Whamcloud - gitweb
204ebd5425561b352e5585c5e979edc73788def5
[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 #include <lustre_ioctl.h>
57 /* struct ptlrpc_request */
58 #include <lustre_net.h>
59 /* struct obd_export */
60 #include <lustre_export.h>
61 /* struct obd_device */
62 #include <obd.h>
63 /* lu2dt_dev() */
64 #include <dt_object.h>
65 #include <lustre_mds.h>
66 #include <lustre_log.h>
67 #include "mdt_internal.h"
68 #include <lustre_acl.h>
69 #include <lustre_param.h>
70 #include <lustre_quota.h>
71 #include <lustre_lfsck.h>
72
73 mdl_mode_t mdt_mdl_lock_modes[] = {
74         [LCK_MINMODE] = MDL_MINMODE,
75         [LCK_EX]      = MDL_EX,
76         [LCK_PW]      = MDL_PW,
77         [LCK_PR]      = MDL_PR,
78         [LCK_CW]      = MDL_CW,
79         [LCK_CR]      = MDL_CR,
80         [LCK_NL]      = MDL_NL,
81         [LCK_GROUP]   = MDL_GROUP
82 };
83
84 ldlm_mode_t mdt_dlm_lock_modes[] = {
85         [MDL_MINMODE] = LCK_MINMODE,
86         [MDL_EX]      = LCK_EX,
87         [MDL_PW]      = LCK_PW,
88         [MDL_PR]      = LCK_PR,
89         [MDL_CW]      = LCK_CW,
90         [MDL_CR]      = LCK_CR,
91         [MDL_NL]      = LCK_NL,
92         [MDL_GROUP]   = LCK_GROUP
93 };
94
95 static struct mdt_device *mdt_dev(struct lu_device *d);
96 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
97
98 static const struct lu_object_operations mdt_obj_ops;
99
100 /* Slab for MDT object allocation */
101 static struct kmem_cache *mdt_object_kmem;
102
103 /* For HSM restore handles */
104 struct kmem_cache *mdt_hsm_cdt_kmem;
105
106 /* For HSM request handles */
107 struct kmem_cache *mdt_hsm_car_kmem;
108
109 static struct lu_kmem_descr mdt_caches[] = {
110         {
111                 .ckd_cache = &mdt_object_kmem,
112                 .ckd_name  = "mdt_obj",
113                 .ckd_size  = sizeof(struct mdt_object)
114         },
115         {
116                 .ckd_cache      = &mdt_hsm_cdt_kmem,
117                 .ckd_name       = "mdt_cdt_restore_handle",
118                 .ckd_size       = sizeof(struct cdt_restore_handle)
119         },
120         {
121                 .ckd_cache      = &mdt_hsm_car_kmem,
122                 .ckd_name       = "mdt_cdt_agent_req",
123                 .ckd_size       = sizeof(struct cdt_agent_req)
124         },
125         {
126                 .ckd_cache = NULL
127         }
128 };
129
130 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
131 {
132         if (!rep)
133                 return 0;
134         return (rep->lock_policy_res1 & flag);
135 }
136
137 void mdt_clear_disposition(struct mdt_thread_info *info,
138                            struct ldlm_reply *rep, int flag)
139 {
140         if (info) {
141                 info->mti_opdata &= ~flag;
142                 tgt_opdata_clear(info->mti_env, flag);
143         }
144         if (rep)
145                 rep->lock_policy_res1 &= ~flag;
146 }
147
148 void mdt_set_disposition(struct mdt_thread_info *info,
149                          struct ldlm_reply *rep, int flag)
150 {
151         if (info) {
152                 info->mti_opdata |= flag;
153                 tgt_opdata_set(info->mti_env, flag);
154         }
155         if (rep)
156                 rep->lock_policy_res1 |= flag;
157 }
158
159 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
160 {
161         lh->mlh_pdo_hash = 0;
162         lh->mlh_reg_mode = lm;
163         lh->mlh_rreg_mode = lm;
164         lh->mlh_type = MDT_REG_LOCK;
165 }
166
167 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
168                        const struct lu_name *lname)
169 {
170         lh->mlh_reg_mode = lock_mode;
171         lh->mlh_rreg_mode = lock_mode;
172         lh->mlh_type = MDT_PDO_LOCK;
173
174         if (lu_name_is_valid(lname)) {
175                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
176                                                   lname->ln_namelen);
177                 /* XXX Workaround for LU-2856
178                  *
179                  * Zero is a valid return value of full_name_hash, but
180                  * several users of mlh_pdo_hash assume a non-zero
181                  * hash value. We therefore map zero onto an
182                  * arbitrary, but consistent value (1) to avoid
183                  * problems further down the road. */
184                 if (unlikely(lh->mlh_pdo_hash == 0))
185                         lh->mlh_pdo_hash = 1;
186         } else {
187                 lh->mlh_pdo_hash = 0;
188         }
189 }
190
191 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
192                               struct mdt_lock_handle *lh)
193 {
194         mdl_mode_t mode;
195         ENTRY;
196
197         /*
198          * Any dir access needs couple of locks:
199          *
200          * 1) on part of dir we gonna take lookup/modify;
201          *
202          * 2) on whole dir to protect it from concurrent splitting and/or to
203          * flush client's cache for readdir().
204          *
205          * so, for a given mode and object this routine decides what lock mode
206          * to use for lock #2:
207          *
208          * 1) if caller's gonna lookup in dir then we need to protect dir from
209          * being splitted only - LCK_CR
210          *
211          * 2) if caller's gonna modify dir then we need to protect dir from
212          * being splitted and to flush cache - LCK_CW
213          *
214          * 3) if caller's gonna modify dir and that dir seems ready for
215          * splitting then we need to protect it from any type of access
216          * (lookup/modify/split) - LCK_EX --bzzz
217          */
218
219         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
220         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
221
222         /*
223          * Ask underlaying level its opinion about preferable PDO lock mode
224          * having access type passed as regular lock mode:
225          *
226          * - MDL_MINMODE means that lower layer does not want to specify lock
227          * mode;
228          *
229          * - MDL_NL means that no PDO lock should be taken. This is used in some
230          * cases. Say, for non-splittable directories no need to use PDO locks
231          * at all.
232          */
233         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
234                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
235
236         if (mode != MDL_MINMODE) {
237                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
238         } else {
239                 /*
240                  * Lower layer does not want to specify locking mode. We do it
241                  * our selves. No special protection is needed, just flush
242                  * client's cache on modification and allow concurrent
243                  * mondification.
244                  */
245                 switch (lh->mlh_reg_mode) {
246                 case LCK_EX:
247                         lh->mlh_pdo_mode = LCK_EX;
248                         break;
249                 case LCK_PR:
250                         lh->mlh_pdo_mode = LCK_CR;
251                         break;
252                 case LCK_PW:
253                         lh->mlh_pdo_mode = LCK_CW;
254                         break;
255                 default:
256                         CERROR("Not expected lock type (0x%x)\n",
257                                (int)lh->mlh_reg_mode);
258                         LBUG();
259                 }
260         }
261
262         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
263         EXIT;
264 }
265
266 static int mdt_getstatus(struct tgt_session_info *tsi)
267 {
268         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
269         struct mdt_device       *mdt = info->mti_mdt;
270         struct mdt_body         *repbody;
271         int                      rc;
272
273         ENTRY;
274
275         rc = mdt_check_ucred(info);
276         if (rc)
277                 GOTO(out, rc = err_serious(rc));
278
279         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
280                 GOTO(out, rc = err_serious(-ENOMEM));
281
282         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
283         repbody->mbo_fid1 = mdt->mdt_md_root_fid;
284         repbody->mbo_valid |= OBD_MD_FLID;
285
286         if (tsi->tsi_tgt->lut_mds_capa &&
287             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
288                 struct mdt_object       *root;
289                 struct lustre_capa      *capa;
290
291                 root = mdt_object_find(info->mti_env, mdt, &repbody->mbo_fid1);
292                 if (IS_ERR(root))
293                         GOTO(out, rc = PTR_ERR(root));
294
295                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
296                 LASSERT(capa);
297                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
298                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
299                                  0);
300                 mdt_object_put(info->mti_env, root);
301                 if (rc == 0)
302                         repbody->mbo_valid |= OBD_MD_FLMDSCAPA;
303         }
304         EXIT;
305 out:
306         mdt_thread_info_fini(info);
307         return rc;
308 }
309
310 static int mdt_statfs(struct tgt_session_info *tsi)
311 {
312         struct ptlrpc_request           *req = tgt_ses_req(tsi);
313         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
314         struct md_device                *next = info->mti_mdt->mdt_child;
315         struct ptlrpc_service_part      *svcpt;
316         struct obd_statfs               *osfs;
317         int                             rc;
318
319         ENTRY;
320
321         svcpt = req->rq_rqbd->rqbd_svcpt;
322
323         /* This will trigger a watchdog timeout */
324         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
325                          (MDT_SERVICE_WATCHDOG_FACTOR *
326                           at_get(&svcpt->scp_at_estimate)) + 1);
327
328         rc = mdt_check_ucred(info);
329         if (rc)
330                 GOTO(out, rc = err_serious(rc));
331
332         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
333                 GOTO(out, rc = err_serious(-ENOMEM));
334
335         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
336         if (!osfs)
337                 GOTO(out, rc = -EPROTO);
338
339         /** statfs information are cached in the mdt_device */
340         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
341                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
342                 /** statfs data is too old, get up-to-date one */
343                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
344                 if (rc)
345                         GOTO(out, rc);
346                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
347                 info->mti_mdt->mdt_osfs = *osfs;
348                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
349                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
350         } else {
351                 /** use cached statfs data */
352                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
353                 *osfs = info->mti_mdt->mdt_osfs;
354                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
355         }
356
357         if (rc == 0)
358                 mdt_counter_incr(req, LPROC_MDT_STATFS);
359 out:
360         mdt_thread_info_fini(info);
361         RETURN(rc);
362 }
363
364 /**
365  * Pack SOM attributes into the reply.
366  * Call under a DLM UPDATE lock.
367  */
368 static void mdt_pack_size2body(struct mdt_thread_info *info,
369                                struct mdt_object *mo)
370 {
371         struct mdt_body *b;
372         struct md_attr *ma = &info->mti_attr;
373
374         LASSERT(ma->ma_attr.la_valid & LA_MODE);
375         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
376
377         /* Check if Size-on-MDS is supported, if this is a regular file,
378          * if SOM is enabled on the object and if SOM cache exists and valid.
379          * Otherwise do not pack Size-on-MDS attributes to the reply. */
380         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
381             !S_ISREG(ma->ma_attr.la_mode) ||
382             !mdt_object_is_som_enabled(mo) ||
383             !(ma->ma_valid & MA_SOM))
384                 return;
385
386         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
387         b->mbo_size = ma->ma_som->msd_size;
388         b->mbo_blocks = ma->ma_som->msd_blocks;
389 }
390
391 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
392                         const struct lu_attr *attr, const struct lu_fid *fid)
393 {
394         struct md_attr *ma = &info->mti_attr;
395
396         LASSERT(ma->ma_valid & MA_INODE);
397
398         b->mbo_atime      = attr->la_atime;
399         b->mbo_mtime      = attr->la_mtime;
400         b->mbo_ctime      = attr->la_ctime;
401         b->mbo_mode       = attr->la_mode;
402         b->mbo_size       = attr->la_size;
403         b->mbo_blocks     = attr->la_blocks;
404         b->mbo_uid      = attr->la_uid;
405         b->mbo_gid      = attr->la_gid;
406         b->mbo_flags      = attr->la_flags;
407         b->mbo_nlink      = attr->la_nlink;
408         b->mbo_rdev       = attr->la_rdev;
409
410         /* XXX should pack the reply body according to lu_valid */
411         b->mbo_valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
412                         OBD_MD_FLGID   | OBD_MD_FLTYPE  |
413                         OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
414                         OBD_MD_FLATIME | OBD_MD_FLMTIME ;
415
416         if (!S_ISREG(attr->la_mode)) {
417                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
418         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
419                 /* means no objects are allocated on osts. */
420                 LASSERT(!(ma->ma_valid & MA_LOV));
421                 /* just ignore blocks occupied by extend attributes on MDS */
422                 b->mbo_blocks = 0;
423                 /* if no object is allocated on osts, the size on mds is valid.
424                  * b=22272 */
425                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
426         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
427                    ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
428                 /* A released file stores its size on MDS. */
429                 /* But return 1 block for released file, unless tools like tar
430                  * will consider it fully sparse. (LU-3864)
431                  */
432                 if (unlikely(b->mbo_size == 0))
433                         b->mbo_blocks = 0;
434                 else
435                         b->mbo_blocks = 1;
436                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
437         }
438
439         if (fid) {
440                 b->mbo_fid1 = *fid;
441                 b->mbo_valid |= OBD_MD_FLID;
442                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
443                                 PFID(fid), b->mbo_nlink,
444                        b->mbo_mode, b->mbo_size);
445         }
446
447         if (info)
448                 mdt_body_reverse_idmap(info, b);
449
450         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
451                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
452                        PFID(fid), (unsigned long long)b->mbo_size);
453 }
454
455 static inline int mdt_body_has_lov(const struct lu_attr *la,
456                                    const struct mdt_body *body)
457 {
458         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
459                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
460 }
461
462 void mdt_client_compatibility(struct mdt_thread_info *info)
463 {
464         struct mdt_body       *body;
465         struct ptlrpc_request *req = mdt_info_req(info);
466         struct obd_export     *exp = req->rq_export;
467         struct md_attr        *ma = &info->mti_attr;
468         struct lu_attr        *la = &ma->ma_attr;
469         ENTRY;
470
471         if (exp_connect_layout(exp))
472                 /* the client can deal with 16-bit lmm_stripe_count */
473                 RETURN_EXIT;
474
475         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
476
477         if (!mdt_body_has_lov(la, body))
478                 RETURN_EXIT;
479
480         /* now we have a reply with a lov for a client not compatible with the
481          * layout lock so we have to clean the layout generation number */
482         if (S_ISREG(la->la_mode))
483                 ma->ma_lmm->lmm_layout_gen = 0;
484         EXIT;
485 }
486
487 int mdt_attr_get_eabuf_size(struct mdt_thread_info *info, struct mdt_object *o)
488 {
489         const struct lu_env *env = info->mti_env;
490         int rc, rc2;
491
492         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
493                           XATTR_NAME_LOV);
494
495         if (rc == -ENODATA)
496                 rc = 0;
497
498         if (rc < 0)
499                 goto out;
500
501         /* Is it a directory? Let's check for the LMV as well */
502         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
503                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
504                                    XATTR_NAME_LMV);
505
506                 if (rc2 == -ENODATA)
507                         rc2 = mo_xattr_get(env, mdt_object_child(o),
508                                            &LU_BUF_NULL,
509                                            XATTR_NAME_DEFAULT_LMV);
510
511                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
512                         rc = rc2;
513         }
514
515 out:
516         return rc;
517 }
518
519 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
520                       const char *name)
521 {
522         const struct lu_env *env = info->mti_env;
523         int rc;
524         ENTRY;
525
526         LASSERT(info->mti_big_lmm_used == 0);
527         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
528         if (rc < 0)
529                 RETURN(rc);
530
531         /* big_lmm may need to be grown */
532         if (info->mti_big_lmmsize < rc) {
533                 int size = size_roundup_power2(rc);
534
535                 if (info->mti_big_lmmsize > 0) {
536                         /* free old buffer */
537                         LASSERT(info->mti_big_lmm);
538                         OBD_FREE_LARGE(info->mti_big_lmm,
539                                        info->mti_big_lmmsize);
540                         info->mti_big_lmm = NULL;
541                         info->mti_big_lmmsize = 0;
542                 }
543
544                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
545                 if (info->mti_big_lmm == NULL)
546                         RETURN(-ENOMEM);
547                 info->mti_big_lmmsize = size;
548         }
549         LASSERT(info->mti_big_lmmsize >= rc);
550
551         info->mti_buf.lb_buf = info->mti_big_lmm;
552         info->mti_buf.lb_len = info->mti_big_lmmsize;
553         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
554
555         RETURN(rc);
556 }
557
558 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
559                    struct md_attr *ma, const char *name)
560 {
561         struct md_object *next = mdt_object_child(o);
562         struct lu_buf    *buf = &info->mti_buf;
563         int rc;
564
565         if (strcmp(name, XATTR_NAME_LOV) == 0) {
566                 buf->lb_buf = ma->ma_lmm;
567                 buf->lb_len = ma->ma_lmm_size;
568                 LASSERT(!(ma->ma_valid & MA_LOV));
569         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
570                 buf->lb_buf = ma->ma_lmv;
571                 buf->lb_len = ma->ma_lmv_size;
572                 LASSERT(!(ma->ma_valid & MA_LMV));
573         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
574                 buf->lb_buf = ma->ma_lmv;
575                 buf->lb_len = ma->ma_lmv_size;
576                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
577         } else {
578                 return -EINVAL;
579         }
580
581         rc = mo_xattr_get(info->mti_env, next, buf, name);
582         if (rc > 0) {
583
584 got:
585                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
586                         if (info->mti_big_lmm_used)
587                                 ma->ma_lmm = info->mti_big_lmm;
588
589                         /* NOT return LOV EA with hole to old client. */
590                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
591                                      LOV_PATTERN_F_HOLE) &&
592                             !(exp_connect_flags(info->mti_exp) &
593                               OBD_CONNECT_LFSCK)) {
594                                 return -EIO;
595                         } else {
596                                 ma->ma_lmm_size = rc;
597                                 ma->ma_valid |= MA_LOV;
598                         }
599                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
600                         if (info->mti_big_lmm_used)
601                                 ma->ma_lmv = info->mti_big_lmm;
602
603                         ma->ma_lmv_size = rc;
604                         ma->ma_valid |= MA_LMV;
605                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
606                         ma->ma_lmv_size = rc;
607                         ma->ma_valid |= MA_LMV_DEF;
608                 }
609
610                 /* Update mdt_max_mdsize so all clients will be aware that */
611                 if (info->mti_mdt->mdt_max_mdsize < rc)
612                         info->mti_mdt->mdt_max_mdsize = rc;
613
614                 rc = 0;
615         } else if (rc == -ENODATA) {
616                 /* no LOV EA */
617                 rc = 0;
618         } else if (rc == -ERANGE) {
619                 /* Default LMV has fixed size, so it must be able to fit
620                  * in the original buffer */
621                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
622                         return rc;
623                 rc = mdt_big_xattr_get(info, o, name);
624                 if (rc > 0) {
625                         info->mti_big_lmm_used = 1;
626                         goto got;
627                 }
628         }
629
630         return rc;
631 }
632
633 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
634                              struct mdt_object *o, struct lu_fid *pfid)
635 {
636         struct lu_buf           *buf = &info->mti_buf;
637         struct link_ea_header   *leh;
638         struct link_ea_entry    *lee;
639         int                      rc;
640         ENTRY;
641
642         buf->lb_buf = info->mti_big_lmm;
643         buf->lb_len = info->mti_big_lmmsize;
644         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
645                           buf, XATTR_NAME_LINK);
646         /* ignore errors, MA_PFID won't be set and it is
647          * up to the caller to treat this as an error */
648         if (rc == -ERANGE || buf->lb_len == 0) {
649                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
650                 buf->lb_buf = info->mti_big_lmm;
651                 buf->lb_len = info->mti_big_lmmsize;
652         }
653
654         if (rc < 0)
655                 RETURN(rc);
656         if (rc < sizeof(*leh)) {
657                 CERROR("short LinkEA on "DFID": rc = %d\n",
658                        PFID(mdt_object_fid(o)), rc);
659                 RETURN(-ENODATA);
660         }
661
662         leh = (struct link_ea_header *) buf->lb_buf;
663         lee = (struct link_ea_entry *)(leh + 1);
664         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
665                 leh->leh_magic = LINK_EA_MAGIC;
666                 leh->leh_reccount = __swab32(leh->leh_reccount);
667                 leh->leh_len = __swab64(leh->leh_len);
668         }
669         if (leh->leh_magic != LINK_EA_MAGIC)
670                 RETURN(-EINVAL);
671         if (leh->leh_reccount == 0)
672                 RETURN(-ENODATA);
673
674         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
675         fid_be_to_cpu(pfid, pfid);
676
677         RETURN(0);
678 }
679
680 int mdt_attr_get_complex(struct mdt_thread_info *info,
681                          struct mdt_object *o, struct md_attr *ma)
682 {
683         const struct lu_env *env = info->mti_env;
684         struct md_object    *next = mdt_object_child(o);
685         struct lu_buf       *buf = &info->mti_buf;
686         int                  need = ma->ma_need;
687         int                  rc = 0, rc2;
688         u32                  mode;
689         ENTRY;
690
691         ma->ma_valid = 0;
692
693         if (mdt_object_exists(o) == 0)
694                 GOTO(out, rc = -ENOENT);
695         mode = lu_object_attr(&next->mo_lu);
696
697         if (need & MA_INODE) {
698                 ma->ma_need = MA_INODE;
699                 rc = mo_attr_get(env, next, ma);
700                 if (rc)
701                         GOTO(out, rc);
702                 ma->ma_valid |= MA_INODE;
703         }
704
705         if (need & MA_PFID) {
706                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
707                 if (rc == 0)
708                         ma->ma_valid |= MA_PFID;
709                 /* ignore this error, parent fid is not mandatory */
710                 rc = 0;
711         }
712
713         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
714                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
715                 if (rc)
716                         GOTO(out, rc);
717         }
718
719         if (need & MA_LMV && S_ISDIR(mode)) {
720                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
721                 if (rc != 0)
722                         GOTO(out, rc);
723         }
724
725         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
726                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
727                 if (rc != 0)
728                         GOTO(out, rc);
729         }
730
731         if (need & MA_SOM && S_ISREG(mode)) {
732                 buf->lb_buf = info->mti_xattr_buf;
733                 buf->lb_len = sizeof(info->mti_xattr_buf);
734                 CLASSERT(sizeof(struct som_attrs) <=
735                          sizeof(info->mti_xattr_buf));
736                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
737                 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
738                 if (rc2 == 0)
739                         ma->ma_valid |= MA_SOM;
740                 else if (rc2 < 0 && rc2 != -ENODATA)
741                         GOTO(out, rc = rc2);
742         }
743
744         if (need & MA_HSM && S_ISREG(mode)) {
745                 buf->lb_buf = info->mti_xattr_buf;
746                 buf->lb_len = sizeof(info->mti_xattr_buf);
747                 CLASSERT(sizeof(struct hsm_attrs) <=
748                          sizeof(info->mti_xattr_buf));
749                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
750                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
751                 if (rc2 == 0)
752                         ma->ma_valid |= MA_HSM;
753                 else if (rc2 < 0 && rc2 != -ENODATA)
754                         GOTO(out, rc = rc2);
755         }
756
757 #ifdef CONFIG_FS_POSIX_ACL
758         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
759                 buf->lb_buf = ma->ma_acl;
760                 buf->lb_len = ma->ma_acl_size;
761                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
762                 if (rc2 > 0) {
763                         ma->ma_acl_size = rc2;
764                         ma->ma_valid |= MA_ACL_DEF;
765                 } else if (rc2 == -ENODATA) {
766                         /* no ACLs */
767                         ma->ma_acl_size = 0;
768                 } else
769                         GOTO(out, rc = rc2);
770         }
771 #endif
772 out:
773         ma->ma_need = need;
774         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
775                rc, ma->ma_valid, ma->ma_lmm);
776         RETURN(rc);
777 }
778
779 static int mdt_getattr_internal(struct mdt_thread_info *info,
780                                 struct mdt_object *o, int ma_need)
781 {
782         struct md_object        *next = mdt_object_child(o);
783         const struct mdt_body   *reqbody = info->mti_body;
784         struct ptlrpc_request   *req = mdt_info_req(info);
785         struct md_attr          *ma = &info->mti_attr;
786         struct lu_attr          *la = &ma->ma_attr;
787         struct req_capsule      *pill = info->mti_pill;
788         const struct lu_env     *env = info->mti_env;
789         struct mdt_body         *repbody;
790         struct lu_buf           *buffer = &info->mti_buf;
791         int                     rc;
792         int                     is_root;
793         ENTRY;
794
795         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
796                 RETURN(err_serious(-ENOMEM));
797
798         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
799
800         ma->ma_valid = 0;
801
802         if (mdt_object_remote(o)) {
803                 /* This object is located on remote node.*/
804                 /* Return -ENOTSUPP for old client */
805                 if (!mdt_is_dne_client(req->rq_export))
806                         GOTO(out, rc = -ENOTSUPP);
807
808                 repbody->mbo_fid1 = *mdt_object_fid(o);
809                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
810                 GOTO(out, rc = 0);
811         }
812
813         if (reqbody->mbo_eadatasize > 0) {
814                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
815                 if (buffer->lb_buf == NULL)
816                         GOTO(out, rc = -EPROTO);
817                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
818                                                       RCL_SERVER);
819         } else {
820                 buffer->lb_buf = NULL;
821                 buffer->lb_len = 0;
822                 ma_need &= ~(MA_LOV | MA_LMV);
823                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
824                        mdt_obd_name(info->mti_mdt),
825                        req->rq_export->exp_client_uuid.uuid);
826         }
827
828         /* If it is dir object and client require MEA, then we got MEA */
829         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
830             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
831                 /* Assumption: MDT_MD size is enough for lmv size. */
832                 ma->ma_lmv = buffer->lb_buf;
833                 ma->ma_lmv_size = buffer->lb_len;
834                 ma->ma_need = MA_INODE;
835                 if (ma->ma_lmv_size > 0) {
836                         if (reqbody->mbo_valid & OBD_MD_MEA)
837                                 ma->ma_need |= MA_LMV;
838                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
839                                 ma->ma_need |= MA_LMV_DEF;
840                 }
841         } else {
842                 ma->ma_lmm = buffer->lb_buf;
843                 ma->ma_lmm_size = buffer->lb_len;
844                 ma->ma_need = MA_INODE | MA_HSM;
845                 if (ma->ma_lmm_size > 0)
846                         ma->ma_need |= MA_LOV;
847         }
848
849         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
850             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
851             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
852                 /* get default stripe info for this dir. */
853                 ma->ma_need |= MA_LOV_DEF;
854         }
855         ma->ma_need |= ma_need;
856         if (ma->ma_need & MA_SOM)
857                 ma->ma_som = &info->mti_u.som.data;
858
859         rc = mdt_attr_get_complex(info, o, ma);
860         if (unlikely(rc)) {
861                 CERROR("%s: getattr error for "DFID": rc = %d\n",
862                        mdt_obd_name(info->mti_mdt),
863                        PFID(mdt_object_fid(o)), rc);
864                 RETURN(rc);
865         }
866
867         /* if file is released, check if a restore is running */
868         if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
869             mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
870                 repbody->mbo_t_state = MS_RESTORE;
871                 repbody->mbo_valid |= OBD_MD_TSTATE;
872         }
873
874         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
875
876         /* the Lustre protocol supposes to return default striping
877          * on the user-visible root if explicitly requested */
878         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
879             (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
880                 struct lu_fid      rootfid;
881                 struct mdt_object *root;
882                 struct mdt_device *mdt = info->mti_mdt;
883
884                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
885                 if (rc)
886                         RETURN(rc);
887                 root = mdt_object_find(env, mdt, &rootfid);
888                 if (IS_ERR(root))
889                         RETURN(PTR_ERR(root));
890                 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
891                 mdt_object_put(info->mti_env, root);
892                 if (unlikely(rc)) {
893                         CERROR("%s: getattr error for "DFID": rc = %d\n",
894                                mdt_obd_name(info->mti_mdt),
895                                PFID(mdt_object_fid(o)), rc);
896                         RETURN(rc);
897                 }
898         }
899
900         if (likely(ma->ma_valid & MA_INODE))
901                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
902         else
903                 RETURN(-EFAULT);
904
905         if (mdt_body_has_lov(la, reqbody)) {
906                 if (ma->ma_valid & MA_LOV) {
907                         LASSERT(ma->ma_lmm_size);
908                         repbody->mbo_eadatasize = ma->ma_lmm_size;
909                         if (S_ISDIR(la->la_mode))
910                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
911                         else
912                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
913                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
914                 }
915                 if (ma->ma_valid & MA_LMV) {
916                         /* Return -ENOTSUPP for old client */
917                         if (!mdt_is_striped_client(req->rq_export))
918                                 RETURN(-ENOTSUPP);
919
920                         LASSERT(S_ISDIR(la->la_mode));
921                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
922                         repbody->mbo_eadatasize = ma->ma_lmv_size;
923                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
924                 }
925                 if (ma->ma_valid & MA_LMV_DEF) {
926                         /* Return -ENOTSUPP for old client */
927                         if (!mdt_is_striped_client(req->rq_export))
928                                 RETURN(-ENOTSUPP);
929                         LASSERT(S_ISDIR(la->la_mode));
930                         repbody->mbo_eadatasize = ma->ma_lmv_size;
931                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
932                                                OBD_MD_DEFAULT_MEA);
933                 }
934         } else if (S_ISLNK(la->la_mode) &&
935                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
936                 buffer->lb_buf = ma->ma_lmm;
937                 /* eadatasize from client includes NULL-terminator, so
938                  * there is no need to read it */
939                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
940                 rc = mo_readlink(env, next, buffer);
941                 if (unlikely(rc <= 0)) {
942                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
943                                mdt_obd_name(info->mti_mdt),
944                                PFID(mdt_object_fid(o)), rc);
945                         rc = -EFAULT;
946                 } else {
947                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
948
949                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
950                                 rc -= 2;
951                         repbody->mbo_valid |= OBD_MD_LINKNAME;
952                         /* we need to report back size with NULL-terminator
953                          * because client expects that */
954                         repbody->mbo_eadatasize = rc + 1;
955                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
956                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
957                                        "on "DFID ", expected %d\n",
958                                        mdt_obd_name(info->mti_mdt),
959                                        rc, PFID(mdt_object_fid(o)),
960                                        reqbody->mbo_eadatasize - 1);
961                         /* NULL terminate */
962                         ((char *)ma->ma_lmm)[rc] = 0;
963
964                         /* If the total CDEBUG() size is larger than a page, it
965                          * will print a warning to the console, avoid this by
966                          * printing just the last part of the symlink. */
967                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
968                                print_limit < rc ? "..." : "", print_limit,
969                                (char *)ma->ma_lmm + rc - print_limit, rc);
970                         rc = 0;
971                 }
972         }
973
974         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
975                 repbody->mbo_max_cookiesize = 0;
976                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
977                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
978                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
979                        "MAX_COOKIE to : %d:%d\n", repbody->mbo_max_mdsize,
980                        repbody->mbo_max_cookiesize);
981         }
982
983         if (exp_connect_rmtclient(info->mti_exp) &&
984             reqbody->mbo_valid & OBD_MD_FLRMTPERM) {
985                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
986
987                 /* mdt_getattr_lock only */
988                 rc = mdt_pack_remote_perm(info, o, buf);
989                 if (rc) {
990                         repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
991                         repbody->mbo_aclsize = 0;
992                         RETURN(rc);
993                 } else {
994                         repbody->mbo_valid |= OBD_MD_FLRMTPERM;
995                         repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
996                 }
997         }
998 #ifdef CONFIG_FS_POSIX_ACL
999         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1000                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1001                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
1002                 buffer->lb_len = req_capsule_get_size(pill,
1003                                                       &RMF_ACL, RCL_SERVER);
1004                 if (buffer->lb_len > 0) {
1005                         rc = mo_xattr_get(env, next, buffer,
1006                                           XATTR_NAME_ACL_ACCESS);
1007                         if (rc < 0) {
1008                                 if (rc == -ENODATA) {
1009                                         repbody->mbo_aclsize = 0;
1010                                         repbody->mbo_valid |= OBD_MD_FLACL;
1011                                         rc = 0;
1012                                 } else if (rc == -EOPNOTSUPP) {
1013                                         rc = 0;
1014                                 } else {
1015                                         CERROR("%s: unable to read "DFID
1016                                                " ACL: rc = %d\n",
1017                                                mdt_obd_name(info->mti_mdt),
1018                                                PFID(mdt_object_fid(o)), rc);
1019                                 }
1020                         } else {
1021                                 repbody->mbo_aclsize = rc;
1022                                 repbody->mbo_valid |= OBD_MD_FLACL;
1023                                 rc = 0;
1024                         }
1025                 }
1026         }
1027 #endif
1028
1029         if (reqbody->mbo_valid & OBD_MD_FLMDSCAPA &&
1030             info->mti_mdt->mdt_lut.lut_mds_capa &&
1031             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
1032                 struct lustre_capa *capa;
1033
1034                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
1035                 LASSERT(capa);
1036                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
1037                 rc = mo_capa_get(env, next, capa, 0);
1038                 if (rc)
1039                         RETURN(rc);
1040                 repbody->mbo_valid |= OBD_MD_FLMDSCAPA;
1041         }
1042
1043 out:
1044         if (rc == 0)
1045                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1046
1047         RETURN(rc);
1048 }
1049
1050 static int mdt_renew_capa(struct mdt_thread_info *info)
1051 {
1052         struct mdt_object  *obj = info->mti_object;
1053         struct mdt_body    *body;
1054         struct lustre_capa *capa, *c;
1055         int rc;
1056         ENTRY;
1057
1058         /* if object doesn't exist, or server has disabled capability,
1059          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
1060          * flag not set.
1061          */
1062         if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa ||
1063             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
1064                 RETURN(0);
1065
1066         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1067         LASSERT(body != NULL);
1068
1069         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
1070         LASSERT(c);
1071
1072         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
1073         LASSERT(capa);
1074
1075         *capa = *c;
1076         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
1077         if (rc == 0)
1078                 body->mbo_valid |= OBD_MD_FLOSSCAPA;
1079         RETURN(rc);
1080 }
1081
1082 static int mdt_getattr(struct tgt_session_info *tsi)
1083 {
1084         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1085         struct mdt_object       *obj = info->mti_object;
1086         struct req_capsule      *pill = info->mti_pill;
1087         struct mdt_body         *reqbody;
1088         struct mdt_body         *repbody;
1089         int rc, rc2;
1090         ENTRY;
1091
1092         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1093         LASSERT(reqbody);
1094
1095         if (reqbody->mbo_valid & OBD_MD_FLOSSCAPA) {
1096                 rc = req_capsule_server_pack(pill);
1097                 if (unlikely(rc))
1098                         RETURN(err_serious(rc));
1099                 rc = mdt_renew_capa(info);
1100                 GOTO(out_shrink, rc);
1101         }
1102
1103         LASSERT(obj != NULL);
1104         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1105
1106         /* Unlike intent case where we need to pre-fill out buffers early on
1107          * in intent policy for ldlm reasons, here we can have a much better
1108          * guess at EA size by just reading it from disk.
1109          * Exceptions are readdir and (missing) directory striping */
1110         /* Readlink */
1111         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1112                 /* No easy way to know how long is the symlink, but it cannot
1113                  * be more than PATH_MAX, so we allocate +1 */
1114                 rc = PATH_MAX + 1;
1115
1116         /* A special case for fs ROOT: getattr there might fetch
1117          * default EA for entire fs, not just for this dir!
1118          */
1119         } else if (lu_fid_eq(mdt_object_fid(obj),
1120                              &info->mti_mdt->mdt_md_root_fid) &&
1121                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1122                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1123                                                                  MDS_GETATTR)) {
1124                 /* Should the default strping be bigger, mdt_fix_reply
1125                  * will reallocate */
1126                 rc = DEF_REP_MD_SIZE;
1127         } else {
1128                 /* Read the actual EA size from disk */
1129                 rc = mdt_attr_get_eabuf_size(info, obj);
1130         }
1131
1132         if (rc < 0)
1133                 GOTO(out, rc = err_serious(rc));
1134
1135         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1136
1137         rc = req_capsule_server_pack(pill);
1138         if (unlikely(rc != 0))
1139                 GOTO(out, rc = err_serious(rc));
1140
1141         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1142         LASSERT(repbody != NULL);
1143         repbody->mbo_eadatasize = 0;
1144         repbody->mbo_aclsize = 0;
1145
1146         if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1147                 rc = mdt_init_ucred(info, reqbody);
1148         else
1149                 rc = mdt_check_ucred(info);
1150         if (unlikely(rc))
1151                 GOTO(out_shrink, rc);
1152
1153         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1154
1155         /*
1156          * Don't check capability at all, because rename might getattr for
1157          * remote obj, and at that time no capability is available.
1158          */
1159         mdt_set_capainfo(info, 1, &reqbody->mbo_fid1, BYPASS_CAPA);
1160         rc = mdt_getattr_internal(info, obj, 0);
1161         if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1162                 mdt_exit_ucred(info);
1163         EXIT;
1164 out_shrink:
1165         mdt_client_compatibility(info);
1166         rc2 = mdt_fix_reply(info);
1167         if (rc == 0)
1168                 rc = rc2;
1169 out:
1170         mdt_thread_info_fini(info);
1171         return rc;
1172 }
1173
1174 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1175 {
1176         struct mdt_thread_info  *info;
1177         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1178         struct obd_export       *exp = req->rq_export;
1179         struct mdt_object       *o1, *o2, *o;
1180         struct mdt_lock_handle  *lh1, *lh2;
1181         struct mdc_swap_layouts *msl;
1182         int                      rc;
1183         ENTRY;
1184
1185         /* client does not support layout lock, so layout swaping
1186          * is disabled.
1187          * FIXME: there is a problem for old clients which don't support
1188          * layout lock yet. If those clients have already opened the file
1189          * they won't be notified at all so that old layout may still be
1190          * used to do IO. This can be fixed after file release is landed by
1191          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1192         if (!exp_connect_layout(exp))
1193                 RETURN(-EOPNOTSUPP);
1194
1195         info = tsi2mdt_info(tsi);
1196
1197         if (info->mti_dlm_req != NULL)
1198                 ldlm_request_cancel(req, info->mti_dlm_req, 0);
1199
1200         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1201                 mdt_set_capainfo(info, 0, &info->mti_body->mbo_fid1,
1202                                  req_capsule_client_get(info->mti_pill,
1203                                                         &RMF_CAPA1));
1204
1205         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1206                 mdt_set_capainfo(info, 1, &info->mti_body->mbo_fid2,
1207                                  req_capsule_client_get(info->mti_pill,
1208                                                         &RMF_CAPA2));
1209
1210         o1 = info->mti_object;
1211         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1212                                 &info->mti_body->mbo_fid2);
1213         if (IS_ERR(o))
1214                 GOTO(out, rc = PTR_ERR(o));
1215
1216         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1217                 GOTO(put, rc = -ENOENT);
1218
1219         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1220         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1221                 GOTO(put, rc);
1222
1223         if (rc < 0)
1224                 swap(o1, o2);
1225
1226         /* permission check. Make sure the calling process having permission
1227          * to write both files. */
1228         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1229                                 MAY_WRITE);
1230         if (rc < 0)
1231                 GOTO(put, rc);
1232
1233         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1234                                 MAY_WRITE);
1235         if (rc < 0)
1236                 GOTO(put, rc);
1237
1238         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1239         if (msl == NULL)
1240                 GOTO(put, rc = -EPROTO);
1241
1242         lh1 = &info->mti_lh[MDT_LH_NEW];
1243         mdt_lock_reg_init(lh1, LCK_EX);
1244         lh2 = &info->mti_lh[MDT_LH_OLD];
1245         mdt_lock_reg_init(lh2, LCK_EX);
1246
1247         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1248                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1249         if (rc < 0)
1250                 GOTO(put, rc);
1251
1252         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1253                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1254         if (rc < 0)
1255                 GOTO(unlock1, rc);
1256
1257         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1258                              mdt_object_child(o2), msl->msl_flags);
1259         GOTO(unlock2, rc);
1260 unlock2:
1261         mdt_object_unlock(info, o2, lh2, rc);
1262 unlock1:
1263         mdt_object_unlock(info, o1, lh1, rc);
1264 put:
1265         mdt_object_put(info->mti_env, o);
1266 out:
1267         mdt_thread_info_fini(info);
1268         RETURN(rc);
1269 }
1270
1271 static int mdt_raw_lookup(struct mdt_thread_info *info,
1272                           struct mdt_object *parent,
1273                           const struct lu_name *lname,
1274                           struct ldlm_reply *ldlm_rep)
1275 {
1276         struct md_object *next = mdt_object_child(info->mti_object);
1277         const struct mdt_body *reqbody = info->mti_body;
1278         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1279         struct mdt_body *repbody;
1280         int rc;
1281         ENTRY;
1282
1283         if (reqbody->mbo_valid != OBD_MD_FLID)
1284                 RETURN(0);
1285
1286         LASSERT(!info->mti_cross_ref);
1287
1288         /* Only got the fid of this obj by name */
1289         fid_zero(child_fid);
1290         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1291                         &info->mti_spec);
1292 #if 0
1293         /* XXX is raw_lookup possible as intent operation? */
1294         if (rc != 0) {
1295                 if (rc == -ENOENT)
1296                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1297                 RETURN(rc);
1298         } else
1299                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1300
1301         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1302 #endif
1303         if (rc == 0) {
1304                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1305                 repbody->mbo_fid1 = *child_fid;
1306                 repbody->mbo_valid = OBD_MD_FLID;
1307         }
1308         RETURN(1);
1309 }
1310
1311 /*
1312  * UPDATE lock should be taken against parent, and be release before exit;
1313  * child_bits lock should be taken against child, and be returned back:
1314  *            (1)normal request should release the child lock;
1315  *            (2)intent request will grant the lock to client.
1316  */
1317 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1318                                  struct mdt_lock_handle *lhc,
1319                                  __u64 child_bits,
1320                                  struct ldlm_reply *ldlm_rep)
1321 {
1322         struct ptlrpc_request  *req       = mdt_info_req(info);
1323         struct mdt_body        *reqbody   = NULL;
1324         struct mdt_object      *parent    = info->mti_object;
1325         struct mdt_object      *child;
1326         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1327         struct lu_name         *lname     = NULL;
1328         struct mdt_lock_handle *lhp       = NULL;
1329         struct ldlm_lock       *lock;
1330         bool                    is_resent;
1331         bool                    try_layout;
1332         int                     ma_need = 0;
1333         int                     rc;
1334         ENTRY;
1335
1336         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1337         LASSERT(ergo(is_resent,
1338                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1339
1340         if (parent == NULL)
1341                 RETURN(-ENOENT);
1342
1343         if (info->mti_cross_ref) {
1344                 /* Only getattr on the child. Parent is on another node. */
1345                 mdt_set_disposition(info, ldlm_rep,
1346                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1347                 child = parent;
1348                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1349                        "ldlm_rep = %p\n",
1350                        PFID(mdt_object_fid(child)), ldlm_rep);
1351
1352                 rc = mdt_check_resent_lock(info, child, lhc);
1353                 if (rc < 0) {
1354                         RETURN(-EPROTO);
1355                 } else if (rc > 0) {
1356                         mdt_lock_handle_init(lhc);
1357                         mdt_lock_reg_init(lhc, LCK_PR);
1358
1359                         /*
1360                          * Object's name is on another MDS, no lookup or layout
1361                          * lock is needed here but update lock is.
1362                          */
1363                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1364                                         MDS_INODELOCK_LAYOUT);
1365                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1366
1367                         rc = mdt_object_lock(info, child, lhc, child_bits,
1368                                              MDT_LOCAL_LOCK);
1369                         if (rc < 0)
1370                                 RETURN(rc);
1371                 }
1372
1373                 /* Finally, we can get attr for child. */
1374                 if (!mdt_object_exists(child)) {
1375                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1376                                         &child->mot_obj,
1377                                         "remote object doesn't exist.\n");
1378                         mdt_object_unlock(info, child, lhc, 1);
1379                         RETURN(-ENOENT);
1380                 }
1381
1382                 mdt_set_capainfo(info, 0, mdt_object_fid(child), BYPASS_CAPA);
1383                 rc = mdt_getattr_internal(info, child, 0);
1384                 if (unlikely(rc != 0))
1385                         mdt_object_unlock(info, child, lhc, 1);
1386
1387                 RETURN(rc);
1388         }
1389
1390         lname = &info->mti_name;
1391         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1392
1393         if (lu_name_is_valid(lname)) {
1394                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1395                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1396                        PNAME(lname), ldlm_rep);
1397         } else {
1398                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1399                 if (unlikely(reqbody == NULL))
1400                         RETURN(err_serious(-EPROTO));
1401
1402                 *child_fid = reqbody->mbo_fid2;
1403
1404                 if (unlikely(!fid_is_sane(child_fid)))
1405                         RETURN(err_serious(-EINVAL));
1406
1407                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1408                        "ldlm_rep = %p\n",
1409                        PFID(mdt_object_fid(parent)),
1410                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1411         }
1412
1413         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1414
1415         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1416                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1417                                 &parent->mot_obj,
1418                                 "Parent doesn't exist!\n");
1419                 RETURN(-ESTALE);
1420         }
1421
1422         if (mdt_object_remote(parent)) {
1423                 CERROR("%s: parent "DFID" is on remote target\n",
1424                        mdt_obd_name(info->mti_mdt),
1425                        PFID(mdt_object_fid(parent)));
1426                 RETURN(-EIO);
1427         }
1428
1429         if (lu_name_is_valid(lname)) {
1430                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1431                 if (rc != 0) {
1432                         if (rc > 0)
1433                                 rc = 0;
1434                         RETURN(rc);
1435                 }
1436
1437                 /* step 1: lock parent only if parent is a directory */
1438                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1439                         lhp = &info->mti_lh[MDT_LH_PARENT];
1440                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1441                         rc = mdt_object_lock(info, parent, lhp,
1442                                              MDS_INODELOCK_UPDATE,
1443                                              MDT_LOCAL_LOCK);
1444                         if (unlikely(rc != 0))
1445                                 RETURN(rc);
1446                 }
1447
1448                 /* step 2: lookup child's fid by name */
1449                 fid_zero(child_fid);
1450                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1451                                 child_fid, &info->mti_spec);
1452                 if (rc == -ENOENT)
1453                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1454
1455                 if (rc != 0)
1456                         GOTO(out_parent, rc);
1457         }
1458
1459         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1460
1461         /*
1462          *step 3: find the child object by fid & lock it.
1463          *        regardless if it is local or remote.
1464          *
1465          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1466          *      set parent dir fid the same as child fid in getattr by fid case
1467          *      we should not lu_object_find() the object again, could lead
1468          *      to hung if there is a concurrent unlink destroyed the object.
1469          */
1470         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1471                 mdt_object_get(info->mti_env, parent);
1472                 child = parent;
1473         } else {
1474                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1475                                         child_fid);
1476         }
1477
1478         if (unlikely(IS_ERR(child)))
1479                 GOTO(out_parent, rc = PTR_ERR(child));
1480
1481         rc = mdt_check_resent_lock(info, child, lhc);
1482         if (rc < 0) {
1483                 GOTO(out_child, rc);
1484         } else if (rc > 0) {
1485                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1486                 mdt_lock_handle_init(lhc);
1487                 mdt_lock_reg_init(lhc, LCK_PR);
1488                 try_layout = false;
1489
1490                 if (!mdt_object_exists(child)) {
1491                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1492                                         &child->mot_obj,
1493                                         "Object doesn't exist!\n");
1494                         GOTO(out_child, rc = -ENOENT);
1495                 }
1496
1497                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1498                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1499                         struct md_attr *ma = &info->mti_attr;
1500
1501                         ma->ma_valid = 0;
1502                         ma->ma_need = MA_INODE;
1503                         rc = mdt_attr_get_complex(info, child, ma);
1504                         if (unlikely(rc != 0))
1505                                 GOTO(out_child, rc);
1506
1507                         /* If the file has not been changed for some time, we
1508                          * return not only a LOOKUP lock, but also an UPDATE
1509                          * lock and this might save us RPC on later STAT. For
1510                          * directories, it also let negative dentry cache start
1511                          * working for this dir. */
1512                         if (ma->ma_valid & MA_INODE &&
1513                             ma->ma_attr.la_valid & LA_CTIME &&
1514                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1515                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1516                                 child_bits |= MDS_INODELOCK_UPDATE;
1517                 }
1518
1519                 /* layout lock must be granted in a best-effort way
1520                  * for IT operations */
1521                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1522                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1523                     exp_connect_layout(info->mti_exp) &&
1524                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1525                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1526                         /* try to grant layout lock for regular file. */
1527                         try_layout = true;
1528                 }
1529
1530                 rc = 0;
1531                 if (try_layout) {
1532                         child_bits |= MDS_INODELOCK_LAYOUT;
1533                         /* try layout lock, it may fail to be granted due to
1534                          * contention at LOOKUP or UPDATE */
1535                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1536                                                  MDT_CROSS_LOCK)) {
1537                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1538                                 LASSERT(child_bits != 0);
1539                                 rc = mdt_object_lock(info, child, lhc,
1540                                                 child_bits, MDT_CROSS_LOCK);
1541                         } else {
1542                                 ma_need |= MA_LOV;
1543                         }
1544                 } else {
1545                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1546                          * client will enqueue the lock to the remote MDT */
1547                         if (mdt_object_remote(child))
1548                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1549                         rc = mdt_object_lock(info, child, lhc, child_bits,
1550                                                 MDT_CROSS_LOCK);
1551                 }
1552                 if (unlikely(rc != 0))
1553                         GOTO(out_child, rc);
1554         }
1555
1556         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1557         /* Get MA_SOM attributes if update lock is given. */
1558         if (lock &&
1559             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1560             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1561                 ma_need |= MA_SOM;
1562
1563         /* finally, we can get attr for child. */
1564         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1565         rc = mdt_getattr_internal(info, child, ma_need);
1566         if (unlikely(rc != 0)) {
1567                 mdt_object_unlock(info, child, lhc, 1);
1568         } else if (lock) {
1569                 /* Debugging code. */
1570                 LDLM_DEBUG(lock, "Returning lock to client");
1571                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1572                                          &lock->l_resource->lr_name),
1573                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1574                          PLDLMRES(lock->l_resource),
1575                          PFID(mdt_object_fid(child)));
1576                 if (mdt_object_exists(child) && !mdt_object_remote(child))
1577                         mdt_pack_size2body(info, child);
1578         }
1579         if (lock)
1580                 LDLM_LOCK_PUT(lock);
1581
1582         EXIT;
1583 out_child:
1584         mdt_object_put(info->mti_env, child);
1585 out_parent:
1586         if (lhp)
1587                 mdt_object_unlock(info, parent, lhp, 1);
1588         return rc;
1589 }
1590
1591 /* normal handler: should release the child lock */
1592 static int mdt_getattr_name(struct tgt_session_info *tsi)
1593 {
1594         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1595         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1596         struct mdt_body        *reqbody;
1597         struct mdt_body        *repbody;
1598         int rc, rc2;
1599         ENTRY;
1600
1601         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1602         LASSERT(reqbody != NULL);
1603         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1604         LASSERT(repbody != NULL);
1605
1606         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1607         repbody->mbo_eadatasize = 0;
1608         repbody->mbo_aclsize = 0;
1609
1610         rc = mdt_init_ucred(info, reqbody);
1611         if (unlikely(rc))
1612                 GOTO(out_shrink, rc);
1613
1614         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1615         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1616                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1617                 lhc->mlh_reg_lh.cookie = 0;
1618         }
1619         mdt_exit_ucred(info);
1620         EXIT;
1621 out_shrink:
1622         mdt_client_compatibility(info);
1623         rc2 = mdt_fix_reply(info);
1624         if (rc == 0)
1625                 rc = rc2;
1626         mdt_thread_info_fini(info);
1627         return rc;
1628 }
1629
1630 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1631                          void *karg, void *uarg);
1632
1633 static int mdt_set_info(struct tgt_session_info *tsi)
1634 {
1635         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1636         char                    *key;
1637         void                    *val;
1638         int                      keylen, vallen, rc = 0;
1639
1640         ENTRY;
1641
1642         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1643         if (key == NULL) {
1644                 DEBUG_REQ(D_HA, req, "no set_info key");
1645                 RETURN(err_serious(-EFAULT));
1646         }
1647
1648         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1649                                       RCL_CLIENT);
1650
1651         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1652         if (val == NULL) {
1653                 DEBUG_REQ(D_HA, req, "no set_info val");
1654                 RETURN(err_serious(-EFAULT));
1655         }
1656
1657         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1658                                       RCL_CLIENT);
1659
1660         /* Swab any part of val you need to here */
1661         if (KEY_IS(KEY_READ_ONLY)) {
1662                 spin_lock(&req->rq_export->exp_lock);
1663                 if (*(__u32 *)val)
1664                         *exp_connect_flags_ptr(req->rq_export) |=
1665                                 OBD_CONNECT_RDONLY;
1666                 else
1667                         *exp_connect_flags_ptr(req->rq_export) &=
1668                                 ~OBD_CONNECT_RDONLY;
1669                 spin_unlock(&req->rq_export->exp_lock);
1670         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1671                 struct changelog_setinfo *cs = val;
1672
1673                 if (vallen != sizeof(*cs)) {
1674                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1675                                tgt_name(tsi->tsi_tgt), vallen);
1676                         RETURN(-EINVAL);
1677                 }
1678                 if (ptlrpc_req_need_swab(req)) {
1679                         __swab64s(&cs->cs_recno);
1680                         __swab32s(&cs->cs_id);
1681                 }
1682
1683                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1684                                    vallen, val, NULL);
1685         } else {
1686                 RETURN(-EINVAL);
1687         }
1688         RETURN(rc);
1689 }
1690
1691 static int mdt_readpage(struct tgt_session_info *tsi)
1692 {
1693         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1694         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1695         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1696         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1697         struct mdt_body         *repbody;
1698         int                      rc;
1699         int                      i;
1700
1701         ENTRY;
1702
1703         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1704                 RETURN(err_serious(-ENOMEM));
1705
1706         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1707         if (repbody == NULL || reqbody == NULL)
1708                 RETURN(err_serious(-EFAULT));
1709
1710         /*
1711          * prepare @rdpg before calling lower layers and transfer itself. Here
1712          * reqbody->size contains offset of where to start to read and
1713          * reqbody->nlink contains number bytes to read.
1714          */
1715         rdpg->rp_hash = reqbody->mbo_size;
1716         if (rdpg->rp_hash != reqbody->mbo_size) {
1717                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1718                        rdpg->rp_hash, reqbody->mbo_size);
1719                 RETURN(-EFAULT);
1720         }
1721
1722         rdpg->rp_attrs = reqbody->mbo_mode;
1723         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1724                 rdpg->rp_attrs |= LUDA_64BITHASH;
1725         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1726                                 exp_max_brw_size(tsi->tsi_exp));
1727         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1728                           PAGE_CACHE_SHIFT;
1729         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1730         if (rdpg->rp_pages == NULL)
1731                 RETURN(-ENOMEM);
1732
1733         for (i = 0; i < rdpg->rp_npages; ++i) {
1734                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1735                 if (rdpg->rp_pages[i] == NULL)
1736                         GOTO(free_rdpg, rc = -ENOMEM);
1737         }
1738
1739         /* call lower layers to fill allocated pages with directory data */
1740         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1741         if (rc < 0)
1742                 GOTO(free_rdpg, rc);
1743
1744         /* send pages to client */
1745         rc = tgt_sendpage(tsi, rdpg, rc);
1746
1747         EXIT;
1748 free_rdpg:
1749
1750         for (i = 0; i < rdpg->rp_npages; i++)
1751                 if (rdpg->rp_pages[i] != NULL)
1752                         __free_page(rdpg->rp_pages[i]);
1753         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1754
1755         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1756                 RETURN(0);
1757
1758         return rc;
1759 }
1760
1761 static int mdt_reint_internal(struct mdt_thread_info *info,
1762                               struct mdt_lock_handle *lhc,
1763                               __u32 op)
1764 {
1765         struct req_capsule      *pill = info->mti_pill;
1766         struct mdt_body         *repbody;
1767         int                      rc = 0, rc2;
1768
1769         ENTRY;
1770
1771         rc = mdt_reint_unpack(info, op);
1772         if (rc != 0) {
1773                 CERROR("Can't unpack reint, rc %d\n", rc);
1774                 RETURN(err_serious(rc));
1775         }
1776
1777         /* for replay (no_create) lmm is not needed, client has it already */
1778         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1779                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1780                                      DEF_REP_MD_SIZE);
1781
1782         /* llog cookies are always 0, the field is kept for compatibility */
1783         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1784                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1785
1786         rc = req_capsule_server_pack(pill);
1787         if (rc != 0) {
1788                 CERROR("Can't pack response, rc %d\n", rc);
1789                 RETURN(err_serious(rc));
1790         }
1791
1792         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1793                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1794                 LASSERT(repbody);
1795                 repbody->mbo_eadatasize = 0;
1796                 repbody->mbo_aclsize = 0;
1797         }
1798
1799         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1800
1801         /* for replay no cookkie / lmm need, because client have this already */
1802         if (info->mti_spec.no_create)
1803                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1804                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1805
1806         rc = mdt_init_ucred_reint(info);
1807         if (rc)
1808                 GOTO(out_shrink, rc);
1809
1810         rc = mdt_fix_attr_ucred(info, op);
1811         if (rc != 0)
1812                 GOTO(out_ucred, rc = err_serious(rc));
1813
1814         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1815                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1816                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1817                 GOTO(out_ucred, rc);
1818         }
1819         rc = mdt_reint_rec(info, lhc);
1820         EXIT;
1821 out_ucred:
1822         mdt_exit_ucred(info);
1823 out_shrink:
1824         mdt_client_compatibility(info);
1825         rc2 = mdt_fix_reply(info);
1826         if (rc == 0)
1827                 rc = rc2;
1828         return rc;
1829 }
1830
1831 static long mdt_reint_opcode(struct ptlrpc_request *req,
1832                              const struct req_format **fmt)
1833 {
1834         struct mdt_device       *mdt;
1835         struct mdt_rec_reint    *rec;
1836         long                     opc;
1837
1838         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1839         if (rec != NULL) {
1840                 opc = rec->rr_opcode;
1841                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1842                 if (opc < REINT_MAX && fmt[opc] != NULL)
1843                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1844                 else {
1845                         mdt = mdt_exp2dev(req->rq_export);
1846                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1847                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1848                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1849                         opc = err_serious(-EFAULT);
1850                 }
1851         } else {
1852                 opc = err_serious(-EFAULT);
1853         }
1854         return opc;
1855 }
1856
1857 static int mdt_reint(struct tgt_session_info *tsi)
1858 {
1859         long opc;
1860         int  rc;
1861         static const struct req_format *reint_fmts[REINT_MAX] = {
1862                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1863                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1864                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1865                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1866                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1867                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1868                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1869                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1870                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
1871         };
1872
1873         ENTRY;
1874
1875         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1876         if (opc >= 0) {
1877                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1878                 /*
1879                  * No lock possible here from client to pass it to reint code
1880                  * path.
1881                  */
1882                 rc = mdt_reint_internal(info, NULL, opc);
1883                 mdt_thread_info_fini(info);
1884         } else {
1885                 rc = opc;
1886         }
1887
1888         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1889         RETURN(rc);
1890 }
1891
1892 /* this should sync the whole device */
1893 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1894 {
1895         struct dt_device *dt = mdt->mdt_bottom;
1896         int rc;
1897         ENTRY;
1898
1899         rc = dt->dd_ops->dt_sync(env, dt);
1900         RETURN(rc);
1901 }
1902
1903 /* this should sync this object */
1904 static int mdt_object_sync(struct mdt_thread_info *info)
1905 {
1906         struct md_object *next;
1907         int rc;
1908         ENTRY;
1909
1910         if (!mdt_object_exists(info->mti_object)) {
1911                 CWARN("%s: non existing object "DFID": rc = %d\n",
1912                       mdt_obd_name(info->mti_mdt),
1913                       PFID(mdt_object_fid(info->mti_object)), -ESTALE);
1914                 RETURN(-ESTALE);
1915         }
1916         next = mdt_object_child(info->mti_object);
1917         rc = mo_object_sync(info->mti_env, next);
1918
1919         RETURN(rc);
1920 }
1921
1922 static int mdt_sync(struct tgt_session_info *tsi)
1923 {
1924         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1925         struct req_capsule      *pill = tsi->tsi_pill;
1926         struct mdt_body         *body;
1927         int                      rc;
1928
1929         ENTRY;
1930
1931         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1932                 RETURN(err_serious(-ENOMEM));
1933
1934         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
1935                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1936         } else {
1937                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1938
1939                 /* sync an object */
1940                 rc = mdt_object_sync(info);
1941                 if (rc == 0) {
1942                         const struct lu_fid *fid;
1943                         struct lu_attr *la = &info->mti_attr.ma_attr;
1944
1945                         info->mti_attr.ma_need = MA_INODE;
1946                         info->mti_attr.ma_valid = 0;
1947                         rc = mdt_attr_get_complex(info, info->mti_object,
1948                                                   &info->mti_attr);
1949                         if (rc == 0) {
1950                                 body = req_capsule_server_get(pill,
1951                                                               &RMF_MDT_BODY);
1952                                 fid = mdt_object_fid(info->mti_object);
1953                                 mdt_pack_attr2body(info, body, la, fid);
1954                         }
1955                 }
1956                 mdt_thread_info_fini(info);
1957         }
1958         if (rc == 0)
1959                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1960
1961         RETURN(rc);
1962 }
1963
1964 /*
1965  * Handle quota control requests to consult current usage/limit, but also
1966  * to configure quota enforcement
1967  */
1968 static int mdt_quotactl(struct tgt_session_info *tsi)
1969 {
1970         struct obd_export       *exp  = tsi->tsi_exp;
1971         struct req_capsule      *pill = tsi->tsi_pill;
1972         struct obd_quotactl     *oqctl, *repoqc;
1973         int                      id, rc;
1974         struct mdt_device       *mdt = mdt_exp2dev(exp);
1975         struct lu_device        *qmt = mdt->mdt_qmt_dev;
1976         ENTRY;
1977
1978         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1979         if (oqctl == NULL)
1980                 RETURN(err_serious(-EPROTO));
1981
1982         rc = req_capsule_server_pack(pill);
1983         if (rc)
1984                 RETURN(err_serious(rc));
1985
1986         switch (oqctl->qc_cmd) {
1987         case Q_QUOTACHECK:
1988         case LUSTRE_Q_INVALIDATE:
1989         case LUSTRE_Q_FINVALIDATE:
1990         case Q_QUOTAON:
1991         case Q_QUOTAOFF:
1992         case Q_INITQUOTA:
1993                 /* deprecated, not used any more */
1994                 RETURN(-EOPNOTSUPP);
1995                 /* master quotactl */
1996         case Q_GETINFO:
1997         case Q_SETINFO:
1998         case Q_SETQUOTA:
1999         case Q_GETQUOTA:
2000                 if (qmt == NULL)
2001                         RETURN(-EOPNOTSUPP);
2002                 /* slave quotactl */
2003         case Q_GETOINFO:
2004         case Q_GETOQUOTA:
2005                 break;
2006         default:
2007                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2008                 RETURN(-EFAULT);
2009         }
2010
2011         /* map uid/gid for remote client */
2012         id = oqctl->qc_id;
2013         if (exp_connect_rmtclient(exp)) {
2014                 struct lustre_idmap_table *idmap;
2015
2016                 idmap = exp->exp_mdt_data.med_idmap;
2017
2018                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2019                              oqctl->qc_cmd != Q_GETINFO))
2020                         RETURN(-EPERM);
2021
2022                 if (oqctl->qc_type == USRQUOTA)
2023                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2024                                                      oqctl->qc_id);
2025                 else if (oqctl->qc_type == GRPQUOTA)
2026                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2027                                                      oqctl->qc_id);
2028                 else
2029                         RETURN(-EINVAL);
2030
2031                 if (id == CFS_IDMAP_NOTFOUND) {
2032                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2033                         RETURN(-EACCES);
2034                 }
2035         }
2036
2037         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2038         if (repoqc == NULL)
2039                 RETURN(err_serious(-EFAULT));
2040
2041         if (oqctl->qc_id != id)
2042                 swap(oqctl->qc_id, id);
2043
2044         switch (oqctl->qc_cmd) {
2045
2046         case Q_GETINFO:
2047         case Q_SETINFO:
2048         case Q_SETQUOTA:
2049         case Q_GETQUOTA:
2050                 /* forward quotactl request to QMT */
2051                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2052                 break;
2053
2054         case Q_GETOINFO:
2055         case Q_GETOQUOTA:
2056                 /* slave quotactl */
2057                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2058                                    oqctl);
2059                 break;
2060
2061         default:
2062                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2063                 RETURN(-EFAULT);
2064         }
2065
2066         if (oqctl->qc_id != id)
2067                 swap(oqctl->qc_id, id);
2068
2069         *repoqc = *oqctl;
2070         RETURN(rc);
2071 }
2072
2073 /** clone llog ctxt from child (mdd)
2074  * This allows remote llog (replicator) access.
2075  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2076  * context was originally set up, or we can handle them directly.
2077  * I choose the latter, but that means I need any llog
2078  * contexts set up by child to be accessable by the mdt.  So we clone the
2079  * context into our context list here.
2080  */
2081 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2082                                int idx)
2083 {
2084         struct md_device  *next = mdt->mdt_child;
2085         struct llog_ctxt *ctxt;
2086         int rc;
2087
2088         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2089                 return 0;
2090
2091         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2092         if (rc || ctxt == NULL) {
2093                 return 0;
2094         }
2095
2096         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2097         if (rc)
2098                 CERROR("Can't set mdt ctxt %d\n", rc);
2099
2100         return rc;
2101 }
2102
2103 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2104                                  struct mdt_device *mdt, int idx)
2105 {
2106         struct llog_ctxt *ctxt;
2107
2108         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2109         if (ctxt == NULL)
2110                 return 0;
2111         /* Put once for the get we just did, and once for the clone */
2112         llog_ctxt_put(ctxt);
2113         llog_ctxt_put(ctxt);
2114         return 0;
2115 }
2116
2117 /*
2118  * sec context handlers
2119  */
2120 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2121 {
2122         int rc;
2123
2124         rc = mdt_handle_idmap(tsi);
2125         if (unlikely(rc)) {
2126                 struct ptlrpc_request   *req = tgt_ses_req(tsi);
2127                 __u32                    opc;
2128
2129                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2130                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2131                         sptlrpc_svc_ctx_invalidate(req);
2132         }
2133
2134         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2135
2136         return rc;
2137 }
2138
2139 /*
2140  * quota request handlers
2141  */
2142 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2143 {
2144         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2145         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2146         int                      rc;
2147         ENTRY;
2148
2149         if (qmt == NULL)
2150                 RETURN(err_serious(-EOPNOTSUPP));
2151
2152         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2153         RETURN(rc);
2154 }
2155
2156 struct mdt_object *mdt_object_new(const struct lu_env *env,
2157                                   struct mdt_device *d,
2158                                   const struct lu_fid *f)
2159 {
2160         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2161         struct lu_object *o;
2162         struct mdt_object *m;
2163         ENTRY;
2164
2165         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2166         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2167         if (unlikely(IS_ERR(o)))
2168                 m = (struct mdt_object *)o;
2169         else
2170                 m = mdt_obj(o);
2171         RETURN(m);
2172 }
2173
2174 struct mdt_object *mdt_object_find(const struct lu_env *env,
2175                                    struct mdt_device *d,
2176                                    const struct lu_fid *f)
2177 {
2178         struct lu_object *o;
2179         struct mdt_object *m;
2180         ENTRY;
2181
2182         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2183         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2184         if (unlikely(IS_ERR(o)))
2185                 m = (struct mdt_object *)o;
2186         else
2187                 m = mdt_obj(o);
2188
2189         RETURN(m);
2190 }
2191
2192 /**
2193  * Asyncronous commit for mdt device.
2194  *
2195  * Pass asynchonous commit call down the MDS stack.
2196  *
2197  * \param env environment
2198  * \param mdt the mdt device
2199  */
2200 static void mdt_device_commit_async(const struct lu_env *env,
2201                                     struct mdt_device *mdt)
2202 {
2203         struct dt_device *dt = mdt->mdt_bottom;
2204         int rc;
2205
2206         rc = dt->dd_ops->dt_commit_async(env, dt);
2207         if (unlikely(rc != 0))
2208                 CWARN("%s: async commit start failed: rc = %d\n",
2209                       mdt_obd_name(mdt), rc);
2210 }
2211
2212 /**
2213  * Mark the lock as "synchonous".
2214  *
2215  * Mark the lock to deffer transaction commit to the unlock time.
2216  *
2217  * \param lock the lock to mark as "synchonous"
2218  *
2219  * \see mdt_is_lock_sync
2220  * \see mdt_save_lock
2221  */
2222 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2223 {
2224         lock->l_ast_data = (void*)1;
2225 }
2226
2227 /**
2228  * Check whehter the lock "synchonous" or not.
2229  *
2230  * \param lock the lock to check
2231  * \retval 1 the lock is "synchonous"
2232  * \retval 0 the lock isn't "synchronous"
2233  *
2234  * \see mdt_set_lock_sync
2235  * \see mdt_save_lock
2236  */
2237 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2238 {
2239         return lock->l_ast_data != NULL;
2240 }
2241
2242 /**
2243  * Blocking AST for mdt locks.
2244  *
2245  * Starts transaction commit if in case of COS lock conflict or
2246  * deffers such a commit to the mdt_save_lock.
2247  *
2248  * \param lock the lock which blocks a request or cancelling lock
2249  * \param desc unused
2250  * \param data unused
2251  * \param flag indicates whether this cancelling or blocking callback
2252  * \retval 0
2253  * \see ldlm_blocking_ast_nocheck
2254  */
2255 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2256                      void *data, int flag)
2257 {
2258         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2259         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2260         int rc;
2261         ENTRY;
2262
2263         if (flag == LDLM_CB_CANCELING)
2264                 RETURN(0);
2265         lock_res_and_lock(lock);
2266         if (lock->l_blocking_ast != mdt_blocking_ast) {
2267                 unlock_res_and_lock(lock);
2268                 RETURN(0);
2269         }
2270         if (mdt_cos_is_enabled(mdt) &&
2271             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2272             lock->l_blocking_lock != NULL &&
2273             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2274                 mdt_set_lock_sync(lock);
2275         }
2276         rc = ldlm_blocking_ast_nocheck(lock);
2277
2278         /* There is no lock conflict if l_blocking_lock == NULL,
2279          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2280          * when the last reference to a local lock was released */
2281         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2282                 struct lu_env env;
2283
2284                 rc = lu_env_init(&env, LCT_LOCAL);
2285                 if (unlikely(rc != 0))
2286                         CWARN("%s: lu_env initialization failed, cannot "
2287                               "start asynchronous commit: rc = %d\n",
2288                               obd->obd_name, rc);
2289                 else
2290                         mdt_device_commit_async(&env, mdt);
2291                 lu_env_fini(&env);
2292         }
2293         RETURN(rc);
2294 }
2295
2296 /* Used for cross-MDT lock */
2297 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2298                             void *data, int flag)
2299 {
2300         struct lustre_handle lockh;
2301         int               rc;
2302
2303         switch (flag) {
2304         case LDLM_CB_BLOCKING:
2305                 ldlm_lock2handle(lock, &lockh);
2306                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2307                 if (rc < 0) {
2308                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2309                         RETURN(rc);
2310                 }
2311                 break;
2312         case LDLM_CB_CANCELING:
2313                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2314                 break;
2315         default:
2316                 LBUG();
2317         }
2318         RETURN(0);
2319 }
2320
2321 int mdt_check_resent_lock(struct mdt_thread_info *info,
2322                           struct mdt_object *mo,
2323                           struct mdt_lock_handle *lhc)
2324 {
2325         /* the lock might already be gotten in ldlm_handle_enqueue() */
2326         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
2327                 struct ptlrpc_request *req = mdt_info_req(info);
2328                 struct ldlm_lock      *lock;
2329
2330                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2331                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2332                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
2333                          lhc->mlh_reg_lh.cookie);
2334
2335                 if (!fid_res_name_eq(mdt_object_fid(mo),
2336                                      &lock->l_resource->lr_name)) {
2337                         CWARN("%s: Although resent, but still not "
2338                               "get child lock:"DFID"\n",
2339                               info->mti_exp->exp_obd->obd_name,
2340                               PFID(mdt_object_fid(mo)));
2341                         LDLM_LOCK_PUT(lock);
2342                         RETURN(-EPROTO);
2343                 }
2344                 LDLM_LOCK_PUT(lock);
2345                 return 0;
2346         }
2347         return 1;
2348 }
2349
2350 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2351                            struct mdt_object *o, const struct lu_fid *fid,
2352                            struct lustre_handle *lh, ldlm_mode_t mode,
2353                            __u64 ibits)
2354 {
2355         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2356         ldlm_policy_data_t *policy = &mti->mti_policy;
2357         struct ldlm_res_id *res_id = &mti->mti_res_id;
2358         int rc = 0;
2359         ENTRY;
2360
2361         LASSERT(mdt_object_remote(o));
2362
2363         LASSERT(ibits == MDS_INODELOCK_UPDATE);
2364
2365         fid_build_reg_res_name(fid, res_id);
2366
2367         memset(einfo, 0, sizeof(*einfo));
2368         einfo->ei_type = LDLM_IBITS;
2369         einfo->ei_mode = mode;
2370         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2371         einfo->ei_cb_cp = ldlm_completion_ast;
2372         einfo->ei_enq_slave = 0;
2373         einfo->ei_res_id = res_id;
2374
2375         memset(policy, 0, sizeof(*policy));
2376         policy->l_inodebits.bits = ibits;
2377
2378         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2379                             policy);
2380         RETURN(rc);
2381 }
2382
2383 static int mdt_object_local_lock(struct mdt_thread_info *info,
2384                                  struct mdt_object *o,
2385                                  struct mdt_lock_handle *lh, __u64 ibits,
2386                                  bool nonblock, int locality)
2387 {
2388         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2389         ldlm_policy_data_t *policy = &info->mti_policy;
2390         struct ldlm_res_id *res_id = &info->mti_res_id;
2391         __u64 dlmflags;
2392         int rc;
2393         ENTRY;
2394
2395         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2396         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2397         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2398         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2399
2400         /* Only enqueue LOOKUP lock for remote object */
2401         if (mdt_object_remote(o))
2402                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2403
2404         if (lh->mlh_type == MDT_PDO_LOCK) {
2405                 /* check for exists after object is locked */
2406                 if (mdt_object_exists(o) == 0) {
2407                         /* Non-existent object shouldn't have PDO lock */
2408                         RETURN(-ESTALE);
2409                 } else {
2410                         /* Non-dir object shouldn't have PDO lock */
2411                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2412                                 RETURN(-ENOTDIR);
2413                 }
2414         }
2415
2416         memset(policy, 0, sizeof(*policy));
2417         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2418
2419         dlmflags = LDLM_FL_ATOMIC_CB;
2420         if (nonblock)
2421                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2422
2423         /*
2424          * Take PDO lock on whole directory and build correct @res_id for lock
2425          * on part of directory.
2426          */
2427         if (lh->mlh_pdo_hash != 0) {
2428                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2429                 mdt_lock_pdo_mode(info, o, lh);
2430                 if (lh->mlh_pdo_mode != LCK_NL) {
2431                         /*
2432                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2433                          * is never going to be sent to client and we do not
2434                          * want it slowed down due to possible cancels.
2435                          */
2436                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2437                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2438                                           policy, res_id, dlmflags,
2439                                           info->mti_exp == NULL ? NULL :
2440                                           &info->mti_exp->exp_handle.h_cookie);
2441                         if (unlikely(rc != 0))
2442                                 GOTO(out_unlock, rc);
2443                 }
2444
2445                 /*
2446                  * Finish res_id initializing by name hash marking part of
2447                  * directory which is taking modification.
2448                  */
2449                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2450         }
2451
2452         policy->l_inodebits.bits = ibits;
2453
2454         /*
2455          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2456          * going to be sent to client. If it is - mdt_intent_policy() path will
2457          * fix it up and turn FL_LOCAL flag off.
2458          */
2459         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2460                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2461                           info->mti_exp == NULL ? NULL :
2462                           &info->mti_exp->exp_handle.h_cookie);
2463 out_unlock:
2464         if (rc != 0)
2465                 mdt_object_unlock(info, o, lh, 1);
2466         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2467                    lh->mlh_pdo_hash != 0 &&
2468                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2469                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2470
2471         RETURN(rc);
2472 }
2473
2474 static int
2475 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2476                          struct mdt_lock_handle *lh, __u64 ibits,
2477                          bool nonblock, int locality)
2478 {
2479         int rc;
2480         ENTRY;
2481
2482         if (!mdt_object_remote(o))
2483                 return mdt_object_local_lock(info, o, lh, ibits, nonblock,
2484                                              locality);
2485
2486         if (locality == MDT_LOCAL_LOCK) {
2487                 CERROR("%s: try to get local lock for remote object"
2488                        DFID".\n", mdt_obd_name(info->mti_mdt),
2489                        PFID(mdt_object_fid(o)));
2490                 RETURN(-EPROTO);
2491         }
2492
2493         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2494         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2495                    MDS_INODELOCK_XATTR);
2496         if (ibits & MDS_INODELOCK_UPDATE) {
2497                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2498                  * enqueue RPC can only request 1 lock, to avoid extra
2499                  * RPC, so it will instead enqueue EX lock for remote
2500                  * object anyway XXX*/
2501                 if (lh->mlh_type == MDT_PDO_LOCK &&
2502                     lh->mlh_pdo_hash != 0) {
2503                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2504                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2505                                PFID(mdt_object_fid(o)));
2506                         lh->mlh_pdo_hash = 0;
2507                         lh->mlh_rreg_mode = LCK_EX;
2508                         lh->mlh_type = MDT_REG_LOCK;
2509                 }
2510                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2511                                             &lh->mlh_rreg_lh,
2512                                             lh->mlh_rreg_mode,
2513                                             MDS_INODELOCK_UPDATE);
2514                 if (rc != ELDLM_OK)
2515                         RETURN(rc);
2516         }
2517
2518         /* Only enqueue LOOKUP lock for remote object */
2519         if (ibits & MDS_INODELOCK_LOOKUP) {
2520                 rc = mdt_object_local_lock(info, o, lh,
2521                                            MDS_INODELOCK_LOOKUP,
2522                                            nonblock, locality);
2523                 if (rc != ELDLM_OK)
2524                         RETURN(rc);
2525         }
2526
2527         RETURN(0);
2528 }
2529
2530 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2531                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2532 {
2533         return mdt_object_lock_internal(info, o, lh, ibits, false, locality);
2534 }
2535
2536 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2537                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2538 {
2539         struct mdt_lock_handle tmp = *lh;
2540         int rc;
2541
2542         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality);
2543         if (rc == 0)
2544                 *lh = tmp;
2545
2546         return rc == 0;
2547 }
2548
2549 /**
2550  * Save a lock within request object.
2551  *
2552  * Keep the lock referenced until whether client ACK or transaction
2553  * commit happens or release the lock immediately depending on input
2554  * parameters. If COS is ON, a write lock is converted to COS lock
2555  * before saving.
2556  *
2557  * \param info thead info object
2558  * \param h lock handle
2559  * \param mode lock mode
2560  * \param decref force immediate lock releasing
2561  */
2562 static
2563 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2564                    ldlm_mode_t mode, int decref)
2565 {
2566         ENTRY;
2567
2568         if (lustre_handle_is_used(h)) {
2569                 if (decref || !info->mti_has_trans ||
2570                     !(mode & (LCK_PW | LCK_EX))){
2571                         mdt_fid_unlock(h, mode);
2572                 } else {
2573                         struct mdt_device *mdt = info->mti_mdt;
2574                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2575                         struct ptlrpc_request *req = mdt_info_req(info);
2576                         int no_ack = 0;
2577
2578                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2579                                  h->cookie);
2580                         /* there is no request if mdt_object_unlock() is called
2581                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2582                         if (likely(req != NULL)) {
2583                                 CDEBUG(D_HA, "request = %p reply state = %p"
2584                                        " transno = "LPD64"\n", req,
2585                                        req->rq_reply_state, req->rq_transno);
2586                                 if (mdt_cos_is_enabled(mdt)) {
2587                                         no_ack = 1;
2588                                         ldlm_lock_downgrade(lock, LCK_COS);
2589                                         mode = LCK_COS;
2590                                 }
2591                                 ptlrpc_save_lock(req, h, mode, no_ack);
2592                         } else {
2593                                 ldlm_lock_decref(h, mode);
2594                         }
2595                         if (mdt_is_lock_sync(lock)) {
2596                                 CDEBUG(D_HA, "found sync-lock,"
2597                                        " async commit started\n");
2598                                 mdt_device_commit_async(info->mti_env,
2599                                                         mdt);
2600                         }
2601                         LDLM_LOCK_PUT(lock);
2602                 }
2603                 h->cookie = 0ull;
2604         }
2605
2606         EXIT;
2607 }
2608
2609 /**
2610  * Unlock mdt object.
2611  *
2612  * Immeditely release the regular lock and the PDO lock or save the
2613  * lock in reqeuest and keep them referenced until client ACK or
2614  * transaction commit.
2615  *
2616  * \param info thread info object
2617  * \param o mdt object
2618  * \param lh mdt lock handle referencing regular and PDO locks
2619  * \param decref force immediate lock releasing
2620  */
2621 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2622                        struct mdt_lock_handle *lh, int decref)
2623 {
2624         ENTRY;
2625
2626         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2627         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2628
2629         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2630                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2631
2632         EXIT;
2633 }
2634
2635 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2636                                         const struct lu_fid *f,
2637                                         struct mdt_lock_handle *lh,
2638                                         __u64 ibits)
2639 {
2640         struct mdt_object *o;
2641
2642         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2643         if (!IS_ERR(o)) {
2644                 int rc;
2645
2646                 rc = mdt_object_lock(info, o, lh, ibits,
2647                                      MDT_LOCAL_LOCK);
2648                 if (rc != 0) {
2649                         mdt_object_put(info->mti_env, o);
2650                         o = ERR_PTR(rc);
2651                 }
2652         }
2653         return o;
2654 }
2655
2656 void mdt_object_unlock_put(struct mdt_thread_info * info,
2657                            struct mdt_object * o,
2658                            struct mdt_lock_handle *lh,
2659                            int decref)
2660 {
2661         mdt_object_unlock(info, o, lh, decref);
2662         mdt_object_put(info->mti_env, o);
2663 }
2664
2665 /*
2666  * Generic code handling requests that have struct mdt_body passed in:
2667  *
2668  *  - extract mdt_body from request and save it in @info, if present;
2669  *
2670  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2671  *  @info;
2672  *
2673  *  - if HABEO_CORPUS flag is set for this request type check whether object
2674  *  actually exists on storage (lu_object_exists()).
2675  *
2676  */
2677 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2678 {
2679         const struct mdt_body    *body;
2680         struct mdt_object        *obj;
2681         const struct lu_env      *env;
2682         struct req_capsule       *pill;
2683         int                       rc;
2684         ENTRY;
2685
2686         env = info->mti_env;
2687         pill = info->mti_pill;
2688
2689         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2690         if (body == NULL)
2691                 RETURN(-EFAULT);
2692
2693         if (!(body->mbo_valid & OBD_MD_FLID))
2694                 RETURN(0);
2695
2696         if (!fid_is_sane(&body->mbo_fid1)) {
2697                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2698                 RETURN(-EINVAL);
2699         }
2700
2701         /*
2702          * Do not get size or any capa fields before we check that request
2703          * contains capa actually. There are some requests which do not, for
2704          * instance MDS_IS_SUBDIR.
2705          */
2706         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2707             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2708                 mdt_set_capainfo(info, 0, &body->mbo_fid1,
2709                                  req_capsule_client_get(pill, &RMF_CAPA1));
2710
2711         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2712         if (!IS_ERR(obj)) {
2713                 if ((flags & HABEO_CORPUS) &&
2714                     !mdt_object_exists(obj)) {
2715                         mdt_object_put(env, obj);
2716                         /* for capability renew ENOENT will be handled in
2717                          * mdt_renew_capa */
2718                         if (body->mbo_valid & OBD_MD_FLOSSCAPA)
2719                                 rc = 0;
2720                         else
2721                                 rc = -ENOENT;
2722                 } else {
2723                         info->mti_object = obj;
2724                         rc = 0;
2725                 }
2726         } else
2727                 rc = PTR_ERR(obj);
2728
2729         RETURN(rc);
2730 }
2731
2732 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2733 {
2734         struct req_capsule *pill = info->mti_pill;
2735         int rc;
2736         ENTRY;
2737
2738         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2739                 rc = mdt_body_unpack(info, flags);
2740         else
2741                 rc = 0;
2742
2743         if (rc == 0 && (flags & HABEO_REFERO)) {
2744                 /* Pack reply. */
2745                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2746                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2747                                              DEF_REP_MD_SIZE);
2748                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2749                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2750                                              RCL_SERVER, 0);
2751
2752                 rc = req_capsule_server_pack(pill);
2753         }
2754         RETURN(rc);
2755 }
2756
2757 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2758 {
2759         struct md_device *next = m->mdt_child;
2760
2761         return next->md_ops->mdo_init_capa_ctxt(env, next,
2762                                                 m->mdt_lut.lut_mds_capa,
2763                                                 m->mdt_capa_timeout,
2764                                                 m->mdt_capa_alg,
2765                                                 m->mdt_capa_keys);
2766 }
2767
2768 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2769 {
2770         lh->mlh_type = MDT_NUL_LOCK;
2771         lh->mlh_reg_lh.cookie = 0ull;
2772         lh->mlh_reg_mode = LCK_MINMODE;
2773         lh->mlh_pdo_lh.cookie = 0ull;
2774         lh->mlh_pdo_mode = LCK_MINMODE;
2775         lh->mlh_rreg_lh.cookie = 0ull;
2776         lh->mlh_rreg_mode = LCK_MINMODE;
2777 }
2778
2779 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2780 {
2781         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2782         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2783 }
2784
2785 /*
2786  * Initialize fields of struct mdt_thread_info. Other fields are left in
2787  * uninitialized state, because it's too expensive to zero out whole
2788  * mdt_thread_info (> 1K) on each request arrival.
2789  */
2790 void mdt_thread_info_init(struct ptlrpc_request *req,
2791                           struct mdt_thread_info *info)
2792 {
2793         int i;
2794
2795         info->mti_pill = &req->rq_pill;
2796
2797         /* lock handle */
2798         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2799                 mdt_lock_handle_init(&info->mti_lh[i]);
2800
2801         /* mdt device: it can be NULL while CONNECT */
2802         if (req->rq_export) {
2803                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2804                 info->mti_exp = req->rq_export;
2805         } else
2806                 info->mti_mdt = NULL;
2807         info->mti_env = req->rq_svc_thread->t_env;
2808         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2809
2810         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2811         info->mti_big_buf = LU_BUF_NULL;
2812         info->mti_body = NULL;
2813         info->mti_object = NULL;
2814         info->mti_dlm_req = NULL;
2815         info->mti_has_trans = 0;
2816         info->mti_cross_ref = 0;
2817         info->mti_opdata = 0;
2818         info->mti_big_lmm_used = 0;
2819
2820         info->mti_spec.no_create = 0;
2821         info->mti_spec.sp_rm_entry = 0;
2822
2823         info->mti_spec.u.sp_ea.eadata = NULL;
2824         info->mti_spec.u.sp_ea.eadatalen = 0;
2825 }
2826
2827 void mdt_thread_info_fini(struct mdt_thread_info *info)
2828 {
2829         int i;
2830
2831         if (info->mti_object != NULL) {
2832                 mdt_object_put(info->mti_env, info->mti_object);
2833                 info->mti_object = NULL;
2834         }
2835
2836         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2837                 mdt_lock_handle_fini(&info->mti_lh[i]);
2838         info->mti_env = NULL;
2839         info->mti_pill = NULL;
2840         info->mti_exp = NULL;
2841
2842         if (unlikely(info->mti_big_buf.lb_buf != NULL))
2843                 lu_buf_free(&info->mti_big_buf);
2844 }
2845
2846 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2847 {
2848         struct mdt_thread_info  *mti;
2849         struct lustre_capa      *lc;
2850
2851         mti = mdt_th_info(tsi->tsi_env);
2852         LASSERT(mti != NULL);
2853
2854         mdt_thread_info_init(tgt_ses_req(tsi), mti);
2855         if (tsi->tsi_corpus != NULL) {
2856                 struct req_capsule *pill = tsi->tsi_pill;
2857
2858                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2859                 lu_object_get(tsi->tsi_corpus);
2860
2861                 /*
2862                  * XXX: must be part of tgt_mdt_body_unpack but moved here
2863                  * due to mdt_set_capainfo().
2864                  */
2865                 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2866                     req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) {
2867                         lc = req_capsule_client_get(pill, &RMF_CAPA1);
2868                         mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->mbo_fid1,
2869                                          lc);
2870                 }
2871         }
2872         mti->mti_body = tsi->tsi_mdt_body;
2873         mti->mti_dlm_req = tsi->tsi_dlm_req;
2874
2875         return mti;
2876 }
2877
2878 static int mdt_tgt_connect(struct tgt_session_info *tsi)
2879 {
2880         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2881         int                      rc;
2882
2883         ENTRY;
2884
2885         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
2886             cfs_fail_val ==
2887             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
2888                 schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE, HZ * 3);
2889
2890         rc = tgt_connect(tsi);
2891         if (rc != 0)
2892                 RETURN(rc);
2893
2894         rc = mdt_init_idmap(tsi);
2895         if (rc != 0)
2896                 GOTO(err, rc);
2897         RETURN(0);
2898 err:
2899         obd_disconnect(class_export_get(req->rq_export));
2900         return rc;
2901 }
2902
2903 enum mdt_it_code {
2904         MDT_IT_OPEN,
2905         MDT_IT_OCREAT,
2906         MDT_IT_CREATE,
2907         MDT_IT_GETATTR,
2908         MDT_IT_READDIR,
2909         MDT_IT_LOOKUP,
2910         MDT_IT_UNLINK,
2911         MDT_IT_TRUNC,
2912         MDT_IT_GETXATTR,
2913         MDT_IT_LAYOUT,
2914         MDT_IT_QUOTA,
2915         MDT_IT_NR
2916 };
2917
2918 static int mdt_intent_getattr(enum mdt_it_code opcode,
2919                               struct mdt_thread_info *info,
2920                               struct ldlm_lock **,
2921                               __u64);
2922
2923 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2924                                 struct mdt_thread_info *info,
2925                                 struct ldlm_lock **lockp,
2926                                 __u64 flags);
2927
2928 static int mdt_intent_layout(enum mdt_it_code opcode,
2929                              struct mdt_thread_info *info,
2930                              struct ldlm_lock **,
2931                              __u64);
2932 static int mdt_intent_reint(enum mdt_it_code opcode,
2933                             struct mdt_thread_info *info,
2934                             struct ldlm_lock **,
2935                             __u64);
2936
2937 static struct mdt_it_flavor {
2938         const struct req_format *it_fmt;
2939         __u32                    it_flags;
2940         int                    (*it_act)(enum mdt_it_code ,
2941                                          struct mdt_thread_info *,
2942                                          struct ldlm_lock **,
2943                                          __u64);
2944         long                     it_reint;
2945 } mdt_it_flavor[] = {
2946         [MDT_IT_OPEN]     = {
2947                 .it_fmt   = &RQF_LDLM_INTENT,
2948                 /*.it_flags = HABEO_REFERO,*/
2949                 .it_flags = 0,
2950                 .it_act   = mdt_intent_reint,
2951                 .it_reint = REINT_OPEN
2952         },
2953         [MDT_IT_OCREAT]   = {
2954                 .it_fmt   = &RQF_LDLM_INTENT,
2955                 /*
2956                  * OCREAT is not a MUTABOR request as if the file
2957                  * already exists.
2958                  * We do the extra check of OBD_CONNECT_RDONLY in
2959                  * mdt_reint_open() when we really need to create
2960                  * the object.
2961                  */
2962                 .it_flags = 0,
2963                 .it_act   = mdt_intent_reint,
2964                 .it_reint = REINT_OPEN
2965         },
2966         [MDT_IT_CREATE]   = {
2967                 .it_fmt   = &RQF_LDLM_INTENT,
2968                 .it_flags = MUTABOR,
2969                 .it_act   = mdt_intent_reint,
2970                 .it_reint = REINT_CREATE
2971         },
2972         [MDT_IT_GETATTR]  = {
2973                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2974                 .it_flags = HABEO_REFERO,
2975                 .it_act   = mdt_intent_getattr
2976         },
2977         [MDT_IT_READDIR]  = {
2978                 .it_fmt   = NULL,
2979                 .it_flags = 0,
2980                 .it_act   = NULL
2981         },
2982         [MDT_IT_LOOKUP]   = {
2983                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2984                 .it_flags = HABEO_REFERO,
2985                 .it_act   = mdt_intent_getattr
2986         },
2987         [MDT_IT_UNLINK]   = {
2988                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2989                 .it_flags = MUTABOR,
2990                 .it_act   = NULL,
2991                 .it_reint = REINT_UNLINK
2992         },
2993         [MDT_IT_TRUNC]    = {
2994                 .it_fmt   = NULL,
2995                 .it_flags = MUTABOR,
2996                 .it_act   = NULL
2997         },
2998         [MDT_IT_GETXATTR] = {
2999                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
3000                 .it_flags = HABEO_CORPUS,
3001                 .it_act   = mdt_intent_getxattr
3002         },
3003         [MDT_IT_LAYOUT] = {
3004                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3005                 .it_flags = 0,
3006                 .it_act   = mdt_intent_layout
3007         }
3008 };
3009
3010 static int
3011 mdt_intent_lock_replace(struct mdt_thread_info *info,
3012                         struct ldlm_lock **lockp,
3013                         struct mdt_lock_handle *lh,
3014                         __u64 flags)
3015 {
3016         struct ptlrpc_request  *req = mdt_info_req(info);
3017         struct ldlm_lock       *lock = *lockp;
3018         struct ldlm_lock       *new_lock;
3019
3020         /* If possible resent found a lock, @lh is set to its handle */
3021         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3022
3023         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3024                 lh->mlh_reg_lh.cookie = 0;
3025                 RETURN(0);
3026         }
3027
3028         LASSERTF(new_lock != NULL,
3029                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3030
3031         /*
3032          * If we've already given this lock to a client once, then we should
3033          * have no readers or writers.  Otherwise, we should have one reader
3034          * _or_ writer ref (which will be zeroed below) before returning the
3035          * lock to a client.
3036          */
3037         if (new_lock->l_export == req->rq_export) {
3038                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3039         } else {
3040                 LASSERT(new_lock->l_export == NULL);
3041                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3042         }
3043
3044         *lockp = new_lock;
3045
3046         if (new_lock->l_export == req->rq_export) {
3047                 /*
3048                  * Already gave this to the client, which means that we
3049                  * reconstructed a reply.
3050                  */
3051                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3052                         MSG_RESENT);
3053
3054                 LDLM_LOCK_RELEASE(new_lock);
3055                 lh->mlh_reg_lh.cookie = 0;
3056                 RETURN(ELDLM_LOCK_REPLACED);
3057         }
3058
3059         /*
3060          * Fixup the lock to be given to the client.
3061          */
3062         lock_res_and_lock(new_lock);
3063         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3064          * possible blocking AST. */
3065         while (new_lock->l_readers > 0) {
3066                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3067                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3068                 new_lock->l_readers--;
3069         }
3070         while (new_lock->l_writers > 0) {
3071                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3072                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3073                 new_lock->l_writers--;
3074         }
3075
3076         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3077         new_lock->l_blocking_ast = lock->l_blocking_ast;
3078         new_lock->l_completion_ast = lock->l_completion_ast;
3079         new_lock->l_remote_handle = lock->l_remote_handle;
3080         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3081
3082         unlock_res_and_lock(new_lock);
3083
3084         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3085                      &new_lock->l_remote_handle,
3086                      &new_lock->l_exp_hash);
3087
3088         LDLM_LOCK_RELEASE(new_lock);
3089         lh->mlh_reg_lh.cookie = 0;
3090
3091         RETURN(ELDLM_LOCK_REPLACED);
3092 }
3093
3094 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3095                                     struct ldlm_lock *new_lock,
3096                                     struct mdt_lock_handle *lh,
3097                                     __u64 flags)
3098 {
3099         struct ptlrpc_request  *req = mdt_info_req(info);
3100         struct ldlm_request    *dlmreq;
3101
3102         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3103                 return;
3104
3105         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3106
3107         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3108          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3109          * initialized. */
3110         if (flags & LDLM_FL_RESENT) {
3111                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3112                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3113
3114                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3115                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64,
3116                           lh->mlh_reg_lh.cookie);
3117                 return;
3118         }
3119
3120         /*
3121          * If the xid matches, then we know this is a resent request, and allow
3122          * it. (It's probably an OPEN, for which we don't send a lock.
3123          */
3124         if (req_xid_is_last(req))
3125                 return;
3126
3127         /*
3128          * This remote handle isn't enqueued, so we never received or processed
3129          * this request.  Clear MSG_RESENT, because it can be handled like any
3130          * normal request now.
3131          */
3132         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3133
3134         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3135                   dlmreq->lock_handle[0].cookie);
3136 }
3137
3138 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3139                                 struct mdt_thread_info *info,
3140                                 struct ldlm_lock **lockp,
3141                                 __u64 flags)
3142 {
3143         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3144         struct ldlm_reply      *ldlm_rep = NULL;
3145         int rc, grc;
3146
3147         /*
3148          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3149          * (for the resend case) or a new lock. Below we will use it to
3150          * replace the original lock.
3151          */
3152         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3153         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3154                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3155                 rc = mdt_object_lock(info, info->mti_object, lhc,
3156                                         MDS_INODELOCK_XATTR,
3157                                         MDT_LOCAL_LOCK);
3158                 if (rc)
3159                         return rc;
3160         }
3161
3162         grc = mdt_getxattr(info);
3163
3164         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3165
3166         if (mdt_info_req(info)->rq_repmsg != NULL)
3167                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3168         if (ldlm_rep == NULL)
3169                 RETURN(err_serious(-EFAULT));
3170
3171         ldlm_rep->lock_policy_res2 = grc;
3172
3173         return rc;
3174 }
3175
3176 static int mdt_intent_getattr(enum mdt_it_code opcode,
3177                               struct mdt_thread_info *info,
3178                               struct ldlm_lock **lockp,
3179                               __u64 flags)
3180 {
3181         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3182         __u64                   child_bits;
3183         struct ldlm_reply      *ldlm_rep;
3184         struct mdt_body        *reqbody;
3185         struct mdt_body        *repbody;
3186         int                     rc, rc2;
3187         ENTRY;
3188
3189         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3190         LASSERT(reqbody);
3191
3192         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3193         LASSERT(repbody);
3194
3195         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3196         repbody->mbo_eadatasize = 0;
3197         repbody->mbo_aclsize = 0;
3198
3199         switch (opcode) {
3200         case MDT_IT_LOOKUP:
3201                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3202                 break;
3203         case MDT_IT_GETATTR:
3204                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3205                              MDS_INODELOCK_PERM;
3206                 break;
3207         default:
3208                 CERROR("Unsupported intent (%d)\n", opcode);
3209                 GOTO(out_shrink, rc = -EINVAL);
3210         }
3211
3212         rc = mdt_init_ucred(info, reqbody);
3213         if (rc)
3214                 GOTO(out_shrink, rc);
3215
3216         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3217         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3218
3219         /* Get lock from request for possible resent case. */
3220         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3221
3222         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3223         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3224
3225         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3226                 ldlm_rep->lock_policy_res2 = 0;
3227         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3228             ldlm_rep->lock_policy_res2) {
3229                 lhc->mlh_reg_lh.cookie = 0ull;
3230                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3231         }
3232
3233         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3234         EXIT;
3235 out_ucred:
3236         mdt_exit_ucred(info);
3237 out_shrink:
3238         mdt_client_compatibility(info);
3239         rc2 = mdt_fix_reply(info);
3240         if (rc == 0)
3241                 rc = rc2;
3242         return rc;
3243 }
3244
3245 static int mdt_intent_layout(enum mdt_it_code opcode,
3246                              struct mdt_thread_info *info,
3247                              struct ldlm_lock **lockp,
3248                              __u64 flags)
3249 {
3250         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3251         struct layout_intent *layout;
3252         struct lu_fid *fid;
3253         struct mdt_object *obj = NULL;
3254         int rc = 0;
3255         ENTRY;
3256
3257         if (opcode != MDT_IT_LAYOUT) {
3258                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3259                         opcode);
3260                 RETURN(-EINVAL);
3261         }
3262
3263         fid = &info->mti_tmp_fid2;
3264         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3265
3266         /* Get lock from request for possible resent case. */
3267         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3268
3269         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3270         if (IS_ERR(obj))
3271                 RETURN(PTR_ERR(obj));
3272
3273         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3274                 /* get the length of lsm */
3275                 rc = mdt_attr_get_eabuf_size(info, obj);
3276                 if (rc < 0) {
3277                         mdt_object_put(info->mti_env, obj);
3278                         RETURN(rc);
3279                 }
3280
3281                 if (rc > info->mti_mdt->mdt_max_mdsize)
3282                         info->mti_mdt->mdt_max_mdsize = rc;
3283         }
3284
3285         mdt_object_put(info->mti_env, obj);
3286
3287         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3288         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, rc);
3289         rc = req_capsule_server_pack(info->mti_pill);
3290         if (rc != 0)
3291                 RETURN(-EINVAL);
3292
3293         if (lustre_handle_is_used(&lhc->mlh_reg_lh))
3294                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3295
3296         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3297         LASSERT(layout != NULL);
3298         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3299                 /* return to normal/resent ldlm handling */
3300                 RETURN(rc);
3301
3302         CERROR("%s: Unsupported layout intent (%d)\n",
3303                 mdt_obd_name(info->mti_mdt), layout->li_opc);
3304         RETURN(-EINVAL);
3305 }
3306
3307 static int mdt_intent_reint(enum mdt_it_code opcode,
3308                             struct mdt_thread_info *info,
3309                             struct ldlm_lock **lockp,
3310                             __u64 flags)
3311 {
3312         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3313         struct ldlm_reply      *rep = NULL;
3314         long                    opc;
3315         int                     rc;
3316
3317         static const struct req_format *intent_fmts[REINT_MAX] = {
3318                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3319                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3320         };
3321
3322         ENTRY;
3323
3324         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3325         if (opc < 0)
3326                 RETURN(opc);
3327
3328         if (mdt_it_flavor[opcode].it_reint != opc) {
3329                 CERROR("Reint code %ld doesn't match intent: %d\n",
3330                        opc, opcode);
3331                 RETURN(err_serious(-EPROTO));
3332         }
3333
3334         /* Get lock from request for possible resent case. */
3335         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3336
3337         rc = mdt_reint_internal(info, lhc, opc);
3338
3339         /* Check whether the reply has been packed successfully. */
3340         if (mdt_info_req(info)->rq_repmsg != NULL)
3341                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3342         if (rep == NULL)
3343                 RETURN(err_serious(-EFAULT));
3344
3345         /* MDC expects this in any case */
3346         if (rc != 0)
3347                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3348
3349         /* the open lock or the lock for cross-ref object should be
3350          * returned to the client */
3351         if (rc == -MDT_EREMOTE_OPEN || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3352                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3353                 rep->lock_policy_res2 = 0;
3354                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3355                 RETURN(rc);
3356         }
3357
3358         rep->lock_policy_res2 = clear_serious(rc);
3359
3360         if (rep->lock_policy_res2 == -ENOENT &&
3361             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3362                 rep->lock_policy_res2 = 0;
3363
3364         if (rc == -ENOTCONN || rc == -ENODEV ||
3365             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3366                 /*
3367                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3368                  * will be returned by rq_status, and client at ptlrpc layer
3369                  * will detect this, then disconnect, reconnect the import
3370                  * immediately, instead of impacting the following the rpc.
3371                  */
3372                 lhc->mlh_reg_lh.cookie = 0ull;
3373                 RETURN(rc);
3374         } else {
3375                 /*
3376                  * For other cases, the error will be returned by intent.
3377                  * and client will retrieve the result from intent.
3378                  */
3379                  /*
3380                   * FIXME: when open lock is finished, that should be
3381                   * checked here.
3382                   */
3383                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3384                         LASSERTF(rc == 0, "Error occurred but lock handle "
3385                                  "is still in use, rc = %d\n", rc);
3386                         rep->lock_policy_res2 = 0;
3387                         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3388                         RETURN(rc);
3389                 } else {
3390                         lhc->mlh_reg_lh.cookie = 0ull;
3391                         RETURN(ELDLM_LOCK_ABORTED);
3392                 }
3393         }
3394 }
3395
3396 static int mdt_intent_code(long itcode)
3397 {
3398         int rc;
3399
3400         switch(itcode) {
3401         case IT_OPEN:
3402                 rc = MDT_IT_OPEN;
3403                 break;
3404         case IT_OPEN|IT_CREAT:
3405                 rc = MDT_IT_OCREAT;
3406                 break;
3407         case IT_CREAT:
3408                 rc = MDT_IT_CREATE;
3409                 break;
3410         case IT_READDIR:
3411                 rc = MDT_IT_READDIR;
3412                 break;
3413         case IT_GETATTR:
3414                 rc = MDT_IT_GETATTR;
3415                 break;
3416         case IT_LOOKUP:
3417                 rc = MDT_IT_LOOKUP;
3418                 break;
3419         case IT_UNLINK:
3420                 rc = MDT_IT_UNLINK;
3421                 break;
3422         case IT_TRUNC:
3423                 rc = MDT_IT_TRUNC;
3424                 break;
3425         case IT_GETXATTR:
3426                 rc = MDT_IT_GETXATTR;
3427                 break;
3428         case IT_LAYOUT:
3429                 rc = MDT_IT_LAYOUT;
3430                 break;
3431         case IT_QUOTA_DQACQ:
3432         case IT_QUOTA_CONN:
3433                 rc = MDT_IT_QUOTA;
3434                 break;
3435         default:
3436                 CERROR("Unknown intent opcode: %ld\n", itcode);
3437                 rc = -EINVAL;
3438                 break;
3439         }
3440         return rc;
3441 }
3442
3443 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3444                           struct ldlm_lock **lockp, __u64 flags)
3445 {
3446         struct req_capsule      *pill = info->mti_pill;
3447         struct ptlrpc_request   *req = mdt_info_req(info);
3448         struct mdt_it_flavor    *flv;
3449         int opc;
3450         int rc;
3451         ENTRY;
3452
3453         opc = mdt_intent_code(itopc);
3454         if (opc < 0)
3455                 RETURN(-EINVAL);
3456
3457         if (opc == MDT_IT_QUOTA) {
3458                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3459
3460                 if (qmt == NULL)
3461                         RETURN(-EOPNOTSUPP);
3462
3463                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3464                 /* pass the request to quota master */
3465                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3466                                                  mdt_info_req(info), lockp,
3467                                                  flags);
3468                 RETURN(rc);
3469         }
3470
3471         flv = &mdt_it_flavor[opc];
3472         if (flv->it_fmt != NULL)
3473                 req_capsule_extend(pill, flv->it_fmt);
3474
3475         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3476         if (rc < 0)
3477                 RETURN(rc);
3478
3479         if (flv->it_flags & MUTABOR &&
3480             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3481                 RETURN(-EROFS);
3482
3483         if (flv->it_act != NULL) {
3484                 struct ldlm_reply *rep;
3485
3486                 /* execute policy */
3487                 rc = flv->it_act(opc, info, lockp, flags);
3488
3489                 /* Check whether the reply has been packed successfully. */
3490                 if (req->rq_repmsg != NULL) {
3491                         rep = req_capsule_server_get(info->mti_pill,
3492                                                      &RMF_DLM_REP);
3493                         rep->lock_policy_res2 =
3494                                 ptlrpc_status_hton(rep->lock_policy_res2);
3495                 }
3496         }
3497
3498         RETURN(rc);
3499 }
3500
3501 static int mdt_intent_policy(struct ldlm_namespace *ns,
3502                              struct ldlm_lock **lockp, void *req_cookie,
3503                              ldlm_mode_t mode, __u64 flags, void *data)
3504 {
3505         struct tgt_session_info *tsi;
3506         struct mdt_thread_info  *info;
3507         struct ptlrpc_request   *req  =  req_cookie;
3508         struct ldlm_intent      *it;
3509         struct req_capsule      *pill;
3510         int rc;
3511
3512         ENTRY;
3513
3514         LASSERT(req != NULL);
3515
3516         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3517
3518         info = tsi2mdt_info(tsi);
3519         LASSERT(info != NULL);
3520         pill = info->mti_pill;
3521         LASSERT(pill->rc_req == req);
3522
3523         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3524                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3525                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3526                 if (it != NULL) {
3527                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3528                         if (rc == 0)
3529                                 rc = ELDLM_OK;
3530
3531                         /* Lock without inodebits makes no sense and will oops
3532                          * later in ldlm. Let's check it now to see if we have
3533                          * ibits corrupted somewhere in mdt_intent_opc().
3534                          * The case for client miss to set ibits has been
3535                          * processed by others. */
3536                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3537                                         lr_type == LDLM_IBITS,
3538                                      info->mti_dlm_req->lock_desc.\
3539                                         l_policy_data.l_inodebits.bits != 0));
3540                 } else
3541                         rc = err_serious(-EFAULT);
3542         } else {
3543                 /* No intent was provided */
3544                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3545                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3546                 rc = req_capsule_server_pack(pill);
3547                 if (rc)
3548                         rc = err_serious(rc);
3549         }
3550         mdt_thread_info_fini(info);
3551         RETURN(rc);
3552 }
3553
3554 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3555 {
3556         struct seq_server_site  *ss = mdt_seq_site(mdt);
3557
3558         if (ss->ss_node_id == 0)
3559                 return;
3560
3561         if (ss->ss_client_seq != NULL) {
3562                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3563                 ss->ss_client_seq->lcs_exp = NULL;
3564         }
3565
3566         if (ss->ss_server_fld != NULL) {
3567                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3568                 ss->ss_server_fld->lsf_control_exp = NULL;
3569         }
3570 }
3571
3572 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3573 {
3574         struct seq_server_site *ss = mdt_seq_site(mdt);
3575
3576         if (ss == NULL)
3577                 return;
3578
3579         if (ss->ss_server_seq != NULL)
3580                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3581 }
3582
3583 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3584 {
3585         mdt_seq_fini_cli(mdt);
3586         mdt_deregister_seq_exp(mdt);
3587
3588         return seq_site_fini(env, mdt_seq_site(mdt));
3589 }
3590
3591 /**
3592  * It will retrieve its FLDB entries from MDT0, and it only happens
3593  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3594  * and it needs to refresh FLDB from the MDT0.
3595  **/
3596 static int mdt_register_lwp_callback(void *data)
3597 {
3598         struct lu_env           env;
3599         struct mdt_device       *mdt = data;
3600         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3601         int                     rc;
3602         ENTRY;
3603
3604         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3605
3606         if (!likely(fld->lsf_new))
3607                 RETURN(0);
3608
3609         rc = lu_env_init(&env, LCT_MD_THREAD);
3610         if (rc) {
3611                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3612                 RETURN(rc);
3613         }
3614
3615         rc = fld_update_from_controller(&env, fld);
3616         if (rc != 0) {
3617                 CERROR("%s: cannot update controller: rc = %d\n",
3618                        mdt_obd_name(mdt), rc);
3619                 GOTO(out, rc);
3620         }
3621 out:
3622         lu_env_fini(&env);
3623         RETURN(rc);
3624 }
3625
3626 static int mdt_register_seq_exp(struct mdt_device *mdt)
3627 {
3628         struct seq_server_site  *ss = mdt_seq_site(mdt);
3629         char                    *lwp_name = NULL;
3630         int                     rc;
3631
3632         if (ss->ss_node_id == 0)
3633                 return 0;
3634
3635         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3636         if (lwp_name == NULL)
3637                 GOTO(out_free, rc = -ENOMEM);
3638
3639         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3640         if (rc != 0)
3641                 GOTO(out_free, rc);
3642
3643         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3644                                       NULL, NULL);
3645         if (rc != 0)
3646                 GOTO(out_free, rc);
3647
3648         rc = lustre_register_lwp_item(lwp_name,
3649                                       &ss->ss_server_fld->lsf_control_exp,
3650                                       mdt_register_lwp_callback, mdt);
3651         if (rc != 0) {
3652                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3653                 ss->ss_client_seq->lcs_exp = NULL;
3654                 GOTO(out_free, rc);
3655         }
3656 out_free:
3657         if (lwp_name != NULL)
3658                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3659
3660         return rc;
3661 }
3662
3663 /*
3664  * Init client sequence manager which is used by local MDS to talk to sequence
3665  * controller on remote node.
3666  */
3667 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3668 {
3669         struct seq_server_site  *ss = mdt_seq_site(mdt);
3670         int                     rc;
3671         char                    *prefix;
3672         ENTRY;
3673
3674         /* check if this is adding the first MDC and controller is not yet
3675          * initialized. */
3676         OBD_ALLOC_PTR(ss->ss_client_seq);
3677         if (ss->ss_client_seq == NULL)
3678                 RETURN(-ENOMEM);
3679
3680         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3681         if (prefix == NULL) {
3682                 OBD_FREE_PTR(ss->ss_client_seq);
3683                 ss->ss_client_seq = NULL;
3684                 RETURN(-ENOMEM);
3685         }
3686
3687         /* Note: seq_client_fini will be called in seq_site_fini */
3688         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3689         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3690                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3691                                                             NULL);
3692         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3693         if (rc != 0) {