsi);
2894         if (rc != 0)
2895                 RETURN(rc);
2896
2897         rc = mdt_init_idmap(tsi);
2898         if (rc != 0)
2899                 GOTO(err, rc);
2900         RETURN(0);
2901 err:
2902         obd_disconnect(class_export_get(req->rq_export));
2903         return rc;
2904 }
2905
2906 enum mdt_it_code {
2907         MDT_IT_OPEN,
2908         MDT_IT_OCREAT,
2909         MDT_IT_CREATE,
2910         MDT_IT_GETATTR,
2911         MDT_IT_READDIR,
2912         MDT_IT_LOOKUP,
2913         MDT_IT_UNLINK,
2914         MDT_IT_TRUNC,
2915         MDT_IT_GETXATTR,
2916         MDT_IT_LAYOUT,
2917         MDT_IT_QUOTA,
2918         MDT_IT_NR
2919 };
2920
2921 static int mdt_intent_getattr(enum mdt_it_code opcode,
2922                               struct mdt_thread_info *info,
2923                               struct ldlm_lock **,
2924                               __u64);
2925
2926 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2927                                 struct mdt_thread_info *info,
2928                                 struct ldlm_lock **lockp,
2929                                 __u64 flags);
2930
2931 static int mdt_intent_layout(enum mdt_it_code opcode,
2932                              struct mdt_thread_info *info,
2933                              struct ldlm_lock **,
2934                              __u64);
2935 static int mdt_intent_reint(enum mdt_it_code opcode,
2936                             struct mdt_thread_info *info,
2937                             struct ldlm_lock **,
2938                             __u64);
2939
2940 static struct mdt_it_flavor {
2941         const struct req_format *it_fmt;
2942         __u32                    it_flags;
2943         int                    (*it_act)(enum mdt_it_code ,
2944                                          struct mdt_thread_info *,
2945                                          struct ldlm_lock **,
2946                                          __u64);
2947         long                     it_reint;
2948 } mdt_it_flavor[] = {
2949         [MDT_IT_OPEN]     = {
2950                 .it_fmt   = &RQF_LDLM_INTENT,
2951                 /*.it_flags = HABEO_REFERO,*/
2952                 .it_flags = 0,
2953                 .it_act   = mdt_intent_reint,
2954                 .it_reint = REINT_OPEN
2955         },
2956         [MDT_IT_OCREAT]   = {
2957                 .it_fmt   = &RQF_LDLM_INTENT,
2958                 /*
2959                  * OCREAT is not a MUTABOR request as if the file
2960                  * already exists.
2961                  * We do the extra check of OBD_CONNECT_RDONLY in
2962                  * mdt_reint_open() when we really need to create
2963                  * the object.
2964                  */
2965                 .it_flags = 0,
2966                 .it_act   = mdt_intent_reint,
2967                 .it_reint = REINT_OPEN
2968         },
2969         [MDT_IT_CREATE]   = {
2970                 .it_fmt   = &RQF_LDLM_INTENT,
2971                 .it_flags = MUTABOR,
2972                 .it_act   = mdt_intent_reint,
2973                 .it_reint = REINT_CREATE
2974         },
2975         [MDT_IT_GETATTR]  = {
2976                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2977                 .it_flags = HABEO_REFERO,
2978                 .it_act   = mdt_intent_getattr
2979         },
2980         [MDT_IT_READDIR]  = {
2981                 .it_fmt   = NULL,
2982                 .it_flags = 0,
2983                 .it_act   = NULL
2984         },
2985         [MDT_IT_LOOKUP]   = {
2986                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2987                 .it_flags = HABEO_REFERO,
2988                 .it_act   = mdt_intent_getattr
2989         },
2990         [MDT_IT_UNLINK]   = {
2991                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2992                 .it_flags = MUTABOR,
2993                 .it_act   = NULL,
2994                 .it_reint = REINT_UNLINK
2995         },
2996         [MDT_IT_TRUNC]    = {
2997                 .it_fmt   = NULL,
2998                 .it_flags = MUTABOR,
2999                 .it_act   = NULL
3000         },
3001         [MDT_IT_GETXATTR] = {
3002                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
3003                 .it_flags = HABEO_CORPUS,
3004                 .it_act   = mdt_intent_getxattr
3005         },
3006         [MDT_IT_LAYOUT] = {
3007                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3008                 .it_flags = 0,
3009                 .it_act   = mdt_intent_layout
3010         }
3011 };
3012
3013 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3014                             struct ldlm_lock **lockp,
3015                             struct ldlm_lock *new_lock,
3016                             struct mdt_lock_handle *lh,
3017                             __u64 flags)
3018 {
3019         struct ptlrpc_request  *req = mdt_info_req(info);
3020         struct ldlm_lock       *lock = *lockp;
3021
3022         /*
3023          * Get new lock only for cases when possible resent did not find any
3024          * lock.
3025          */
3026         if (new_lock == NULL)
3027                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3028
3029         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3030                 lh->mlh_reg_lh.cookie = 0;
3031                 RETURN(0);
3032         }
3033
3034         LASSERTF(new_lock != NULL,
3035                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3036
3037         /*
3038          * If we've already given this lock to a client once, then we should
3039          * have no readers or writers.  Otherwise, we should have one reader
3040          * _or_ writer ref (which will be zeroed below) before returning the
3041          * lock to a client.
3042          */
3043         if (new_lock->l_export == req->rq_export) {
3044                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3045         } else {
3046                 LASSERT(new_lock->l_export == NULL);
3047                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3048         }
3049
3050         *lockp = new_lock;
3051
3052         if (new_lock->l_export == req->rq_export) {
3053                 /*
3054                  * Already gave this to the client, which means that we
3055                  * reconstructed a reply.
3056                  */
3057                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3058                         MSG_RESENT);
3059                 lh->mlh_reg_lh.cookie = 0;
3060                 RETURN(ELDLM_LOCK_REPLACED);
3061         }
3062
3063         /*
3064          * Fixup the lock to be given to the client.
3065          */
3066         lock_res_and_lock(new_lock);
3067         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3068          * possible blocking AST. */
3069         while (new_lock->l_readers > 0) {
3070                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3071                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3072                 new_lock->l_readers--;
3073         }
3074         while (new_lock->l_writers > 0) {
3075                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3076                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3077                 new_lock->l_writers--;
3078         }
3079
3080         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3081         new_lock->l_blocking_ast = lock->l_blocking_ast;
3082         new_lock->l_completion_ast = lock->l_completion_ast;
3083         new_lock->l_remote_handle = lock->l_remote_handle;
3084         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3085
3086         unlock_res_and_lock(new_lock);
3087
3088         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3089                      &new_lock->l_remote_handle,
3090                      &new_lock->l_exp_hash);
3091
3092         LDLM_LOCK_RELEASE(new_lock);
3093         lh->mlh_reg_lh.cookie = 0;
3094
3095         RETURN(ELDLM_LOCK_REPLACED);
3096 }
3097
3098 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3099                                     struct ldlm_lock *new_lock,
3100                                     struct ldlm_lock **old_lock,
3101                                     struct mdt_lock_handle *lh,
3102                                     enum mdt_it_code opcode)
3103 {
3104         struct ptlrpc_request  *req = mdt_info_req(info);
3105         struct obd_export      *exp = req->rq_export;
3106         struct lustre_handle    remote_hdl;
3107         struct ldlm_request    *dlmreq;
3108         struct ldlm_lock       *lock;
3109
3110         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3111                 return;
3112
3113         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3114         remote_hdl = dlmreq->lock_handle[0];
3115         /* If the client does not require open lock, it does not need to
3116          * search lock in exp_lock_hash, since the server thread will
3117          * make sure the lock will be released, and the resend request
3118          * can always re-enqueue the lock */
3119         if ((opcode != MDT_IT_OPEN) || (opcode == MDT_IT_OPEN &&
3120             info->mti_spec.sp_cr_flags & MDS_OPEN_LOCK)) {
3121                 /* In the function below, .hs_keycmp resolves to
3122                  * ldlm_export_lock_keycmp() */
3123                 /* coverity[overrun-buffer-val] */
3124                 lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3125                 if (lock) {
3126                         lock_res_and_lock(lock);
3127                         if (lock != new_lock) {
3128                                 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3129                                 lh->mlh_reg_mode = lock->l_granted_mode;
3130
3131                                 LDLM_DEBUG(lock, "Restoring lock cookie");
3132                                 DEBUG_REQ(D_DLMTRACE, req,
3133                                           "restoring lock cookie "LPX64,
3134                                           lh->mlh_reg_lh.cookie);
3135                                 if (old_lock)
3136                                         *old_lock = LDLM_LOCK_GET(lock);
3137                                 cfs_hash_put(exp->exp_lock_hash,
3138                                              &lock->l_exp_hash);
3139                                 unlock_res_and_lock(lock);
3140                                 return;
3141                         }
3142                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3143                         unlock_res_and_lock(lock);
3144                 }
3145         }
3146         /*
3147          * If the xid matches, then we know this is a resent request, and allow
3148          * it. (It's probably an OPEN, for which we don't send a lock.
3149          */
3150         if (req_xid_is_last(req))
3151                 return;
3152
3153         /*
3154          * This remote handle isn't enqueued, so we never received or processed
3155          * this request.  Clear MSG_RESENT, because it can be handled like any
3156          * normal request now.
3157          */
3158         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3159
3160         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3161                   remote_hdl.cookie);
3162 }
3163
3164 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3165                                 struct mdt_thread_info *info,
3166                                 struct ldlm_lock **lockp,
3167                                 __u64 flags)
3168 {
3169         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3170         struct ldlm_reply      *ldlm_rep = NULL;
3171         int rc, grc;
3172
3173         /*
3174          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3175          * (for the resend case) or a new lock. Below we will use it to
3176          * replace the original lock.
3177          */
3178         mdt_intent_fixup_resent(info, *lockp, NULL, lhc, opcode);
3179         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3180                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3181                 rc = mdt_object_lock(info, info->mti_object, lhc,
3182                                         MDS_INODELOCK_XATTR,
3183                                         MDT_LOCAL_LOCK);
3184                 if (rc)
3185                         return rc;
3186         }
3187
3188         grc = mdt_getxattr(info);
3189
3190         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3191
3192         if (mdt_info_req(info)->rq_repmsg != NULL)
3193                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3194         if (ldlm_rep == NULL)
3195                 RETURN(err_serious(-EFAULT));
3196
3197         ldlm_rep->lock_policy_res2 = grc;
3198
3199         return rc;
3200 }
3201
3202 static int mdt_intent_getattr(enum mdt_it_code opcode,
3203                               struct mdt_thread_info *info,
3204                               struct ldlm_lock **lockp,
3205                               __u64 flags)
3206 {
3207         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3208         struct ldlm_lock       *new_lock = NULL;
3209         __u64                   child_bits;
3210         struct ldlm_reply      *ldlm_rep;
3211         struct ptlrpc_request  *req;
3212         struct mdt_body        *reqbody;
3213         struct mdt_body        *repbody;
3214         int                     rc, rc2;
3215         ENTRY;
3216
3217         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3218         LASSERT(reqbody);
3219
3220         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3221         LASSERT(repbody);
3222
3223         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3224         repbody->eadatasize = 0;
3225         repbody->aclsize = 0;
3226
3227         switch (opcode) {
3228         case MDT_IT_LOOKUP:
3229                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3230                 break;
3231         case MDT_IT_GETATTR:
3232                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3233                              MDS_INODELOCK_PERM;
3234                 break;
3235         default:
3236                 CERROR("Unsupported intent (%d)\n", opcode);
3237                 GOTO(out_shrink, rc = -EINVAL);
3238         }
3239
3240         rc = mdt_init_ucred(info, reqbody);
3241         if (rc)
3242                 GOTO(out_shrink, rc);
3243
3244         req = info->mti_pill->rc_req;
3245         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3246         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3247
3248         /* Get lock from request for possible resent case. */
3249         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc, opcode);
3250
3251         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3252         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3253
3254         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3255                 ldlm_rep->lock_policy_res2 = 0;
3256         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3257             ldlm_rep->lock_policy_res2) {
3258                 lhc->mlh_reg_lh.cookie = 0ull;
3259                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3260         }
3261
3262         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3263         EXIT;
3264 out_ucred:
3265         mdt_exit_ucred(info);
3266 out_shrink:
3267         mdt_client_compatibility(info);
3268         rc2 = mdt_fix_reply(info);
3269         if (rc == 0)
3270                 rc = rc2;
3271         return rc;
3272 }
3273
3274 static int mdt_intent_layout(enum mdt_it_code opcode,
3275                              struct mdt_thread_info *info,
3276                              struct ldlm_lock **lockp,
3277                              __u64 flags)
3278 {
3279         struct layout_intent *layout;
3280         struct lu_fid *fid;
3281         struct mdt_object *obj = NULL;
3282         int rc = 0;
3283         ENTRY;
3284
3285         if (opcode != MDT_IT_LAYOUT) {
3286                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3287                         opcode);
3288                 RETURN(-EINVAL);
3289         }
3290
3291         fid = &info->mti_tmp_fid2;
3292         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3293
3294         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3295         if (IS_ERR(obj))
3296                 RETURN(PTR_ERR(obj));
3297
3298         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3299                 /* get the length of lsm */
3300                 rc = mdt_attr_get_eabuf_size(info, obj);
3301                 if (rc < 0)
3302                         RETURN(rc);
3303
3304                 if (rc > info->mti_mdt->mdt_max_mdsize)
3305                         info->mti_mdt->mdt_max_mdsize = rc;
3306         }
3307
3308         mdt_object_put(info->mti_env, obj);
3309
3310         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3311         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, rc);
3312         rc = req_capsule_server_pack(info->mti_pill);
3313         if (rc != 0)
3314                 RETURN(-EINVAL);
3315
3316         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3317         LASSERT(layout != NULL);
3318         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3319                 /* return to normal ldlm handling */
3320                 RETURN(0);
3321
3322         CERROR("%s: Unsupported layout intent (%d)\n",
3323                 mdt_obd_name(info->mti_mdt), layout->li_opc);
3324         RETURN(-EINVAL);
3325 }
3326
3327 static int mdt_intent_reint(enum mdt_it_code opcode,
3328                             struct mdt_thread_info *info,
3329                             struct ldlm_lock **lockp,
3330                             __u64 flags)
3331 {
3332         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3333         struct ldlm_reply      *rep = NULL;
3334         long                    opc;
3335         int                     rc;
3336
3337         static const struct req_format *intent_fmts[REINT_MAX] = {
3338                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3339                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3340         };
3341
3342         ENTRY;
3343
3344         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3345         if (opc < 0)
3346                 RETURN(opc);
3347
3348         if (mdt_it_flavor[opcode].it_reint != opc) {
3349                 CERROR("Reint code %ld doesn't match intent: %d\n",
3350                        opc, opcode);
3351                 RETURN(err_serious(-EPROTO));
3352         }
3353
3354         /* Get lock from request for possible resent case. */
3355         mdt_intent_fixup_resent(info, *lockp, NULL, lhc, opcode);
3356
3357         rc = mdt_reint_internal(info, lhc, opc);
3358
3359         /* Check whether the reply has been packed successfully. */
3360         if (mdt_info_req(info)->rq_repmsg != NULL)
3361                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3362         if (rep == NULL)
3363                 RETURN(err_serious(-EFAULT));
3364
3365         /* MDC expects this in any case */
3366         if (rc != 0)
3367                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3368
3369         /* the open lock or the lock for cross-ref object should be
3370          * returned to the client */
3371         if (rc == -EREMOTE || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3372                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3373                 rep->lock_policy_res2 = 0;
3374                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3375                 RETURN(rc);
3376         }
3377
3378         rep->lock_policy_res2 = clear_serious(rc);
3379
3380         if (rep->lock_policy_res2 == -ENOENT &&
3381             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3382                 rep->lock_policy_res2 = 0;
3383
3384         if (rc == -ENOTCONN || rc == -ENODEV ||
3385             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3386                 /*
3387                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3388                  * will be returned by rq_status, and client at ptlrpc layer
3389                  * will detect this, then disconnect, reconnect the import
3390                  * immediately, instead of impacting the following the rpc.
3391                  */
3392                 lhc->mlh_reg_lh.cookie = 0ull;
3393                 RETURN(rc);
3394         } else {
3395                 /*
3396                  * For other cases, the error will be returned by intent.
3397                  * and client will retrieve the result from intent.
3398                  */
3399                  /*
3400                   * FIXME: when open lock is finished, that should be
3401                   * checked here.
3402                   */
3403                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3404                         LASSERTF(rc == 0, "Error occurred but lock handle "
3405                                  "is still in use, rc = %d\n", rc);
3406                         rep->lock_policy_res2 = 0;
3407                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3408                         RETURN(rc);
3409                 } else {
3410                         lhc->mlh_reg_lh.cookie = 0ull;
3411                         RETURN(ELDLM_LOCK_ABORTED);
3412                 }
3413         }
3414 }
3415
3416 static int mdt_intent_code(long itcode)
3417 {
3418         int rc;
3419
3420         switch(itcode) {
3421         case IT_OPEN:
3422                 rc = MDT_IT_OPEN;
3423                 break;
3424         case IT_OPEN|IT_CREAT:
3425                 rc = MDT_IT_OCREAT;
3426                 break;
3427         case IT_CREAT:
3428                 rc = MDT_IT_CREATE;
3429                 break;
3430         case IT_READDIR:
3431                 rc = MDT_IT_READDIR;
3432                 break;
3433         case IT_GETATTR:
3434                 rc = MDT_IT_GETATTR;
3435                 break;
3436         case IT_LOOKUP:
3437                 rc = MDT_IT_LOOKUP;
3438                 break;
3439         case IT_UNLINK:
3440                 rc = MDT_IT_UNLINK;
3441                 break;
3442         case IT_TRUNC:
3443                 rc = MDT_IT_TRUNC;
3444                 break;
3445         case IT_GETXATTR:
3446                 rc = MDT_IT_GETXATTR;
3447                 break;
3448         case IT_LAYOUT:
3449                 rc = MDT_IT_LAYOUT;
3450                 break;
3451         case IT_QUOTA_DQACQ:
3452         case IT_QUOTA_CONN:
3453                 rc = MDT_IT_QUOTA;
3454                 break;
3455         default:
3456                 CERROR("Unknown intent opcode: %ld\n", itcode);
3457                 rc = -EINVAL;
3458                 break;
3459         }
3460         return rc;
3461 }
3462
3463 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3464                           struct ldlm_lock **lockp, __u64 flags)
3465 {
3466         struct req_capsule   *pill;
3467         struct mdt_it_flavor *flv;
3468         int opc;
3469         int rc;
3470         ENTRY;
3471
3472         opc = mdt_intent_code(itopc);
3473         if (opc < 0)
3474                 RETURN(-EINVAL);
3475
3476         pill = info->mti_pill;
3477
3478         if (opc == MDT_IT_QUOTA) {
3479                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3480
3481                 if (qmt == NULL)
3482                         RETURN(-EOPNOTSUPP);
3483
3484                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3485                 /* pass the request to quota master */
3486                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3487                                                  mdt_info_req(info), lockp,
3488                                                  flags);
3489                 RETURN(rc);
3490         }
3491
3492         flv  = &mdt_it_flavor[opc];
3493         if (flv->it_fmt != NULL)
3494                 req_capsule_extend(pill, flv->it_fmt);
3495
3496         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3497         if (rc == 0) {
3498                 struct ptlrpc_request *req = mdt_info_req(info);
3499                 if (flv->it_flags & MUTABOR &&
3500                     exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3501                         RETURN(-EROFS);
3502         }
3503         if (rc == 0 && flv->it_act != NULL) {
3504                 struct ldlm_reply *rep;
3505
3506                 /* execute policy */
3507                 rc = flv->it_act(opc, info, lockp, flags);
3508
3509                 /* Check whether the reply has been packed successfully. */
3510                 if (mdt_info_req(info)->rq_repmsg != NULL) {
3511                         rep = req_capsule_server_get(info->mti_pill,
3512                                                      &RMF_DLM_REP);
3513                         rep->lock_policy_res2 =
3514                                 ptlrpc_status_hton(rep->lock_policy_res2);
3515                 }
3516         } else {
3517                 rc = -EPROTO;
3518         }
3519         RETURN(rc);
3520 }
3521
3522 static int mdt_intent_policy(struct ldlm_namespace *ns,
3523                              struct ldlm_lock **lockp, void *req_cookie,
3524                              ldlm_mode_t mode, __u64 flags, void *data)
3525 {
3526         struct tgt_session_info *tsi;
3527         struct mdt_thread_info  *info;
3528         struct ptlrpc_request   *req  =  req_cookie;
3529         struct ldlm_intent      *it;
3530         struct req_capsule      *pill;
3531         int rc;
3532
3533         ENTRY;
3534
3535         LASSERT(req != NULL);
3536
3537         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3538
3539         info = tsi2mdt_info(tsi);
3540         LASSERT(info != NULL);
3541         pill = info->mti_pill;
3542         LASSERT(pill->rc_req == req);
3543
3544         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3545                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3546                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3547                 if (it != NULL) {
3548                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3549                         if (rc == 0)
3550                                 rc = ELDLM_OK;
3551
3552                         /* Lock without inodebits makes no sense and will oops
3553                          * later in ldlm. Let's check it now to see if we have
3554                          * ibits corrupted somewhere in mdt_intent_opc().
3555                          * The case for client miss to set ibits has been
3556                          * processed by others. */
3557                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3558                                         lr_type == LDLM_IBITS,
3559                                      info->mti_dlm_req->lock_desc.\
3560                                         l_policy_data.l_inodebits.bits != 0));
3561                 } else
3562                         rc = err_serious(-EFAULT);
3563         } else {
3564                 /* No intent was provided */
3565                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3566                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3567                 rc = req_capsule_server_pack(pill);
3568                 if (rc)
3569                         rc = err_serious(rc);
3570         }
3571         mdt_thread_info_fini(info);
3572         RETURN(rc);
3573 }
3574
3575 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3576 {
3577         struct seq_server_site  *ss = mdt_seq_site(mdt);
3578
3579         if (ss->ss_node_id == 0)
3580                 return;
3581
3582         if (ss->ss_client_seq != NULL) {
3583                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3584                 ss->ss_client_seq->lcs_exp = NULL;
3585         }
3586
3587         if (ss->ss_server_fld != NULL) {
3588                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3589                 ss->ss_server_fld->lsf_control_exp = NULL;
3590         }
3591 }
3592
3593 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3594 {
3595         struct seq_server_site *ss = mdt_seq_site(mdt);
3596
3597         if (ss == NULL)
3598                 return;
3599
3600         if (ss->ss_server_seq == NULL)
3601                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3602
3603         return;
3604 }
3605
3606 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3607 {
3608         mdt_seq_fini_cli(mdt);
3609         mdt_deregister_seq_exp(mdt);
3610
3611         return seq_site_fini(env, mdt_seq_site(mdt));
3612 }
3613
3614 /**
3615  * It will retrieve its FLDB entries from MDT0, and it only happens
3616  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3617  * and it needs to refresh FLDB from the MDT0.
3618  **/
3619 static int mdt_register_lwp_callback(void *data)
3620 {
3621         struct lu_env           env;
3622         struct mdt_device       *mdt = data;
3623         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3624         int                     rc;
3625         ENTRY;
3626
3627         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3628
3629         if (!likely(fld->lsf_new))
3630                 RETURN(0);
3631
3632         rc = lu_env_init(&env, LCT_MD_THREAD);
3633         if (rc) {
3634                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3635                 RETURN(rc);
3636         }
3637
3638         rc = fld_update_from_controller(&env, fld);
3639         if (rc != 0) {
3640                 CERROR("%s: cannot update controller: rc = %d\n",
3641                        mdt_obd_name(mdt), rc);
3642                 GOTO(out, rc);
3643         }
3644 out:
3645         lu_env_fini(&env);
3646         RETURN(rc);
3647 }
3648
3649 static int mdt_register_seq_exp(struct mdt_device *mdt)
3650 {
3651         struct seq_server_site  *ss = mdt_seq_site(mdt);
3652         char                    *lwp_name = NULL;
3653         int                     rc;
3654
3655         if (ss->ss_node_id == 0)
3656                 return 0;
3657
3658         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3659         if (lwp_name == NULL)
3660                 GOTO(out_free, rc = -ENOMEM);
3661
3662         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3663         if (rc != 0)
3664                 GOTO(out_free, rc);
3665
3666         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3667                                       NULL, NULL);
3668         if (rc != 0)
3669                 GOTO(out_free, rc);
3670
3671         rc = lustre_register_lwp_item(lwp_name,
3672                                       &ss->ss_server_fld->lsf_control_exp,
3673                                       mdt_register_lwp_callback, mdt);
3674         if (rc != 0) {