Whamcloud - gitweb
LU-12212 mdt: fix SECCTX reply buffer handling
[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.gnu.org/licenses/gpl-2.0.html
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2010, 2017, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  * Lustre is a trademark of Sun Microsystems, Inc.
31  *
32  * lustre/mdt/mdt_handler.c
33  *
34  * Lustre Metadata Target (mdt) request handler
35  *
36  * Author: Peter Braam <braam@clusterfs.com>
37  * Author: Andreas Dilger <adilger@clusterfs.com>
38  * Author: Phil Schwan <phil@clusterfs.com>
39  * Author: Mike Shaver <shaver@clusterfs.com>
40  * Author: Nikita Danilov <nikita@clusterfs.com>
41  * Author: Huang Hua <huanghua@clusterfs.com>
42  * Author: Yury Umanets <umka@clusterfs.com>
43  */
44
45 #define DEBUG_SUBSYSTEM S_MDS
46
47 #include <linux/module.h>
48 #include <linux/pagemap.h>
49
50 #include <dt_object.h>
51 #include <lustre_acl.h>
52 #include <lustre_export.h>
53 #include <uapi/linux/lustre/lustre_ioctl.h>
54 #include <lustre_lfsck.h>
55 #include <lustre_log.h>
56 #include <lustre_nodemap.h>
57 #include <lustre_mds.h>
58 #include <uapi/linux/lustre/lustre_param.h>
59 #include <lustre_quota.h>
60 #include <lustre_swab.h>
61 #include <obd.h>
62 #include <obd_support.h>
63 #include <lustre_barrier.h>
64 #include <obd_cksum.h>
65 #include <llog_swab.h>
66
67 #include "mdt_internal.h"
68
69 static unsigned int max_mod_rpcs_per_client = 8;
70 module_param(max_mod_rpcs_per_client, uint, 0644);
71 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
72
73 mdl_mode_t mdt_mdl_lock_modes[] = {
74         [LCK_MINMODE] = MDL_MINMODE,
75         [LCK_EX]      = MDL_EX,
76         [LCK_PW]      = MDL_PW,
77         [LCK_PR]      = MDL_PR,
78         [LCK_CW]      = MDL_CW,
79         [LCK_CR]      = MDL_CR,
80         [LCK_NL]      = MDL_NL,
81         [LCK_GROUP]   = MDL_GROUP
82 };
83
84 enum ldlm_mode mdt_dlm_lock_modes[] = {
85         [MDL_MINMODE]   = LCK_MINMODE,
86         [MDL_EX]        = LCK_EX,
87         [MDL_PW]        = LCK_PW,
88         [MDL_PR]        = LCK_PR,
89         [MDL_CW]        = LCK_CW,
90         [MDL_CR]        = LCK_CR,
91         [MDL_NL]        = LCK_NL,
92         [MDL_GROUP]     = LCK_GROUP
93 };
94
95 static struct mdt_device *mdt_dev(struct lu_device *d);
96
97 static const struct lu_object_operations mdt_obj_ops;
98
99 /* Slab for MDT object allocation */
100 static struct kmem_cache *mdt_object_kmem;
101
102 /* For HSM restore handles */
103 struct kmem_cache *mdt_hsm_cdt_kmem;
104
105 /* For HSM request handles */
106 struct kmem_cache *mdt_hsm_car_kmem;
107
108 static struct lu_kmem_descr mdt_caches[] = {
109         {
110                 .ckd_cache = &mdt_object_kmem,
111                 .ckd_name  = "mdt_obj",
112                 .ckd_size  = sizeof(struct mdt_object)
113         },
114         {
115                 .ckd_cache      = &mdt_hsm_cdt_kmem,
116                 .ckd_name       = "mdt_cdt_restore_handle",
117                 .ckd_size       = sizeof(struct cdt_restore_handle)
118         },
119         {
120                 .ckd_cache      = &mdt_hsm_car_kmem,
121                 .ckd_name       = "mdt_cdt_agent_req",
122                 .ckd_size       = sizeof(struct cdt_agent_req)
123         },
124         {
125                 .ckd_cache = NULL
126         }
127 };
128
129 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
130 {
131         if (!rep)
132                 return 0;
133         return rep->lock_policy_res1 & op_flag;
134 }
135
136 void mdt_clear_disposition(struct mdt_thread_info *info,
137                            struct ldlm_reply *rep, __u64 op_flag)
138 {
139         if (info) {
140                 info->mti_opdata &= ~op_flag;
141                 tgt_opdata_clear(info->mti_env, op_flag);
142         }
143         if (rep)
144                 rep->lock_policy_res1 &= ~op_flag;
145 }
146
147 void mdt_set_disposition(struct mdt_thread_info *info,
148                          struct ldlm_reply *rep, __u64 op_flag)
149 {
150         if (info) {
151                 info->mti_opdata |= op_flag;
152                 tgt_opdata_set(info->mti_env, op_flag);
153         }
154         if (rep)
155                 rep->lock_policy_res1 |= op_flag;
156 }
157
158 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
159 {
160         lh->mlh_pdo_hash = 0;
161         lh->mlh_reg_mode = lm;
162         lh->mlh_rreg_mode = lm;
163         lh->mlh_type = MDT_REG_LOCK;
164 }
165
166 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
167                        const struct lu_name *lname)
168 {
169         lh->mlh_reg_mode = lock_mode;
170         lh->mlh_pdo_mode = LCK_MINMODE;
171         lh->mlh_rreg_mode = lock_mode;
172         lh->mlh_type = MDT_PDO_LOCK;
173
174         if (lu_name_is_valid(lname)) {
175                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
176                                                      lname->ln_namelen);
177                 /* XXX Workaround for LU-2856
178                  *
179                  * Zero is a valid return value of full_name_hash, but
180                  * several users of mlh_pdo_hash assume a non-zero
181                  * hash value. We therefore map zero onto an
182                  * arbitrary, but consistent value (1) to avoid
183                  * problems further down the road. */
184                 if (unlikely(lh->mlh_pdo_hash == 0))
185                         lh->mlh_pdo_hash = 1;
186         } else {
187                 lh->mlh_pdo_hash = 0;
188         }
189 }
190
191 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
192                               struct mdt_lock_handle *lh)
193 {
194         mdl_mode_t mode;
195         ENTRY;
196
197         /*
198          * Any dir access needs couple of locks:
199          *
200          * 1) on part of dir we gonna take lookup/modify;
201          *
202          * 2) on whole dir to protect it from concurrent splitting and/or to
203          * flush client's cache for readdir().
204          *
205          * so, for a given mode and object this routine decides what lock mode
206          * to use for lock #2:
207          *
208          * 1) if caller's gonna lookup in dir then we need to protect dir from
209          * being splitted only - LCK_CR
210          *
211          * 2) if caller's gonna modify dir then we need to protect dir from
212          * being splitted and to flush cache - LCK_CW
213          *
214          * 3) if caller's gonna modify dir and that dir seems ready for
215          * splitting then we need to protect it from any type of access
216          * (lookup/modify/split) - LCK_EX --bzzz
217          */
218
219         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
220         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
221
222         /*
223          * Ask underlaying level its opinion about preferable PDO lock mode
224          * having access type passed as regular lock mode:
225          *
226          * - MDL_MINMODE means that lower layer does not want to specify lock
227          * mode;
228          *
229          * - MDL_NL means that no PDO lock should be taken. This is used in some
230          * cases. Say, for non-splittable directories no need to use PDO locks
231          * at all.
232          */
233         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
234                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
235
236         if (mode != MDL_MINMODE) {
237                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
238         } else {
239                 /*
240                  * Lower layer does not want to specify locking mode. We do it
241                  * our selves. No special protection is needed, just flush
242                  * client's cache on modification and allow concurrent
243                  * mondification.
244                  */
245                 switch (lh->mlh_reg_mode) {
246                 case LCK_EX:
247                         lh->mlh_pdo_mode = LCK_EX;
248                         break;
249                 case LCK_PR:
250                         lh->mlh_pdo_mode = LCK_CR;
251                         break;
252                 case LCK_PW:
253                         lh->mlh_pdo_mode = LCK_CW;
254                         break;
255                 default:
256                         CERROR("Not expected lock type (0x%x)\n",
257                                (int)lh->mlh_reg_mode);
258                         LBUG();
259                 }
260         }
261
262         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
263         EXIT;
264 }
265
266 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
267                               struct lu_fid *fid)
268 {
269         struct mdt_device *mdt = info->mti_mdt;
270         struct lu_name *lname = &info->mti_name;
271         char *filename = info->mti_filename;
272         struct mdt_object *parent;
273         u32 mode;
274         int rc = 0;
275
276         LASSERT(!info->mti_cross_ref);
277
278         /*
279          * We may want to allow this to mount a completely separate
280          * fileset from the MDT in the future, but keeping it to
281          * ROOT/ only for now avoid potential security issues.
282          */
283         *fid = mdt->mdt_md_root_fid;
284
285         while (rc == 0 && fileset != NULL && *fileset != '\0') {
286                 const char *s1 = fileset;
287                 const char *s2;
288
289                 while (*++s1 == '/')
290                         ;
291                 s2 = s1;
292                 while (*s2 != '/' && *s2 != '\0')
293                         s2++;
294
295                 if (s2 == s1)
296                         break;
297
298                 fileset = s2;
299
300                 lname->ln_namelen = s2 - s1;
301                 if (lname->ln_namelen > NAME_MAX) {
302                         rc = -EINVAL;
303                         break;
304                 }
305
306                 /* reject .. as a path component */
307                 if (lname->ln_namelen == 2 &&
308                     strncmp(s1, "..", 2) == 0) {
309                         rc = -EINVAL;
310                         break;
311                 }
312
313                 strncpy(filename, s1, lname->ln_namelen);
314                 filename[lname->ln_namelen] = '\0';
315                 lname->ln_name = filename;
316
317                 parent = mdt_object_find(info->mti_env, mdt, fid);
318                 if (IS_ERR(parent)) {
319                         rc = PTR_ERR(parent);
320                         break;
321                 }
322                 /* Only got the fid of this obj by name */
323                 fid_zero(fid);
324                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
325                                 fid, &info->mti_spec);
326                 mdt_object_put(info->mti_env, parent);
327         }
328         if (!rc) {
329                 parent = mdt_object_find(info->mti_env, mdt, fid);
330                 if (IS_ERR(parent))
331                         rc = PTR_ERR(parent);
332                 else {
333                         mode = lu_object_attr(&parent->mot_obj);
334                         mdt_object_put(info->mti_env, parent);
335                         if (!S_ISDIR(mode))
336                                 rc = -ENOTDIR;
337                 }
338         }
339
340         return rc;
341 }
342
343 static int mdt_get_root(struct tgt_session_info *tsi)
344 {
345         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
346         struct mdt_device       *mdt = info->mti_mdt;
347         struct mdt_body         *repbody;
348         char                    *fileset = NULL, *buffer = NULL;
349         int                      rc;
350         struct obd_export       *exp = info->mti_exp;
351         char                    *nodemap_fileset;
352
353         ENTRY;
354
355         rc = mdt_check_ucred(info);
356         if (rc)
357                 GOTO(out, rc = err_serious(rc));
358
359         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
360                 GOTO(out, rc = err_serious(-ENOMEM));
361
362         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
363         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
364                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
365                 if (fileset == NULL)
366                         GOTO(out, rc = err_serious(-EFAULT));
367         }
368
369         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
370         if (nodemap_fileset && nodemap_fileset[0]) {
371                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
372                 if (fileset) {
373                         /* consider fileset from client as a sub-fileset
374                          * of the nodemap one */
375                         OBD_ALLOC(buffer, PATH_MAX + 1);
376                         if (buffer == NULL)
377                                 GOTO(out, rc = err_serious(-ENOMEM));
378                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
379                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
380                                 GOTO(out, rc = err_serious(-EINVAL));
381                         fileset = buffer;
382                 } else {
383                         /* enforce fileset as specified in the nodemap */
384                         fileset = nodemap_fileset;
385                 }
386         }
387
388         if (fileset) {
389                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
390                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
391                 if (rc < 0)
392                         GOTO(out, rc = err_serious(rc));
393         } else {
394                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
395         }
396         repbody->mbo_valid |= OBD_MD_FLID;
397
398         EXIT;
399 out:
400         mdt_thread_info_fini(info);
401         if (buffer)
402                 OBD_FREE(buffer, PATH_MAX+1);
403         return rc;
404 }
405
406 static int mdt_statfs(struct tgt_session_info *tsi)
407 {
408         struct ptlrpc_request *req = tgt_ses_req(tsi);
409         struct mdt_thread_info *info = tsi2mdt_info(tsi);
410         struct mdt_device *mdt = info->mti_mdt;
411         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
412         struct md_device *next = mdt->mdt_child;
413         struct ptlrpc_service_part *svcpt;
414         struct obd_statfs *osfs;
415         struct mdt_body *reqbody = NULL;
416         struct mdt_statfs_cache *msf;
417         int rc;
418
419         ENTRY;
420
421         svcpt = req->rq_rqbd->rqbd_svcpt;
422
423         /* This will trigger a watchdog timeout */
424         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
425                          (MDT_SERVICE_WATCHDOG_FACTOR *
426                           at_get(&svcpt->scp_at_estimate)) + 1);
427
428         rc = mdt_check_ucred(info);
429         if (rc)
430                 GOTO(out, rc = err_serious(rc));
431
432         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
433                 GOTO(out, rc = err_serious(-ENOMEM));
434
435         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
436         if (!osfs)
437                 GOTO(out, rc = -EPROTO);
438
439         if (mdt_is_sum_statfs_client(req->rq_export) &&
440                 lustre_packed_msg_size(req->rq_reqmsg) ==
441                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
442                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
443                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
444                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
445         }
446
447         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
448                 msf = &mdt->mdt_sum_osfs;
449         else
450                 msf = &mdt->mdt_osfs;
451
452         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
453                         /** statfs data is too old, get up-to-date one */
454                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
455                                 rc = next->md_ops->mdo_statfs(info->mti_env,
456                                                               next, osfs);
457                         else
458                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
459                                                osfs);
460                         if (rc)
461                                 GOTO(out, rc);
462                         spin_lock(&mdt->mdt_lock);
463                         msf->msf_osfs = *osfs;
464                         msf->msf_age = ktime_get_seconds();
465                         spin_unlock(&mdt->mdt_lock);
466         } else {
467                         /** use cached statfs data */
468                         spin_lock(&mdt->mdt_lock);
469                         *osfs = msf->msf_osfs;
470                         spin_unlock(&mdt->mdt_lock);
471         }
472
473         /* at least try to account for cached pages.  its still racy and
474          * might be under-reporting if clients haven't announced their
475          * caches with brw recently */
476         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
477                " pending %llu free %llu avail %llu\n",
478                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
479                tgd->tgd_tot_pending,
480                osfs->os_bfree << tgd->tgd_blockbits,
481                osfs->os_bavail << tgd->tgd_blockbits);
482
483         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
484                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
485                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
486
487         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
488         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
489                "%llu objects: %llu free; state %x\n",
490                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
491                osfs->os_files, osfs->os_ffree, osfs->os_state);
492
493         if (!exp_grant_param_supp(tsi->tsi_exp) &&
494             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
495                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
496                  * should not see a block size > page size, otherwise
497                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
498                  * block size which is the biggest block size known to work
499                  * with all client's page size. */
500                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
501                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
502                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
503                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
504         }
505         if (rc == 0)
506                 mdt_counter_incr(req, LPROC_MDT_STATFS);
507 out:
508         mdt_thread_info_fini(info);
509         RETURN(rc);
510 }
511
512 /**
513  * Pack size attributes into the reply.
514  */
515 int mdt_pack_size2body(struct mdt_thread_info *info,
516                         const struct lu_fid *fid, struct lustre_handle *lh)
517 {
518         struct mdt_body *b;
519         struct md_attr *ma = &info->mti_attr;
520         int dom_stripe;
521         bool dom_lock = false;
522
523         ENTRY;
524
525         LASSERT(ma->ma_attr.la_valid & LA_MODE);
526
527         if (!S_ISREG(ma->ma_attr.la_mode) ||
528             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
529                 RETURN(-ENODATA);
530
531         dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
532         /* no DoM stripe, no size in reply */
533         if (dom_stripe == LMM_NO_DOM)
534                 RETURN(-ENOENT);
535
536         if (lustre_handle_is_used(lh)) {
537                 struct ldlm_lock *lock;
538
539                 lock = ldlm_handle2lock(lh);
540                 if (lock != NULL) {
541                         dom_lock = ldlm_has_dom(lock);
542                         LDLM_LOCK_PUT(lock);
543                 }
544         }
545
546         /* no DoM lock, no size in reply */
547         if (!dom_lock)
548                 RETURN(0);
549
550         /* Either DoM lock exists or LMM has only DoM stripe then
551          * return size on body. */
552         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
553
554         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
555         RETURN(0);
556 }
557
558 #ifdef CONFIG_FS_POSIX_ACL
559 /*
560  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
561  *
562  * \param       info    thread info object
563  * \param       repbody reply to pack ACLs into
564  * \param       o       mdt object of file to examine
565  * \param       nodemap nodemap of client to reply to
566  * \retval      0       success
567  * \retval      -errno  error getting or parsing ACL from disk
568  */
569 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
570                       struct mdt_object *o, struct lu_nodemap *nodemap)
571 {
572         const struct lu_env     *env = info->mti_env;
573         struct md_object        *next = mdt_object_child(o);
574         struct lu_buf           *buf = &info->mti_buf;
575         struct mdt_device       *mdt = info->mti_mdt;
576         struct req_capsule *pill = info->mti_pill;
577         int rc;
578
579         ENTRY;
580
581         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
582         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
583         if (buf->lb_len == 0)
584                 RETURN(0);
585
586 again:
587         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
588         if (rc < 0) {
589                 if (rc == -ENODATA) {
590                         repbody->mbo_aclsize = 0;
591                         repbody->mbo_valid |= OBD_MD_FLACL;
592                         rc = 0;
593                 } else if (rc == -EOPNOTSUPP) {
594                         rc = 0;
595                 } else {
596                         if (rc == -ERANGE &&
597                             exp_connect_large_acl(info->mti_exp) &&
598                             buf->lb_buf != info->mti_big_acl) {
599                                 if (info->mti_big_acl == NULL) {
600                                         info->mti_big_aclsize =
601                                                         MIN(mdt->mdt_max_ea_size,
602                                                             XATTR_SIZE_MAX);
603                                         OBD_ALLOC_LARGE(info->mti_big_acl,
604                                                         info->mti_big_aclsize);
605                                         if (info->mti_big_acl == NULL) {
606                                                 info->mti_big_aclsize = 0;
607                                                 CERROR("%s: unable to grow "
608                                                        DFID" ACL buffer\n",
609                                                        mdt_obd_name(mdt),
610                                                        PFID(mdt_object_fid(o)));
611                                                 RETURN(-ENOMEM);
612                                         }
613                                 }
614
615                                 CDEBUG(D_INODE, "%s: grow the "DFID
616                                        " ACL buffer to size %d\n",
617                                        mdt_obd_name(mdt),
618                                        PFID(mdt_object_fid(o)),
619                                        info->mti_big_aclsize);
620
621                                 buf->lb_buf = info->mti_big_acl;
622                                 buf->lb_len = info->mti_big_aclsize;
623
624                                 goto again;
625                         }
626
627                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
628                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
629                 }
630         } else {
631                 int client;
632                 int server;
633                 int acl_buflen;
634                 int lmm_buflen = 0;
635                 int lmmsize = 0;
636
637                 acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
638                 if (acl_buflen >= rc)
639                         goto map;
640
641                 /* If LOV/LMA EA is small, we can reuse part of their buffer */
642                 client = ptlrpc_req_get_repsize(pill->rc_req);
643                 server = lustre_packed_msg_size(pill->rc_req->rq_repmsg);
644                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
645                         lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD,
646                                                           RCL_SERVER);
647                         lmmsize = repbody->mbo_eadatasize;
648                 }
649
650                 if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) {
651                         CDEBUG(D_INODE, "%s: client prepared buffer size %d "
652                                "is not big enough with the ACL size %d (%d)\n",
653                                mdt_obd_name(mdt), client, rc,
654                                server - acl_buflen - lmm_buflen + rc + lmmsize);
655                         repbody->mbo_aclsize = 0;
656                         repbody->mbo_valid &= ~OBD_MD_FLACL;
657                         RETURN(-ERANGE);
658                 }
659
660 map:
661                 if (buf->lb_buf == info->mti_big_acl)
662                         info->mti_big_acl_used = 1;
663
664                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
665                                      rc, NODEMAP_FS_TO_CLIENT);
666                 /* if all ACLs mapped out, rc is still >= 0 */
667                 if (rc < 0) {
668                         CERROR("%s: nodemap_map_acl unable to parse "DFID
669                                " ACL: rc = %d\n", mdt_obd_name(mdt),
670                                PFID(mdt_object_fid(o)), rc);
671                         repbody->mbo_aclsize = 0;
672                         repbody->mbo_valid &= ~OBD_MD_FLACL;
673                 } else {
674                         repbody->mbo_aclsize = rc;
675                         repbody->mbo_valid |= OBD_MD_FLACL;
676                         rc = 0;
677                 }
678         }
679
680         RETURN(rc);
681 }
682 #endif
683
684 /* XXX Look into layout in MDT layer. */
685 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
686 {
687         struct lov_comp_md_v1   *comp_v1;
688         struct lov_mds_md       *v1;
689         int                      i;
690
691         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
692                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
693
694                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
695                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
696                                 comp_v1->lcm_entries[i].lcme_offset);
697                         /* We don't support partial release for now */
698                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
699                                 return false;
700                 }
701                 return true;
702         } else {
703                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
704                         true : false;
705         }
706 }
707
708 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
709                         const struct lu_attr *attr, const struct lu_fid *fid)
710 {
711         struct md_attr *ma = &info->mti_attr;
712         struct obd_export *exp = info->mti_exp;
713         struct lu_nodemap *nodemap = NULL;
714
715         LASSERT(ma->ma_valid & MA_INODE);
716
717         if (attr->la_valid & LA_ATIME) {
718                 b->mbo_atime = attr->la_atime;
719                 b->mbo_valid |= OBD_MD_FLATIME;
720         }
721         if (attr->la_valid & LA_MTIME) {
722                 b->mbo_mtime = attr->la_mtime;
723                 b->mbo_valid |= OBD_MD_FLMTIME;
724         }
725         if (attr->la_valid & LA_CTIME) {
726                 b->mbo_ctime = attr->la_ctime;
727                 b->mbo_valid |= OBD_MD_FLCTIME;
728         }
729         if (attr->la_valid & LA_FLAGS) {
730                 b->mbo_flags = attr->la_flags;
731                 b->mbo_valid |= OBD_MD_FLFLAGS;
732         }
733         if (attr->la_valid & LA_NLINK) {
734                 b->mbo_nlink = attr->la_nlink;
735                 b->mbo_valid |= OBD_MD_FLNLINK;
736         }
737         if (attr->la_valid & (LA_UID|LA_GID)) {
738                 nodemap = nodemap_get_from_exp(exp);
739                 if (IS_ERR(nodemap))
740                         goto out;
741         }
742         if (attr->la_valid & LA_UID) {
743                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
744                                             NODEMAP_FS_TO_CLIENT,
745                                             attr->la_uid);
746                 b->mbo_valid |= OBD_MD_FLUID;
747         }
748         if (attr->la_valid & LA_GID) {
749                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
750                                             NODEMAP_FS_TO_CLIENT,
751                                             attr->la_gid);
752                 b->mbo_valid |= OBD_MD_FLGID;
753         }
754
755         if (attr->la_valid & LA_PROJID) {
756                 /* TODO, nodemap for project id */
757                 b->mbo_projid = attr->la_projid;
758                 b->mbo_valid |= OBD_MD_FLPROJID;
759         }
760
761         b->mbo_mode = attr->la_mode;
762         if (attr->la_valid & LA_MODE)
763                 b->mbo_valid |= OBD_MD_FLMODE;
764         if (attr->la_valid & LA_TYPE)
765                 b->mbo_valid |= OBD_MD_FLTYPE;
766
767         if (fid != NULL) {
768                 b->mbo_fid1 = *fid;
769                 b->mbo_valid |= OBD_MD_FLID;
770                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
771                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
772         }
773
774         if (!(attr->la_valid & LA_TYPE))
775                 return;
776
777         b->mbo_rdev   = attr->la_rdev;
778         b->mbo_size   = attr->la_size;
779         b->mbo_blocks = attr->la_blocks;
780
781         if (!S_ISREG(attr->la_mode)) {
782                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
783         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
784                 /* means no objects are allocated on osts. */
785                 LASSERT(!(ma->ma_valid & MA_LOV));
786                 /* just ignore blocks occupied by extend attributes on MDS */
787                 b->mbo_blocks = 0;
788                 /* if no object is allocated on osts, the size on mds is valid.
789                  * b=22272 */
790                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
791         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
792                 if (mdt_hsm_is_released(ma->ma_lmm)) {
793                         /* A released file stores its size on MDS. */
794                         /* But return 1 block for released file, unless tools
795                          * like tar will consider it fully sparse. (LU-3864)
796                          */
797                         if (unlikely(b->mbo_size == 0))
798                                 b->mbo_blocks = 0;
799                         else
800                                 b->mbo_blocks = 1;
801                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
802                 } else if (info->mti_som_valid) { /* som is valid */
803                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
804                 }
805         }
806
807         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
808                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
809                        PFID(fid), (unsigned long long)b->mbo_size);
810
811 out:
812         if (!IS_ERR_OR_NULL(nodemap))
813                 nodemap_putref(nodemap);
814 }
815
816 static inline int mdt_body_has_lov(const struct lu_attr *la,
817                                    const struct mdt_body *body)
818 {
819         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
820                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
821 }
822
823 void mdt_client_compatibility(struct mdt_thread_info *info)
824 {
825         struct mdt_body       *body;
826         struct ptlrpc_request *req = mdt_info_req(info);
827         struct obd_export     *exp = req->rq_export;
828         struct md_attr        *ma = &info->mti_attr;
829         struct lu_attr        *la = &ma->ma_attr;
830         ENTRY;
831
832         if (exp_connect_layout(exp))
833                 /* the client can deal with 16-bit lmm_stripe_count */
834                 RETURN_EXIT;
835
836         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
837
838         if (!mdt_body_has_lov(la, body))
839                 RETURN_EXIT;
840
841         /* now we have a reply with a lov for a client not compatible with the
842          * layout lock so we have to clean the layout generation number */
843         if (S_ISREG(la->la_mode))
844                 ma->ma_lmm->lmm_layout_gen = 0;
845         EXIT;
846 }
847
848 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
849                                    struct mdt_object *o)
850 {
851         const struct lu_env *env = info->mti_env;
852         int rc, rc2;
853
854         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
855                           XATTR_NAME_LOV);
856
857         if (rc == -ENODATA)
858                 rc = 0;
859
860         if (rc < 0)
861                 goto out;
862
863         /* Is it a directory? Let's check for the LMV as well */
864         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
865                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
866                                    XATTR_NAME_LMV);
867
868                 if (rc2 == -ENODATA)
869                         rc2 = mo_xattr_get(env, mdt_object_child(o),
870                                            &LU_BUF_NULL,
871                                            XATTR_NAME_DEFAULT_LMV);
872
873                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
874                         rc = rc2;
875         }
876
877 out:
878         return rc;
879 }
880
881 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
882                       const char *name)
883 {
884         const struct lu_env *env = info->mti_env;
885         int rc;
886         ENTRY;
887
888         LASSERT(info->mti_big_lmm_used == 0);
889         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
890         if (rc < 0)
891                 RETURN(rc);
892
893         /* big_lmm may need to be grown */
894         if (info->mti_big_lmmsize < rc) {
895                 int size = size_roundup_power2(rc);
896
897                 if (info->mti_big_lmmsize > 0) {
898                         /* free old buffer */
899                         LASSERT(info->mti_big_lmm);
900                         OBD_FREE_LARGE(info->mti_big_lmm,
901                                        info->mti_big_lmmsize);
902                         info->mti_big_lmm = NULL;
903                         info->mti_big_lmmsize = 0;
904                 }
905
906                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
907                 if (info->mti_big_lmm == NULL)
908                         RETURN(-ENOMEM);
909                 info->mti_big_lmmsize = size;
910         }
911         LASSERT(info->mti_big_lmmsize >= rc);
912
913         info->mti_buf.lb_buf = info->mti_big_lmm;
914         info->mti_buf.lb_len = info->mti_big_lmmsize;
915         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
916
917         RETURN(rc);
918 }
919
920 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
921                    struct md_attr *ma, const char *name)
922 {
923         struct md_object *next = mdt_object_child(o);
924         struct lu_buf    *buf = &info->mti_buf;
925         int rc;
926
927         if (strcmp(name, XATTR_NAME_LOV) == 0) {
928                 buf->lb_buf = ma->ma_lmm;
929                 buf->lb_len = ma->ma_lmm_size;
930                 LASSERT(!(ma->ma_valid & MA_LOV));
931         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
932                 buf->lb_buf = ma->ma_lmv;
933                 buf->lb_len = ma->ma_lmv_size;
934                 LASSERT(!(ma->ma_valid & MA_LMV));
935         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
936                 buf->lb_buf = ma->ma_lmv;
937                 buf->lb_len = ma->ma_lmv_size;
938                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
939         } else {
940                 return -EINVAL;
941         }
942
943         LASSERT(buf->lb_buf);
944
945         rc = mo_xattr_get(info->mti_env, next, buf, name);
946         if (rc > 0) {
947
948 got:
949                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
950                         if (info->mti_big_lmm_used)
951                                 ma->ma_lmm = info->mti_big_lmm;
952
953                         /* NOT return LOV EA with hole to old client. */
954                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
955                                      LOV_PATTERN_F_HOLE) &&
956                             !(exp_connect_flags(info->mti_exp) &
957                               OBD_CONNECT_LFSCK)) {
958                                 return -EIO;
959                         } else {
960                                 ma->ma_lmm_size = rc;
961                                 ma->ma_valid |= MA_LOV;
962                         }
963                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
964                         if (info->mti_big_lmm_used)
965                                 ma->ma_lmv = info->mti_big_lmm;
966
967                         ma->ma_lmv_size = rc;
968                         ma->ma_valid |= MA_LMV;
969                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
970                         ma->ma_lmv_size = rc;
971                         ma->ma_valid |= MA_LMV_DEF;
972                 }
973
974                 /* Update mdt_max_mdsize so all clients will be aware that */
975                 if (info->mti_mdt->mdt_max_mdsize < rc)
976                         info->mti_mdt->mdt_max_mdsize = rc;
977
978                 rc = 0;
979         } else if (rc == -ENODATA) {
980                 /* no LOV EA */
981                 rc = 0;
982         } else if (rc == -ERANGE) {
983                 /* Default LMV has fixed size, so it must be able to fit
984                  * in the original buffer */
985                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
986                         return rc;
987                 rc = mdt_big_xattr_get(info, o, name);
988                 if (rc > 0) {
989                         info->mti_big_lmm_used = 1;
990                         goto got;
991                 }
992         }
993
994         return rc;
995 }
996
997 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
998                       struct lu_fid *pfid)
999 {
1000         struct lu_buf           *buf = &info->mti_buf;
1001         struct link_ea_header   *leh;
1002         struct link_ea_entry    *lee;
1003         int                      rc;
1004         ENTRY;
1005
1006         buf->lb_buf = info->mti_big_lmm;
1007         buf->lb_len = info->mti_big_lmmsize;
1008         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1009                           buf, XATTR_NAME_LINK);
1010         /* ignore errors, MA_PFID won't be set and it is
1011          * up to the caller to treat this as an error */
1012         if (rc == -ERANGE || buf->lb_len == 0) {
1013                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1014                 buf->lb_buf = info->mti_big_lmm;
1015                 buf->lb_len = info->mti_big_lmmsize;
1016         }
1017
1018         if (rc < 0)
1019                 RETURN(rc);
1020         if (rc < sizeof(*leh)) {
1021                 CERROR("short LinkEA on "DFID": rc = %d\n",
1022                        PFID(mdt_object_fid(o)), rc);
1023                 RETURN(-ENODATA);
1024         }
1025
1026         leh = (struct link_ea_header *) buf->lb_buf;
1027         lee = (struct link_ea_entry *)(leh + 1);
1028         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1029                 leh->leh_magic = LINK_EA_MAGIC;
1030                 leh->leh_reccount = __swab32(leh->leh_reccount);
1031                 leh->leh_len = __swab64(leh->leh_len);
1032         }
1033         if (leh->leh_magic != LINK_EA_MAGIC)
1034                 RETURN(-EINVAL);
1035         if (leh->leh_reccount == 0)
1036                 RETURN(-ENODATA);
1037
1038         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1039         fid_be_to_cpu(pfid, pfid);
1040
1041         RETURN(0);
1042 }
1043
1044 int mdt_attr_get_complex(struct mdt_thread_info *info,
1045                          struct mdt_object *o, struct md_attr *ma)
1046 {
1047         const struct lu_env *env = info->mti_env;
1048         struct md_object    *next = mdt_object_child(o);
1049         struct lu_buf       *buf = &info->mti_buf;
1050         int                  need = ma->ma_need;
1051         int                  rc = 0, rc2;
1052         u32                  mode;
1053         ENTRY;
1054
1055         ma->ma_valid = 0;
1056
1057         if (mdt_object_exists(o) == 0)
1058                 GOTO(out, rc = -ENOENT);
1059         mode = lu_object_attr(&next->mo_lu);
1060
1061         if (need & MA_INODE) {
1062                 ma->ma_need = MA_INODE;
1063                 rc = mo_attr_get(env, next, ma);
1064                 if (rc)
1065                         GOTO(out, rc);
1066
1067                 if (S_ISREG(mode))
1068                         (void) mdt_get_som(info, o, ma);
1069                 ma->ma_valid |= MA_INODE;
1070         }
1071
1072         if (need & MA_PFID) {
1073                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1074                 if (rc == 0)
1075                         ma->ma_valid |= MA_PFID;
1076                 /* ignore this error, parent fid is not mandatory */
1077                 rc = 0;
1078         }
1079
1080         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1081                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1082                 if (rc)
1083                         GOTO(out, rc);
1084         }
1085
1086         if (need & MA_LMV && S_ISDIR(mode)) {
1087                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1088                 if (rc != 0)
1089                         GOTO(out, rc);
1090         }
1091
1092         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1093                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1094                 if (rc != 0)
1095                         GOTO(out, rc);
1096         }
1097
1098         /*
1099          * In the handle of MA_INODE, we may already get the SOM attr.
1100          */
1101         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1102                 rc = mdt_get_som(info, o, ma);
1103                 if (rc != 0)
1104                         GOTO(out, rc);
1105         }
1106
1107         if (need & MA_HSM && S_ISREG(mode)) {
1108                 buf->lb_buf = info->mti_xattr_buf;
1109                 buf->lb_len = sizeof(info->mti_xattr_buf);
1110                 CLASSERT(sizeof(struct hsm_attrs) <=
1111                          sizeof(info->mti_xattr_buf));
1112                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1113                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1114                 if (rc2 == 0)
1115                         ma->ma_valid |= MA_HSM;
1116                 else if (rc2 < 0 && rc2 != -ENODATA)
1117                         GOTO(out, rc = rc2);
1118         }
1119
1120 #ifdef CONFIG_FS_POSIX_ACL
1121         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1122                 buf->lb_buf = ma->ma_acl;
1123                 buf->lb_len = ma->ma_acl_size;
1124                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1125                 if (rc2 > 0) {
1126                         ma->ma_acl_size = rc2;
1127                         ma->ma_valid |= MA_ACL_DEF;
1128                 } else if (rc2 == -ENODATA) {
1129                         /* no ACLs */
1130                         ma->ma_acl_size = 0;
1131                 } else
1132                         GOTO(out, rc = rc2);
1133         }
1134 #endif
1135 out:
1136         ma->ma_need = need;
1137         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1138                rc, ma->ma_valid, ma->ma_lmm);
1139         RETURN(rc);
1140 }
1141
1142 static int mdt_getattr_internal(struct mdt_thread_info *info,
1143                                 struct mdt_object *o, int ma_need)
1144 {
1145         struct md_object        *next = mdt_object_child(o);
1146         const struct mdt_body   *reqbody = info->mti_body;
1147         struct ptlrpc_request   *req = mdt_info_req(info);
1148         struct md_attr          *ma = &info->mti_attr;
1149         struct lu_attr          *la = &ma->ma_attr;
1150         struct req_capsule      *pill = info->mti_pill;
1151         const struct lu_env     *env = info->mti_env;
1152         struct mdt_body         *repbody;
1153         struct lu_buf           *buffer = &info->mti_buf;
1154         struct obd_export       *exp = info->mti_exp;
1155         int                      rc;
1156         ENTRY;
1157
1158         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1159                 RETURN(err_serious(-ENOMEM));
1160
1161         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1162
1163         ma->ma_valid = 0;
1164
1165         if (mdt_object_remote(o)) {
1166                 /* This object is located on remote node.*/
1167                 /* Return -ENOTSUPP for old client */
1168                 if (!mdt_is_dne_client(req->rq_export))
1169                         GOTO(out, rc = -ENOTSUPP);
1170
1171                 repbody->mbo_fid1 = *mdt_object_fid(o);
1172                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1173                 GOTO(out, rc = 0);
1174         }
1175
1176         if (reqbody->mbo_eadatasize > 0) {
1177                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1178                 if (buffer->lb_buf == NULL)
1179                         GOTO(out, rc = -EPROTO);
1180                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1181                                                       RCL_SERVER);
1182         } else {
1183                 buffer->lb_buf = NULL;
1184                 buffer->lb_len = 0;
1185                 ma_need &= ~(MA_LOV | MA_LMV);
1186                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1187                        mdt_obd_name(info->mti_mdt),
1188                        req->rq_export->exp_client_uuid.uuid);
1189         }
1190
1191         /* If it is dir object and client require MEA, then we got MEA */
1192         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1193             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1194                 /* Assumption: MDT_MD size is enough for lmv size. */
1195                 ma->ma_lmv = buffer->lb_buf;
1196                 ma->ma_lmv_size = buffer->lb_len;
1197                 ma->ma_need = MA_INODE;
1198                 if (ma->ma_lmv_size > 0) {
1199                         if (reqbody->mbo_valid & OBD_MD_MEA)
1200                                 ma->ma_need |= MA_LMV;
1201                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1202                                 ma->ma_need |= MA_LMV_DEF;
1203                 }
1204         } else {
1205                 ma->ma_lmm = buffer->lb_buf;
1206                 ma->ma_lmm_size = buffer->lb_len;
1207                 ma->ma_need = MA_INODE | MA_HSM;
1208                 if (ma->ma_lmm_size > 0)
1209                         ma->ma_need |= MA_LOV;
1210         }
1211
1212         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1213             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1214             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1215                 /* get default stripe info for this dir. */
1216                 ma->ma_need |= MA_LOV_DEF;
1217         }
1218         ma->ma_need |= ma_need;
1219
1220         rc = mdt_attr_get_complex(info, o, ma);
1221         if (unlikely(rc)) {
1222                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1223                        "%s: getattr error for "DFID": rc = %d\n",
1224                        mdt_obd_name(info->mti_mdt),
1225                        PFID(mdt_object_fid(o)), rc);
1226                 RETURN(rc);
1227         }
1228
1229         /* if file is released, check if a restore is running */
1230         if (ma->ma_valid & MA_HSM) {
1231                 repbody->mbo_valid |= OBD_MD_TSTATE;
1232                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1233                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1234                         repbody->mbo_t_state = MS_RESTORE;
1235         }
1236
1237         if (likely(ma->ma_valid & MA_INODE))
1238                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1239         else
1240                 RETURN(-EFAULT);
1241
1242         if (mdt_body_has_lov(la, reqbody)) {
1243                 if (ma->ma_valid & MA_LOV) {
1244                         LASSERT(ma->ma_lmm_size);
1245                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1246                         if (S_ISDIR(la->la_mode))
1247                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1248                         else
1249                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1250                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1251                 }
1252                 if (ma->ma_valid & MA_LMV) {
1253                         /* Return -ENOTSUPP for old client */
1254                         if (!mdt_is_striped_client(req->rq_export))
1255                                 RETURN(-ENOTSUPP);
1256
1257                         LASSERT(S_ISDIR(la->la_mode));
1258                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1259                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1260                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1261                 }
1262                 if (ma->ma_valid & MA_LMV_DEF) {
1263                         /* Return -ENOTSUPP for old client */
1264                         if (!mdt_is_striped_client(req->rq_export))
1265                                 RETURN(-ENOTSUPP);
1266                         LASSERT(S_ISDIR(la->la_mode));
1267                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1268                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1269                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1270                                                OBD_MD_DEFAULT_MEA);
1271                 }
1272         } else if (S_ISLNK(la->la_mode) &&
1273                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1274                 buffer->lb_buf = ma->ma_lmm;
1275                 /* eadatasize from client includes NULL-terminator, so
1276                  * there is no need to read it */
1277                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1278                 rc = mo_readlink(env, next, buffer);
1279                 if (unlikely(rc <= 0)) {
1280                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1281                                mdt_obd_name(info->mti_mdt),
1282                                PFID(mdt_object_fid(o)), rc);
1283                         rc = -EFAULT;
1284                 } else {
1285                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1286
1287                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1288                                 rc -= 2;
1289                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1290                         /* we need to report back size with NULL-terminator
1291                          * because client expects that */
1292                         repbody->mbo_eadatasize = rc + 1;
1293                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1294                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1295                                        "on "DFID ", expected %d\n",
1296                                        mdt_obd_name(info->mti_mdt),
1297                                        rc, PFID(mdt_object_fid(o)),
1298                                        reqbody->mbo_eadatasize - 1);
1299                         /* NULL terminate */
1300                         ((char *)ma->ma_lmm)[rc] = 0;
1301
1302                         /* If the total CDEBUG() size is larger than a page, it
1303                          * will print a warning to the console, avoid this by
1304                          * printing just the last part of the symlink. */
1305                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1306                                print_limit < rc ? "..." : "", print_limit,
1307                                (char *)ma->ma_lmm + rc - print_limit, rc);
1308                         rc = 0;
1309                 }
1310         }
1311
1312         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1313                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1314                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1315                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1316                        repbody->mbo_max_mdsize);
1317         }
1318
1319 #ifdef CONFIG_FS_POSIX_ACL
1320         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1321                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1322                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1323                 if (IS_ERR(nodemap))
1324                         RETURN(PTR_ERR(nodemap));
1325
1326                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1327                 nodemap_putref(nodemap);
1328         }
1329 #endif
1330
1331 out:
1332         if (rc == 0)
1333                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1334
1335         RETURN(rc);
1336 }
1337
1338 static int mdt_getattr(struct tgt_session_info *tsi)
1339 {
1340         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1341         struct mdt_object       *obj = info->mti_object;
1342         struct req_capsule      *pill = info->mti_pill;
1343         struct mdt_body         *reqbody;
1344         struct mdt_body         *repbody;
1345         int rc, rc2;
1346         ENTRY;
1347
1348         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1349         LASSERT(reqbody);
1350         LASSERT(obj != NULL);
1351         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1352
1353         /* Special case for Data-on-MDT files to get data version */
1354         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1355                 rc = mdt_data_version_get(tsi);
1356                 GOTO(out, rc);
1357         }
1358
1359         /* Unlike intent case where we need to pre-fill out buffers early on
1360          * in intent policy for ldlm reasons, here we can have a much better
1361          * guess at EA size by just reading it from disk.
1362          * Exceptions are readdir and (missing) directory striping */
1363         /* Readlink */
1364         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1365                 /* No easy way to know how long is the symlink, but it cannot
1366                  * be more than PATH_MAX, so we allocate +1 */
1367                 rc = PATH_MAX + 1;
1368         /* A special case for fs ROOT: getattr there might fetch
1369          * default EA for entire fs, not just for this dir!
1370          */
1371         } else if (lu_fid_eq(mdt_object_fid(obj),
1372                              &info->mti_mdt->mdt_md_root_fid) &&
1373                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1374                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1375                                                                  MDS_GETATTR)) {
1376                 /* Should the default strping be bigger, mdt_fix_reply
1377                  * will reallocate */
1378                 rc = DEF_REP_MD_SIZE;
1379         } else {
1380                 /* Read the actual EA size from disk */
1381                 rc = mdt_attr_get_eabuf_size(info, obj);
1382         }
1383
1384         if (rc < 0)
1385                 GOTO(out, rc = err_serious(rc));
1386
1387         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1388
1389         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1390          * by default. If the target object has more ACL entries, then
1391          * enlarge the buffer when necessary. */
1392         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1393                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1394
1395         rc = req_capsule_server_pack(pill);
1396         if (unlikely(rc != 0))
1397                 GOTO(out, rc = err_serious(rc));
1398
1399         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1400         LASSERT(repbody != NULL);
1401         repbody->mbo_eadatasize = 0;
1402         repbody->mbo_aclsize = 0;
1403
1404         rc = mdt_check_ucred(info);
1405         if (unlikely(rc))
1406                 GOTO(out_shrink, rc);
1407
1408         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1409
1410         rc = mdt_getattr_internal(info, obj, 0);
1411         EXIT;
1412 out_shrink:
1413         mdt_client_compatibility(info);
1414         rc2 = mdt_fix_reply(info);
1415         if (rc == 0)
1416                 rc = rc2;
1417 out:
1418         mdt_thread_info_fini(info);
1419         return rc;
1420 }
1421
1422 /**
1423  * Handler of layout intent RPC requiring the layout modification
1424  *
1425  * \param[in] info      thread environment
1426  * \param[in] obj       object
1427  * \param[in] layout    layout change descriptor
1428  *
1429  * \retval 0    on success
1430  * \retval < 0  error code
1431  */
1432 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1433                       struct md_layout_change *layout)
1434 {
1435         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1436         int rc;
1437         ENTRY;
1438
1439         if (!mdt_object_exists(obj))
1440                 GOTO(out, rc = -ENOENT);
1441
1442         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1443                 GOTO(out, rc = -EINVAL);
1444
1445         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1446                            MAY_WRITE);
1447         if (rc)
1448                 GOTO(out, rc);
1449
1450         /* take layout lock to prepare layout change */
1451         mdt_lock_reg_init(lh, LCK_EX);
1452         rc = mdt_object_lock(info, obj, lh, MDS_INODELOCK_LAYOUT);
1453         if (rc)
1454                 GOTO(out, rc);
1455
1456         mutex_lock(&obj->mot_som_mutex);
1457         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1458         mutex_unlock(&obj->mot_som_mutex);
1459         mdt_object_unlock(info, obj, lh, 1);
1460 out:
1461         RETURN(rc);
1462 }
1463
1464 /**
1465  * Exchange MOF_LOV_CREATED flags between two objects after a
1466  * layout swap. No assumption is made on whether o1 or o2 have
1467  * created objects or not.
1468  *
1469  * \param[in,out] o1    First swap layout object
1470  * \param[in,out] o2    Second swap layout object
1471  */
1472 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1473 {
1474         unsigned int o1_lov_created = o1->mot_lov_created;
1475
1476         mutex_lock(&o1->mot_lov_mutex);
1477         mutex_lock(&o2->mot_lov_mutex);
1478
1479         o1->mot_lov_created = o2->mot_lov_created;
1480         o2->mot_lov_created = o1_lov_created;
1481
1482         mutex_unlock(&o2->mot_lov_mutex);
1483         mutex_unlock(&o1->mot_lov_mutex);
1484 }
1485
1486 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1487 {
1488         struct mdt_thread_info  *info;
1489         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1490         struct obd_export       *exp = req->rq_export;
1491         struct mdt_object       *o1, *o2, *o;
1492         struct mdt_lock_handle  *lh1, *lh2;
1493         struct mdc_swap_layouts *msl;
1494         int                      rc;
1495         ENTRY;
1496
1497         /* client does not support layout lock, so layout swaping
1498          * is disabled.
1499          * FIXME: there is a problem for old clients which don't support
1500          * layout lock yet. If those clients have already opened the file
1501          * they won't be notified at all so that old layout may still be
1502          * used to do IO. This can be fixed after file release is landed by
1503          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1504         if (!exp_connect_layout(exp))
1505                 RETURN(-EOPNOTSUPP);
1506
1507         info = tsi2mdt_info(tsi);
1508
1509         if (info->mti_dlm_req != NULL)
1510                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1511
1512         o1 = info->mti_object;
1513         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1514                                 &info->mti_body->mbo_fid2);
1515         if (IS_ERR(o))
1516                 GOTO(out, rc = PTR_ERR(o));
1517
1518         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1519                 GOTO(put, rc = -ENOENT);
1520
1521         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1522         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1523                 GOTO(put, rc);
1524
1525         if (rc < 0)
1526                 swap(o1, o2);
1527
1528         /* permission check. Make sure the calling process having permission
1529          * to write both files. */
1530         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1531                            MAY_WRITE);
1532         if (rc < 0)
1533                 GOTO(put, rc);
1534
1535         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1536                            MAY_WRITE);
1537         if (rc < 0)
1538                 GOTO(put, rc);
1539
1540         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1541         if (msl == NULL)
1542                 GOTO(put, rc = -EPROTO);
1543
1544         lh1 = &info->mti_lh[MDT_LH_NEW];
1545         mdt_lock_reg_init(lh1, LCK_EX);
1546         lh2 = &info->mti_lh[MDT_LH_OLD];
1547         mdt_lock_reg_init(lh2, LCK_EX);
1548
1549         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1550                              MDS_INODELOCK_XATTR);
1551         if (rc < 0)
1552                 GOTO(put, rc);
1553
1554         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1555                              MDS_INODELOCK_XATTR);
1556         if (rc < 0)
1557                 GOTO(unlock1, rc);
1558
1559         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1560                              mdt_object_child(o2), msl->msl_flags);
1561         if (rc < 0)
1562                 GOTO(unlock2, rc);
1563
1564         mdt_swap_lov_flag(o1, o2);
1565
1566 unlock2:
1567         mdt_object_unlock(info, o2, lh2, rc);
1568 unlock1:
1569         mdt_object_unlock(info, o1, lh1, rc);
1570 put:
1571         mdt_object_put(info->mti_env, o);
1572 out:
1573         mdt_thread_info_fini(info);
1574         RETURN(rc);
1575 }
1576
1577 static int mdt_raw_lookup(struct mdt_thread_info *info,
1578                           struct mdt_object *parent,
1579                           const struct lu_name *lname,
1580                           struct ldlm_reply *ldlm_rep)
1581 {
1582         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1583         int              rc;
1584         ENTRY;
1585
1586         LASSERT(!info->mti_cross_ref);
1587
1588         /* Only got the fid of this obj by name */
1589         fid_zero(child_fid);
1590         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1591                         lname, child_fid, &info->mti_spec);
1592         if (rc == 0) {
1593                 struct mdt_body *repbody;
1594
1595                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1596                 repbody->mbo_fid1 = *child_fid;
1597                 repbody->mbo_valid = OBD_MD_FLID;
1598                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1599         } else if (rc == -ENOENT) {
1600                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1601         }
1602
1603         RETURN(rc);
1604 }
1605
1606 /*
1607  * UPDATE lock should be taken against parent, and be released before exit;
1608  * child_bits lock should be taken against child, and be returned back:
1609  *            (1)normal request should release the child lock;
1610  *            (2)intent request will grant the lock to client.
1611  */
1612 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1613                                  struct mdt_lock_handle *lhc,
1614                                  __u64 child_bits,
1615                                  struct ldlm_reply *ldlm_rep)
1616 {
1617         struct ptlrpc_request  *req = mdt_info_req(info);
1618         struct mdt_body        *reqbody = NULL;
1619         struct mdt_object      *parent = info->mti_object;
1620         struct mdt_object      *child;
1621         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1622         struct lu_name         *lname = NULL;
1623         struct mdt_lock_handle *lhp = NULL;
1624         struct ldlm_lock       *lock;
1625         struct req_capsule *pill = info->mti_pill;
1626         __u64 try_bits = 0;
1627         bool is_resent;
1628         int ma_need = 0;
1629         int rc;
1630
1631         ENTRY;
1632
1633         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1634         LASSERT(ergo(is_resent,
1635                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1636
1637         if (parent == NULL)
1638                 RETURN(-ENOENT);
1639
1640         if (info->mti_cross_ref) {
1641                 /* Only getattr on the child. Parent is on another node. */
1642                 mdt_set_disposition(info, ldlm_rep,
1643                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1644                 child = parent;
1645                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1646                        "ldlm_rep = %p\n",
1647                        PFID(mdt_object_fid(child)), ldlm_rep);
1648
1649                 rc = mdt_check_resent_lock(info, child, lhc);
1650                 if (rc < 0) {
1651                         RETURN(rc);
1652                 } else if (rc > 0) {
1653                         mdt_lock_handle_init(lhc);
1654                         mdt_lock_reg_init(lhc, LCK_PR);
1655
1656                         /*
1657                          * Object's name entry is on another MDS, it will
1658                          * request PERM lock only because LOOKUP lock is owned
1659                          * by the MDS where name entry resides.
1660                          *
1661                          * TODO: it should try layout lock too. - Jinshan
1662                          */
1663                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1664                                         MDS_INODELOCK_LAYOUT);
1665                         child_bits |= MDS_INODELOCK_PERM;
1666
1667                         rc = mdt_object_lock(info, child, lhc, child_bits);
1668                         if (rc < 0)
1669                                 RETURN(rc);
1670                 }
1671
1672                 /* Finally, we can get attr for child. */
1673                 if (!mdt_object_exists(child)) {
1674                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1675                                         &child->mot_obj,
1676                                         "remote object doesn't exist.");
1677                         mdt_object_unlock(info, child, lhc, 1);
1678                         RETURN(-ENOENT);
1679                 }
1680
1681                 rc = mdt_getattr_internal(info, child, 0);
1682                 if (unlikely(rc != 0))
1683                         mdt_object_unlock(info, child, lhc, 1);
1684
1685                 mdt_pack_secctx_in_reply(info, child);
1686
1687                 RETURN(rc);
1688         }
1689
1690         lname = &info->mti_name;
1691         mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
1692
1693         if (lu_name_is_valid(lname)) {
1694                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1695                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1696                        PNAME(lname), ldlm_rep);
1697         } else {
1698                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1699                 if (unlikely(reqbody == NULL))
1700                         RETURN(err_serious(-EPROTO));
1701
1702                 *child_fid = reqbody->mbo_fid2;
1703
1704                 if (unlikely(!fid_is_sane(child_fid)))
1705                         RETURN(err_serious(-EINVAL));
1706
1707                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1708                        "ldlm_rep = %p\n",
1709                        PFID(mdt_object_fid(parent)),
1710                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1711         }
1712
1713         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1714
1715         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1716                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1717                                 &parent->mot_obj,
1718                                 "Parent doesn't exist!");
1719                 RETURN(-ESTALE);
1720         }
1721
1722         if (mdt_object_remote(parent)) {
1723                 CERROR("%s: parent "DFID" is on remote target\n",
1724                        mdt_obd_name(info->mti_mdt),
1725                        PFID(mdt_object_fid(parent)));
1726                 RETURN(-EIO);
1727         }
1728
1729         if (lu_name_is_valid(lname)) {
1730                 /* Always allow to lookup ".." */
1731                 if (unlikely(lname->ln_namelen == 2 &&
1732                              lname->ln_name[0] == '.' &&
1733                              lname->ln_name[1] == '.'))
1734                         info->mti_spec.sp_permitted = 1;
1735
1736                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1737                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1738
1739                         RETURN(rc);
1740                 }
1741
1742                 /* step 1: lock parent only if parent is a directory */
1743                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1744                         lhp = &info->mti_lh[MDT_LH_PARENT];
1745                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1746                         rc = mdt_object_lock(info, parent, lhp,
1747                                              MDS_INODELOCK_UPDATE);
1748                         if (unlikely(rc != 0))
1749                                 RETURN(rc);
1750                 }
1751
1752                 /* step 2: lookup child's fid by name */
1753                 fid_zero(child_fid);
1754                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1755                                 child_fid, &info->mti_spec);
1756                 if (rc == -ENOENT)
1757                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1758
1759                 if (rc != 0)
1760                         GOTO(out_parent, rc);
1761         }
1762
1763         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1764
1765         /*
1766          *step 3: find the child object by fid & lock it.
1767          *        regardless if it is local or remote.
1768          *
1769          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1770          *      set parent dir fid the same as child fid in getattr by fid case
1771          *      we should not lu_object_find() the object again, could lead
1772          *      to hung if there is a concurrent unlink destroyed the object.
1773          */
1774         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1775                 mdt_object_get(info->mti_env, parent);
1776                 child = parent;
1777         } else {
1778                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1779                                         child_fid);
1780         }
1781
1782         if (unlikely(IS_ERR(child)))
1783                 GOTO(out_parent, rc = PTR_ERR(child));
1784
1785         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1786         if (!mdt_object_exists(child)) {
1787                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1788                                 &child->mot_obj,
1789                                 "Object doesn't exist!");
1790                 GOTO(out_child, rc = -ENOENT);
1791         }
1792
1793         rc = mdt_check_resent_lock(info, child, lhc);
1794         if (rc < 0) {
1795                 GOTO(out_child, rc);
1796         } else if (rc > 0) {
1797                 mdt_lock_handle_init(lhc);
1798                 mdt_lock_reg_init(lhc, LCK_PR);
1799
1800                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1801                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1802                         struct md_attr *ma = &info->mti_attr;
1803
1804                         ma->ma_valid = 0;
1805                         ma->ma_need = MA_INODE;
1806                         rc = mdt_attr_get_complex(info, child, ma);
1807                         if (unlikely(rc != 0))
1808                                 GOTO(out_child, rc);
1809
1810                         /* If the file has not been changed for some time, we
1811                          * return not only a LOOKUP lock, but also an UPDATE
1812                          * lock and this might save us RPC on later STAT. For
1813                          * directories, it also let negative dentry cache start
1814                          * working for this dir. */
1815                         if (ma->ma_valid & MA_INODE &&
1816                             ma->ma_attr.la_valid & LA_CTIME &&
1817                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1818                                 ma->ma_attr.la_ctime < ktime_get_real_seconds())
1819                                 child_bits |= MDS_INODELOCK_UPDATE;
1820                 }
1821
1822                 /* layout lock must be granted in a best-effort way
1823                  * for IT operations */
1824                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1825                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1826                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1827                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1828                             exp_connect_layout(info->mti_exp)) {
1829                                 /* try to grant layout lock for regular file. */
1830                                 try_bits = MDS_INODELOCK_LAYOUT;
1831                         }
1832                         /* Acquire DOM lock in advance for data-on-mdt file */
1833                         if (child != parent)
1834                                 try_bits |= MDS_INODELOCK_DOM;
1835                 }
1836
1837                 if (try_bits != 0) {
1838                         /* try layout lock, it may fail to be granted due to
1839                          * contention at LOOKUP or UPDATE */
1840                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1841                                                  try_bits, false);
1842                         if (child_bits & MDS_INODELOCK_LAYOUT)
1843                                 ma_need |= MA_LOV;
1844                 } else {
1845                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1846                          * client will enqueue the lock to the remote MDT */
1847                         if (mdt_object_remote(child))
1848                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1849                         rc = mdt_object_lock(info, child, lhc, child_bits);
1850                 }
1851                 if (unlikely(rc != 0))
1852                         GOTO(out_child, rc);
1853         }
1854
1855         /* finally, we can get attr for child. */
1856         rc = mdt_getattr_internal(info, child, ma_need);
1857         if (unlikely(rc != 0)) {
1858                 mdt_object_unlock(info, child, lhc, 1);
1859                 GOTO(out_child, rc);
1860         }
1861
1862         mdt_pack_secctx_in_reply(info, child);
1863
1864         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1865         if (lock) {
1866                 /* Debugging code. */
1867                 LDLM_DEBUG(lock, "Returning lock to client");
1868                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1869                                          &lock->l_resource->lr_name),
1870                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1871                          PLDLMRES(lock->l_resource),
1872                          PFID(mdt_object_fid(child)));
1873
1874                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1875                     mdt_object_exists(child) && !mdt_object_remote(child) &&
1876                     child != parent) {
1877                         mdt_object_put(info->mti_env, child);
1878                         rc = mdt_pack_size2body(info, child_fid,
1879                                                 &lhc->mlh_reg_lh);
1880                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1881                                 /* DOM lock was taken in advance but this is
1882                                  * not DoM file. Drop the lock.
1883                                  */
1884                                 lock_res_and_lock(lock);
1885                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1886                                 unlock_res_and_lock(lock);
1887                         }
1888                         LDLM_LOCK_PUT(lock);
1889                         GOTO(out_parent, rc = 0);
1890                 }
1891                 LDLM_LOCK_PUT(lock);
1892         }
1893
1894         EXIT;
1895 out_child:
1896         mdt_object_put(info->mti_env, child);
1897 out_parent:
1898         if (lhp)
1899                 mdt_object_unlock(info, parent, lhp, 1);
1900         return rc;
1901 }
1902
1903 /* normal handler: should release the child lock */
1904 static int mdt_getattr_name(struct tgt_session_info *tsi)
1905 {
1906         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1907         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1908         struct mdt_body        *reqbody;
1909         struct mdt_body        *repbody;
1910         int rc, rc2;
1911         ENTRY;
1912
1913         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1914         LASSERT(reqbody != NULL);
1915         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1916         LASSERT(repbody != NULL);
1917
1918         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1919         repbody->mbo_eadatasize = 0;
1920         repbody->mbo_aclsize = 0;
1921
1922         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1923         if (unlikely(rc))
1924                 GOTO(out_shrink, rc);
1925
1926         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1927         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1928                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1929                 lhc->mlh_reg_lh.cookie = 0;
1930         }
1931         mdt_exit_ucred(info);
1932         EXIT;
1933 out_shrink:
1934         mdt_client_compatibility(info);
1935         rc2 = mdt_fix_reply(info);
1936         if (rc == 0)
1937                 rc = rc2;
1938         mdt_thread_info_fini(info);
1939         return rc;
1940 }
1941
1942 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1943                          void *karg, void __user *uarg);
1944
1945 static int mdt_set_info(struct tgt_session_info *tsi)
1946 {
1947         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1948         char                    *key;
1949         void                    *val;
1950         int                      keylen, vallen, rc = 0;
1951
1952         ENTRY;
1953
1954         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1955         if (key == NULL) {
1956                 DEBUG_REQ(D_HA, req, "no set_info key");
1957                 RETURN(err_serious(-EFAULT));
1958         }
1959
1960         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1961                                       RCL_CLIENT);
1962
1963         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1964         if (val == NULL) {
1965                 DEBUG_REQ(D_HA, req, "no set_info val");
1966                 RETURN(err_serious(-EFAULT));
1967         }
1968
1969         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1970                                       RCL_CLIENT);
1971
1972         /* Swab any part of val you need to here */
1973         if (KEY_IS(KEY_READ_ONLY)) {
1974                 spin_lock(&req->rq_export->exp_lock);
1975                 if (*(__u32 *)val)
1976                         *exp_connect_flags_ptr(req->rq_export) |=
1977                                 OBD_CONNECT_RDONLY;
1978                 else
1979                         *exp_connect_flags_ptr(req->rq_export) &=
1980                                 ~OBD_CONNECT_RDONLY;
1981                 spin_unlock(&req->rq_export->exp_lock);
1982         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1983                 struct changelog_setinfo *cs = val;
1984
1985                 if (vallen != sizeof(*cs)) {
1986                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1987                                tgt_name(tsi->tsi_tgt), vallen);
1988                         RETURN(-EINVAL);
1989                 }
1990                 if (ptlrpc_req_need_swab(req)) {
1991                         __swab64s(&cs->cs_recno);
1992                         __swab32s(&cs->cs_id);
1993                 }
1994
1995                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1996                                    vallen, val, NULL);
1997         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1998                 if (vallen > 0)
1999                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2000         } else {
2001                 RETURN(-EINVAL);
2002         }
2003         RETURN(rc);
2004 }
2005
2006 static int mdt_readpage(struct tgt_session_info *tsi)
2007 {
2008         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2009         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2010         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2011         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2012         struct mdt_body         *repbody;
2013         int                      rc;
2014         int                      i;
2015
2016         ENTRY;
2017
2018         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2019                 RETURN(err_serious(-ENOMEM));
2020
2021         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2022         if (repbody == NULL || reqbody == NULL)
2023                 RETURN(err_serious(-EFAULT));
2024
2025         /*
2026          * prepare @rdpg before calling lower layers and transfer itself. Here
2027          * reqbody->size contains offset of where to start to read and
2028          * reqbody->nlink contains number bytes to read.
2029          */
2030         rdpg->rp_hash = reqbody->mbo_size;
2031         if (rdpg->rp_hash != reqbody->mbo_size) {
2032                 CERROR("Invalid hash: %#llx != %#llx\n",
2033                        rdpg->rp_hash, reqbody->mbo_size);
2034                 RETURN(-EFAULT);
2035         }
2036
2037         rdpg->rp_attrs = reqbody->mbo_mode;
2038         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2039                 rdpg->rp_attrs |= LUDA_64BITHASH;
2040         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2041                                 exp_max_brw_size(tsi->tsi_exp));
2042         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2043                           PAGE_SHIFT;
2044         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2045         if (rdpg->rp_pages == NULL)
2046                 RETURN(-ENOMEM);
2047
2048         for (i = 0; i < rdpg->rp_npages; ++i) {
2049                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2050                 if (rdpg->rp_pages[i] == NULL)
2051                         GOTO(free_rdpg, rc = -ENOMEM);
2052         }
2053
2054         /* call lower layers to fill allocated pages with directory data */
2055         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2056         if (rc < 0)
2057                 GOTO(free_rdpg, rc);
2058
2059         /* send pages to client */
2060         rc = tgt_sendpage(tsi, rdpg, rc);
2061
2062         EXIT;
2063 free_rdpg:
2064
2065         for (i = 0; i < rdpg->rp_npages; i++)
2066                 if (rdpg->rp_pages[i] != NULL)
2067                         __free_page(rdpg->rp_pages[i]);
2068         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2069
2070         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2071                 RETURN(0);
2072
2073         return rc;
2074 }
2075
2076 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2077 {
2078         struct lu_ucred *uc = mdt_ucred_check(info);
2079         struct lu_attr *attr = &info->mti_attr.ma_attr;
2080
2081         if (uc == NULL)
2082                 return -EINVAL;
2083
2084         if (op != REINT_SETATTR) {
2085                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2086                         attr->la_uid = uc->uc_fsuid;
2087                 /* for S_ISGID, inherit gid from his parent, such work will be
2088                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2089                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2090                         attr->la_gid = uc->uc_fsgid;
2091         }
2092
2093         return 0;
2094 }
2095
2096 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2097 {
2098         return op == REINT_OPEN &&
2099              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2100 }
2101
2102 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2103 {
2104         struct req_capsule *pill = info->mti_pill;
2105
2106         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2107                                   RCL_SERVER) &&
2108             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2109                                   RCL_CLIENT)) {
2110                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2111                                          RCL_CLIENT) != 0) {
2112                         /* pre-set size in server part with max size */
2113                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2114                                              RCL_SERVER,
2115                                              info->mti_mdt->mdt_max_ea_size);
2116                 } else {
2117                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2118                                              RCL_SERVER, 0);
2119                 }
2120         }
2121
2122 }
2123
2124 static int mdt_reint_internal(struct mdt_thread_info *info,
2125                               struct mdt_lock_handle *lhc,
2126                               __u32 op)
2127 {
2128         struct req_capsule      *pill = info->mti_pill;
2129         struct mdt_body         *repbody;
2130         int                      rc = 0, rc2;
2131
2132         ENTRY;
2133
2134         rc = mdt_reint_unpack(info, op);
2135         if (rc != 0) {
2136                 CERROR("Can't unpack reint, rc %d\n", rc);
2137                 RETURN(err_serious(rc));
2138         }
2139
2140
2141         /* check if the file system is set to readonly. O_RDONLY open
2142          * is still allowed even the file system is set to readonly mode */
2143         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2144                 RETURN(err_serious(-EROFS));
2145
2146         /* for replay (no_create) lmm is not needed, client has it already */
2147         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2148                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2149                                      DEF_REP_MD_SIZE);
2150
2151         /* llog cookies are always 0, the field is kept for compatibility */
2152         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2153                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2154
2155         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2156          * by default. If the target object has more ACL entries, then
2157          * enlarge the buffer when necessary. */
2158         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2159                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2160                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2161
2162         mdt_preset_secctx_size(info);
2163
2164         rc = req_capsule_server_pack(pill);
2165         if (rc != 0) {
2166                 CERROR("Can't pack response, rc %d\n", rc);
2167                 RETURN(err_serious(rc));
2168         }
2169
2170         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2171                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2172                 LASSERT(repbody);
2173                 repbody->mbo_eadatasize = 0;
2174                 repbody->mbo_aclsize = 0;
2175         }
2176
2177         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2178
2179         /* for replay no cookkie / lmm need, because client have this already */
2180         if (info->mti_spec.no_create)
2181                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2182                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2183
2184         rc = mdt_init_ucred_reint(info);
2185         if (rc)
2186                 GOTO(out_shrink, rc);
2187
2188         rc = mdt_fix_attr_ucred(info, op);
2189         if (rc != 0)
2190                 GOTO(out_ucred, rc = err_serious(rc));
2191
2192         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2193         if (rc < 0) {
2194                 GOTO(out_ucred, rc);
2195         } else if (rc == 1) {
2196                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
2197                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2198                 GOTO(out_ucred, rc);
2199         }
2200         rc = mdt_reint_rec(info, lhc);
2201         EXIT;
2202 out_ucred:
2203         mdt_exit_ucred(info);
2204 out_shrink:
2205         mdt_client_compatibility(info);
2206
2207         rc2 = mdt_fix_reply(info);
2208         if (rc == 0)
2209                 rc = rc2;
2210
2211         /*
2212          * Data-on-MDT optimization - read data along with OPEN and return it
2213          * in reply. Do that only if we have both DOM and LAYOUT locks.
2214          */
2215         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req) &&
2216             info->mti_attr.ma_lmm != NULL &&
2217             mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
2218                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2219                                           &lhc->mlh_reg_lh);
2220         }
2221
2222         return rc;
2223 }
2224
2225 static long mdt_reint_opcode(struct ptlrpc_request *req,
2226                              const struct req_format **fmt)
2227 {
2228         struct mdt_device       *mdt;
2229         struct mdt_rec_reint    *rec;
2230         long                     opc;
2231
2232         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2233         if (rec != NULL) {
2234                 opc = rec->rr_opcode;
2235                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2236                 if (opc < REINT_MAX && fmt[opc] != NULL)
2237                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2238                 else {
2239                         mdt = mdt_exp2dev(req->rq_export);
2240                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2241                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2242                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2243                         opc = err_serious(-EFAULT);
2244                 }
2245         } else {
2246                 opc = err_serious(-EFAULT);
2247         }
2248         return opc;
2249 }
2250
2251 static int mdt_reint(struct tgt_session_info *tsi)
2252 {
2253         long opc;
2254         int  rc;
2255         static const struct req_format *reint_fmts[REINT_MAX] = {
2256                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2257                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2258                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2259                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2260                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2261                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2262                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2263                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2264                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2265                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2266         };
2267
2268         ENTRY;
2269
2270         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2271         if (opc >= 0) {
2272                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2273                 /*
2274                  * No lock possible here from client to pass it to reint code
2275                  * path.
2276                  */
2277                 rc = mdt_reint_internal(info, NULL, opc);
2278                 mdt_thread_info_fini(info);
2279         } else {
2280                 rc = opc;
2281         }
2282
2283         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2284         RETURN(rc);
2285 }
2286
2287 /* this should sync the whole device */
2288 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2289 {
2290         struct dt_device *dt = mdt->mdt_bottom;
2291         int rc;
2292         ENTRY;
2293
2294         rc = dt->dd_ops->dt_sync(env, dt);
2295         RETURN(rc);
2296 }
2297
2298 /* this should sync this object */
2299 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2300                            struct mdt_object *mo)
2301 {
2302         int rc;
2303
2304         ENTRY;
2305
2306         if (!mdt_object_exists(mo)) {
2307                 CWARN("%s: non existing object "DFID": rc = %d\n",
2308                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2309                       -ESTALE);
2310                 RETURN(-ESTALE);
2311         }
2312
2313         rc = mo_object_sync(env, mdt_object_child(mo));
2314
2315         RETURN(rc);
2316 }
2317
2318 static int mdt_sync(struct tgt_session_info *tsi)
2319 {
2320         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2321         struct req_capsule      *pill = tsi->tsi_pill;
2322         struct mdt_body         *body;
2323         int                      rc;
2324
2325         ENTRY;
2326
2327         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2328                 RETURN(err_serious(-ENOMEM));
2329
2330         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2331                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2332         } else {
2333                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2334
2335                 /* sync an object */
2336                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2337                                      info->mti_object);
2338                 if (rc == 0) {
2339                         const struct lu_fid *fid;
2340                         struct lu_attr *la = &info->mti_attr.ma_attr;
2341
2342                         info->mti_attr.ma_need = MA_INODE;
2343                         info->mti_attr.ma_valid = 0;
2344                         rc = mdt_attr_get_complex(info, info->mti_object,
2345                                                   &info->mti_attr);
2346                         if (rc == 0) {
2347                                 body = req_capsule_server_get(pill,
2348                                                               &RMF_MDT_BODY);
2349                                 fid = mdt_object_fid(info->mti_object);
2350                                 mdt_pack_attr2body(info, body, la, fid);
2351                         }
2352                 }
2353                 mdt_thread_info_fini(info);
2354         }
2355         if (rc == 0)
2356                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2357
2358         RETURN(rc);
2359 }
2360
2361 static int mdt_data_sync(struct tgt_session_info *tsi)
2362 {
2363         struct mdt_thread_info *info;
2364         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2365         struct ost_body *body = tsi->tsi_ost_body;
2366         struct ost_body *repbody;
2367         struct mdt_object *mo = NULL;
2368         struct md_attr *ma;
2369         int rc = 0;
2370
2371         ENTRY;
2372
2373         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2374
2375         /* if no fid is specified then do nothing,
2376          * device sync is done via MDS_SYNC */
2377         if (fid_is_zero(&tsi->tsi_fid))
2378                 RETURN(0);
2379
2380         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2381         if (IS_ERR(mo))
2382                 RETURN(PTR_ERR(mo));
2383
2384         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2385         if (rc)
2386                 GOTO(put, rc);
2387
2388         repbody->oa.o_oi = body->oa.o_oi;
2389         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2390
2391         info = tsi2mdt_info(tsi);
2392         ma = &info->mti_attr;
2393         ma->ma_need = MA_INODE;
2394         ma->ma_valid = 0;
2395         rc = mdt_attr_get_complex(info, mo, ma);
2396         if (rc == 0)
2397                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2398         else
2399                 rc = 0;
2400         mdt_thread_info_fini(info);
2401
2402         EXIT;
2403 put:
2404         if (mo != NULL)
2405                 mdt_object_put(tsi->tsi_env, mo);
2406         return rc;
2407 }
2408
2409 /*
2410  * Handle quota control requests to consult current usage/limit, but also
2411  * to configure quota enforcement
2412  */
2413 static int mdt_quotactl(struct tgt_session_info *tsi)
2414 {
2415         struct obd_export       *exp  = tsi->tsi_exp;
2416         struct req_capsule      *pill = tsi->tsi_pill;
2417         struct obd_quotactl     *oqctl, *repoqc;
2418         int                      id, rc;
2419         struct mdt_device       *mdt = mdt_exp2dev(exp);
2420         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2421         struct lu_nodemap       *nodemap;
2422         ENTRY;
2423
2424         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2425         if (oqctl == NULL)
2426                 RETURN(err_serious(-EPROTO));
2427
2428         rc = req_capsule_server_pack(pill);
2429         if (rc)
2430                 RETURN(err_serious(rc));
2431
2432         nodemap = nodemap_get_from_exp(exp);
2433         if (IS_ERR(nodemap))
2434                 RETURN(PTR_ERR(nodemap));
2435
2436         switch (oqctl->qc_cmd) {
2437                 /* master quotactl */
2438         case Q_SETINFO:
2439         case Q_SETQUOTA:
2440         case LUSTRE_Q_SETDEFAULT:
2441                 if (!nodemap_can_setquota(nodemap))
2442                         GOTO(out_nodemap, rc = -EPERM);
2443         case Q_GETINFO:
2444         case Q_GETQUOTA:
2445         case LUSTRE_Q_GETDEFAULT:
2446                 if (qmt == NULL)
2447                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2448                 /* slave quotactl */
2449         case Q_GETOINFO:
2450         case Q_GETOQUOTA:
2451                 break;
2452         default:
2453                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2454                 GOTO(out_nodemap, rc = -EFAULT);
2455         }
2456
2457         id = oqctl->qc_id;
2458         switch (oqctl->qc_type) {
2459         case USRQUOTA:
2460                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2461                                     NODEMAP_CLIENT_TO_FS, id);
2462                 break;
2463         case GRPQUOTA:
2464                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2465                                     NODEMAP_CLIENT_TO_FS, id);
2466                 break;
2467         case PRJQUOTA:
2468                 /* todo: check/map project id */
2469                 id = oqctl->qc_id;
2470                 break;
2471         default:
2472                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2473         }
2474         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2475         if (repoqc == NULL)
2476                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2477
2478         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2479                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2480
2481         if (oqctl->qc_id != id)
2482                 swap(oqctl->qc_id, id);
2483
2484         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2485                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2486                         RETURN(-EINPROGRESS);
2487         }
2488
2489         switch (oqctl->qc_cmd) {
2490
2491         case Q_GETINFO:
2492         case Q_SETINFO:
2493         case Q_SETQUOTA:
2494         case Q_GETQUOTA:
2495         case LUSTRE_Q_SETDEFAULT:
2496         case LUSTRE_Q_GETDEFAULT:
2497                 /* forward quotactl request to QMT */
2498                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2499                 break;
2500
2501         case Q_GETOINFO:
2502         case Q_GETOQUOTA:
2503                 /* slave quotactl */
2504                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2505                                    oqctl);
2506                 break;
2507
2508         default:
2509                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2510                 GOTO(out_nodemap, rc = -EFAULT);
2511         }
2512
2513         if (oqctl->qc_id != id)
2514                 swap(oqctl->qc_id, id);
2515
2516         *repoqc = *oqctl;
2517
2518         EXIT;
2519
2520 out_nodemap:
2521         nodemap_putref(nodemap);
2522
2523         return rc;
2524 }
2525
2526 /** clone llog ctxt from child (mdd)
2527  * This allows remote llog (replicator) access.
2528  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2529  * context was originally set up, or we can handle them directly.
2530  * I choose the latter, but that means I need any llog
2531  * contexts set up by child to be accessable by the mdt.  So we clone the
2532  * context into our context list here.
2533  */
2534 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2535                                int idx)
2536 {
2537         struct md_device  *next = mdt->mdt_child;
2538         struct llog_ctxt *ctxt;
2539         int rc;
2540
2541         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2542                 return 0;
2543
2544         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2545         if (rc || ctxt == NULL) {
2546                 return 0;
2547         }
2548
2549         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2550         if (rc)
2551                 CERROR("Can't set mdt ctxt %d\n", rc);
2552
2553         return rc;
2554 }
2555
2556 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2557                                  struct mdt_device *mdt, int idx)
2558 {
2559         struct llog_ctxt *ctxt;
2560
2561         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2562         if (ctxt == NULL)
2563                 return 0;
2564         /* Put once for the get we just did, and once for the clone */
2565         llog_ctxt_put(ctxt);
2566         llog_ctxt_put(ctxt);
2567         return 0;
2568 }
2569
2570 /*
2571  * sec context handlers
2572  */
2573 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2574 {
2575         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2576
2577         return 0;
2578 }
2579
2580 /*
2581  * quota request handlers
2582  */
2583 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2584 {
2585         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2586         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2587         int                      rc;
2588         ENTRY;
2589
2590         if (qmt == NULL)
2591                 RETURN(err_serious(-EOPNOTSUPP));
2592
2593         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2594         RETURN(rc);
2595 }
2596
2597 struct mdt_object *mdt_object_new(const struct lu_env *env,
2598                                   struct mdt_device *d,
2599                                   const struct lu_fid *f)
2600 {
2601         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2602         struct lu_object *o;
2603         struct mdt_object *m;
2604         ENTRY;
2605
2606         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2607         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2608         if (unlikely(IS_ERR(o)))
2609                 m = (struct mdt_object *)o;
2610         else
2611                 m = mdt_obj(o);
2612         RETURN(m);
2613 }
2614
2615 struct mdt_object *mdt_object_find(const struct lu_env *env,
2616                                    struct mdt_device *d,
2617                                    const struct lu_fid *f)
2618 {
2619         struct lu_object *o;
2620         struct mdt_object *m;
2621         ENTRY;
2622
2623         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2624         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2625         if (unlikely(IS_ERR(o)))
2626                 m = (struct mdt_object *)o;
2627         else
2628                 m = mdt_obj(o);
2629
2630         RETURN(m);
2631 }
2632
2633 /**
2634  * Asyncronous commit for mdt device.
2635  *
2636  * Pass asynchonous commit call down the MDS stack.
2637  *
2638  * \param env environment
2639  * \param mdt the mdt device
2640  */
2641 static void mdt_device_commit_async(const struct lu_env *env,
2642                                     struct mdt_device *mdt)
2643 {
2644         struct dt_device *dt = mdt->mdt_bottom;
2645         int rc;
2646         ENTRY;
2647
2648         rc = dt->dd_ops->dt_commit_async(env, dt);
2649         if (unlikely(rc != 0))
2650                 CWARN("%s: async commit start failed: rc = %d\n",
2651                       mdt_obd_name(mdt), rc);
2652         atomic_inc(&mdt->mdt_async_commit_count);
2653         EXIT;
2654 }
2655
2656 /**
2657  * Mark the lock as "synchonous".
2658  *
2659  * Mark the lock to deffer transaction commit to the unlock time.
2660  *
2661  * \param lock the lock to mark as "synchonous"
2662  *
2663  * \see mdt_is_lock_sync
2664  * \see mdt_save_lock
2665  */
2666 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2667 {
2668         lock->l_ast_data = (void*)1;
2669 }
2670
2671 /**
2672  * Check whehter the lock "synchonous" or not.
2673  *
2674  * \param lock the lock to check
2675  * \retval 1 the lock is "synchonous"
2676  * \retval 0 the lock isn't "synchronous"
2677  *
2678  * \see mdt_set_lock_sync
2679  * \see mdt_save_lock
2680  */
2681 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2682 {
2683         return lock->l_ast_data != NULL;
2684 }
2685
2686 /**
2687  * Blocking AST for mdt locks.
2688  *
2689  * Starts transaction commit if in case of COS lock conflict or
2690  * deffers such a commit to the mdt_save_lock.
2691  *
2692  * \param lock the lock which blocks a request or cancelling lock
2693  * \param desc unused
2694  * \param data unused
2695  * \param flag indicates whether this cancelling or blocking callback
2696  * \retval 0
2697  * \see ldlm_blocking_ast_nocheck
2698  */
2699 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2700                      void *data, int flag)
2701 {
2702         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2703         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2704         struct ldlm_cb_set_arg *arg = data;
2705         bool commit_async = false;
2706         int rc;
2707         ENTRY;
2708
2709         if (flag == LDLM_CB_CANCELING)
2710                 RETURN(0);
2711
2712         lock_res_and_lock(lock);
2713         if (lock->l_blocking_ast != mdt_blocking_ast) {
2714                 unlock_res_and_lock(lock);
2715                 RETURN(0);
2716         }
2717
2718         /* A blocking ast may be sent from ldlm_lock_decref_internal
2719          * when the last reference to a local lock was released and
2720          * during blocking event from ldlm_work_bl_ast_lock().
2721          * The 'data' parameter is l_ast_data in the first case and
2722          * callback arguments in the second one. Distinguish them by that.
2723          */
2724         if (!data || data == lock->l_ast_data || !arg->bl_desc)
2725                 goto skip_cos_checks;
2726
2727         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
2728                 if (mdt_cos_is_enabled(mdt)) {
2729                         if (!arg->bl_desc->bl_same_client)
2730                                 mdt_set_lock_sync(lock);
2731                 } else if (mdt_slc_is_enabled(mdt) &&
2732                            arg->bl_desc->bl_cos_incompat) {
2733                         mdt_set_lock_sync(lock);
2734                         /*
2735                          * we may do extra commit here, but there is a small
2736                          * window to miss a commit: lock was unlocked (saved),
2737                          * then a conflict lock queued and we come here, but
2738                          * REP-ACK not received, so lock was not converted to
2739                          * COS mode yet.
2740                          * Fortunately this window is quite small, so the
2741                          * extra commit should be rare (not to say distributed
2742                          * operation is rare too).
2743                          */
2744                         commit_async = true;
2745                 }
2746         } else if (lock->l_req_mode == LCK_COS) {
2747                 commit_async = true;
2748         }
2749
2750 skip_cos_checks:
2751         rc = ldlm_blocking_ast_nocheck(lock);
2752
2753         if (commit_async) {
2754                 struct lu_env env;
2755
2756                 rc = lu_env_init(&env, LCT_LOCAL);
2757                 if (unlikely(rc != 0))
2758                         CWARN("%s: lu_env initialization failed, cannot "
2759                               "start asynchronous commit: rc = %d\n",
2760                               obd->obd_name, rc);
2761                 else
2762                         mdt_device_commit_async(&env, mdt);
2763                 lu_env_fini(&env);
2764         }
2765         RETURN(rc);
2766 }
2767
2768 /*
2769  * Blocking AST for cross-MDT lock
2770  *
2771  * Discard lock from uncommitted_slc_locks and cancel it.
2772  *
2773  * \param lock  the lock which blocks a request or cancelling lock
2774  * \param desc  unused
2775  * \param data  unused
2776  * \param flag  indicates whether this cancelling or blocking callback
2777  * \retval      0 on success
2778  * \retval      negative number on error
2779  */
2780 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2781                             void *data, int flag)
2782 {
2783         int rc = 0;
2784         ENTRY;
2785
2786         switch (flag) {
2787         case LDLM_CB_BLOCKING: {
2788                 struct lustre_handle lockh;
2789
2790                 ldlm_lock2handle(lock, &lockh);
2791                 rc = ldlm_cli_cancel(&lockh,
2792                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2793                 if (rc < 0) {
2794                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2795                         RETURN(rc);
2796                 }
2797                 break;
2798         }
2799         case LDLM_CB_CANCELING: {
2800                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2801                 struct mdt_device *mdt =
2802                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2803
2804                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2805
2806                 /* discard slc lock here so that it can be cleaned anytime,
2807                  * especially for cleanup_resource() */
2808                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2809
2810                 /* once we cache lock, l_ast_data is set to mdt_object */
2811                 if (lock->l_ast_data != NULL) {
2812                         struct mdt_object *mo = lock->l_ast_data;
2813                         struct lu_env env;
2814
2815                         rc = lu_env_init(&env, LCT_MD_THREAD);
2816                         if (unlikely(rc != 0)) {
2817                                 CWARN("%s: lu_env initialization failed, object"
2818                                       "%p "DFID" is leaked!\n",
2819                                       obd->obd_name, mo,
2820                                       PFID(mdt_object_fid(mo)));
2821                                 RETURN(rc);
2822                         }
2823
2824                         if (lock->l_policy_data.l_inodebits.bits &
2825                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2826                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2827                                 mo->mot_cache_attr = 0;
2828                         }
2829                         mdt_object_put(&env, mo);
2830                         lu_env_fini(&env);
2831                 }
2832                 break;
2833         }
2834         default:
2835                 LBUG();
2836         }
2837
2838         RETURN(rc);
2839 }
2840
2841 int mdt_check_resent_lock(struct mdt_thread_info *info,
2842                           struct mdt_object *mo,
2843                           struct mdt_lock_handle *lhc)
2844 {
2845         /* the lock might already be gotten in ldlm_handle_enqueue() */
2846         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2847                 struct ptlrpc_request *req = mdt_info_req(info);
2848                 struct ldlm_lock      *lock;
2849
2850                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2851                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2852                 if (lock == NULL) {
2853                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2854                          * a resend case, however, it could be already destroyed
2855                          * due to client eviction or a raced cancel RPC. */
2856                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2857                                           lhc->mlh_reg_lh.cookie);
2858                         RETURN(-ESTALE);
2859                 }
2860
2861                 if (!fid_res_name_eq(mdt_object_fid(mo),
2862                                      &lock->l_resource->lr_name)) {
2863                         CWARN("%s: Although resent, but still not "
2864                               "get child lock:"DFID"\n",
2865                               info->mti_exp->exp_obd->obd_name,
2866                               PFID(mdt_object_fid(mo)));
2867                         LDLM_LOCK_PUT(lock);
2868                         RETURN(-EPROTO);
2869                 }
2870                 LDLM_LOCK_PUT(lock);
2871                 return 0;
2872         }
2873         return 1;
2874 }
2875
2876 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
2877 {
2878         mdt_object_get(NULL, lock->l_ast_data);
2879 }
2880
2881 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
2882                                struct mdt_object *o, const struct lu_fid *fid,
2883                                struct lustre_handle *lh, enum ldlm_mode mode,
2884                                __u64 *ibits, __u64 trybits, bool cache)
2885 {
2886         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
2887         union ldlm_policy_data *policy = &mti->mti_policy;
2888         struct ldlm_res_id *res_id = &mti->mti_res_id;
2889         int rc = 0;
2890         ENTRY;
2891
2892         LASSERT(mdt_object_remote(o));
2893
2894         fid_build_reg_res_name(fid, res_id);
2895
2896         memset(einfo, 0, sizeof(*einfo));
2897         einfo->ei_type = LDLM_IBITS;
2898         einfo->ei_mode = mode;
2899         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2900         einfo->ei_cb_cp = ldlm_completion_ast;
2901         einfo->ei_enq_slave = 0;
2902         einfo->ei_res_id = res_id;
2903
2904         if (cache) {
2905                 /*
2906                  * if we cache lock, couple lock with mdt_object, so that object
2907                  * can be easily found in lock ASTs.
2908                  */
2909                 einfo->ei_cbdata = o;
2910                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
2911         }
2912
2913         memset(policy, 0, sizeof(*policy));
2914         policy->l_inodebits.bits = *ibits;
2915         policy->l_inodebits.try_bits = trybits;
2916
2917         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2918                             policy);
2919
2920         /* Return successfully acquired bits to a caller */
2921         if (rc == 0) {
2922                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
2923
2924                 LASSERT(lock);
2925                 *ibits = lock->l_policy_data.l_inodebits.bits;
2926                 LDLM_LOCK_PUT(lock);
2927         }
2928         RETURN(rc);
2929 }
2930
2931 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2932                            const struct lu_fid *fid, struct lustre_handle *lh,
2933                            enum ldlm_mode mode, __u64 ibits, bool cache)
2934 {
2935         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
2936                                           cache);
2937 }
2938
2939 static int mdt_object_local_lock(struct mdt_thread_info *info,
2940                                  struct mdt_object *o,
2941                                  struct mdt_lock_handle *lh, __u64 *ibits,
2942                                  __u64 trybits, bool cos_incompat)
2943 {
2944         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2945         union ldlm_policy_data *policy = &info->mti_policy;
2946         struct ldlm_res_id *res_id = &info->mti_res_id;
2947         __u64 dlmflags = 0, *cookie = NULL;
2948         int rc;
2949         ENTRY;
2950
2951         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2952         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2953         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2954         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2955
2956         if (cos_incompat) {
2957                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2958                         lh->mlh_reg_mode == LCK_EX);
2959                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2960         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2961                 dlmflags |= LDLM_FL_COS_ENABLED;
2962         }
2963
2964         /* Only enqueue LOOKUP lock for remote object */
2965         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
2966
2967         /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
2968         if (lh->mlh_type == MDT_REG_LOCK && lh->mlh_reg_mode == LCK_EX &&
2969             *ibits == MDS_INODELOCK_OPEN)
2970                 dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
2971
2972         if (lh->mlh_type == MDT_PDO_LOCK) {
2973                 /* check for exists after object is locked */
2974                 if (mdt_object_exists(o) == 0) {
2975                         /* Non-existent object shouldn't have PDO lock */
2976                         RETURN(-ESTALE);
2977                 } else {
2978                         /* Non-dir object shouldn't have PDO lock */
2979                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2980                                 RETURN(-ENOTDIR);
2981                 }
2982         }
2983
2984         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2985         dlmflags |= LDLM_FL_ATOMIC_CB;
2986
2987         if (info->mti_exp)
2988                 cookie = &info->mti_exp->exp_handle.h_cookie;
2989
2990         /*
2991          * Take PDO lock on whole directory and build correct @res_id for lock
2992          * on part of directory.
2993          */
2994         if (lh->mlh_pdo_hash != 0) {
2995                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2996                 mdt_lock_pdo_mode(info, o, lh);
2997                 if (lh->mlh_pdo_mode != LCK_NL) {
2998                         /*
2999                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
3000                          * is never going to be sent to client and we do not
3001                          * want it slowed down due to possible cancels.
3002                          */
3003                         policy->l_inodebits.bits =
3004                                 *ibits & MDS_INODELOCK_UPDATE;
3005                         policy->l_inodebits.try_bits =
3006                                 trybits & MDS_INODELOCK_UPDATE;
3007                         /* at least one of them should be set */
3008                         LASSERT(policy->l_inodebits.bits |
3009                                 policy->l_inodebits.try_bits);
3010                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
3011                                           lh->mlh_pdo_mode, policy, res_id,
3012                                           dlmflags, cookie);
3013                         if (unlikely(rc != 0))
3014                                 GOTO(out_unlock, rc);
3015                 }
3016
3017                 /*
3018                  * Finish res_id initializing by name hash marking part of
3019                  * directory which is taking modification.
3020                  */
3021                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3022         }
3023
3024         policy->l_inodebits.bits = *ibits;
3025         policy->l_inodebits.try_bits = trybits;
3026
3027         /*
3028          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
3029          * going to be sent to client. If it is - mdt_intent_policy() path will
3030          * fix it up and turn FL_LOCAL flag off.
3031          */
3032         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
3033                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
3034                           cookie);
3035 out_unlock:
3036         if (rc != 0)
3037                 mdt_object_unlock(info, o, lh, 1);
3038         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
3039                    lh->mlh_pdo_hash != 0 &&
3040                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3041                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3042
3043         /* Return successfully acquired bits to a caller */
3044         if (rc == 0) {
3045                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
3046
3047                 LASSERT(lock);
3048                 *ibits = lock->l_policy_data.l_inodebits.bits;
3049                 LDLM_LOCK_PUT(lock);
3050         }
3051         RETURN(rc);
3052 }
3053
3054 static int
3055 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
3056                          struct mdt_lock_handle *lh, __u64 *ibits,
3057                          __u64 trybits, bool cos_incompat)
3058 {
3059         struct mdt_lock_handle *local_lh = NULL;
3060         int rc;
3061         ENTRY;
3062
3063         if (!mdt_object_remote(o)) {
3064                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3065                                            cos_incompat);
3066                 RETURN(rc);
3067         }
3068
3069         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3070         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3071                     MDS_INODELOCK_XATTR);
3072
3073         /* Only enqueue LOOKUP lock for remote object */
3074         if (*ibits & MDS_INODELOCK_LOOKUP) {
3075                 __u64 local = MDS_INODELOCK_LOOKUP;
3076
3077                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3078                                            cos_incompat);
3079                 if (rc != ELDLM_OK)
3080                         RETURN(rc);
3081
3082                 local_lh = lh;
3083         }
3084
3085         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3086                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3087                  * enqueue RPC can only request 1 lock, to avoid extra
3088                  * RPC, so it will instead enqueue EX lock for remote
3089                  * object anyway XXX*/
3090                 if (lh->mlh_type == MDT_PDO_LOCK &&
3091                     lh->mlh_pdo_hash != 0) {
3092                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
3093                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
3094                                PFID(mdt_object_fid(o)));
3095                         lh->mlh_pdo_hash = 0;
3096                         lh->mlh_rreg_mode = LCK_EX;
3097                         lh->mlh_type = MDT_REG_LOCK;
3098                 }
3099
3100                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3101                                                 &lh->mlh_rreg_lh,
3102                                                 lh->mlh_rreg_mode,
3103                                                 ibits, trybits, false);
3104                 if (rc != ELDLM_OK) {
3105                         if (local_lh != NULL)
3106                                 mdt_object_unlock(info, o, local_lh, rc);
3107                         RETURN(rc);
3108                 }
3109         }
3110
3111         /* other components like LFSCK can use lockless access
3112          * and populate cache, so we better invalidate it */
3113         mo_invalidate(info->mti_env, mdt_object_child(o));
3114
3115         RETURN(0);
3116 }
3117
3118 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3119                     struct mdt_lock_handle *lh, __u64 ibits)
3120 {
3121         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3122 }
3123
3124 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3125                           struct mdt_lock_handle *lh, __u64 ibits,
3126                           bool cos_incompat)
3127 {
3128         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3129         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3130                                         cos_incompat);
3131 }
3132
3133 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3134                         struct mdt_lock_handle *lh, __u64 *ibits,
3135                         __u64 trybits, bool cos_incompat)
3136 {
3137         bool trylock_only = *ibits == 0;
3138         int rc;
3139
3140         LASSERT(!(*ibits & trybits));
3141         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3142                                       cos_incompat);
3143         if (rc && trylock_only) { /* clear error for try ibits lock only */
3144                 LASSERT(*ibits == 0);
3145                 rc = 0;
3146         }
3147         return rc;
3148 }
3149
3150 /**
3151  * Save a lock within request object.
3152  *
3153  * Keep the lock referenced until whether client ACK or transaction
3154  * commit happens or release the lock immediately depending on input
3155  * parameters. If COS is ON, a write lock is converted to COS lock
3156  * before saving.
3157  *
3158  * \param info thead info object
3159  * \param h lock handle
3160  * \param mode lock mode
3161  * \param decref force immediate lock releasing
3162  */
3163 void m