Whamcloud - gitweb
LU-3540 lod: update recovery thread
[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, 2014, 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 #include <lustre_nodemap.h>
73
74 mdl_mode_t mdt_mdl_lock_modes[] = {
75         [LCK_MINMODE] = MDL_MINMODE,
76         [LCK_EX]      = MDL_EX,
77         [LCK_PW]      = MDL_PW,
78         [LCK_PR]      = MDL_PR,
79         [LCK_CW]      = MDL_CW,
80         [LCK_CR]      = MDL_CR,
81         [LCK_NL]      = MDL_NL,
82         [LCK_GROUP]   = MDL_GROUP
83 };
84
85 ldlm_mode_t mdt_dlm_lock_modes[] = {
86         [MDL_MINMODE] = LCK_MINMODE,
87         [MDL_EX]      = LCK_EX,
88         [MDL_PW]      = LCK_PW,
89         [MDL_PR]      = LCK_PR,
90         [MDL_CW]      = LCK_CW,
91         [MDL_CR]      = LCK_CR,
92         [MDL_NL]      = LCK_NL,
93         [MDL_GROUP]   = LCK_GROUP
94 };
95
96 static struct mdt_device *mdt_dev(struct lu_device *d);
97 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
98
99 static const struct lu_object_operations mdt_obj_ops;
100
101 /* Slab for MDT object allocation */
102 static struct kmem_cache *mdt_object_kmem;
103
104 /* For HSM restore handles */
105 struct kmem_cache *mdt_hsm_cdt_kmem;
106
107 /* For HSM request handles */
108 struct kmem_cache *mdt_hsm_car_kmem;
109
110 static struct lu_kmem_descr mdt_caches[] = {
111         {
112                 .ckd_cache = &mdt_object_kmem,
113                 .ckd_name  = "mdt_obj",
114                 .ckd_size  = sizeof(struct mdt_object)
115         },
116         {
117                 .ckd_cache      = &mdt_hsm_cdt_kmem,
118                 .ckd_name       = "mdt_cdt_restore_handle",
119                 .ckd_size       = sizeof(struct cdt_restore_handle)
120         },
121         {
122                 .ckd_cache      = &mdt_hsm_car_kmem,
123                 .ckd_name       = "mdt_cdt_agent_req",
124                 .ckd_size       = sizeof(struct cdt_agent_req)
125         },
126         {
127                 .ckd_cache = NULL
128         }
129 };
130
131 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
132 {
133         if (!rep)
134                 return 0;
135         return rep->lock_policy_res1 & op_flag;
136 }
137
138 void mdt_clear_disposition(struct mdt_thread_info *info,
139                            struct ldlm_reply *rep, __u64 op_flag)
140 {
141         if (info) {
142                 info->mti_opdata &= ~op_flag;
143                 tgt_opdata_clear(info->mti_env, op_flag);
144         }
145         if (rep)
146                 rep->lock_policy_res1 &= ~op_flag;
147 }
148
149 void mdt_set_disposition(struct mdt_thread_info *info,
150                          struct ldlm_reply *rep, __u64 op_flag)
151 {
152         if (info) {
153                 info->mti_opdata |= op_flag;
154                 tgt_opdata_set(info->mti_env, op_flag);
155         }
156         if (rep)
157                 rep->lock_policy_res1 |= op_flag;
158 }
159
160 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
161 {
162         lh->mlh_pdo_hash = 0;
163         lh->mlh_reg_mode = lm;
164         lh->mlh_rreg_mode = lm;
165         lh->mlh_type = MDT_REG_LOCK;
166 }
167
168 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
169                        const struct lu_name *lname)
170 {
171         lh->mlh_reg_mode = lock_mode;
172         lh->mlh_rreg_mode = lock_mode;
173         lh->mlh_type = MDT_PDO_LOCK;
174
175         if (lu_name_is_valid(lname)) {
176                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
177                                                   lname->ln_namelen);
178                 /* XXX Workaround for LU-2856
179                  *
180                  * Zero is a valid return value of full_name_hash, but
181                  * several users of mlh_pdo_hash assume a non-zero
182                  * hash value. We therefore map zero onto an
183                  * arbitrary, but consistent value (1) to avoid
184                  * problems further down the road. */
185                 if (unlikely(lh->mlh_pdo_hash == 0))
186                         lh->mlh_pdo_hash = 1;
187         } else {
188                 lh->mlh_pdo_hash = 0;
189         }
190 }
191
192 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
193                               struct mdt_lock_handle *lh)
194 {
195         mdl_mode_t mode;
196         ENTRY;
197
198         /*
199          * Any dir access needs couple of locks:
200          *
201          * 1) on part of dir we gonna take lookup/modify;
202          *
203          * 2) on whole dir to protect it from concurrent splitting and/or to
204          * flush client's cache for readdir().
205          *
206          * so, for a given mode and object this routine decides what lock mode
207          * to use for lock #2:
208          *
209          * 1) if caller's gonna lookup in dir then we need to protect dir from
210          * being splitted only - LCK_CR
211          *
212          * 2) if caller's gonna modify dir then we need to protect dir from
213          * being splitted and to flush cache - LCK_CW
214          *
215          * 3) if caller's gonna modify dir and that dir seems ready for
216          * splitting then we need to protect it from any type of access
217          * (lookup/modify/split) - LCK_EX --bzzz
218          */
219
220         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
221         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
222
223         /*
224          * Ask underlaying level its opinion about preferable PDO lock mode
225          * having access type passed as regular lock mode:
226          *
227          * - MDL_MINMODE means that lower layer does not want to specify lock
228          * mode;
229          *
230          * - MDL_NL means that no PDO lock should be taken. This is used in some
231          * cases. Say, for non-splittable directories no need to use PDO locks
232          * at all.
233          */
234         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
235                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
236
237         if (mode != MDL_MINMODE) {
238                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
239         } else {
240                 /*
241                  * Lower layer does not want to specify locking mode. We do it
242                  * our selves. No special protection is needed, just flush
243                  * client's cache on modification and allow concurrent
244                  * mondification.
245                  */
246                 switch (lh->mlh_reg_mode) {
247                 case LCK_EX:
248                         lh->mlh_pdo_mode = LCK_EX;
249                         break;
250                 case LCK_PR:
251                         lh->mlh_pdo_mode = LCK_CR;
252                         break;
253                 case LCK_PW:
254                         lh->mlh_pdo_mode = LCK_CW;
255                         break;
256                 default:
257                         CERROR("Not expected lock type (0x%x)\n",
258                                (int)lh->mlh_reg_mode);
259                         LBUG();
260                 }
261         }
262
263         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
264         EXIT;
265 }
266
267 static int mdt_getstatus(struct tgt_session_info *tsi)
268 {
269         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
270         struct mdt_device       *mdt = info->mti_mdt;
271         struct mdt_body         *repbody;
272         int                      rc;
273
274         ENTRY;
275
276         rc = mdt_check_ucred(info);
277         if (rc)
278                 GOTO(out, rc = err_serious(rc));
279
280         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
281                 GOTO(out, rc = err_serious(-ENOMEM));
282
283         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
284         repbody->mbo_fid1 = mdt->mdt_md_root_fid;
285         repbody->mbo_valid |= OBD_MD_FLID;
286
287         EXIT;
288 out:
289         mdt_thread_info_fini(info);
290         return rc;
291 }
292
293 static int mdt_statfs(struct tgt_session_info *tsi)
294 {
295         struct ptlrpc_request           *req = tgt_ses_req(tsi);
296         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
297         struct md_device                *next = info->mti_mdt->mdt_child;
298         struct ptlrpc_service_part      *svcpt;
299         struct obd_statfs               *osfs;
300         int                             rc;
301
302         ENTRY;
303
304         svcpt = req->rq_rqbd->rqbd_svcpt;
305
306         /* This will trigger a watchdog timeout */
307         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
308                          (MDT_SERVICE_WATCHDOG_FACTOR *
309                           at_get(&svcpt->scp_at_estimate)) + 1);
310
311         rc = mdt_check_ucred(info);
312         if (rc)
313                 GOTO(out, rc = err_serious(rc));
314
315         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
316                 GOTO(out, rc = err_serious(-ENOMEM));
317
318         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
319         if (!osfs)
320                 GOTO(out, rc = -EPROTO);
321
322         /** statfs information are cached in the mdt_device */
323         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
324                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
325                 /** statfs data is too old, get up-to-date one */
326                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
327                 if (rc)
328                         GOTO(out, rc);
329                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
330                 info->mti_mdt->mdt_osfs = *osfs;
331                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
332                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
333         } else {
334                 /** use cached statfs data */
335                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
336                 *osfs = info->mti_mdt->mdt_osfs;
337                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
338         }
339
340         if (rc == 0)
341                 mdt_counter_incr(req, LPROC_MDT_STATFS);
342 out:
343         mdt_thread_info_fini(info);
344         RETURN(rc);
345 }
346
347 #ifdef CONFIG_FS_POSIX_ACL
348 /*
349  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
350  *
351  * \param       info    thread info object
352  * \param       repbody reply to pack ACLs into
353  * \param       o       mdt object of file to examine
354  * \param       nodemap nodemap of client to reply to
355  * \retval      0       success
356  * \retval      -errno  error getting or parsing ACL from disk
357  */
358 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
359                       struct mdt_object *o, struct lu_nodemap *nodemap)
360 {
361         const struct lu_env     *env = info->mti_env;
362         struct md_object        *next = mdt_object_child(o);
363         struct lu_buf           *buf = &info->mti_buf;
364         int rc;
365
366         buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
367         buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
368                                            RCL_SERVER);
369         if (buf->lb_len == 0)
370                 return 0;
371
372         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
373         if (rc < 0) {
374                 if (rc == -ENODATA) {
375                         repbody->mbo_aclsize = 0;
376                         repbody->mbo_valid |= OBD_MD_FLACL;
377                         rc = 0;
378                 } else if (rc == -EOPNOTSUPP) {
379                         rc = 0;
380                 } else {
381                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
382                                mdt_obd_name(info->mti_mdt),
383                                PFID(mdt_object_fid(o)), rc);
384                 }
385         } else {
386                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
387                                      rc, NODEMAP_FS_TO_CLIENT);
388                 /* if all ACLs mapped out, rc is still >= 0 */
389                 if (rc < 0) {
390                         CERROR("%s: nodemap_map_acl unable to parse "DFID
391                                " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
392                                PFID(mdt_object_fid(o)), rc);
393                 } else {
394                         repbody->mbo_aclsize = rc;
395                         repbody->mbo_valid |= OBD_MD_FLACL;
396                         rc = 0;
397                 }
398         }
399         return rc;
400 }
401 #endif
402
403 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
404                         const struct lu_attr *attr, const struct lu_fid *fid)
405 {
406         struct md_attr          *ma = &info->mti_attr;
407         struct obd_export       *exp = info->mti_exp;
408         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
409
410         LASSERT(ma->ma_valid & MA_INODE);
411
412         if (attr->la_valid & LA_ATIME) {
413                 b->mbo_atime = attr->la_atime;
414                 b->mbo_valid |= OBD_MD_FLATIME;
415         }
416         if (attr->la_valid & LA_MTIME) {
417                 b->mbo_mtime = attr->la_mtime;
418                 b->mbo_valid |= OBD_MD_FLMTIME;
419         }
420         if (attr->la_valid & LA_CTIME) {
421                 b->mbo_ctime = attr->la_ctime;
422                 b->mbo_valid |= OBD_MD_FLCTIME;
423         }
424         if (attr->la_valid & LA_FLAGS) {
425                 b->mbo_flags = attr->la_flags;
426                 b->mbo_valid |= OBD_MD_FLFLAGS;
427         }
428         if (attr->la_valid & LA_NLINK) {
429                 b->mbo_nlink = attr->la_nlink;
430                 b->mbo_valid |= OBD_MD_FLNLINK;
431         }
432         if (attr->la_valid & LA_UID) {
433                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
434                                             NODEMAP_FS_TO_CLIENT,
435                                             attr->la_uid);
436                 b->mbo_valid |= OBD_MD_FLUID;
437         }
438         if (attr->la_valid & LA_GID) {
439                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
440                                             NODEMAP_FS_TO_CLIENT,
441                                             attr->la_gid);
442                 b->mbo_valid |= OBD_MD_FLGID;
443         }
444         b->mbo_mode = attr->la_mode;
445         if (attr->la_valid & LA_MODE)
446                 b->mbo_valid |= OBD_MD_FLMODE;
447         if (attr->la_valid & LA_TYPE)
448                 b->mbo_valid |= OBD_MD_FLTYPE;
449
450         if (fid != NULL) {
451                 b->mbo_fid1 = *fid;
452                 b->mbo_valid |= OBD_MD_FLID;
453                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid="LPX64"\n",
454                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
455         }
456
457         if (info != NULL)
458                 mdt_body_reverse_idmap(info, b);
459
460         if (!(attr->la_valid & LA_TYPE))
461                 return;
462
463         b->mbo_rdev   = attr->la_rdev;
464         b->mbo_size   = attr->la_size;
465         b->mbo_blocks = attr->la_blocks;
466
467         if (!S_ISREG(attr->la_mode)) {
468                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
469         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
470                 /* means no objects are allocated on osts. */
471                 LASSERT(!(ma->ma_valid & MA_LOV));
472                 /* just ignore blocks occupied by extend attributes on MDS */
473                 b->mbo_blocks = 0;
474                 /* if no object is allocated on osts, the size on mds is valid.
475                  * b=22272 */
476                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
477         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
478                    ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
479                 /* A released file stores its size on MDS. */
480                 /* But return 1 block for released file, unless tools like tar
481                  * will consider it fully sparse. (LU-3864)
482                  */
483                 if (unlikely(b->mbo_size == 0))
484                         b->mbo_blocks = 0;
485                 else
486                         b->mbo_blocks = 1;
487                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
488         }
489
490         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
491                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
492                        PFID(fid), (unsigned long long)b->mbo_size);
493 }
494
495 static inline int mdt_body_has_lov(const struct lu_attr *la,
496                                    const struct mdt_body *body)
497 {
498         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
499                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
500 }
501
502 void mdt_client_compatibility(struct mdt_thread_info *info)
503 {
504         struct mdt_body       *body;
505         struct ptlrpc_request *req = mdt_info_req(info);
506         struct obd_export     *exp = req->rq_export;
507         struct md_attr        *ma = &info->mti_attr;
508         struct lu_attr        *la = &ma->ma_attr;
509         ENTRY;
510
511         if (exp_connect_layout(exp))
512                 /* the client can deal with 16-bit lmm_stripe_count */
513                 RETURN_EXIT;
514
515         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
516
517         if (!mdt_body_has_lov(la, body))
518                 RETURN_EXIT;
519
520         /* now we have a reply with a lov for a client not compatible with the
521          * layout lock so we have to clean the layout generation number */
522         if (S_ISREG(la->la_mode))
523                 ma->ma_lmm->lmm_layout_gen = 0;
524         EXIT;
525 }
526
527 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
528                                    struct mdt_object *o)
529 {
530         const struct lu_env *env = info->mti_env;
531         int rc, rc2;
532
533         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
534                           XATTR_NAME_LOV);
535
536         if (rc == -ENODATA)
537                 rc = 0;
538
539         if (rc < 0)
540                 goto out;
541
542         /* Is it a directory? Let's check for the LMV as well */
543         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
544                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
545                                    XATTR_NAME_LMV);
546
547                 if (rc2 == -ENODATA)
548                         rc2 = mo_xattr_get(env, mdt_object_child(o),
549                                            &LU_BUF_NULL,
550                                            XATTR_NAME_DEFAULT_LMV);
551
552                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
553                         rc = rc2;
554         }
555
556 out:
557         return rc;
558 }
559
560 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
561                       const char *name)
562 {
563         const struct lu_env *env = info->mti_env;
564         int rc;
565         ENTRY;
566
567         LASSERT(info->mti_big_lmm_used == 0);
568         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
569         if (rc < 0)
570                 RETURN(rc);
571
572         /* big_lmm may need to be grown */
573         if (info->mti_big_lmmsize < rc) {
574                 int size = size_roundup_power2(rc);
575
576                 if (info->mti_big_lmmsize > 0) {
577                         /* free old buffer */
578                         LASSERT(info->mti_big_lmm);
579                         OBD_FREE_LARGE(info->mti_big_lmm,
580                                        info->mti_big_lmmsize);
581                         info->mti_big_lmm = NULL;
582                         info->mti_big_lmmsize = 0;
583                 }
584
585                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
586                 if (info->mti_big_lmm == NULL)
587                         RETURN(-ENOMEM);
588                 info->mti_big_lmmsize = size;
589         }
590         LASSERT(info->mti_big_lmmsize >= rc);
591
592         info->mti_buf.lb_buf = info->mti_big_lmm;
593         info->mti_buf.lb_len = info->mti_big_lmmsize;
594         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
595
596         RETURN(rc);
597 }
598
599 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
600                    struct md_attr *ma, const char *name)
601 {
602         struct md_object *next = mdt_object_child(o);
603         struct lu_buf    *buf = &info->mti_buf;
604         int rc;
605
606         if (strcmp(name, XATTR_NAME_LOV) == 0) {
607                 buf->lb_buf = ma->ma_lmm;
608                 buf->lb_len = ma->ma_lmm_size;
609                 LASSERT(!(ma->ma_valid & MA_LOV));
610         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
611                 buf->lb_buf = ma->ma_lmv;
612                 buf->lb_len = ma->ma_lmv_size;
613                 LASSERT(!(ma->ma_valid & MA_LMV));
614         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
615                 buf->lb_buf = ma->ma_lmv;
616                 buf->lb_len = ma->ma_lmv_size;
617                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
618         } else {
619                 return -EINVAL;
620         }
621
622         rc = mo_xattr_get(info->mti_env, next, buf, name);
623         if (rc > 0) {
624
625 got:
626                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
627                         if (info->mti_big_lmm_used)
628                                 ma->ma_lmm = info->mti_big_lmm;
629
630                         /* NOT return LOV EA with hole to old client. */
631                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
632                                      LOV_PATTERN_F_HOLE) &&
633                             !(exp_connect_flags(info->mti_exp) &
634                               OBD_CONNECT_LFSCK)) {
635                                 return -EIO;
636                         } else {
637                                 ma->ma_lmm_size = rc;
638                                 ma->ma_valid |= MA_LOV;
639                         }
640                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
641                         if (info->mti_big_lmm_used)
642                                 ma->ma_lmv = info->mti_big_lmm;
643
644                         ma->ma_lmv_size = rc;
645                         ma->ma_valid |= MA_LMV;
646                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
647                         ma->ma_lmv_size = rc;
648                         ma->ma_valid |= MA_LMV_DEF;
649                 }
650
651                 /* Update mdt_max_mdsize so all clients will be aware that */
652                 if (info->mti_mdt->mdt_max_mdsize < rc)
653                         info->mti_mdt->mdt_max_mdsize = rc;
654
655                 rc = 0;
656         } else if (rc == -ENODATA) {
657                 /* no LOV EA */
658                 rc = 0;
659         } else if (rc == -ERANGE) {
660                 /* Default LMV has fixed size, so it must be able to fit
661                  * in the original buffer */
662                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
663                         return rc;
664                 rc = mdt_big_xattr_get(info, o, name);
665                 if (rc > 0) {
666                         info->mti_big_lmm_used = 1;
667                         goto got;
668                 }
669         }
670
671         return rc;
672 }
673
674 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
675                              struct mdt_object *o, struct lu_fid *pfid)
676 {
677         struct lu_buf           *buf = &info->mti_buf;
678         struct link_ea_header   *leh;
679         struct link_ea_entry    *lee;
680         int                      rc;
681         ENTRY;
682
683         buf->lb_buf = info->mti_big_lmm;
684         buf->lb_len = info->mti_big_lmmsize;
685         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
686                           buf, XATTR_NAME_LINK);
687         /* ignore errors, MA_PFID won't be set and it is
688          * up to the caller to treat this as an error */
689         if (rc == -ERANGE || buf->lb_len == 0) {
690                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
691                 buf->lb_buf = info->mti_big_lmm;
692                 buf->lb_len = info->mti_big_lmmsize;
693         }
694
695         if (rc < 0)
696                 RETURN(rc);
697         if (rc < sizeof(*leh)) {
698                 CERROR("short LinkEA on "DFID": rc = %d\n",
699                        PFID(mdt_object_fid(o)), rc);
700                 RETURN(-ENODATA);
701         }
702
703         leh = (struct link_ea_header *) buf->lb_buf;
704         lee = (struct link_ea_entry *)(leh + 1);
705         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
706                 leh->leh_magic = LINK_EA_MAGIC;
707                 leh->leh_reccount = __swab32(leh->leh_reccount);
708                 leh->leh_len = __swab64(leh->leh_len);
709         }
710         if (leh->leh_magic != LINK_EA_MAGIC)
711                 RETURN(-EINVAL);
712         if (leh->leh_reccount == 0)
713                 RETURN(-ENODATA);
714
715         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
716         fid_be_to_cpu(pfid, pfid);
717
718         RETURN(0);
719 }
720
721 int mdt_attr_get_complex(struct mdt_thread_info *info,
722                          struct mdt_object *o, struct md_attr *ma)
723 {
724         const struct lu_env *env = info->mti_env;
725         struct md_object    *next = mdt_object_child(o);
726         struct lu_buf       *buf = &info->mti_buf;
727         int                  need = ma->ma_need;
728         int                  rc = 0, rc2;
729         u32                  mode;
730         ENTRY;
731
732         ma->ma_valid = 0;
733
734         if (mdt_object_exists(o) == 0)
735                 GOTO(out, rc = -ENOENT);
736         mode = lu_object_attr(&next->mo_lu);
737
738         if (need & MA_INODE) {
739                 ma->ma_need = MA_INODE;
740                 rc = mo_attr_get(env, next, ma);
741                 if (rc)
742                         GOTO(out, rc);
743                 ma->ma_valid |= MA_INODE;
744         }
745
746         if (need & MA_PFID) {
747                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
748                 if (rc == 0)
749                         ma->ma_valid |= MA_PFID;
750                 /* ignore this error, parent fid is not mandatory */
751                 rc = 0;
752         }
753
754         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
755                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
756                 if (rc)
757                         GOTO(out, rc);
758         }
759
760         if (need & MA_LMV && S_ISDIR(mode)) {
761                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
762                 if (rc != 0)
763                         GOTO(out, rc);
764         }
765
766         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
767                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
768                 if (rc != 0)
769                         GOTO(out, rc);
770         }
771
772         if (need & MA_HSM && S_ISREG(mode)) {
773                 buf->lb_buf = info->mti_xattr_buf;
774                 buf->lb_len = sizeof(info->mti_xattr_buf);
775                 CLASSERT(sizeof(struct hsm_attrs) <=
776                          sizeof(info->mti_xattr_buf));
777                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
778                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
779                 if (rc2 == 0)
780                         ma->ma_valid |= MA_HSM;
781                 else if (rc2 < 0 && rc2 != -ENODATA)
782                         GOTO(out, rc = rc2);
783         }
784
785 #ifdef CONFIG_FS_POSIX_ACL
786         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
787                 buf->lb_buf = ma->ma_acl;
788                 buf->lb_len = ma->ma_acl_size;
789                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
790                 if (rc2 > 0) {
791                         ma->ma_acl_size = rc2;
792                         ma->ma_valid |= MA_ACL_DEF;
793                 } else if (rc2 == -ENODATA) {
794                         /* no ACLs */
795                         ma->ma_acl_size = 0;
796                 } else
797                         GOTO(out, rc = rc2);
798         }
799 #endif
800 out:
801         ma->ma_need = need;
802         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
803                rc, ma->ma_valid, ma->ma_lmm);
804         RETURN(rc);
805 }
806
807 static int mdt_getattr_internal(struct mdt_thread_info *info,
808                                 struct mdt_object *o, int ma_need)
809 {
810         struct md_object        *next = mdt_object_child(o);
811         const struct mdt_body   *reqbody = info->mti_body;
812         struct ptlrpc_request   *req = mdt_info_req(info);
813         struct md_attr          *ma = &info->mti_attr;
814         struct lu_attr          *la = &ma->ma_attr;
815         struct req_capsule      *pill = info->mti_pill;
816         const struct lu_env     *env = info->mti_env;
817         struct mdt_body         *repbody;
818         struct lu_buf           *buffer = &info->mti_buf;
819         struct obd_export       *exp = info->mti_exp;
820         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
821         int                      rc;
822         int                      is_root;
823         ENTRY;
824
825         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
826                 RETURN(err_serious(-ENOMEM));
827
828         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
829
830         ma->ma_valid = 0;
831
832         if (mdt_object_remote(o)) {
833                 /* This object is located on remote node.*/
834                 /* Return -ENOTSUPP for old client */
835                 if (!mdt_is_dne_client(req->rq_export))
836                         GOTO(out, rc = -ENOTSUPP);
837
838                 repbody->mbo_fid1 = *mdt_object_fid(o);
839                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
840                 GOTO(out, rc = 0);
841         }
842
843         if (reqbody->mbo_eadatasize > 0) {
844                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
845                 if (buffer->lb_buf == NULL)
846                         GOTO(out, rc = -EPROTO);
847                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
848                                                       RCL_SERVER);
849         } else {
850                 buffer->lb_buf = NULL;
851                 buffer->lb_len = 0;
852                 ma_need &= ~(MA_LOV | MA_LMV);
853                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
854                        mdt_obd_name(info->mti_mdt),
855                        req->rq_export->exp_client_uuid.uuid);
856         }
857
858         /* If it is dir object and client require MEA, then we got MEA */
859         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
860             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
861                 /* Assumption: MDT_MD size is enough for lmv size. */
862                 ma->ma_lmv = buffer->lb_buf;
863                 ma->ma_lmv_size = buffer->lb_len;
864                 ma->ma_need = MA_INODE;
865                 if (ma->ma_lmv_size > 0) {
866                         if (reqbody->mbo_valid & OBD_MD_MEA)
867                                 ma->ma_need |= MA_LMV;
868                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
869                                 ma->ma_need |= MA_LMV_DEF;
870                 }
871         } else {
872                 ma->ma_lmm = buffer->lb_buf;
873                 ma->ma_lmm_size = buffer->lb_len;
874                 ma->ma_need = MA_INODE | MA_HSM;
875                 if (ma->ma_lmm_size > 0)
876                         ma->ma_need |= MA_LOV;
877         }
878
879         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
880             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
881             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
882                 /* get default stripe info for this dir. */
883                 ma->ma_need |= MA_LOV_DEF;
884         }
885         ma->ma_need |= ma_need;
886
887         rc = mdt_attr_get_complex(info, o, ma);
888         if (unlikely(rc)) {
889                 CERROR("%s: getattr error for "DFID": rc = %d\n",
890                        mdt_obd_name(info->mti_mdt),
891                        PFID(mdt_object_fid(o)), rc);
892                 RETURN(rc);
893         }
894
895         /* if file is released, check if a restore is running */
896         if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
897             mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
898                 repbody->mbo_t_state = MS_RESTORE;
899                 repbody->mbo_valid |= OBD_MD_TSTATE;
900         }
901
902         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
903
904         /* the Lustre protocol supposes to return default striping
905          * on the user-visible root if explicitly requested */
906         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
907             (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
908                 struct lu_fid      rootfid;
909                 struct mdt_object *root;
910                 struct mdt_device *mdt = info->mti_mdt;
911
912                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
913                 if (rc)
914                         RETURN(rc);
915                 root = mdt_object_find(env, mdt, &rootfid);
916                 if (IS_ERR(root))
917                         RETURN(PTR_ERR(root));
918                 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
919                 mdt_object_put(info->mti_env, root);
920                 if (unlikely(rc)) {
921                         CERROR("%s: getattr error for "DFID": rc = %d\n",
922                                mdt_obd_name(info->mti_mdt),
923                                PFID(mdt_object_fid(o)), rc);
924                         RETURN(rc);
925                 }
926         }
927
928         if (likely(ma->ma_valid & MA_INODE))
929                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
930         else
931                 RETURN(-EFAULT);
932
933         if (mdt_body_has_lov(la, reqbody)) {
934                 if (ma->ma_valid & MA_LOV) {
935                         LASSERT(ma->ma_lmm_size);
936                         repbody->mbo_eadatasize = ma->ma_lmm_size;
937                         if (S_ISDIR(la->la_mode))
938                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
939                         else
940                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
941                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
942                 }
943                 if (ma->ma_valid & MA_LMV) {
944                         /* Return -ENOTSUPP for old client */
945                         if (!mdt_is_striped_client(req->rq_export))
946                                 RETURN(-ENOTSUPP);
947
948                         LASSERT(S_ISDIR(la->la_mode));
949                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
950                         repbody->mbo_eadatasize = ma->ma_lmv_size;
951                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
952                 }
953                 if (ma->ma_valid & MA_LMV_DEF) {
954                         /* Return -ENOTSUPP for old client */
955                         if (!mdt_is_striped_client(req->rq_export))
956                                 RETURN(-ENOTSUPP);
957                         LASSERT(S_ISDIR(la->la_mode));
958                         repbody->mbo_eadatasize = ma->ma_lmv_size;
959                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
960                                                OBD_MD_DEFAULT_MEA);
961                 }
962         } else if (S_ISLNK(la->la_mode) &&
963                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
964                 buffer->lb_buf = ma->ma_lmm;
965                 /* eadatasize from client includes NULL-terminator, so
966                  * there is no need to read it */
967                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
968                 rc = mo_readlink(env, next, buffer);
969                 if (unlikely(rc <= 0)) {
970                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
971                                mdt_obd_name(info->mti_mdt),
972                                PFID(mdt_object_fid(o)), rc);
973                         rc = -EFAULT;
974                 } else {
975                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
976
977                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
978                                 rc -= 2;
979                         repbody->mbo_valid |= OBD_MD_LINKNAME;
980                         /* we need to report back size with NULL-terminator
981                          * because client expects that */
982                         repbody->mbo_eadatasize = rc + 1;
983                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
984                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
985                                        "on "DFID ", expected %d\n",
986                                        mdt_obd_name(info->mti_mdt),
987                                        rc, PFID(mdt_object_fid(o)),
988                                        reqbody->mbo_eadatasize - 1);
989                         /* NULL terminate */
990                         ((char *)ma->ma_lmm)[rc] = 0;
991
992                         /* If the total CDEBUG() size is larger than a page, it
993                          * will print a warning to the console, avoid this by
994                          * printing just the last part of the symlink. */
995                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
996                                print_limit < rc ? "..." : "", print_limit,
997                                (char *)ma->ma_lmm + rc - print_limit, rc);
998                         rc = 0;
999                 }
1000         }
1001
1002         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1003                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1004                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1005                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1006                        repbody->mbo_max_mdsize);
1007         }
1008
1009         if (exp_connect_rmtclient(info->mti_exp) &&
1010             reqbody->mbo_valid & OBD_MD_FLRMTPERM) {
1011                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
1012
1013                 /* mdt_getattr_lock only */
1014                 rc = mdt_pack_remote_perm(info, o, buf);
1015                 if (rc) {
1016                         repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
1017                         repbody->mbo_aclsize = 0;
1018                         RETURN(rc);
1019                 } else {
1020                         repbody->mbo_valid |= OBD_MD_FLRMTPERM;
1021                         repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
1022                 }
1023         }
1024 #ifdef CONFIG_FS_POSIX_ACL
1025         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1026                  (reqbody->mbo_valid & OBD_MD_FLACL))
1027                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1028 #endif
1029
1030 out:
1031         if (rc == 0)
1032                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1033
1034         RETURN(rc);
1035 }
1036
1037 static int mdt_getattr(struct tgt_session_info *tsi)
1038 {
1039         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1040         struct mdt_object       *obj = info->mti_object;
1041         struct req_capsule      *pill = info->mti_pill;
1042         struct mdt_body         *reqbody;
1043         struct mdt_body         *repbody;
1044         int rc, rc2;
1045         ENTRY;
1046
1047         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1048         LASSERT(reqbody);
1049         LASSERT(obj != NULL);
1050         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1051
1052         /* Unlike intent case where we need to pre-fill out buffers early on
1053          * in intent policy for ldlm reasons, here we can have a much better
1054          * guess at EA size by just reading it from disk.
1055          * Exceptions are readdir and (missing) directory striping */
1056         /* Readlink */
1057         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1058                 /* No easy way to know how long is the symlink, but it cannot
1059                  * be more than PATH_MAX, so we allocate +1 */
1060                 rc = PATH_MAX + 1;
1061
1062         /* A special case for fs ROOT: getattr there might fetch
1063          * default EA for entire fs, not just for this dir!
1064          */
1065         } else if (lu_fid_eq(mdt_object_fid(obj),
1066                              &info->mti_mdt->mdt_md_root_fid) &&
1067                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1068                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1069                                                                  MDS_GETATTR)) {
1070                 /* Should the default strping be bigger, mdt_fix_reply
1071                  * will reallocate */
1072                 rc = DEF_REP_MD_SIZE;
1073         } else {
1074                 /* Read the actual EA size from disk */
1075                 rc = mdt_attr_get_eabuf_size(info, obj);
1076         }
1077
1078         if (rc < 0)
1079                 GOTO(out, rc = err_serious(rc));
1080
1081         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1082
1083         rc = req_capsule_server_pack(pill);
1084         if (unlikely(rc != 0))
1085                 GOTO(out, rc = err_serious(rc));
1086
1087         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1088         LASSERT(repbody != NULL);
1089         repbody->mbo_eadatasize = 0;
1090         repbody->mbo_aclsize = 0;
1091
1092         if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1093                 rc = mdt_init_ucred(info, reqbody);
1094         else
1095                 rc = mdt_check_ucred(info);
1096         if (unlikely(rc))
1097                 GOTO(out_shrink, rc);
1098
1099         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1100
1101         rc = mdt_getattr_internal(info, obj, 0);
1102         if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1103                 mdt_exit_ucred(info);
1104         EXIT;
1105 out_shrink:
1106         mdt_client_compatibility(info);
1107         rc2 = mdt_fix_reply(info);
1108         if (rc == 0)
1109                 rc = rc2;
1110 out:
1111         mdt_thread_info_fini(info);
1112         return rc;
1113 }
1114
1115 /**
1116  * Exchange MOF_LOV_CREATED flags between two objects after a
1117  * layout swap. No assumption is made on whether o1 or o2 have
1118  * created objects or not.
1119  *
1120  * \param[in,out] o1    First swap layout object
1121  * \param[in,out] o2    Second swap layout object
1122  */
1123 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1124 {
1125         __u64   o1_flags;
1126
1127         mutex_lock(&o1->mot_lov_mutex);
1128         mutex_lock(&o2->mot_lov_mutex);
1129
1130         o1_flags = o1->mot_flags;
1131         o1->mot_flags = (o1->mot_flags & ~MOF_LOV_CREATED) |
1132                         (o2->mot_flags & MOF_LOV_CREATED);
1133
1134         o2->mot_flags = (o2->mot_flags & ~MOF_LOV_CREATED) |
1135                         (o1_flags & MOF_LOV_CREATED);
1136
1137         mutex_unlock(&o2->mot_lov_mutex);
1138         mutex_unlock(&o1->mot_lov_mutex);
1139 }
1140
1141 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1142 {
1143         struct mdt_thread_info  *info;
1144         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1145         struct obd_export       *exp = req->rq_export;
1146         struct mdt_object       *o1, *o2, *o;
1147         struct mdt_lock_handle  *lh1, *lh2;
1148         struct mdc_swap_layouts *msl;
1149         int                      rc;
1150         ENTRY;
1151
1152         /* client does not support layout lock, so layout swaping
1153          * is disabled.
1154          * FIXME: there is a problem for old clients which don't support
1155          * layout lock yet. If those clients have already opened the file
1156          * they won't be notified at all so that old layout may still be
1157          * used to do IO. This can be fixed after file release is landed by
1158          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1159         if (!exp_connect_layout(exp))
1160                 RETURN(-EOPNOTSUPP);
1161
1162         info = tsi2mdt_info(tsi);
1163
1164         if (info->mti_dlm_req != NULL)
1165                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1166
1167         o1 = info->mti_object;
1168         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1169                                 &info->mti_body->mbo_fid2);
1170         if (IS_ERR(o))
1171                 GOTO(out, rc = PTR_ERR(o));
1172
1173         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1174                 GOTO(put, rc = -ENOENT);
1175
1176         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1177         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1178                 GOTO(put, rc);
1179
1180         if (rc < 0)
1181                 swap(o1, o2);
1182
1183         /* permission check. Make sure the calling process having permission
1184          * to write both files. */
1185         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1186                                 MAY_WRITE);
1187         if (rc < 0)
1188                 GOTO(put, rc);
1189
1190         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1191                                 MAY_WRITE);
1192         if (rc < 0)
1193                 GOTO(put, rc);
1194
1195         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1196         if (msl == NULL)
1197                 GOTO(put, rc = -EPROTO);
1198
1199         lh1 = &info->mti_lh[MDT_LH_NEW];
1200         mdt_lock_reg_init(lh1, LCK_EX);
1201         lh2 = &info->mti_lh[MDT_LH_OLD];
1202         mdt_lock_reg_init(lh2, LCK_EX);
1203
1204         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1205                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1206         if (rc < 0)
1207                 GOTO(put, rc);
1208
1209         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1210                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1211         if (rc < 0)
1212                 GOTO(unlock1, rc);
1213
1214         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1215                              mdt_object_child(o2), msl->msl_flags);
1216         if (rc < 0)
1217                 GOTO(unlock2, rc);
1218
1219         mdt_swap_lov_flag(o1, o2);
1220
1221 unlock2:
1222         mdt_object_unlock(info, o2, lh2, rc);
1223 unlock1:
1224         mdt_object_unlock(info, o1, lh1, rc);
1225 put:
1226         mdt_object_put(info->mti_env, o);
1227 out:
1228         mdt_thread_info_fini(info);
1229         RETURN(rc);
1230 }
1231
1232 static int mdt_raw_lookup(struct mdt_thread_info *info,
1233                           struct mdt_object *parent,
1234                           const struct lu_name *lname,
1235                           struct ldlm_reply *ldlm_rep)
1236 {
1237         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1238         int              rc;
1239         ENTRY;
1240
1241         LASSERT(!info->mti_cross_ref);
1242
1243         /* Only got the fid of this obj by name */
1244         fid_zero(child_fid);
1245         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1246                         lname, child_fid, &info->mti_spec);
1247         if (rc == 0) {
1248                 struct mdt_body *repbody;
1249
1250                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1251                 repbody->mbo_fid1 = *child_fid;
1252                 repbody->mbo_valid = OBD_MD_FLID;
1253                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1254         } else if (rc == -ENOENT) {
1255                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1256         }
1257
1258         RETURN(rc);
1259 }
1260
1261 /*
1262  * UPDATE lock should be taken against parent, and be released before exit;
1263  * child_bits lock should be taken against child, and be returned back:
1264  *            (1)normal request should release the child lock;
1265  *            (2)intent request will grant the lock to client.
1266  */
1267 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1268                                  struct mdt_lock_handle *lhc,
1269                                  __u64 child_bits,
1270                                  struct ldlm_reply *ldlm_rep)
1271 {
1272         struct ptlrpc_request  *req       = mdt_info_req(info);
1273         struct mdt_body        *reqbody   = NULL;
1274         struct mdt_object      *parent    = info->mti_object;
1275         struct mdt_object      *child;
1276         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1277         struct lu_name         *lname     = NULL;
1278         struct mdt_lock_handle *lhp       = NULL;
1279         struct ldlm_lock       *lock;
1280         bool                    is_resent;
1281         bool                    try_layout;
1282         int                     ma_need = 0;
1283         int                     rc;
1284         ENTRY;
1285
1286         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1287         LASSERT(ergo(is_resent,
1288                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1289
1290         if (parent == NULL)
1291                 RETURN(-ENOENT);
1292
1293         if (info->mti_cross_ref) {
1294                 /* Only getattr on the child. Parent is on another node. */
1295                 mdt_set_disposition(info, ldlm_rep,
1296                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1297                 child = parent;
1298                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1299                        "ldlm_rep = %p\n",
1300                        PFID(mdt_object_fid(child)), ldlm_rep);
1301
1302                 rc = mdt_check_resent_lock(info, child, lhc);
1303                 if (rc < 0) {
1304                         RETURN(rc);
1305                 } else if (rc > 0) {
1306                         mdt_lock_handle_init(lhc);
1307                         mdt_lock_reg_init(lhc, LCK_PR);
1308
1309                         /*
1310                          * Object's name is on another MDS, no lookup or layout
1311                          * lock is needed here but update lock is.
1312                          */
1313                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1314                                         MDS_INODELOCK_LAYOUT);
1315                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1316
1317                         rc = mdt_object_lock(info, child, lhc, child_bits,
1318                                              MDT_LOCAL_LOCK);
1319                         if (rc < 0)
1320                                 RETURN(rc);
1321                 }
1322
1323                 /* Finally, we can get attr for child. */
1324                 if (!mdt_object_exists(child)) {
1325                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1326                                         &child->mot_obj,
1327                                         "remote object doesn't exist.\n");
1328                         mdt_object_unlock(info, child, lhc, 1);
1329                         RETURN(-ENOENT);
1330                 }
1331
1332                 rc = mdt_getattr_internal(info, child, 0);
1333                 if (unlikely(rc != 0))
1334                         mdt_object_unlock(info, child, lhc, 1);
1335
1336                 RETURN(rc);
1337         }
1338
1339         lname = &info->mti_name;
1340         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1341
1342         if (lu_name_is_valid(lname)) {
1343                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1344                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1345                        PNAME(lname), ldlm_rep);
1346         } else {
1347                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1348                 if (unlikely(reqbody == NULL))
1349                         RETURN(err_serious(-EPROTO));
1350
1351                 *child_fid = reqbody->mbo_fid2;
1352
1353                 if (unlikely(!fid_is_sane(child_fid)))
1354                         RETURN(err_serious(-EINVAL));
1355
1356                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1357                        "ldlm_rep = %p\n",
1358                        PFID(mdt_object_fid(parent)),
1359                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1360         }
1361
1362         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1363
1364         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1365                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1366                                 &parent->mot_obj,
1367                                 "Parent doesn't exist!\n");
1368                 RETURN(-ESTALE);
1369         }
1370
1371         if (mdt_object_remote(parent)) {
1372                 CERROR("%s: parent "DFID" is on remote target\n",
1373                        mdt_obd_name(info->mti_mdt),
1374                        PFID(mdt_object_fid(parent)));
1375                 RETURN(-EIO);
1376         }
1377
1378         if (lu_name_is_valid(lname)) {
1379                 /* Always allow to lookup ".." */
1380                 if (unlikely(lname->ln_namelen == 2 &&
1381                              lname->ln_name[0] == '.' &&
1382                              lname->ln_name[1] == '.'))
1383                         info->mti_spec.sp_permitted = 1;
1384
1385                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1386                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1387
1388                         RETURN(rc);
1389                 }
1390
1391                 /* step 1: lock parent only if parent is a directory */
1392                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1393                         lhp = &info->mti_lh[MDT_LH_PARENT];
1394                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1395                         rc = mdt_object_lock(info, parent, lhp,
1396                                              MDS_INODELOCK_UPDATE,
1397                                              MDT_LOCAL_LOCK);
1398                         if (unlikely(rc != 0))
1399                                 RETURN(rc);
1400                 }
1401
1402                 /* step 2: lookup child's fid by name */
1403                 fid_zero(child_fid);
1404                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1405                                 child_fid, &info->mti_spec);
1406                 if (rc == -ENOENT)
1407                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1408
1409                 if (rc != 0)
1410                         GOTO(out_parent, rc);
1411         }
1412
1413         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1414
1415         /*
1416          *step 3: find the child object by fid & lock it.
1417          *        regardless if it is local or remote.
1418          *
1419          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1420          *      set parent dir fid the same as child fid in getattr by fid case
1421          *      we should not lu_object_find() the object again, could lead
1422          *      to hung if there is a concurrent unlink destroyed the object.
1423          */
1424         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1425                 mdt_object_get(info->mti_env, parent);
1426                 child = parent;
1427         } else {
1428                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1429                                         child_fid);
1430         }
1431
1432         if (unlikely(IS_ERR(child)))
1433                 GOTO(out_parent, rc = PTR_ERR(child));
1434
1435         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1436         rc = mdt_check_resent_lock(info, child, lhc);
1437         if (rc < 0) {
1438                 GOTO(out_child, rc);
1439         } else if (rc > 0) {
1440                 mdt_lock_handle_init(lhc);
1441                 mdt_lock_reg_init(lhc, LCK_PR);
1442                 try_layout = false;
1443
1444                 if (!mdt_object_exists(child)) {
1445                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1446                                         &child->mot_obj,
1447                                         "Object doesn't exist!\n");
1448                         GOTO(out_child, rc = -ENOENT);
1449                 }
1450
1451                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1452                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1453                         struct md_attr *ma = &info->mti_attr;
1454
1455                         ma->ma_valid = 0;
1456                         ma->ma_need = MA_INODE;
1457                         rc = mdt_attr_get_complex(info, child, ma);
1458                         if (unlikely(rc != 0))
1459                                 GOTO(out_child, rc);
1460
1461                         /* If the file has not been changed for some time, we
1462                          * return not only a LOOKUP lock, but also an UPDATE
1463                          * lock and this might save us RPC on later STAT. For
1464                          * directories, it also let negative dentry cache start
1465                          * working for this dir. */
1466                         if (ma->ma_valid & MA_INODE &&
1467                             ma->ma_attr.la_valid & LA_CTIME &&
1468                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1469                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1470                                 child_bits |= MDS_INODELOCK_UPDATE;
1471                 }
1472
1473                 /* layout lock must be granted in a best-effort way
1474                  * for IT operations */
1475                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1476                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1477                     exp_connect_layout(info->mti_exp) &&
1478                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1479                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1480                         /* try to grant layout lock for regular file. */
1481                         try_layout = true;
1482                 }
1483
1484                 rc = 0;
1485                 if (try_layout) {
1486                         child_bits |= MDS_INODELOCK_LAYOUT;
1487                         /* try layout lock, it may fail to be granted due to
1488                          * contention at LOOKUP or UPDATE */
1489                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1490                                                  MDT_CROSS_LOCK)) {
1491                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1492                                 LASSERT(child_bits != 0);
1493                                 rc = mdt_object_lock(info, child, lhc,
1494                                                 child_bits, MDT_CROSS_LOCK);
1495                         } else {
1496                                 ma_need |= MA_LOV;
1497                         }
1498                 } else {
1499                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1500                          * client will enqueue the lock to the remote MDT */
1501                         if (mdt_object_remote(child))
1502                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1503                         rc = mdt_object_lock(info, child, lhc, child_bits,
1504                                                 MDT_CROSS_LOCK);
1505                 }
1506                 if (unlikely(rc != 0))
1507                         GOTO(out_child, rc);
1508         }
1509
1510         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1511
1512         /* finally, we can get attr for child. */
1513         rc = mdt_getattr_internal(info, child, ma_need);
1514         if (unlikely(rc != 0)) {
1515                 mdt_object_unlock(info, child, lhc, 1);
1516         } else if (lock) {
1517                 /* Debugging code. */
1518                 LDLM_DEBUG(lock, "Returning lock to client");
1519                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1520                                          &lock->l_resource->lr_name),
1521                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1522                          PLDLMRES(lock->l_resource),
1523                          PFID(mdt_object_fid(child)));
1524         }
1525         if (lock)
1526                 LDLM_LOCK_PUT(lock);
1527
1528         EXIT;
1529 out_child:
1530         mdt_object_put(info->mti_env, child);
1531 out_parent:
1532         if (lhp)
1533                 mdt_object_unlock(info, parent, lhp, 1);
1534         return rc;
1535 }
1536
1537 /* normal handler: should release the child lock */
1538 static int mdt_getattr_name(struct tgt_session_info *tsi)
1539 {
1540         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1541         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1542         struct mdt_body        *reqbody;
1543         struct mdt_body        *repbody;
1544         int rc, rc2;
1545         ENTRY;
1546
1547         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1548         LASSERT(reqbody != NULL);
1549         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1550         LASSERT(repbody != NULL);
1551
1552         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1553         repbody->mbo_eadatasize = 0;
1554         repbody->mbo_aclsize = 0;
1555
1556         rc = mdt_init_ucred(info, reqbody);
1557         if (unlikely(rc))
1558                 GOTO(out_shrink, rc);
1559
1560         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1561         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1562                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1563                 lhc->mlh_reg_lh.cookie = 0;
1564         }
1565         mdt_exit_ucred(info);
1566         EXIT;
1567 out_shrink:
1568         mdt_client_compatibility(info);
1569         rc2 = mdt_fix_reply(info);
1570         if (rc == 0)
1571                 rc = rc2;
1572         mdt_thread_info_fini(info);
1573         return rc;
1574 }
1575
1576 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1577                          void *karg, void *uarg);
1578
1579 static int mdt_set_info(struct tgt_session_info *tsi)
1580 {
1581         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1582         char                    *key;
1583         void                    *val;
1584         int                      keylen, vallen, rc = 0;
1585
1586         ENTRY;
1587
1588         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1589         if (key == NULL) {
1590                 DEBUG_REQ(D_HA, req, "no set_info key");
1591                 RETURN(err_serious(-EFAULT));
1592         }
1593
1594         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1595                                       RCL_CLIENT);
1596
1597         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1598         if (val == NULL) {
1599                 DEBUG_REQ(D_HA, req, "no set_info val");
1600                 RETURN(err_serious(-EFAULT));
1601         }
1602
1603         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1604                                       RCL_CLIENT);
1605
1606         /* Swab any part of val you need to here */
1607         if (KEY_IS(KEY_READ_ONLY)) {
1608                 spin_lock(&req->rq_export->exp_lock);
1609                 if (*(__u32 *)val)
1610                         *exp_connect_flags_ptr(req->rq_export) |=
1611                                 OBD_CONNECT_RDONLY;
1612                 else
1613                         *exp_connect_flags_ptr(req->rq_export) &=
1614                                 ~OBD_CONNECT_RDONLY;
1615                 spin_unlock(&req->rq_export->exp_lock);
1616         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1617                 struct changelog_setinfo *cs = val;
1618
1619                 if (vallen != sizeof(*cs)) {
1620                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1621                                tgt_name(tsi->tsi_tgt), vallen);
1622                         RETURN(-EINVAL);
1623                 }
1624                 if (ptlrpc_req_need_swab(req)) {
1625                         __swab64s(&cs->cs_recno);
1626                         __swab32s(&cs->cs_id);
1627                 }
1628
1629                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1630                                    vallen, val, NULL);
1631         } else {
1632                 RETURN(-EINVAL);
1633         }
1634         RETURN(rc);
1635 }
1636
1637 static int mdt_readpage(struct tgt_session_info *tsi)
1638 {
1639         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1640         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1641         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1642         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1643         struct mdt_body         *repbody;
1644         int                      rc;
1645         int                      i;
1646
1647         ENTRY;
1648
1649         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1650                 RETURN(err_serious(-ENOMEM));
1651
1652         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1653         if (repbody == NULL || reqbody == NULL)
1654                 RETURN(err_serious(-EFAULT));
1655
1656         /*
1657          * prepare @rdpg before calling lower layers and transfer itself. Here
1658          * reqbody->size contains offset of where to start to read and
1659          * reqbody->nlink contains number bytes to read.
1660          */
1661         rdpg->rp_hash = reqbody->mbo_size;
1662         if (rdpg->rp_hash != reqbody->mbo_size) {
1663                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1664                        rdpg->rp_hash, reqbody->mbo_size);
1665                 RETURN(-EFAULT);
1666         }
1667
1668         rdpg->rp_attrs = reqbody->mbo_mode;
1669         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1670                 rdpg->rp_attrs |= LUDA_64BITHASH;
1671         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1672                                 exp_max_brw_size(tsi->tsi_exp));
1673         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1674                           PAGE_CACHE_SHIFT;
1675         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1676         if (rdpg->rp_pages == NULL)
1677                 RETURN(-ENOMEM);
1678
1679         for (i = 0; i < rdpg->rp_npages; ++i) {
1680                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1681                 if (rdpg->rp_pages[i] == NULL)
1682                         GOTO(free_rdpg, rc = -ENOMEM);
1683         }
1684
1685         /* call lower layers to fill allocated pages with directory data */
1686         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1687         if (rc < 0)
1688                 GOTO(free_rdpg, rc);
1689
1690         /* send pages to client */
1691         rc = tgt_sendpage(tsi, rdpg, rc);
1692
1693         EXIT;
1694 free_rdpg:
1695
1696         for (i = 0; i < rdpg->rp_npages; i++)
1697                 if (rdpg->rp_pages[i] != NULL)
1698                         __free_page(rdpg->rp_pages[i]);
1699         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1700
1701         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1702                 RETURN(0);
1703
1704         return rc;
1705 }
1706
1707 static int mdt_reint_internal(struct mdt_thread_info *info,
1708                               struct mdt_lock_handle *lhc,
1709                               __u32 op)
1710 {
1711         struct req_capsule      *pill = info->mti_pill;
1712         struct mdt_body         *repbody;
1713         int                      rc = 0, rc2;
1714
1715         ENTRY;
1716
1717         rc = mdt_reint_unpack(info, op);
1718         if (rc != 0) {
1719                 CERROR("Can't unpack reint, rc %d\n", rc);
1720                 RETURN(err_serious(rc));
1721         }
1722
1723         /* for replay (no_create) lmm is not needed, client has it already */
1724         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1725                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1726                                      DEF_REP_MD_SIZE);
1727
1728         /* llog cookies are always 0, the field is kept for compatibility */
1729         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1730                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1731
1732         rc = req_capsule_server_pack(pill);
1733         if (rc != 0) {
1734                 CERROR("Can't pack response, rc %d\n", rc);
1735                 RETURN(err_serious(rc));
1736         }
1737
1738         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1739                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1740                 LASSERT(repbody);
1741                 repbody->mbo_eadatasize = 0;
1742                 repbody->mbo_aclsize = 0;
1743         }
1744
1745         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1746
1747         /* for replay no cookkie / lmm need, because client have this already */
1748         if (info->mti_spec.no_create)
1749                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1750                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1751
1752         rc = mdt_init_ucred_reint(info);
1753         if (rc)
1754                 GOTO(out_shrink, rc);
1755
1756         rc = mdt_fix_attr_ucred(info, op);
1757         if (rc != 0)
1758                 GOTO(out_ucred, rc = err_serious(rc));
1759
1760         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1761                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1762                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1763                 GOTO(out_ucred, rc);
1764         }
1765         rc = mdt_reint_rec(info, lhc);
1766         EXIT;
1767 out_ucred:
1768         mdt_exit_ucred(info);
1769 out_shrink:
1770         mdt_client_compatibility(info);
1771         rc2 = mdt_fix_reply(info);
1772         if (rc == 0)
1773                 rc = rc2;
1774         return rc;
1775 }
1776
1777 static long mdt_reint_opcode(struct ptlrpc_request *req,
1778                              const struct req_format **fmt)
1779 {
1780         struct mdt_device       *mdt;
1781         struct mdt_rec_reint    *rec;
1782         long                     opc;
1783
1784         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1785         if (rec != NULL) {
1786                 opc = rec->rr_opcode;
1787                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1788                 if (opc < REINT_MAX && fmt[opc] != NULL)
1789                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1790                 else {
1791                         mdt = mdt_exp2dev(req->rq_export);
1792                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1793                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1794                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1795                         opc = err_serious(-EFAULT);
1796                 }
1797         } else {
1798                 opc = err_serious(-EFAULT);
1799         }
1800         return opc;
1801 }
1802
1803 static int mdt_reint(struct tgt_session_info *tsi)
1804 {
1805         long opc;
1806         int  rc;
1807         static const struct req_format *reint_fmts[REINT_MAX] = {
1808                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1809                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1810                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1811                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1812                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1813                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1814                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1815                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1816                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
1817         };
1818
1819         ENTRY;
1820
1821         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1822         if (opc >= 0) {
1823                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1824                 /*
1825                  * No lock possible here from client to pass it to reint code
1826                  * path.
1827                  */
1828                 rc = mdt_reint_internal(info, NULL, opc);
1829                 mdt_thread_info_fini(info);
1830         } else {
1831                 rc = opc;
1832         }
1833
1834         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1835         RETURN(rc);
1836 }
1837
1838 /* this should sync the whole device */
1839 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1840 {
1841         struct dt_device *dt = mdt->mdt_bottom;
1842         int rc;
1843         ENTRY;
1844
1845         rc = dt->dd_ops->dt_sync(env, dt);
1846         RETURN(rc);
1847 }
1848
1849 /* this should sync this object */
1850 static int mdt_object_sync(struct mdt_thread_info *info)
1851 {
1852         struct md_object *next;
1853         int rc;
1854         ENTRY;
1855
1856         if (!mdt_object_exists(info->mti_object)) {
1857                 CWARN("%s: non existing object "DFID": rc = %d\n",
1858                       mdt_obd_name(info->mti_mdt),
1859                       PFID(mdt_object_fid(info->mti_object)), -ESTALE);
1860                 RETURN(-ESTALE);
1861         }
1862         next = mdt_object_child(info->mti_object);
1863         rc = mo_object_sync(info->mti_env, next);
1864
1865         RETURN(rc);
1866 }
1867
1868 static int mdt_sync(struct tgt_session_info *tsi)
1869 {
1870         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1871         struct req_capsule      *pill = tsi->tsi_pill;
1872         struct mdt_body         *body;
1873         int                      rc;
1874
1875         ENTRY;
1876
1877         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1878                 RETURN(err_serious(-ENOMEM));
1879
1880         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
1881                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1882         } else {
1883                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1884
1885                 /* sync an object */
1886                 rc = mdt_object_sync(info);
1887                 if (rc == 0) {
1888                         const struct lu_fid *fid;
1889                         struct lu_attr *la = &info->mti_attr.ma_attr;
1890
1891                         info->mti_attr.ma_need = MA_INODE;
1892                         info->mti_attr.ma_valid = 0;
1893                         rc = mdt_attr_get_complex(info, info->mti_object,
1894                                                   &info->mti_attr);
1895                         if (rc == 0) {
1896                                 body = req_capsule_server_get(pill,
1897                                                               &RMF_MDT_BODY);
1898                                 fid = mdt_object_fid(info->mti_object);
1899                                 mdt_pack_attr2body(info, body, la, fid);
1900                         }
1901                 }
1902                 mdt_thread_info_fini(info);
1903         }
1904         if (rc == 0)
1905                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1906
1907         RETURN(rc);
1908 }
1909
1910 /*
1911  * Handle quota control requests to consult current usage/limit, but also
1912  * to configure quota enforcement
1913  */
1914 static int mdt_quotactl(struct tgt_session_info *tsi)
1915 {
1916         struct obd_export       *exp  = tsi->tsi_exp;
1917         struct req_capsule      *pill = tsi->tsi_pill;
1918         struct obd_quotactl     *oqctl, *repoqc;
1919         int                      id, rc;
1920         struct mdt_device       *mdt = mdt_exp2dev(exp);
1921         struct lu_device        *qmt = mdt->mdt_qmt_dev;
1922         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
1923         ENTRY;
1924
1925         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1926         if (oqctl == NULL)
1927                 RETURN(err_serious(-EPROTO));
1928
1929         rc = req_capsule_server_pack(pill);
1930         if (rc)
1931                 RETURN(err_serious(rc));
1932
1933         switch (oqctl->qc_cmd) {
1934                 /* master quotactl */
1935         case Q_SETINFO:
1936         case Q_SETQUOTA:
1937                 if (!nodemap_can_setquota(nodemap))
1938                         RETURN(-EPERM);
1939         case Q_GETINFO:
1940         case Q_GETQUOTA:
1941                 if (qmt == NULL)
1942                         RETURN(-EOPNOTSUPP);
1943                 /* slave quotactl */
1944         case Q_GETOINFO:
1945         case Q_GETOQUOTA:
1946                 break;
1947         default:
1948                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1949                 RETURN(-EFAULT);
1950         }
1951
1952         /* map uid/gid for remote client */
1953         id = oqctl->qc_id;
1954         if (exp_connect_rmtclient(exp)) {
1955                 struct lustre_idmap_table *idmap;
1956
1957                 idmap = exp->exp_mdt_data.med_idmap;
1958
1959                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1960                              oqctl->qc_cmd != Q_GETINFO))
1961                         RETURN(-EPERM);
1962
1963                 if (oqctl->qc_type == USRQUOTA)
1964                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1965                                                      oqctl->qc_id);
1966                 else if (oqctl->qc_type == GRPQUOTA)
1967                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1968                                                      oqctl->qc_id);
1969                 else
1970                         RETURN(-EINVAL);
1971
1972                 if (id == CFS_IDMAP_NOTFOUND) {
1973                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
1974                         RETURN(-EACCES);
1975                 }
1976         }
1977
1978         if (oqctl->qc_type == USRQUOTA)
1979                 id = nodemap_map_id(nodemap, NODEMAP_UID,
1980                                     NODEMAP_CLIENT_TO_FS, id);
1981         else if (oqctl->qc_type == GRPQUOTA)
1982                 id = nodemap_map_id(nodemap, NODEMAP_UID,
1983                                     NODEMAP_CLIENT_TO_FS, id);
1984
1985         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1986         if (repoqc == NULL)
1987                 RETURN(err_serious(-EFAULT));
1988
1989         if (oqctl->qc_id != id)
1990                 swap(oqctl->qc_id, id);
1991
1992         switch (oqctl->qc_cmd) {
1993
1994         case Q_GETINFO:
1995         case Q_SETINFO:
1996         case Q_SETQUOTA:
1997         case Q_GETQUOTA:
1998                 /* forward quotactl request to QMT */
1999                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2000                 break;
2001
2002         case Q_GETOINFO:
2003         case Q_GETOQUOTA:
2004                 /* slave quotactl */
2005                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2006                                    oqctl);
2007                 break;
2008
2009         default:
2010                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2011                 RETURN(-EFAULT);
2012         }
2013
2014         if (oqctl->qc_id != id)
2015                 swap(oqctl->qc_id, id);
2016
2017         *repoqc = *oqctl;
2018         RETURN(rc);
2019 }
2020
2021 /** clone llog ctxt from child (mdd)
2022  * This allows remote llog (replicator) access.
2023  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2024  * context was originally set up, or we can handle them directly.
2025  * I choose the latter, but that means I need any llog
2026  * contexts set up by child to be accessable by the mdt.  So we clone the
2027  * context into our context list here.
2028  */
2029 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2030                                int idx)
2031 {
2032         struct md_device  *next = mdt->mdt_child;
2033         struct llog_ctxt *ctxt;
2034         int rc;
2035
2036         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2037                 return 0;
2038
2039         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2040         if (rc || ctxt == NULL) {
2041                 return 0;
2042         }
2043
2044         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2045         if (rc)
2046                 CERROR("Can't set mdt ctxt %d\n", rc);
2047
2048         return rc;
2049 }
2050
2051 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2052                                  struct mdt_device *mdt, int idx)
2053 {
2054         struct llog_ctxt *ctxt;
2055
2056         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2057         if (ctxt == NULL)
2058                 return 0;
2059         /* Put once for the get we just did, and once for the clone */
2060         llog_ctxt_put(ctxt);
2061         llog_ctxt_put(ctxt);
2062         return 0;
2063 }
2064
2065 /*
2066  * sec context handlers
2067  */
2068 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2069 {
2070         int rc;
2071
2072         rc = mdt_handle_idmap(tsi);
2073         if (unlikely(rc)) {
2074                 struct ptlrpc_request   *req = tgt_ses_req(tsi);
2075                 __u32                    opc;
2076
2077                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2078                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2079                         sptlrpc_svc_ctx_invalidate(req);
2080         }
2081
2082         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2083
2084         return rc;
2085 }
2086
2087 /*
2088  * quota request handlers
2089  */
2090 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2091 {
2092         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2093         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2094         int                      rc;
2095         ENTRY;
2096
2097         if (qmt == NULL)
2098                 RETURN(err_serious(-EOPNOTSUPP));
2099
2100         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2101         RETURN(rc);
2102 }
2103
2104 struct mdt_object *mdt_object_new(const struct lu_env *env,
2105                                   struct mdt_device *d,
2106                                   const struct lu_fid *f)
2107 {
2108         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2109         struct lu_object *o;
2110         struct mdt_object *m;
2111         ENTRY;
2112
2113         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2114         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2115         if (unlikely(IS_ERR(o)))
2116                 m = (struct mdt_object *)o;
2117         else
2118                 m = mdt_obj(o);
2119         RETURN(m);
2120 }
2121
2122 struct mdt_object *mdt_object_find(const struct lu_env *env,
2123                                    struct mdt_device *d,
2124                                    const struct lu_fid *f)
2125 {
2126         struct lu_object *o;
2127         struct mdt_object *m;
2128         ENTRY;
2129
2130         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2131         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2132         if (unlikely(IS_ERR(o)))
2133                 m = (struct mdt_object *)o;
2134         else
2135                 m = mdt_obj(o);
2136
2137         RETURN(m);
2138 }
2139
2140 /**
2141  * Asyncronous commit for mdt device.
2142  *
2143  * Pass asynchonous commit call down the MDS stack.
2144  *
2145  * \param env environment
2146  * \param mdt the mdt device
2147  */
2148 static void mdt_device_commit_async(const struct lu_env *env,
2149                                     struct mdt_device *mdt)
2150 {
2151         struct dt_device *dt = mdt->mdt_bottom;
2152         int rc;
2153
2154         rc = dt->dd_ops->dt_commit_async(env, dt);
2155         if (unlikely(rc != 0))
2156                 CWARN("%s: async commit start failed: rc = %d\n",
2157                       mdt_obd_name(mdt), rc);
2158 }
2159
2160 /**
2161  * Mark the lock as "synchonous".
2162  *
2163  * Mark the lock to deffer transaction commit to the unlock time.
2164  *
2165  * \param lock the lock to mark as "synchonous"
2166  *
2167  * \see mdt_is_lock_sync
2168  * \see mdt_save_lock
2169  */
2170 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2171 {
2172         lock->l_ast_data = (void*)1;
2173 }
2174
2175 /**
2176  * Check whehter the lock "synchonous" or not.
2177  *
2178  * \param lock the lock to check
2179  * \retval 1 the lock is "synchonous"
2180  * \retval 0 the lock isn't "synchronous"
2181  *
2182  * \see mdt_set_lock_sync
2183  * \see mdt_save_lock
2184  */
2185 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2186 {
2187         return lock->l_ast_data != NULL;
2188 }
2189
2190 /**
2191  * Blocking AST for mdt locks.
2192  *
2193  * Starts transaction commit if in case of COS lock conflict or
2194  * deffers such a commit to the mdt_save_lock.
2195  *
2196  * \param lock the lock which blocks a request or cancelling lock
2197  * \param desc unused
2198  * \param data unused
2199  * \param flag indicates whether this cancelling or blocking callback
2200  * \retval 0
2201  * \see ldlm_blocking_ast_nocheck
2202  */
2203 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2204                      void *data, int flag)
2205 {
2206         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2207         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2208         int rc;
2209         ENTRY;
2210
2211         if (flag == LDLM_CB_CANCELING)
2212                 RETURN(0);
2213         lock_res_and_lock(lock);
2214         if (lock->l_blocking_ast != mdt_blocking_ast) {
2215                 unlock_res_and_lock(lock);
2216                 RETURN(0);
2217         }
2218         if (mdt_cos_is_enabled(mdt) &&
2219             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2220             lock->l_blocking_lock != NULL &&
2221             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2222                 mdt_set_lock_sync(lock);
2223         }
2224         rc = ldlm_blocking_ast_nocheck(lock);
2225
2226         /* There is no lock conflict if l_blocking_lock == NULL,
2227          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2228          * when the last reference to a local lock was released */
2229         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2230                 struct lu_env env;
2231
2232                 rc = lu_env_init(&env, LCT_LOCAL);
2233                 if (unlikely(rc != 0))
2234                         CWARN("%s: lu_env initialization failed, cannot "
2235                               "start asynchronous commit: rc = %d\n",
2236                               obd->obd_name, rc);
2237                 else
2238                         mdt_device_commit_async(&env, mdt);
2239                 lu_env_fini(&env);
2240         }
2241         RETURN(rc);
2242 }
2243
2244 /* Used for cross-MDT lock */
2245 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2246                             void *data, int flag)
2247 {
2248         struct lustre_handle lockh;
2249         int               rc;
2250
2251         switch (flag) {
2252         case LDLM_CB_BLOCKING:
2253                 ldlm_lock2handle(lock, &lockh);
2254                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2255                 if (rc < 0) {
2256                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2257                         RETURN(rc);
2258                 }
2259                 break;
2260         case LDLM_CB_CANCELING:
2261                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2262                 break;
2263         default:
2264                 LBUG();
2265         }
2266         RETURN(0);
2267 }
2268
2269 int mdt_check_resent_lock(struct mdt_thread_info *info,
2270                           struct mdt_object *mo,
2271                           struct mdt_lock_handle *lhc)
2272 {
2273         /* the lock might already be gotten in ldlm_handle_enqueue() */
2274         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2275                 struct ptlrpc_request *req = mdt_info_req(info);
2276                 struct ldlm_lock      *lock;
2277
2278                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2279                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2280                 if (lock == NULL) {
2281                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2282                          * a resend case, however, it could be already destroyed
2283                          * due to client eviction or a raced cancel RPC. */
2284                         LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64"\n",
2285                                           lhc->mlh_reg_lh.cookie);
2286                         RETURN(-ESTALE);
2287                 }
2288
2289                 if (!fid_res_name_eq(mdt_object_fid(mo),
2290                                      &lock->l_resource->lr_name)) {
2291                         CWARN("%s: Although resent, but still not "
2292                               "get child lock:"DFID"\n",
2293                               info->mti_exp->exp_obd->obd_name,
2294                               PFID(mdt_object_fid(mo)));
2295                         LDLM_LOCK_PUT(lock);
2296                         RETURN(-EPROTO);
2297                 }
2298                 LDLM_LOCK_PUT(lock);
2299                 return 0;
2300         }
2301         return 1;
2302 }
2303
2304 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2305                            struct mdt_object *o, const struct lu_fid *fid,
2306                            struct lustre_handle *lh, ldlm_mode_t mode,
2307                            __u64 ibits)
2308 {
2309         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2310         ldlm_policy_data_t *policy = &mti->mti_policy;
2311         struct ldlm_res_id *res_id = &mti->mti_res_id;
2312         int rc = 0;
2313         ENTRY;
2314
2315         LASSERT(mdt_object_remote(o));
2316
2317         fid_build_reg_res_name(fid, res_id);
2318
2319         memset(einfo, 0, sizeof(*einfo));
2320         einfo->ei_type = LDLM_IBITS;
2321         einfo->ei_mode = mode;
2322         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2323         einfo->ei_cb_cp = ldlm_completion_ast;
2324         einfo->ei_enq_slave = 0;
2325         einfo->ei_res_id = res_id;
2326
2327         memset(policy, 0, sizeof(*policy));
2328         policy->l_inodebits.bits = ibits;
2329
2330         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2331                             policy);
2332         RETURN(rc);
2333 }
2334
2335 static int mdt_object_local_lock(struct mdt_thread_info *info,
2336                                  struct mdt_object *o,
2337                                  struct mdt_lock_handle *lh, __u64 ibits,
2338                                  bool nonblock, int locality)
2339 {
2340         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2341         ldlm_policy_data_t *policy = &info->mti_policy;
2342         struct ldlm_res_id *res_id = &info->mti_res_id;
2343         __u64 dlmflags;
2344         int rc;
2345         ENTRY;
2346
2347         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2348         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2349         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2350         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2351
2352         /* Only enqueue LOOKUP lock for remote object */
2353         if (mdt_object_remote(o))
2354                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2355
2356         if (lh->mlh_type == MDT_PDO_LOCK) {
2357                 /* check for exists after object is locked */
2358                 if (mdt_object_exists(o) == 0) {
2359                         /* Non-existent object shouldn't have PDO lock */
2360                         RETURN(-ESTALE);
2361                 } else {
2362                         /* Non-dir object shouldn't have PDO lock */
2363                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2364                                 RETURN(-ENOTDIR);
2365                 }
2366         }
2367
2368         memset(policy, 0, sizeof(*policy));
2369         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2370
2371         dlmflags = LDLM_FL_ATOMIC_CB;
2372         if (nonblock)
2373                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2374
2375         /*
2376          * Take PDO lock on whole directory and build correct @res_id for lock
2377          * on part of directory.
2378          */
2379         if (lh->mlh_pdo_hash != 0) {
2380                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2381                 mdt_lock_pdo_mode(info, o, lh);
2382                 if (lh->mlh_pdo_mode != LCK_NL) {
2383                         /*
2384                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2385                          * is never going to be sent to client and we do not
2386                          * want it slowed down due to possible cancels.
2387                          */
2388                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2389                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2390                                           policy, res_id, dlmflags,
2391                                           info->mti_exp == NULL ? NULL :
2392                                           &info->mti_exp->exp_handle.h_cookie);
2393                         if (unlikely(rc != 0))
2394                                 GOTO(out_unlock, rc);
2395                 }
2396
2397                 /*
2398                  * Finish res_id initializing by name hash marking part of
2399                  * directory which is taking modification.
2400                  */
2401                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2402         }
2403
2404         policy->l_inodebits.bits = ibits;
2405
2406         /*
2407          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2408          * going to be sent to client. If it is - mdt_intent_policy() path will
2409          * fix it up and turn FL_LOCAL flag off.
2410          */
2411         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2412                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2413                           info->mti_exp == NULL ? NULL :
2414                           &info->mti_exp->exp_handle.h_cookie);
2415 out_unlock:
2416         if (rc != 0)
2417                 mdt_object_unlock(info, o, lh, 1);
2418         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2419                    lh->mlh_pdo_hash != 0 &&
2420                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2421                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2422
2423         RETURN(rc);
2424 }
2425
2426 static int
2427 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2428                          struct mdt_lock_handle *lh, __u64 ibits,
2429                          bool nonblock, int locality)
2430 {
2431         int rc;
2432         ENTRY;
2433
2434         if (!mdt_object_remote(o))
2435                 return mdt_object_local_lock(info, o, lh, ibits, nonblock,
2436                                              locality);
2437
2438         if (locality == MDT_LOCAL_LOCK) {
2439                 CERROR("%s: try to get local lock for remote object"
2440                        DFID".\n", mdt_obd_name(info->mti_mdt),
2441                        PFID(mdt_object_fid(o)));
2442                 RETURN(-EPROTO);
2443         }
2444
2445         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2446         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2447                    MDS_INODELOCK_XATTR);
2448         if (ibits & MDS_INODELOCK_UPDATE) {
2449                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2450                  * enqueue RPC can only request 1 lock, to avoid extra
2451                  * RPC, so it will instead enqueue EX lock for remote
2452                  * object anyway XXX*/
2453                 if (lh->mlh_type == MDT_PDO_LOCK &&
2454                     lh->mlh_pdo_hash != 0) {
2455                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2456                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2457                                PFID(mdt_object_fid(o)));
2458                         lh->mlh_pdo_hash = 0;
2459                         lh->mlh_rreg_mode = LCK_EX;
2460                         lh->mlh_type = MDT_REG_LOCK;
2461                 }
2462                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2463                                             &lh->mlh_rreg_lh,
2464                                             lh->mlh_rreg_mode,
2465                                             MDS_INODELOCK_UPDATE);
2466                 if (rc != ELDLM_OK)
2467                         RETURN(rc);
2468         }
2469
2470         /* Only enqueue LOOKUP lock for remote object */
2471         if (ibits & MDS_INODELOCK_LOOKUP) {
2472                 rc = mdt_object_local_lock(info, o, lh,
2473                                            MDS_INODELOCK_LOOKUP,
2474                                            nonblock, locality);
2475                 if (rc != ELDLM_OK)
2476                         RETURN(rc);
2477         }
2478
2479         RETURN(0);
2480 }
2481
2482 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2483                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2484 {
2485         return mdt_object_lock_internal(info, o, lh, ibits, false, locality);
2486 }
2487
2488 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2489                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2490 {
2491         struct mdt_lock_handle tmp = *lh;
2492         int rc;
2493
2494         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality);
2495         if (rc == 0)
2496                 *lh = tmp;
2497
2498         return rc == 0;
2499 }
2500
2501 /**
2502  * Save a lock within request object.
2503  *
2504  * Keep the lock referenced until whether client ACK or transaction
2505  * commit happens or release the lock immediately depending on input
2506  * parameters. If COS is ON, a write lock is converted to COS lock
2507  * before saving.
2508  *
2509  * \param info thead info object
2510  * \param h lock handle
2511  * \param mode lock mode
2512  * \param decref force immediate lock releasing
2513  */
2514 static
2515 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2516                    ldlm_mode_t mode, int decref)
2517 {
2518         ENTRY;
2519
2520         if (lustre_handle_is_used(h)) {
2521                 if (decref || !info->mti_has_trans ||
2522                     !(mode & (LCK_PW | LCK_EX))){
2523                         mdt_fid_unlock(h, mode);
2524                 } else {
2525                         struct mdt_device *mdt = info->mti_mdt;
2526                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2527                         struct ptlrpc_request *req = mdt_info_req(info);
2528                         int no_ack = 0;
2529
2530                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2531                                  h->cookie);
2532                         /* there is no request if mdt_object_unlock() is called
2533                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2534                         if (likely(req != NULL)) {
2535                                 CDEBUG(D_HA, "request = %p reply state = %p"
2536                                        " transno = "LPD64"\n", req,
2537                                        req->rq_reply_state, req->rq_transno);
2538                                 if (mdt_cos_is_enabled(mdt)) {
2539                                         no_ack = 1;
2540                                         ldlm_lock_downgrade(lock, LCK_COS);
2541                                         mode = LCK_COS;
2542                                 }
2543                                 ptlrpc_save_lock(req, h, mode, no_ack);
2544                         } else {
2545                                 ldlm_lock_decref(h, mode);
2546                         }
2547                         if (mdt_is_lock_sync(lock)) {
2548                                 CDEBUG(D_HA, "found sync-lock,"
2549                                        " async commit started\n");
2550                                 mdt_device_commit_async(info->mti_env,
2551                                                         mdt);
2552                         }
2553                         LDLM_LOCK_PUT(lock);
2554                 }
2555                 h->cookie = 0ull;
2556         }
2557
2558         EXIT;
2559 }
2560
2561 /**
2562  * Unlock mdt object.
2563  *
2564  * Immeditely release the regular lock and the PDO lock or save the
2565  * lock in request and keep them referenced until client ACK or
2566  * transaction commit.
2567  *
2568  * \param info thread info object
2569  * \param o mdt object
2570  * \param lh mdt lock handle referencing regular and PDO locks
2571  * \param decref force immediate lock releasing
2572  */
2573 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2574                        struct mdt_lock_handle *lh, int decref)
2575 {
2576         ENTRY;
2577
2578         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2579         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2580
2581         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2582                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2583
2584         EXIT;
2585 }
2586
2587 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2588                                         const struct lu_fid *f,
2589                                         struct mdt_lock_handle *lh,
2590                                         __u64 ibits)
2591 {
2592         struct mdt_object *o;
2593
2594         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2595         if (!IS_ERR(o)) {
2596                 int rc;
2597
2598                 rc = mdt_object_lock(info, o, lh, ibits,
2599                                      MDT_LOCAL_LOCK);
2600                 if (rc != 0) {
2601                         mdt_object_put(info->mti_env, o);
2602                         o = ERR_PTR(rc);
2603                 }
2604         }
2605         return o;
2606 }
2607
2608 void mdt_object_unlock_put(struct mdt_thread_info * info,
2609                            struct mdt_object * o,
2610                            struct mdt_lock_handle *lh,
2611                            int decref)
2612 {
2613         mdt_object_unlock(info, o, lh, decref);
2614         mdt_object_put(info->mti_env, o);
2615 }
2616
2617 /*
2618  * Generic code handling requests that have struct mdt_body passed in:
2619  *
2620  *  - extract mdt_body from request and save it in @info, if present;
2621  *
2622  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2623  *  @info;
2624  *
2625  *  - if HABEO_CORPUS flag is set for this request type check whether object
2626  *  actually exists on storage (lu_object_exists()).
2627  *
2628  */
2629 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2630 {
2631         const struct mdt_body    *body;
2632         struct mdt_object        *obj;
2633         const struct lu_env      *env;
2634         struct req_capsule       *pill;
2635         int                       rc;
2636         ENTRY;
2637
2638         env = info->mti_env;
2639         pill = info->mti_pill;
2640
2641         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2642         if (body == NULL)
2643                 RETURN(-EFAULT);
2644
2645         if (!(body->mbo_valid & OBD_MD_FLID))
2646                 RETURN(0);
2647
2648         if (!fid_is_sane(&body->mbo_fid1)) {
2649                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2650                 RETURN(-EINVAL);
2651         }
2652
2653         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2654         if (!IS_ERR(obj)) {
2655                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
2656                         mdt_object_put(env, obj);
2657                         rc = -ENOENT;
2658                 } else {
2659                         info->mti_object = obj;
2660                         rc = 0;
2661                 }
2662         } else
2663                 rc = PTR_ERR(obj);
2664
2665         RETURN(rc);
2666 }
2667
2668 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2669 {
2670         struct req_capsule *pill = info->mti_pill;
2671         int rc;
2672         ENTRY;
2673
2674         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2675                 rc = mdt_body_unpack(info, flags);
2676         else
2677                 rc = 0;
2678
2679         if (rc == 0 && (flags & HABEO_REFERO)) {
2680                 /* Pack reply. */
2681                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2682                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2683                                              DEF_REP_MD_SIZE);
2684                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2685                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2686                                              RCL_SERVER, 0);
2687
2688                 rc = req_capsule_server_pack(pill);
2689         }
2690         RETURN(rc);
2691 }
2692
2693 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2694 {
2695         lh->mlh_type = MDT_NUL_LOCK;
2696         lh->mlh_reg_lh.cookie = 0ull;
2697         lh->mlh_reg_mode = LCK_MINMODE;
2698         lh->mlh_pdo_lh.cookie = 0ull;
2699         lh->mlh_pdo_mode = LCK_MINMODE;
2700         lh->mlh_rreg_lh.cookie = 0ull;
2701         lh->mlh_rreg_mode = LCK_MINMODE;
2702 }
2703
2704 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2705 {
2706         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2707         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2708 }
2709
2710 /*
2711  * Initialize fields of struct mdt_thread_info. Other fields are left in
2712  * uninitialized state, because it's too expensive to zero out whole
2713  * mdt_thread_info (> 1K) on each request arrival.
2714  */
2715 void mdt_thread_info_init(struct ptlrpc_request *req,
2716                           struct mdt_thread_info *info)
2717 {
2718         int i;
2719
2720         info->mti_pill = &req->rq_pill;
2721
2722         /* lock handle */
2723         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2724                 mdt_lock_handle_init(&info->mti_lh[i]);
2725
2726         /* mdt device: it can be NULL while CONNECT */
2727         if (req->rq_export) {
2728                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2729                 info->mti_exp = req->rq_export;
2730         } else
2731                 info->mti_mdt = NULL;
2732         info->mti_env = req->rq_svc_thread->t_env;
2733         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2734
2735         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2736         info->mti_big_buf = LU_BUF_NULL;
2737         info->mti_body = NULL;
2738         info->mti_object = NULL;
2739         info->mti_dlm_req = NULL;
2740         info->mti_has_trans = 0;
2741         info->mti_cross_ref = 0;
2742         info->mti_opdata = 0;
2743         info->mti_big_lmm_used = 0;
2744
2745         info->mti_spec.no_create = 0;
2746         info->mti_spec.sp_rm_entry = 0;
2747         info->mti_spec.sp_permitted = 0;
2748
2749         info->mti_spec.u.sp_ea.eadata = NULL;
2750         info->mti_spec.u.sp_ea.eadatalen = 0;
2751 }
2752
2753 void mdt_thread_info_fini(struct mdt_thread_info *info)
2754 {
2755         int i;
2756
2757         if (info->mti_object != NULL) {
2758                 mdt_object_put(info->mti_env, info->mti_object);
2759                 info->mti_object = NULL;
2760         }
2761
2762         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2763                 mdt_lock_handle_fini(&info->mti_lh[i]);
2764         info->mti_env = NULL;
2765         info->mti_pill = NULL;
2766         info->mti_exp = NULL;
2767
2768         if (unlikely(info->mti_big_buf.lb_buf != NULL))
2769                 lu_buf_free(&info->mti_big_buf);
2770 }
2771
2772 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2773 {
2774         struct mdt_thread_info  *mti;
2775
2776         mti = mdt_th_info(tsi->tsi_env);
2777         LASSERT(mti != NULL);
2778
2779         mdt_thread_info_init(tgt_ses_req(tsi), mti);
2780         if (tsi->tsi_corpus != NULL) {
2781                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2782                 lu_object_get(tsi->tsi_corpus);
2783         }
2784         mti->mti_body = tsi->tsi_mdt_body;
2785         mti->mti_dlm_req = tsi->tsi_dlm_req;
2786
2787         return mti;
2788 }
2789
2790 static int mdt_tgt_connect(struct tgt_session_info *tsi)
2791 {
2792         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2793         int                      rc;
2794
2795         ENTRY;
2796
2797         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
2798             cfs_fail_val ==
2799             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
2800                 set_current_state(TASK_UNINTERRUPTIBLE);
2801                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
2802         }
2803
2804         rc = tgt_connect(tsi);
2805         if (rc != 0)
2806                 RETURN(rc);
2807
2808         rc = mdt_init_idmap(tsi);
2809         if (rc != 0)
2810                 GOTO(err, rc);
2811         RETURN(0);
2812 err:
2813         obd_disconnect(class_export_get(req->rq_export));
2814         return rc;
2815 }
2816
2817 enum mdt_it_code {
2818         MDT_IT_OPEN,
2819         MDT_IT_OCREAT,
2820         MDT_IT_CREATE,
2821         MDT_IT_GETATTR,
2822         MDT_IT_READDIR,
2823         MDT_IT_LOOKUP,
2824         MDT_IT_UNLINK,
2825         MDT_IT_TRUNC,
2826         MDT_IT_GETXATTR,
2827         MDT_IT_LAYOUT,
2828         MDT_IT_QUOTA,
2829         MDT_IT_NR
2830 };
2831
2832 static int mdt_intent_getattr(enum mdt_it_code opcode,
2833                               struct mdt_thread_info *info,
2834                               struct ldlm_lock **,
2835                               __u64);
2836
2837 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2838                                 struct mdt_thread_info *info,
2839                                 struct ldlm_lock **lockp,
2840                                 __u64 flags);
2841
2842 static int mdt_intent_layout(enum mdt_it_code opcode,
2843                              struct mdt_thread_info *info,
2844                              struct ldlm_lock **,
2845                              __u64);
2846 static int mdt_intent_reint(enum mdt_it_code opcode,
2847                             struct mdt_thread_info *info,
2848                             struct ldlm_lock **,
2849                             __u64);
2850
2851 static struct mdt_it_flavor {
2852         const struct req_format *it_fmt;
2853         __u32                    it_flags;
2854         int                    (*it_act)(enum mdt_it_code ,
2855                                          struct mdt_thread_info *,
2856                                          struct ldlm_lock **,
2857                                          __u64);
2858         long                     it_reint;
2859 } mdt_it_flavor[] = {
2860         [MDT_IT_OPEN]     = {
2861                 .it_fmt   = &RQF_LDLM_INTENT,
2862                 /*.it_flags = HABEO_REFERO,*/
2863                 .it_flags = 0,
2864                 .it_act   = mdt_intent_reint,
2865                 .it_reint = REINT_OPEN
2866         },
2867         [MDT_IT_OCREAT]   = {
2868                 .it_fmt   = &RQF_LDLM_INTENT,
2869                 /*
2870                  * OCREAT is not a MUTABOR request as if the file
2871                  * already exists.
2872                  * We do the extra check of OBD_CONNECT_RDONLY in
2873                  * mdt_reint_open() when we really need to create
2874                  * the object.
2875                  */
2876                 .it_flags = 0,
2877                 .it_act   = mdt_intent_reint,
2878                 .it_reint = REINT_OPEN
2879         },
2880         [MDT_IT_CREATE]   = {
2881                 .it_fmt   = &RQF_LDLM_INTENT,
2882                 .it_flags = MUTABOR,
2883                 .it_act   = mdt_intent_reint,
2884                 .it_reint = REINT_CREATE
2885         },
2886         [MDT_IT_GETATTR]  = {
2887                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2888                 .it_flags = HABEO_REFERO,
2889                 .it_act   = mdt_intent_getattr
2890         },
2891         [MDT_IT_READDIR]  = {
2892                 .it_fmt   = NULL,
2893                 .it_flags = 0,
2894                 .it_act   = NULL
2895         },
2896         [MDT_IT_LOOKUP]   = {
2897                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2898                 .it_flags = HABEO_REFERO,
2899                 .it_act   = mdt_intent_getattr
2900         },
2901         [MDT_IT_UNLINK]   = {
2902                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2903                 .it_flags = MUTABOR,
2904                 .it_act   = NULL,
2905                 .it_reint = REINT_UNLINK
2906         },
2907         [MDT_IT_TRUNC]    = {
2908                 .it_fmt   = NULL,
2909                 .it_flags = MUTABOR,
2910                 .it_act   = NULL
2911         },
2912         [MDT_IT_GETXATTR] = {
2913                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
2914                 .it_flags = HABEO_CORPUS,
2915                 .it_act   = mdt_intent_getxattr
2916         },
2917         [MDT_IT_LAYOUT] = {
2918                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
2919                 .it_flags = 0,
2920                 .it_act   = mdt_intent_layout
2921         }
2922 };
2923
2924 static int
2925 mdt_intent_lock_replace(struct mdt_thread_info *info,
2926                         struct ldlm_lock **lockp,
2927                         struct mdt_lock_handle *lh,
2928                         __u64 flags)
2929 {
2930         struct ptlrpc_request  *req = mdt_info_req(info);
2931         struct ldlm_lock       *lock = *lockp;
2932         struct ldlm_lock       *new_lock;
2933
2934         /* If possible resent found a lock, @lh is set to its handle */
2935         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
2936
2937         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2938                 lh->mlh_reg_lh.cookie = 0;
2939                 RETURN(0);
2940         }
2941
2942         LASSERTF(new_lock != NULL,
2943                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2944
2945         /*
2946          * If we've already given this lock to a client once, then we should
2947          * have no readers or writers.  Otherwise, we should have one reader
2948          * _or_ writer ref (which will be zeroed below) before returning the
2949          * lock to a client.
2950          */
2951         if (new_lock->l_export == req->rq_export) {
2952                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2953         } else {
2954                 LASSERT(new_lock->l_export == NULL);
2955                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2956         }
2957
2958         *lockp = new_lock;
2959
2960         if (new_lock->l_export == req->rq_export) {
2961                 /*
2962                  * Already gave this to the client, which means that we
2963                  * reconstructed a reply.
2964                  */
2965                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2966                         MSG_RESENT);
2967
2968                 LDLM_LOCK_RELEASE(new_lock);
2969                 lh->mlh_reg_lh.cookie = 0;
2970                 RETURN(ELDLM_LOCK_REPLACED);
2971         }
2972
2973         /*
2974          * Fixup the lock to be given to the client.
2975          */
2976         lock_res_and_lock(new_lock);
2977         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
2978          * possible blocking AST. */
2979         while (new_lock->l_readers > 0) {
2980                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
2981                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2982                 new_lock->l_readers--;
2983         }
2984         while (new_lock->l_writers > 0) {
2985                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
2986                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2987                 new_lock->l_writers--;
2988         }
2989
2990         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
2991         new_lock->l_blocking_ast = lock->l_blocking_ast;
2992         new_lock->l_completion_ast = lock->l_completion_ast;
2993         new_lock->l_remote_handle = lock->l_remote_handle;
2994         new_lock->l_flags &= ~LDLM_FL_LOCAL;
2995
2996         unlock_res_and_lock(new_lock);
2997
2998         cfs_hash_add(new_lock->l_export->exp_lock_hash,
2999                      &new_lock->l_remote_handle,
3000                      &new_lock->l_exp_hash);
3001
3002         LDLM_LOCK_RELEASE(new_lock);
3003         lh->mlh_reg_lh.cookie = 0;
3004
3005         RETURN(ELDLM_LOCK_REPLACED);
3006 }
3007
3008 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3009                                     struct ldlm_lock *new_lock,
3010                                     struct mdt_lock_handle *lh,
3011                                     __u64 flags)
3012 {
3013         struct ptlrpc_request  *req = mdt_info_req(info);
3014         struct ldlm_request    *dlmreq;
3015
3016         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3017                 return;
3018
3019         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3020
3021         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3022          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3023          * initialized. */
3024         if (flags & LDLM_FL_RESENT) {
3025                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3026                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3027
3028                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3029                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64,
3030                           lh->mlh_reg_lh.cookie);
3031                 return;
3032         }
3033
3034         /*
3035          * If the xid matches, then we know this is a resent request, and allow
3036          * it. (It's probably an OPEN, for which we don't send a lock.
3037          */
3038         if (req_xid_is_last(req))
3039                 return;
3040
3041         /*
3042          * This remote handle isn't enqueued, so we never received or processed
3043          * this request.  Clear MSG_RESENT, because it can be handled like any
3044          * normal request now.
3045          */
3046         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3047
3048         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3049                   dlmreq->lock_handle[0].cookie);
3050 }
3051
3052 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3053                                 struct mdt_thread_info *info,
3054                                 struct ldlm_lock **lockp,
3055                                 __u64 flags)
3056 {
3057         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3058         struct ldlm_reply      *ldlm_rep = NULL;
3059         int rc, grc;
3060
3061         /*
3062          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3063          * (for the resend case) or a new lock. Below we will use it to
3064          * replace the original lock.
3065          */
3066         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3067         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3068                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3069                 rc = mdt_object_lock(info, info->mti_object, lhc,
3070                                         MDS_INODELOCK_XATTR,
3071                                         MDT_LOCAL_LOCK);
3072                 if (rc)
3073                         return rc;
3074         }
3075
3076         grc = mdt_getxattr(info);
3077
3078         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3079
3080         if (mdt_info_req(info)->rq_repmsg != NULL)
3081                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3082         if (ldlm_rep == NULL)
3083                 RETURN(err_serious(-EFAULT));
3084
3085         ldlm_rep->lock_policy_res2 = grc;
3086
3087         return rc;
3088 }
3089
3090 static int mdt_intent_getattr(enum mdt_it_code opcode,
3091                               struct mdt_thread_info *info,
3092                               struct ldlm_lock **lockp,
3093                               __u64 flags)
3094 {
3095         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3096         __u64                   child_bits;
3097         struct ldlm_reply      *ldlm_rep;
3098         struct mdt_body        *reqbody;
3099         struct mdt_body        *repbody;
3100         int                     rc, rc2;
3101         ENTRY;
3102
3103         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3104         LASSERT(reqbody);
3105
3106         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3107         LASSERT(repbody);
3108
3109         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3110         repbody->mbo_eadatasize = 0;
3111         repbody->mbo_aclsize = 0;
3112
3113         switch (opcode) {
3114         case MDT_IT_LOOKUP:
3115                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3116                 break;
3117         case MDT_IT_GETATTR:
3118                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3119                              MDS_INODELOCK_PERM;
3120                 break;
3121         default:
3122                 CERROR("Unsupported intent (%d)\n", opcode);
3123                 GOTO(out_shrink, rc = -EINVAL);
3124         }
3125
3126         rc = mdt_init_ucred(info, reqbody);
3127         if (rc)
3128                 GOTO(out_shrink, rc);
3129
3130         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3131         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3132
3133         /* Get lock from request for possible resent case. */
3134         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3135
3136         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3137         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3138
3139         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3140                 ldlm_rep->lock_policy_res2 = 0;
3141         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3142             ldlm_rep->lock_policy_res2) {
3143                 lhc->mlh_reg_lh.cookie = 0ull;
3144                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3145         }
3146
3147         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3148         EXIT;
3149 out_ucred:
3150         mdt_exit_ucred(info);
3151 out_shrink:
3152         mdt_client_compatibility(info);
3153         rc2 = mdt_fix_reply(info);
3154         if (rc == 0)
3155                 rc = rc2;
3156         return rc;
3157 }
3158
3159 static int mdt_intent_layout(enum mdt_it_code opcode,
3160                              struct mdt_thread_info *info,
3161                              struct ldlm_lock **lockp,
3162                              __u64 flags)
3163 {
3164         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3165         struct layout_intent *layout;
3166         struct lu_fid *fid;
3167         struct mdt_object *obj = NULL;
3168         int rc = 0;
3169         ENTRY;
3170
3171         if (opcode != MDT_IT_LAYOUT) {
3172                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3173                         opcode);
3174                 RETURN(-EINVAL);
3175         }
3176
3177         fid = &info->mti_tmp_fid2;
3178         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3179
3180         /* Get lock from request for possible resent case. */
3181         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3182
3183         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3184         if (IS_ERR(obj))
3185                 RETURN(PTR_ERR(obj));
3186
3187         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3188                 /* get the length of lsm */
3189                 rc = mdt_attr_get_eabuf_size(info, obj);
3190                 if (rc < 0) {
3191                         mdt_object_put(info->mti_env, obj);
3192                         RETURN(rc);
3193                 }
3194
3195                 if (rc > info->mti_mdt->mdt_max_mdsize)
3196