Whamcloud - gitweb
LU-12930 various: use schedule_timeout_*interruptible
[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_LUSTRE_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_t(unsigned int,
602                                                               mdt->mdt_max_ea_size,
603                                                               XATTR_SIZE_MAX);
604                                         OBD_ALLOC_LARGE(info->mti_big_acl,
605                                                         info->mti_big_aclsize);
606                                         if (info->mti_big_acl == NULL) {
607                                                 info->mti_big_aclsize = 0;
608                                                 CERROR("%s: unable to grow "
609                                                        DFID" ACL buffer\n",
610                                                        mdt_obd_name(mdt),
611                                                        PFID(mdt_object_fid(o)));
612                                                 RETURN(-ENOMEM);
613                                         }
614                                 }
615
616                                 CDEBUG(D_INODE, "%s: grow the "DFID
617                                        " ACL buffer to size %d\n",
618                                        mdt_obd_name(mdt),
619                                        PFID(mdt_object_fid(o)),
620                                        info->mti_big_aclsize);
621
622                                 buf->lb_buf = info->mti_big_acl;
623                                 buf->lb_len = info->mti_big_aclsize;
624
625                                 goto again;
626                         }
627
628                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
629                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
630                 }
631         } else {
632                 int client;
633                 int server;
634                 int acl_buflen;
635                 int lmm_buflen = 0;
636                 int lmmsize = 0;
637
638                 acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
639                 if (acl_buflen >= rc)
640                         goto map;
641
642                 /* If LOV/LMA EA is small, we can reuse part of their buffer */
643                 client = ptlrpc_req_get_repsize(pill->rc_req);
644                 server = lustre_packed_msg_size(pill->rc_req->rq_repmsg);
645                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
646                         lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD,
647                                                           RCL_SERVER);
648                         lmmsize = repbody->mbo_eadatasize;
649                 }
650
651                 if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) {
652                         CDEBUG(D_INODE, "%s: client prepared buffer size %d "
653                                "is not big enough with the ACL size %d (%d)\n",
654                                mdt_obd_name(mdt), client, rc,
655                                server - acl_buflen - lmm_buflen + rc + lmmsize);
656                         repbody->mbo_aclsize = 0;
657                         repbody->mbo_valid &= ~OBD_MD_FLACL;
658                         RETURN(-ERANGE);
659                 }
660
661 map:
662                 if (buf->lb_buf == info->mti_big_acl)
663                         info->mti_big_acl_used = 1;
664
665                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
666                                      rc, NODEMAP_FS_TO_CLIENT);
667                 /* if all ACLs mapped out, rc is still >= 0 */
668                 if (rc < 0) {
669                         CERROR("%s: nodemap_map_acl unable to parse "DFID
670                                " ACL: rc = %d\n", mdt_obd_name(mdt),
671                                PFID(mdt_object_fid(o)), rc);
672                         repbody->mbo_aclsize = 0;
673                         repbody->mbo_valid &= ~OBD_MD_FLACL;
674                 } else {
675                         repbody->mbo_aclsize = rc;
676                         repbody->mbo_valid |= OBD_MD_FLACL;
677                         rc = 0;
678                 }
679         }
680
681         RETURN(rc);
682 }
683 #endif
684
685 /* XXX Look into layout in MDT layer. */
686 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
687 {
688         struct lov_comp_md_v1   *comp_v1;
689         struct lov_mds_md       *v1;
690         int                      i;
691
692         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
693                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
694
695                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
696                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
697                                 comp_v1->lcm_entries[i].lcme_offset);
698                         /* We don't support partial release for now */
699                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
700                                 return false;
701                 }
702                 return true;
703         } else {
704                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
705                         true : false;
706         }
707 }
708
709 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
710                         const struct lu_attr *attr, const struct lu_fid *fid)
711 {
712         struct md_attr *ma = &info->mti_attr;
713         struct obd_export *exp = info->mti_exp;
714         struct lu_nodemap *nodemap = NULL;
715
716         LASSERT(ma->ma_valid & MA_INODE);
717
718         if (attr->la_valid & LA_ATIME) {
719                 b->mbo_atime = attr->la_atime;
720                 b->mbo_valid |= OBD_MD_FLATIME;
721         }
722         if (attr->la_valid & LA_MTIME) {
723                 b->mbo_mtime = attr->la_mtime;
724                 b->mbo_valid |= OBD_MD_FLMTIME;
725         }
726         if (attr->la_valid & LA_CTIME) {
727                 b->mbo_ctime = attr->la_ctime;
728                 b->mbo_valid |= OBD_MD_FLCTIME;
729         }
730         if (attr->la_valid & LA_FLAGS) {
731                 b->mbo_flags = attr->la_flags;
732                 b->mbo_valid |= OBD_MD_FLFLAGS;
733         }
734         if (attr->la_valid & LA_NLINK) {
735                 b->mbo_nlink = attr->la_nlink;
736                 b->mbo_valid |= OBD_MD_FLNLINK;
737         }
738         if (attr->la_valid & (LA_UID|LA_GID)) {
739                 nodemap = nodemap_get_from_exp(exp);
740                 if (IS_ERR(nodemap))
741                         goto out;
742         }
743         if (attr->la_valid & LA_UID) {
744                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
745                                             NODEMAP_FS_TO_CLIENT,
746                                             attr->la_uid);
747                 b->mbo_valid |= OBD_MD_FLUID;
748         }
749         if (attr->la_valid & LA_GID) {
750                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
751                                             NODEMAP_FS_TO_CLIENT,
752                                             attr->la_gid);
753                 b->mbo_valid |= OBD_MD_FLGID;
754         }
755
756         if (attr->la_valid & LA_PROJID) {
757                 /* TODO, nodemap for project id */
758                 b->mbo_projid = attr->la_projid;
759                 b->mbo_valid |= OBD_MD_FLPROJID;
760         }
761
762         b->mbo_mode = attr->la_mode;
763         if (attr->la_valid & LA_MODE)
764                 b->mbo_valid |= OBD_MD_FLMODE;
765         if (attr->la_valid & LA_TYPE)
766                 b->mbo_valid |= OBD_MD_FLTYPE;
767
768         if (fid != NULL) {
769                 b->mbo_fid1 = *fid;
770                 b->mbo_valid |= OBD_MD_FLID;
771                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
772                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
773         }
774
775         if (!(attr->la_valid & LA_TYPE))
776                 return;
777
778         b->mbo_rdev   = attr->la_rdev;
779         b->mbo_size   = attr->la_size;
780         b->mbo_blocks = attr->la_blocks;
781
782         if (!S_ISREG(attr->la_mode)) {
783                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
784         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
785                 /* means no objects are allocated on osts. */
786                 LASSERT(!(ma->ma_valid & MA_LOV));
787                 /* just ignore blocks occupied by extend attributes on MDS */
788                 b->mbo_blocks = 0;
789                 /* if no object is allocated on osts, the size on mds is valid.
790                  * b=22272 */
791                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
792         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
793                 if (mdt_hsm_is_released(ma->ma_lmm)) {
794                         /* A released file stores its size on MDS. */
795                         /* But return 1 block for released file, unless tools
796                          * like tar will consider it fully sparse. (LU-3864)
797                          */
798                         if (unlikely(b->mbo_size == 0))
799                                 b->mbo_blocks = 0;
800                         else
801                                 b->mbo_blocks = 1;
802                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
803                 } else if (info->mti_som_valid) { /* som is valid */
804                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
805                 } else if (ma->ma_valid & MA_SOM) { /* lsom is valid */
806                         b->mbo_valid |= OBD_MD_FLLAZYSIZE | OBD_MD_FLLAZYBLOCKS;
807                         b->mbo_size = ma->ma_som.ms_size;
808                         b->mbo_blocks = ma->ma_som.ms_blocks;
809                 }
810         }
811
812         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE ||
813                             b->mbo_valid & OBD_MD_FLLAZYSIZE))
814                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
815                        PFID(fid), (unsigned long long)b->mbo_size);
816
817 out:
818         if (!IS_ERR_OR_NULL(nodemap))
819                 nodemap_putref(nodemap);
820 }
821
822 static inline int mdt_body_has_lov(const struct lu_attr *la,
823                                    const struct mdt_body *body)
824 {
825         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
826                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
827 }
828
829 void mdt_client_compatibility(struct mdt_thread_info *info)
830 {
831         struct mdt_body       *body;
832         struct ptlrpc_request *req = mdt_info_req(info);
833         struct obd_export     *exp = req->rq_export;
834         struct md_attr        *ma = &info->mti_attr;
835         struct lu_attr        *la = &ma->ma_attr;
836         ENTRY;
837
838         if (exp_connect_layout(exp))
839                 /* the client can deal with 16-bit lmm_stripe_count */
840                 RETURN_EXIT;
841
842         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
843
844         if (!mdt_body_has_lov(la, body))
845                 RETURN_EXIT;
846
847         /* now we have a reply with a lov for a client not compatible with the
848          * layout lock so we have to clean the layout generation number */
849         if (S_ISREG(la->la_mode))
850                 ma->ma_lmm->lmm_layout_gen = 0;
851         EXIT;
852 }
853
854 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
855                                    struct mdt_object *o)
856 {
857         const struct lu_env *env = info->mti_env;
858         int rc, rc2;
859
860         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
861                           XATTR_NAME_LOV);
862
863         if (rc == -ENODATA)
864                 rc = 0;
865
866         if (rc < 0)
867                 goto out;
868
869         /* Is it a directory? Let's check for the LMV as well */
870         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
871                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
872                                    XATTR_NAME_LMV);
873
874                 if (rc2 == -ENODATA)
875                         rc2 = mo_xattr_get(env, mdt_object_child(o),
876                                            &LU_BUF_NULL,
877                                            XATTR_NAME_DEFAULT_LMV);
878
879                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
880                         rc = rc2;
881         }
882
883 out:
884         return rc;
885 }
886
887 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
888                       const char *name)
889 {
890         const struct lu_env *env = info->mti_env;
891         int rc;
892         ENTRY;
893
894         LASSERT(info->mti_big_lmm_used == 0);
895         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
896         if (rc < 0)
897                 RETURN(rc);
898
899         /* big_lmm may need to be grown */
900         if (info->mti_big_lmmsize < rc) {
901                 int size = size_roundup_power2(rc);
902
903                 if (info->mti_big_lmmsize > 0) {
904                         /* free old buffer */
905                         LASSERT(info->mti_big_lmm);
906                         OBD_FREE_LARGE(info->mti_big_lmm,
907                                        info->mti_big_lmmsize);
908                         info->mti_big_lmm = NULL;
909                         info->mti_big_lmmsize = 0;
910                 }
911
912                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
913                 if (info->mti_big_lmm == NULL)
914                         RETURN(-ENOMEM);
915                 info->mti_big_lmmsize = size;
916         }
917         LASSERT(info->mti_big_lmmsize >= rc);
918
919         info->mti_buf.lb_buf = info->mti_big_lmm;
920         info->mti_buf.lb_len = info->mti_big_lmmsize;
921         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
922
923         RETURN(rc);
924 }
925
926 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
927                    struct md_attr *ma, const char *name)
928 {
929         struct md_object *next = mdt_object_child(o);
930         struct lu_buf    *buf = &info->mti_buf;
931         int rc;
932
933         if (strcmp(name, XATTR_NAME_LOV) == 0) {
934                 buf->lb_buf = ma->ma_lmm;
935                 buf->lb_len = ma->ma_lmm_size;
936                 LASSERT(!(ma->ma_valid & MA_LOV));
937         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
938                 buf->lb_buf = ma->ma_lmv;
939                 buf->lb_len = ma->ma_lmv_size;
940                 LASSERT(!(ma->ma_valid & MA_LMV));
941         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
942                 buf->lb_buf = ma->ma_default_lmv;
943                 buf->lb_len = ma->ma_default_lmv_size;
944                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
945         } else {
946                 return -EINVAL;
947         }
948
949         LASSERT(buf->lb_buf);
950
951         rc = mo_xattr_get(info->mti_env, next, buf, name);
952         if (rc > 0) {
953
954 got:
955                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
956                         if (info->mti_big_lmm_used)
957                                 ma->ma_lmm = info->mti_big_lmm;
958
959                         /* NOT return LOV EA with hole to old client. */
960                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
961                                      LOV_PATTERN_F_HOLE) &&
962                             !(exp_connect_flags(info->mti_exp) &
963                               OBD_CONNECT_LFSCK)) {
964                                 return -EIO;
965                         } else {
966                                 ma->ma_lmm_size = rc;
967                                 ma->ma_valid |= MA_LOV;
968                         }
969                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
970                         if (info->mti_big_lmm_used)
971                                 ma->ma_lmv = info->mti_big_lmm;
972
973                         ma->ma_lmv_size = rc;
974                         ma->ma_valid |= MA_LMV;
975                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
976                         ma->ma_default_lmv_size = rc;
977                         ma->ma_valid |= MA_LMV_DEF;
978                 }
979
980                 /* Update mdt_max_mdsize so all clients will be aware that */
981                 if (info->mti_mdt->mdt_max_mdsize < rc)
982                         info->mti_mdt->mdt_max_mdsize = rc;
983
984                 rc = 0;
985         } else if (rc == -ENODATA) {
986                 /* no LOV EA */
987                 rc = 0;
988         } else if (rc == -ERANGE) {
989                 /* Default LMV has fixed size, so it must be able to fit
990                  * in the original buffer */
991                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
992                         return rc;
993                 rc = mdt_big_xattr_get(info, o, name);
994                 if (rc > 0) {
995                         info->mti_big_lmm_used = 1;
996                         goto got;
997                 }
998         }
999
1000         return rc;
1001 }
1002
1003 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
1004                       struct lu_fid *pfid)
1005 {
1006         struct lu_buf           *buf = &info->mti_buf;
1007         struct link_ea_header   *leh;
1008         struct link_ea_entry    *lee;
1009         int                      rc;
1010         ENTRY;
1011
1012         buf->lb_buf = info->mti_big_lmm;
1013         buf->lb_len = info->mti_big_lmmsize;
1014         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1015                           buf, XATTR_NAME_LINK);
1016         /* ignore errors, MA_PFID won't be set and it is
1017          * up to the caller to treat this as an error */
1018         if (rc == -ERANGE || buf->lb_len == 0) {
1019                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1020                 buf->lb_buf = info->mti_big_lmm;
1021                 buf->lb_len = info->mti_big_lmmsize;
1022         }
1023
1024         if (rc < 0)
1025                 RETURN(rc);
1026         if (rc < sizeof(*leh)) {
1027                 CERROR("short LinkEA on "DFID": rc = %d\n",
1028                        PFID(mdt_object_fid(o)), rc);
1029                 RETURN(-ENODATA);
1030         }
1031
1032         leh = (struct link_ea_header *) buf->lb_buf;
1033         lee = (struct link_ea_entry *)(leh + 1);
1034         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1035                 leh->leh_magic = LINK_EA_MAGIC;
1036                 leh->leh_reccount = __swab32(leh->leh_reccount);
1037                 leh->leh_len = __swab64(leh->leh_len);
1038         }
1039         if (leh->leh_magic != LINK_EA_MAGIC)
1040                 RETURN(-EINVAL);
1041         if (leh->leh_reccount == 0)
1042                 RETURN(-ENODATA);
1043
1044         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1045         fid_be_to_cpu(pfid, pfid);
1046
1047         RETURN(0);
1048 }
1049
1050 int mdt_attr_get_complex(struct mdt_thread_info *info,
1051                          struct mdt_object *o, struct md_attr *ma)
1052 {
1053         const struct lu_env *env = info->mti_env;
1054         struct md_object    *next = mdt_object_child(o);
1055         struct lu_buf       *buf = &info->mti_buf;
1056         int                  need = ma->ma_need;
1057         int                  rc = 0, rc2;
1058         u32                  mode;
1059         ENTRY;
1060
1061         ma->ma_valid = 0;
1062
1063         if (mdt_object_exists(o) == 0)
1064                 GOTO(out, rc = -ENOENT);
1065         mode = lu_object_attr(&next->mo_lu);
1066
1067         if (need & MA_INODE) {
1068                 ma->ma_need = MA_INODE;
1069                 rc = mo_attr_get(env, next, ma);
1070                 if (rc)
1071                         GOTO(out, rc);
1072
1073                 if (S_ISREG(mode))
1074                         (void) mdt_get_som(info, o, ma);
1075                 ma->ma_valid |= MA_INODE;
1076         }
1077
1078         if (need & MA_PFID) {
1079                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1080                 if (rc == 0)
1081                         ma->ma_valid |= MA_PFID;
1082                 /* ignore this error, parent fid is not mandatory */
1083                 rc = 0;
1084         }
1085
1086         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1087                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1088                 if (rc)
1089                         GOTO(out, rc);
1090         }
1091
1092         if (need & MA_LMV && S_ISDIR(mode)) {
1093                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1094                 if (rc != 0)
1095                         GOTO(out, rc);
1096         }
1097
1098         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1099                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1100                 if (rc != 0)
1101                         GOTO(out, rc);
1102         }
1103
1104         /*
1105          * In the handle of MA_INODE, we may already get the SOM attr.
1106          */
1107         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1108                 rc = mdt_get_som(info, o, ma);
1109                 if (rc != 0)
1110                         GOTO(out, rc);
1111         }
1112
1113         if (need & MA_HSM && S_ISREG(mode)) {
1114                 buf->lb_buf = info->mti_xattr_buf;
1115                 buf->lb_len = sizeof(info->mti_xattr_buf);
1116                 BUILD_BUG_ON(sizeof(struct hsm_attrs) >
1117                              sizeof(info->mti_xattr_buf));
1118                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1119                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1120                 if (rc2 == 0)
1121                         ma->ma_valid |= MA_HSM;
1122                 else if (rc2 < 0 && rc2 != -ENODATA)
1123                         GOTO(out, rc = rc2);
1124         }
1125
1126 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1127         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1128                 buf->lb_buf = ma->ma_acl;
1129                 buf->lb_len = ma->ma_acl_size;
1130                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1131                 if (rc2 > 0) {
1132                         ma->ma_acl_size = rc2;
1133                         ma->ma_valid |= MA_ACL_DEF;
1134                 } else if (rc2 == -ENODATA) {
1135                         /* no ACLs */
1136                         ma->ma_acl_size = 0;
1137                 } else
1138                         GOTO(out, rc = rc2);
1139         }
1140 #endif
1141 out:
1142         ma->ma_need = need;
1143         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1144                rc, ma->ma_valid, ma->ma_lmm);
1145         RETURN(rc);
1146 }
1147
1148 static int mdt_getattr_internal(struct mdt_thread_info *info,
1149                                 struct mdt_object *o, int ma_need)
1150 {
1151         struct md_object        *next = mdt_object_child(o);
1152         const struct mdt_body   *reqbody = info->mti_body;
1153         struct ptlrpc_request   *req = mdt_info_req(info);
1154         struct md_attr          *ma = &info->mti_attr;
1155         struct lu_attr          *la = &ma->ma_attr;
1156         struct req_capsule      *pill = info->mti_pill;
1157         const struct lu_env     *env = info->mti_env;
1158         struct mdt_body         *repbody;
1159         struct lu_buf           *buffer = &info->mti_buf;
1160         struct obd_export       *exp = info->mti_exp;
1161         int                      rc;
1162         ENTRY;
1163
1164         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1165                 RETURN(err_serious(-ENOMEM));
1166
1167         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1168
1169         ma->ma_valid = 0;
1170
1171         if (mdt_object_remote(o)) {
1172                 /* This object is located on remote node.*/
1173                 /* Return -ENOTSUPP for old client */
1174                 if (!mdt_is_dne_client(req->rq_export))
1175                         GOTO(out, rc = -ENOTSUPP);
1176
1177                 repbody->mbo_fid1 = *mdt_object_fid(o);
1178                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1179                 GOTO(out, rc = 0);
1180         }
1181
1182         if (reqbody->mbo_eadatasize > 0) {
1183                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1184                 if (buffer->lb_buf == NULL)
1185                         GOTO(out, rc = -EPROTO);
1186                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1187                                                       RCL_SERVER);
1188         } else {
1189                 buffer->lb_buf = NULL;
1190                 buffer->lb_len = 0;
1191                 ma_need &= ~(MA_LOV | MA_LMV);
1192                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1193                        mdt_obd_name(info->mti_mdt),
1194                        req->rq_export->exp_client_uuid.uuid);
1195         }
1196
1197         /* from 2.12.58 intent_getattr pack default LMV in reply */
1198         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1199             ((reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) ==
1200                     (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) &&
1201             req_capsule_has_field(&req->rq_pill, &RMF_DEFAULT_MDT_MD,
1202                                   RCL_SERVER)) {
1203                 ma->ma_lmv = buffer->lb_buf;
1204                 ma->ma_lmv_size = buffer->lb_len;
1205                 ma->ma_default_lmv = req_capsule_server_get(pill,
1206                                                 &RMF_DEFAULT_MDT_MD);
1207                 ma->ma_default_lmv_size = req_capsule_get_size(pill,
1208                                                 &RMF_DEFAULT_MDT_MD,
1209                                                 RCL_SERVER);
1210                 ma->ma_need = MA_INODE;
1211                 if (ma->ma_lmv_size > 0)
1212                         ma->ma_need |= MA_LMV;
1213                 if (ma->ma_default_lmv_size > 0)
1214                         ma->ma_need |= MA_LMV_DEF;
1215         } else if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1216                    (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1217                 /* If it is dir and client require MEA, then we got MEA */
1218                 /* Assumption: MDT_MD size is enough for lmv size. */
1219                 ma->ma_lmv = buffer->lb_buf;
1220                 ma->ma_lmv_size = buffer->lb_len;
1221                 ma->ma_need = MA_INODE;
1222                 if (ma->ma_lmv_size > 0) {
1223                         if (reqbody->mbo_valid & OBD_MD_MEA) {
1224                                 ma->ma_need |= MA_LMV;
1225                         } else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA) {
1226                                 ma->ma_need |= MA_LMV_DEF;
1227                                 ma->ma_default_lmv = buffer->lb_buf;
1228                                 ma->ma_lmv = NULL;
1229                                 ma->ma_default_lmv_size = buffer->lb_len;
1230                                 ma->ma_lmv_size = 0;
1231                         }
1232                 }
1233         } else {
1234                 ma->ma_lmm = buffer->lb_buf;
1235                 ma->ma_lmm_size = buffer->lb_len;
1236                 ma->ma_need = MA_INODE | MA_HSM;
1237                 if (ma->ma_lmm_size > 0) {
1238                         ma->ma_need |= MA_LOV;
1239                         /* Older clients may crash if they getattr overstriped
1240                          * files
1241                          */
1242                         if (!exp_connect_overstriping(exp) &&
1243                             mdt_lmm_is_overstriping(ma->ma_lmm))
1244                                 RETURN(-EOPNOTSUPP);
1245                 }
1246         }
1247
1248         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1249             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1250             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1251                 /* get default stripe info for this dir. */
1252                 ma->ma_need |= MA_LOV_DEF;
1253         }
1254         ma->ma_need |= ma_need;
1255
1256         rc = mdt_attr_get_complex(info, o, ma);
1257         if (unlikely(rc)) {
1258                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1259                        "%s: getattr error for "DFID": rc = %d\n",
1260                        mdt_obd_name(info->mti_mdt),
1261                        PFID(mdt_object_fid(o)), rc);
1262                 RETURN(rc);
1263         }
1264
1265         /* if file is released, check if a restore is running */
1266         if (ma->ma_valid & MA_HSM) {
1267                 repbody->mbo_valid |= OBD_MD_TSTATE;
1268                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1269                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1270                         repbody->mbo_t_state = MS_RESTORE;
1271         }
1272
1273         if (likely(ma->ma_valid & MA_INODE))
1274                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1275         else
1276                 RETURN(-EFAULT);
1277
1278         if (mdt_body_has_lov(la, reqbody)) {
1279                 if (ma->ma_valid & MA_LOV) {
1280                         LASSERT(ma->ma_lmm_size);
1281                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1282                         if (S_ISDIR(la->la_mode))
1283                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1284                         else
1285                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1286                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1287                 }
1288                 if (ma->ma_valid & MA_LMV) {
1289                         /* Return -ENOTSUPP for old client */
1290                         if (!mdt_is_striped_client(req->rq_export))
1291                                 RETURN(-ENOTSUPP);
1292
1293                         LASSERT(S_ISDIR(la->la_mode));
1294                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1295                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1296                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1297                 }
1298                 if (ma->ma_valid & MA_LMV_DEF) {
1299                         /* Return -ENOTSUPP for old client */
1300                         if (!mdt_is_striped_client(req->rq_export))
1301                                 RETURN(-ENOTSUPP);
1302                         LASSERT(S_ISDIR(la->la_mode));
1303                         /*
1304                          * when ll_dir_getstripe() gets default LMV, it
1305                          * checks mbo_eadatasize.
1306                          */
1307                         if (!(ma->ma_valid & MA_LMV))
1308                                 repbody->mbo_eadatasize =
1309                                         ma->ma_default_lmv_size;
1310                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1311                                                OBD_MD_DEFAULT_MEA);
1312                 }
1313         } else if (S_ISLNK(la->la_mode) &&
1314                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1315                 buffer->lb_buf = ma->ma_lmm;
1316                 /* eadatasize from client includes NULL-terminator, so
1317                  * there is no need to read it */
1318                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1319                 rc = mo_readlink(env, next, buffer);
1320                 if (unlikely(rc <= 0)) {
1321                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1322                                mdt_obd_name(info->mti_mdt),
1323                                PFID(mdt_object_fid(o)), rc);
1324                         rc = -EFAULT;
1325                 } else {
1326                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1327
1328                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1329                                 rc -= 2;
1330                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1331                         /* we need to report back size with NULL-terminator
1332                          * because client expects that */
1333                         repbody->mbo_eadatasize = rc + 1;
1334                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1335                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1336                                        "on "DFID ", expected %d\n",
1337                                        mdt_obd_name(info->mti_mdt),
1338                                        rc, PFID(mdt_object_fid(o)),
1339                                        reqbody->mbo_eadatasize - 1);
1340                         /* NULL terminate */
1341                         ((char *)ma->ma_lmm)[rc] = 0;
1342
1343                         /* If the total CDEBUG() size is larger than a page, it
1344                          * will print a warning to the console, avoid this by
1345                          * printing just the last part of the symlink. */
1346                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1347                                print_limit < rc ? "..." : "", print_limit,
1348                                (char *)ma->ma_lmm + rc - print_limit, rc);
1349                         rc = 0;
1350                 }
1351         }
1352
1353         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1354                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1355                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1356                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1357                        repbody->mbo_max_mdsize);
1358         }
1359
1360 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1361         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1362                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1363                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1364                 if (IS_ERR(nodemap))
1365                         RETURN(PTR_ERR(nodemap));
1366
1367                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1368                 nodemap_putref(nodemap);
1369         }
1370 #endif
1371
1372 out:
1373         if (rc == 0)
1374                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1375
1376         RETURN(rc);
1377 }
1378
1379 static int mdt_getattr(struct tgt_session_info *tsi)
1380 {
1381         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1382         struct mdt_object       *obj = info->mti_object;
1383         struct req_capsule      *pill = info->mti_pill;
1384         struct mdt_body         *reqbody;
1385         struct mdt_body         *repbody;
1386         int rc, rc2;
1387         ENTRY;
1388
1389         if (unlikely(info->mti_object == NULL))
1390                 RETURN(-EPROTO);
1391
1392         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1393         LASSERT(reqbody);
1394         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1395
1396         /* Special case for Data-on-MDT files to get data version */
1397         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1398                 rc = mdt_data_version_get(tsi);
1399                 GOTO(out, rc);
1400         }
1401
1402         /* Unlike intent case where we need to pre-fill out buffers early on
1403          * in intent policy for ldlm reasons, here we can have a much better
1404          * guess at EA size by just reading it from disk.
1405          * Exceptions are readdir and (missing) directory striping */
1406         /* Readlink */
1407         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1408                 /* No easy way to know how long is the symlink, but it cannot
1409                  * be more than PATH_MAX, so we allocate +1 */
1410                 rc = PATH_MAX + 1;
1411         /* A special case for fs ROOT: getattr there might fetch
1412          * default EA for entire fs, not just for this dir!
1413          */
1414         } else if (lu_fid_eq(mdt_object_fid(obj),
1415                              &info->mti_mdt->mdt_md_root_fid) &&
1416                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1417                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1418                                                                  MDS_GETATTR)) {
1419                 /* Should the default strping be bigger, mdt_fix_reply
1420                  * will reallocate */
1421                 rc = DEF_REP_MD_SIZE;
1422         } else {
1423                 /* Read the actual EA size from disk */
1424                 rc = mdt_attr_get_eabuf_size(info, obj);
1425         }
1426
1427         if (rc < 0)
1428                 GOTO(out, rc = err_serious(rc));
1429
1430         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1431
1432         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1433          * by default. If the target object has more ACL entries, then
1434          * enlarge the buffer when necessary. */
1435         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1436                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1437
1438         rc = req_capsule_server_pack(pill);
1439         if (unlikely(rc != 0))
1440                 GOTO(out, rc = err_serious(rc));
1441
1442         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1443         LASSERT(repbody != NULL);
1444         repbody->mbo_eadatasize = 0;
1445         repbody->mbo_aclsize = 0;
1446
1447         rc = mdt_check_ucred(info);
1448         if (unlikely(rc))
1449                 GOTO(out_shrink, rc);
1450
1451         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1452
1453         rc = mdt_getattr_internal(info, obj, 0);
1454         EXIT;
1455 out_shrink:
1456         mdt_client_compatibility(info);
1457         rc2 = mdt_fix_reply(info);
1458         if (rc == 0)
1459                 rc = rc2;
1460 out:
1461         mdt_thread_info_fini(info);
1462         return rc;
1463 }
1464
1465 /**
1466  * Handler of layout intent RPC requiring the layout modification
1467  *
1468  * \param[in]  info     thread environment
1469  * \param[in]  obj      object
1470  * \param[out] lhc      object ldlm lock handle
1471  * \param[in]  layout   layout change descriptor
1472  *
1473  * \retval 0    on success
1474  * \retval < 0  error code
1475  */
1476 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1477                       struct mdt_lock_handle *lhc,
1478                       struct md_layout_change *layout)
1479 {
1480         int rc;
1481
1482         ENTRY;
1483
1484         if (!mdt_object_exists(obj))
1485                 RETURN(-ENOENT);
1486
1487         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1488                 RETURN(-EINVAL);
1489
1490         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1491                            MAY_WRITE);
1492         if (rc)
1493                 RETURN(rc);
1494
1495         rc = mdt_check_resent_lock(info, obj, lhc);
1496         if (rc < 0)
1497                 RETURN(rc);
1498
1499         if (rc > 0) {
1500                 /* not resent */
1501                 mdt_lock_handle_init(lhc);
1502                 mdt_lock_reg_init(lhc, LCK_EX);
1503                 rc = mdt_reint_object_lock(info, obj, lhc, MDS_INODELOCK_LAYOUT,
1504                                            false);
1505                 if (rc)
1506                         RETURN(rc);
1507         }
1508
1509         mutex_lock(&obj->mot_som_mutex);
1510         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1511         mutex_unlock(&obj->mot_som_mutex);
1512
1513         if (rc)
1514                 mdt_object_unlock(info, obj, lhc, 1);
1515
1516         RETURN(rc);
1517 }
1518
1519 /**
1520  * Exchange MOF_LOV_CREATED flags between two objects after a
1521  * layout swap. No assumption is made on whether o1 or o2 have
1522  * created objects or not.
1523  *
1524  * \param[in,out] o1    First swap layout object
1525  * \param[in,out] o2    Second swap layout object
1526  */
1527 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1528 {
1529         unsigned int o1_lov_created = o1->mot_lov_created;
1530
1531         mutex_lock(&o1->mot_lov_mutex);
1532         mutex_lock(&o2->mot_lov_mutex);
1533
1534         o1->mot_lov_created = o2->mot_lov_created;
1535         o2->mot_lov_created = o1_lov_created;
1536
1537         mutex_unlock(&o2->mot_lov_mutex);
1538         mutex_unlock(&o1->mot_lov_mutex);
1539 }
1540
1541 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1542 {
1543         struct mdt_thread_info  *info;
1544         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1545         struct obd_export       *exp = req->rq_export;
1546         struct mdt_object       *o1, *o2, *o;
1547         struct mdt_lock_handle  *lh1, *lh2;
1548         struct mdc_swap_layouts *msl;
1549         int                      rc;
1550         ENTRY;
1551
1552         /* client does not support layout lock, so layout swaping
1553          * is disabled.
1554          * FIXME: there is a problem for old clients which don't support
1555          * layout lock yet. If those clients have already opened the file
1556          * they won't be notified at all so that old layout may still be
1557          * used to do IO. This can be fixed after file release is landed by
1558          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1559         if (!exp_connect_layout(exp))
1560                 RETURN(-EOPNOTSUPP);
1561
1562         info = tsi2mdt_info(tsi);
1563         if (unlikely(info->mti_object == NULL))
1564                 RETURN(-EPROTO);
1565
1566         if (info->mti_dlm_req != NULL)
1567                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1568
1569         o1 = info->mti_object;
1570         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1571                                 &info->mti_body->mbo_fid2);
1572         if (IS_ERR(o))
1573                 GOTO(out, rc = PTR_ERR(o));
1574
1575         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1576                 GOTO(put, rc = -ENOENT);
1577
1578         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1579         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1580                 GOTO(put, rc);
1581
1582         if (rc < 0)
1583                 swap(o1, o2);
1584
1585         /* permission check. Make sure the calling process having permission
1586          * to write both files. */
1587         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1588                            MAY_WRITE);
1589         if (rc < 0)
1590                 GOTO(put, rc);
1591
1592         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1593                            MAY_WRITE);
1594         if (rc < 0)
1595                 GOTO(put, rc);
1596
1597         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1598         if (msl == NULL)
1599                 GOTO(put, rc = -EPROTO);
1600
1601         lh1 = &info->mti_lh[MDT_LH_NEW];
1602         mdt_lock_reg_init(lh1, LCK_EX);
1603         lh2 = &info->mti_lh[MDT_LH_OLD];
1604         mdt_lock_reg_init(lh2, LCK_EX);
1605
1606         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1607                              MDS_INODELOCK_XATTR);
1608         if (rc < 0)
1609                 GOTO(put, rc);
1610
1611         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1612                              MDS_INODELOCK_XATTR);
1613         if (rc < 0)
1614                 GOTO(unlock1, rc);
1615
1616         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1617                              mdt_object_child(o2), msl->msl_flags);
1618         if (rc < 0)
1619                 GOTO(unlock2, rc);
1620
1621         mdt_swap_lov_flag(o1, o2);
1622
1623 unlock2:
1624         mdt_object_unlock(info, o2, lh2, rc);
1625 unlock1:
1626         mdt_object_unlock(info, o1, lh1, rc);
1627 put:
1628         mdt_object_put(info->mti_env, o);
1629 out:
1630         mdt_thread_info_fini(info);
1631         RETURN(rc);
1632 }
1633
1634 static int mdt_raw_lookup(struct mdt_thread_info *info,
1635                           struct mdt_object *parent,
1636                           const struct lu_name *lname,
1637                           struct ldlm_reply *ldlm_rep)
1638 {
1639         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1640         int              rc;
1641         ENTRY;
1642
1643         LASSERT(!info->mti_cross_ref);
1644
1645         /* Only got the fid of this obj by name */
1646         fid_zero(child_fid);
1647         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1648                         lname, child_fid, &info->mti_spec);
1649         if (rc == 0) {
1650                 struct mdt_body *repbody;
1651
1652                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1653                 repbody->mbo_fid1 = *child_fid;
1654                 repbody->mbo_valid = OBD_MD_FLID;
1655                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1656         } else if (rc == -ENOENT) {
1657                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1658         }
1659
1660         RETURN(rc);
1661 }
1662
1663 /*
1664  * UPDATE lock should be taken against parent, and be released before exit;
1665  * child_bits lock should be taken against child, and be returned back:
1666  *            (1)normal request should release the child lock;
1667  *            (2)intent request will grant the lock to client.
1668  */
1669 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1670                                  struct mdt_lock_handle *lhc,
1671                                  __u64 child_bits,
1672                                  struct ldlm_reply *ldlm_rep)
1673 {
1674         struct ptlrpc_request  *req = mdt_info_req(info);
1675         struct mdt_body        *reqbody = NULL;
1676         struct mdt_object      *parent = info->mti_object;
1677         struct mdt_object      *child;
1678         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1679         struct lu_name         *lname = NULL;
1680         struct mdt_lock_handle *lhp = NULL;
1681         struct ldlm_lock       *lock;
1682         struct req_capsule *pill = info->mti_pill;
1683         __u64 try_bits = 0;
1684         bool is_resent;
1685         int ma_need = 0;
1686         int rc;
1687
1688         ENTRY;
1689
1690         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1691         LASSERT(ergo(is_resent,
1692                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1693
1694         if (parent == NULL)
1695                 RETURN(-ENOENT);
1696
1697         if (info->mti_cross_ref) {
1698                 /* Only getattr on the child. Parent is on another node. */
1699                 mdt_set_disposition(info, ldlm_rep,
1700                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1701                 child = parent;
1702                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1703                        "ldlm_rep = %p\n",
1704                        PFID(mdt_object_fid(child)), ldlm_rep);
1705
1706                 rc = mdt_check_resent_lock(info, child, lhc);
1707                 if (rc < 0) {
1708                         RETURN(rc);
1709                 } else if (rc > 0) {
1710                         mdt_lock_handle_init(lhc);
1711                         mdt_lock_reg_init(lhc, LCK_PR);
1712
1713                         /*
1714                          * Object's name entry is on another MDS, it will
1715                          * request PERM lock only because LOOKUP lock is owned
1716                          * by the MDS where name entry resides.
1717                          *
1718                          * TODO: it should try layout lock too. - Jinshan
1719                          */
1720                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1721                                         MDS_INODELOCK_LAYOUT);
1722                         child_bits |= MDS_INODELOCK_PERM;
1723
1724                         rc = mdt_object_lock(info, child, lhc, child_bits);
1725                         if (rc < 0)
1726                                 RETURN(rc);
1727                 }
1728
1729                 /* Finally, we can get attr for child. */
1730                 if (!mdt_object_exists(child)) {
1731                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1732                                         &child->mot_obj,
1733                                         "remote object doesn't exist.");
1734                         mdt_object_unlock(info, child, lhc, 1);
1735                         RETURN(-ENOENT);
1736                 }
1737
1738                 rc = mdt_getattr_internal(info, child, 0);
1739                 if (unlikely(rc != 0)) {
1740                         mdt_object_unlock(info, child, lhc, 1);
1741                         RETURN(rc);
1742                 }
1743
1744                 rc = mdt_pack_secctx_in_reply(info, child);
1745                 if (unlikely(rc))
1746                         mdt_object_unlock(info, child, lhc, 1);
1747                 RETURN(rc);
1748         }
1749
1750         lname = &info->mti_name;
1751         mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
1752
1753         if (lu_name_is_valid(lname)) {
1754                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1755                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1756                        PNAME(lname), ldlm_rep);
1757         } else {
1758                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1759                 if (unlikely(reqbody == NULL))
1760                         RETURN(err_serious(-EPROTO));
1761
1762                 *child_fid = reqbody->mbo_fid2;
1763
1764                 if (unlikely(!fid_is_sane(child_fid)))
1765                         RETURN(err_serious(-EINVAL));
1766
1767                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1768                        "ldlm_rep = %p\n",
1769                        PFID(mdt_object_fid(parent)),
1770                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1771         }
1772
1773         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1774
1775         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1776                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1777                                 &parent->mot_obj,
1778                                 "Parent doesn't exist!");
1779                 RETURN(-ESTALE);
1780         }
1781
1782         if (mdt_object_remote(parent)) {
1783                 CERROR("%s: parent "DFID" is on remote target\n",
1784                        mdt_obd_name(info->mti_mdt),
1785                        PFID(mdt_object_fid(parent)));
1786                 RETURN(-EIO);
1787         }
1788
1789         if (lu_name_is_valid(lname)) {
1790                 /* Always allow to lookup ".." */
1791                 if (unlikely(lname->ln_namelen == 2 &&
1792                              lname->ln_name[0] == '.' &&
1793                              lname->ln_name[1] == '.'))
1794                         info->mti_spec.sp_permitted = 1;
1795
1796                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1797                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1798
1799                         RETURN(rc);
1800                 }
1801
1802                 /* step 1: lock parent only if parent is a directory */
1803                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1804                         lhp = &info->mti_lh[MDT_LH_PARENT];
1805                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1806                         rc = mdt_object_lock(info, parent, lhp,
1807                                              MDS_INODELOCK_UPDATE);
1808                         if (unlikely(rc != 0))
1809                                 RETURN(rc);
1810                 }
1811
1812                 /* step 2: lookup child's fid by name */
1813                 fid_zero(child_fid);
1814                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1815                                 child_fid, &info->mti_spec);
1816                 if (rc == -ENOENT)
1817                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1818
1819                 if (rc != 0)
1820                         GOTO(out_parent, rc);
1821         }
1822
1823         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1824
1825         /*
1826          *step 3: find the child object by fid & lock it.
1827          *        regardless if it is local or remote.
1828          *
1829          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1830          *      set parent dir fid the same as child fid in getattr by fid case
1831          *      we should not lu_object_find() the object again, could lead
1832          *      to hung if there is a concurrent unlink destroyed the object.
1833          */
1834         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1835                 mdt_object_get(info->mti_env, parent);
1836                 child = parent;
1837         } else {
1838                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1839                                         child_fid);
1840         }
1841
1842         if (unlikely(IS_ERR(child)))
1843                 GOTO(out_parent, rc = PTR_ERR(child));
1844
1845         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1846         if (!mdt_object_exists(child)) {
1847                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1848                                 &child->mot_obj,
1849                                 "Object doesn't exist!");
1850                 GOTO(out_child, rc = -ENOENT);
1851         }
1852
1853         rc = mdt_check_resent_lock(info, child, lhc);
1854         if (rc < 0) {
1855                 GOTO(out_child, rc);
1856         } else if (rc > 0) {
1857                 mdt_lock_handle_init(lhc);
1858                 mdt_lock_reg_init(lhc, LCK_PR);
1859
1860                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1861                     !mdt_object_remote(child)) {
1862                         struct md_attr *ma = &info->mti_attr;
1863
1864                         ma->ma_valid = 0;
1865                         ma->ma_need = MA_INODE;
1866                         rc = mdt_attr_get_complex(info, child, ma);
1867                         if (unlikely(rc != 0))
1868                                 GOTO(out_child, rc);
1869
1870                         /* If the file has not been changed for some time, we
1871                          * return not only a LOOKUP lock, but also an UPDATE
1872                          * lock and this might save us RPC on later STAT. For
1873                          * directories, it also let negative dentry cache start
1874                          * working for this dir. */
1875                         if (ma->ma_valid & MA_INODE &&
1876                             ma->ma_attr.la_valid & LA_CTIME &&
1877                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1878                             ma->ma_attr.la_ctime < ktime_get_real_seconds())
1879                                 child_bits |= MDS_INODELOCK_UPDATE;
1880                 }
1881
1882                 /* layout lock must be granted in a best-effort way
1883                  * for IT operations */
1884                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1885                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1886                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1887                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1888                             exp_connect_layout(info->mti_exp)) {
1889                                 /* try to grant layout lock for regular file. */
1890                                 try_bits = MDS_INODELOCK_LAYOUT;
1891                         }
1892                         /* Acquire DOM lock in advance for data-on-mdt file */
1893                         if (child != parent)
1894                                 try_bits |= MDS_INODELOCK_DOM;
1895                 }
1896
1897                 if (try_bits != 0) {
1898                         /* try layout lock, it may fail to be granted due to
1899                          * contention at LOOKUP or UPDATE */
1900                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1901                                                  try_bits, false);
1902                         if (child_bits & MDS_INODELOCK_LAYOUT)
1903                                 ma_need |= MA_LOV;
1904                 } else {
1905                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1906                          * client will enqueue the lock to the remote MDT */
1907                         if (mdt_object_remote(child))
1908                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1909                         rc = mdt_object_lock(info, child, lhc, child_bits);
1910                 }
1911                 if (unlikely(rc != 0))
1912                         GOTO(out_child, rc);
1913         }
1914
1915         /* finally, we can get attr for child. */
1916         rc = mdt_getattr_internal(info, child, ma_need);
1917         if (unlikely(rc != 0)) {
1918                 mdt_object_unlock(info, child, lhc, 1);
1919                 GOTO(out_child, rc);
1920         }
1921
1922         rc = mdt_pack_secctx_in_reply(info, child);
1923         if (unlikely(rc)) {
1924                 mdt_object_unlock(info, child, lhc, 1);
1925                 GOTO(out_child, rc);
1926         }
1927
1928         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1929         if (lock) {
1930                 /* Debugging code. */
1931                 LDLM_DEBUG(lock, "Returning lock to client");
1932                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1933                                          &lock->l_resource->lr_name),
1934                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1935                          PLDLMRES(lock->l_resource),
1936                          PFID(mdt_object_fid(child)));
1937
1938                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1939                     !mdt_object_remote(child) && child != parent) {
1940                         mdt_object_put(info->mti_env, child);
1941                         rc = mdt_pack_size2body(info, child_fid,
1942                                                 &lhc->mlh_reg_lh);
1943                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1944                                 /* DOM lock was taken in advance but this is
1945                                  * not DoM file. Drop the lock.
1946                                  */
1947                                 lock_res_and_lock(lock);
1948                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1949                                 unlock_res_and_lock(lock);
1950                         }
1951                         LDLM_LOCK_PUT(lock);
1952                         GOTO(out_parent, rc = 0);
1953                 }
1954                 LDLM_LOCK_PUT(lock);
1955         }
1956
1957         EXIT;
1958 out_child:
1959         mdt_object_put(info->mti_env, child);
1960 out_parent:
1961         if (lhp)
1962                 mdt_object_unlock(info, parent, lhp, 1);
1963         return rc;
1964 }
1965
1966 /* normal handler: should release the child lock */
1967 static int mdt_getattr_name(struct tgt_session_info *tsi)
1968 {
1969         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1970         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1971         struct mdt_body        *reqbody;
1972         struct mdt_body        *repbody;
1973         int rc, rc2;
1974         ENTRY;
1975
1976         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1977         LASSERT(reqbody != NULL);
1978         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1979         LASSERT(repbody != NULL);
1980
1981         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1982         repbody->mbo_eadatasize = 0;
1983         repbody->mbo_aclsize = 0;
1984
1985         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1986         if (unlikely(rc))
1987                 GOTO(out_shrink, rc);
1988
1989         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1990         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1991                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1992                 lhc->mlh_reg_lh.cookie = 0;
1993         }
1994         mdt_exit_ucred(info);
1995         EXIT;
1996 out_shrink:
1997         mdt_client_compatibility(info);
1998         rc2 = mdt_fix_reply(info);
1999         if (rc == 0)
2000                 rc = rc2;
2001         mdt_thread_info_fini(info);
2002         return rc;
2003 }
2004
2005 static int mdt_rmfid_unlink(struct mdt_thread_info *info,
2006                             const struct lu_fid *pfid,
2007                             const struct lu_name *name,
2008                             struct mdt_object *obj, s64 ctime)
2009 {
2010         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2011         struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
2012         struct mdt_device *mdt = info->mti_mdt;
2013         struct md_attr *ma = &info->mti_attr;
2014         struct mdt_lock_handle *parent_lh;
2015         struct mdt_lock_handle *child_lh;
2016         struct mdt_object *pobj;
2017         bool cos_incompat = false;
2018         int rc;
2019         ENTRY;
2020
2021         pobj = mdt_object_find(info->mti_env, mdt, pfid);
2022         if (IS_ERR(pobj))
2023                 GOTO(out, rc = PTR_ERR(pobj));
2024
2025         parent_lh = &info->mti_lh[MDT_LH_PARENT];
2026         mdt_lock_pdo_init(parent_lh, LCK_PW, name);
2027         rc = mdt_object_lock(info, pobj, parent_lh, MDS_INODELOCK_UPDATE);
2028         if (rc != 0)
2029                 GOTO(put_parent, rc);
2030
2031         if (mdt_object_remote(pobj))
2032                 cos_incompat = true;
2033
2034         rc = mdo_lookup(info->mti_env, mdt_object_child(pobj),
2035                         name, child_fid, &info->mti_spec);
2036         if (rc != 0)
2037                 GOTO(unlock_parent, rc);
2038
2039         if (!lu_fid_eq(child_fid, mdt_object_fid(obj)))
2040                 GOTO(unlock_parent, rc = -EREMCHG);
2041
2042         child_lh = &info->mti_lh[MDT_LH_CHILD];
2043         mdt_lock_reg_init(child_lh, LCK_EX);
2044         rc = mdt_reint_striped_lock(info, obj, child_lh,
2045                                     MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
2046                                     einfo, cos_incompat);
2047         if (rc != 0)
2048                 GOTO(unlock_parent, rc);
2049
2050         if (atomic_read(&obj->mot_open_count)) {
2051                 CDEBUG(D_OTHER, "object "DFID" open, skip\n",
2052                        PFID(mdt_object_fid(obj)));
2053                 GOTO(unlock_child, rc = -EBUSY);
2054         }
2055
2056         ma->ma_need = 0;
2057         ma->ma_valid = MA_INODE;
2058         ma->ma_attr.la_valid = LA_CTIME;
2059         ma->ma_attr.la_ctime = ctime;
2060
2061         mutex_lock(&obj->mot_lov_mutex);
2062
2063         rc = mdo_unlink(info->mti_env, mdt_object_child(pobj),
2064                         mdt_object_child(obj), name, ma, 0);
2065
2066         mutex_unlock(&obj->mot_lov_mutex);
2067
2068 unlock_child:
2069         mdt_reint_striped_unlock(info, obj, child_lh, einfo, 1);
2070 unlock_parent:
2071         mdt_object_unlock(info, pobj, parent_lh, 1);
2072 put_parent:
2073         mdt_object_put(info->mti_env, pobj);
2074 out:
2075         RETURN(rc);
2076 }
2077
2078 static int mdt_rmfid_check_permission(struct mdt_thread_info *info,
2079                                         struct mdt_object *obj)
2080 {
2081         struct lu_ucred *uc = lu_ucred(info->mti_env);
2082         struct md_attr *ma = &info->mti_attr;
2083         struct lu_attr *la = &ma->ma_attr;
2084         int rc = 0;
2085         ENTRY;
2086
2087         ma->ma_need = MA_INODE;
2088         rc = mo_attr_get(info->mti_env, mdt_object_child(obj), ma);
2089         if (rc)
2090                 GOTO(out, rc);
2091
2092         if (la->la_flags & LUSTRE_IMMUTABLE_FL)
2093                         rc = -EACCES;
2094
2095         if (md_capable(uc, CFS_CAP_DAC_OVERRIDE))
2096                 RETURN(0);
2097         if (uc->uc_fsuid == la->la_uid) {
2098                 if ((la->la_mode & S_IWUSR) == 0)
2099                         rc = -EACCES;
2100         } else if (uc->uc_fsgid == la->la_gid) {
2101                 if ((la->la_mode & S_IWGRP) == 0)
2102                         rc = -EACCES;
2103         } else if ((la->la_mode & S_IWOTH) == 0) {
2104                         rc = -EACCES;
2105         }
2106
2107 out:
2108         RETURN(rc);
2109 }
2110
2111 static int mdt_rmfid_one(struct mdt_thread_info *info, struct lu_fid *fid,
2112                          s64 ctime)
2113 {
2114         struct mdt_device *mdt = info->mti_mdt;
2115         struct mdt_object *obj = NULL;
2116         struct linkea_data ldata = { NULL };
2117         struct lu_buf *buf = &info->mti_big_buf;
2118         struct lu_name *name = &info->mti_name;
2119         struct lu_fid *pfid = &info->mti_tmp_fid1;
2120         struct link_ea_header *leh;
2121         struct link_ea_entry *lee;
2122         int reclen, count, rc = 0;
2123         ENTRY;
2124
2125         if (!fid_is_sane(fid))
2126                 GOTO(out, rc = -EINVAL);
2127
2128         if (!fid_is_namespace_visible(fid))
2129                 GOTO(out, rc = -EINVAL);
2130
2131         obj = mdt_object_find(info->mti_env, mdt, fid);
2132         if (IS_ERR(obj))
2133                 GOTO(out, rc = PTR_ERR(obj));
2134
2135         if (mdt_object_remote(obj))
2136                 GOTO(out, rc = -EREMOTE);
2137         if (!mdt_object_exists(obj) || lu_object_is_dying(&obj->mot_header))
2138                 GOTO(out, rc = -ENOENT);
2139
2140         rc = mdt_rmfid_check_permission(info, obj);
2141         if (rc)
2142                 GOTO(out, rc);
2143
2144         /* take LinkEA */
2145         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2146         if (!buf->lb_buf)
2147                 GOTO(out, rc = -ENOMEM);
2148
2149         ldata.ld_buf = buf;
2150         rc = mdt_links_read(info, obj, &ldata);
2151         if (rc)
2152                 GOTO(out, rc);
2153
2154         leh = buf->lb_buf;
2155         lee = (struct link_ea_entry *)(leh + 1);
2156         for (count = 0; count < leh->leh_reccount; count++) {
2157                 /* remove every hardlink */
2158                 linkea_entry_unpack(lee, &reclen, name, pfid);
2159                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2160                 rc = mdt_rmfid_unlink(info, pfid, name, obj, ctime);
2161                 if (rc)
2162                         break;
2163         }
2164
2165 out:
2166         if (obj && !IS_ERR(obj))
2167                 mdt_object_put(info->mti_env, obj);
2168         if (info->mti_big_buf.lb_buf)
2169                 lu_buf_free(&info->mti_big_buf);
2170
2171         RETURN(rc);
2172 }
2173
2174 static int mdt_rmfid(struct tgt_session_info *tsi)
2175 {
2176         struct mdt_thread_info *mti = tsi2mdt_info(tsi);
2177         struct mdt_body *reqbody;
2178         struct lu_fid *fids, *rfids;
2179         int bufsize, rc;
2180         __u32 *rcs;
2181         int i, nr;
2182         ENTRY;
2183
2184         reqbody = req_capsule_client_get(tsi->tsi_pill, &RMF_MDT_BODY);
2185         if (reqbody == NULL)
2186                 RETURN(-EPROTO);
2187         bufsize = req_capsule_get_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2188                                        RCL_CLIENT);
2189         nr = bufsize / sizeof(struct lu_fid);
2190         if (nr * sizeof(struct lu_fid) != bufsize)
2191                 RETURN(-EINVAL);
2192         req_capsule_set_size(tsi->tsi_pill, &RMF_RCS,
2193                              RCL_SERVER, nr * sizeof(__u32));
2194         req_capsule_set_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2195                              RCL_SERVER, nr * sizeof(struct lu_fid));
2196         rc = req_capsule_server_pack(tsi->tsi_pill);
2197         if (rc)
2198                 GOTO(out, rc = err_serious(rc));
2199         fids = req_capsule_client_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2200         if (fids == NULL)
2201                 RETURN(-EPROTO);
2202         rcs = req_capsule_server_get(tsi->tsi_pill, &RMF_RCS);
2203         LASSERT(rcs);
2204         rfids = req_capsule_server_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2205         LASSERT(rfids);
2206
2207         mdt_init_ucred(mti, reqbody);
2208         for (i = 0; i < nr; i++) {
2209                 rfids[i] = fids[i];
2210                 rcs[i] = mdt_rmfid_one(mti, fids + i, reqbody->mbo_ctime);
2211         }
2212         mdt_exit_ucred(mti);
2213
2214 out:
2215         RETURN(rc);
2216 }
2217
2218 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2219                          void *karg, void __user *uarg);
2220
2221 static int mdt_set_info(struct tgt_session_info *tsi)
2222 {
2223         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2224         char                    *key;
2225         void                    *val;
2226         int                      keylen, vallen, rc = 0;
2227
2228         ENTRY;
2229
2230         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2231         if (key == NULL) {
2232                 DEBUG_REQ(D_HA, req, "no set_info key");
2233                 RETURN(err_serious(-EFAULT));
2234         }
2235
2236         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2237                                       RCL_CLIENT);
2238
2239         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2240         if (val == NULL) {
2241                 DEBUG_REQ(D_HA, req, "no set_info val");
2242                 RETURN(err_serious(-EFAULT));
2243         }
2244
2245         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2246                                       RCL_CLIENT);
2247
2248         /* Swab any part of val you need to here */
2249         if (KEY_IS(KEY_READ_ONLY)) {
2250                 spin_lock(&req->rq_export->exp_lock);
2251                 if (*(__u32 *)val)
2252                         *exp_connect_flags_ptr(req->rq_export) |=
2253                                 OBD_CONNECT_RDONLY;
2254                 else
2255                         *exp_connect_flags_ptr(req->rq_export) &=
2256                                 ~OBD_CONNECT_RDONLY;
2257                 spin_unlock(&req->rq_export->exp_lock);
2258         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2259                 struct changelog_setinfo *cs = val;
2260
2261                 if (vallen != sizeof(*cs)) {
2262                         CERROR("%s: bad changelog_clear setinfo size %d\n",
2263                                tgt_name(tsi->tsi_tgt), vallen);
2264                         RETURN(-EINVAL);
2265                 }
2266                 if (ptlrpc_req_need_swab(req)) {
2267                         __swab64s(&cs->cs_recno);
2268                         __swab32s(&cs->cs_id);
2269                 }
2270
2271                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2272                                    vallen, val, NULL);
2273         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2274                 if (vallen > 0)
2275                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2276         } else {
2277                 RETURN(-EINVAL);
2278         }
2279         RETURN(rc);
2280 }
2281
2282 static int mdt_readpage(struct tgt_session_info *tsi)
2283 {
2284         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2285         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2286         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2287         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2288         struct mdt_body         *repbody;
2289         int                      rc;
2290         int                      i;
2291
2292         ENTRY;
2293
2294         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2295                 RETURN(err_serious(-ENOMEM));
2296
2297         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2298         if (repbody == NULL || reqbody == NULL)
2299                 RETURN(err_serious(-EFAULT));
2300
2301         /*
2302          * prepare @rdpg before calling lower layers and transfer itself. Here
2303          * reqbody->size contains offset of where to start to read and
2304          * reqbody->nlink contains number bytes to read.
2305          */
2306         rdpg->rp_hash = reqbody->mbo_size;
2307         if (rdpg->rp_hash != reqbody->mbo_size) {
2308                 CERROR("Invalid hash: %#llx != %#llx\n",
2309                        rdpg->rp_hash, reqbody->mbo_size);
2310                 RETURN(-EFAULT);
2311         }
2312
2313         rdpg->rp_attrs = reqbody->mbo_mode;
2314         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2315                 rdpg->rp_attrs |= LUDA_64BITHASH;
2316         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2317                                 exp_max_brw_size(tsi->tsi_exp));
2318         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2319                           PAGE_SHIFT;
2320         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2321         if (rdpg->rp_pages == NULL)
2322                 RETURN(-ENOMEM);
2323
2324         for (i = 0; i < rdpg->rp_npages; ++i) {
2325                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2326                 if (rdpg->rp_pages[i] == NULL)
2327                         GOTO(free_rdpg, rc = -ENOMEM);
2328         }
2329
2330         /* call lower layers to fill allocated pages with directory data */
2331         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2332         if (rc < 0)
2333                 GOTO(free_rdpg, rc);
2334
2335         /* send pages to client */
2336         rc = tgt_sendpage(tsi, rdpg, rc);
2337
2338         EXIT;
2339 free_rdpg:
2340
2341         for (i = 0; i < rdpg->rp_npages; i++)
2342                 if (rdpg->rp_pages[i] != NULL)
2343                         __free_page(rdpg->rp_pages[i]);
2344         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2345
2346         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2347                 RETURN(0);
2348
2349         return rc;
2350 }
2351
2352 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2353 {
2354         struct lu_ucred *uc = mdt_ucred_check(info);
2355         struct lu_attr *attr = &info->mti_attr.ma_attr;
2356
2357         if (uc == NULL)
2358                 return -EINVAL;
2359
2360         if (op != REINT_SETATTR) {
2361                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2362                         attr->la_uid = uc->uc_fsuid;
2363                 /* for S_ISGID, inherit gid from his parent, such work will be
2364                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2365                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2366                         attr->la_gid = uc->uc_fsgid;
2367         }
2368
2369         return 0;
2370 }
2371
2372 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2373 {
2374         return op == REINT_OPEN &&
2375              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2376 }
2377
2378 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2379 {
2380         struct req_capsule *pill = info->mti_pill;
2381
2382         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2383                                   RCL_SERVER) &&
2384             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2385                                   RCL_CLIENT)) {
2386                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2387                                          RCL_CLIENT) != 0) {
2388                         /* pre-set size in server part with max size */
2389                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2390                                              RCL_SERVER,
2391                                              info->mti_mdt->mdt_max_ea_size);
2392                 } else {
2393                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2394                                              RCL_SERVER, 0);
2395                 }
2396         }
2397
2398 }
2399
2400 static int mdt_reint_internal(struct mdt_thread_info *info,
2401                               struct mdt_lock_handle *lhc,
2402                               __u32 op)
2403 {
2404         struct req_capsule      *pill = info->mti_pill;
2405         struct mdt_body         *repbody;
2406         int                      rc = 0, rc2;
2407
2408         ENTRY;
2409
2410         rc = mdt_reint_unpack(info, op);
2411         if (rc != 0) {
2412                 CERROR("Can't unpack reint, rc %d\n", rc);
2413                 RETURN(err_serious(rc));
2414         }
2415
2416
2417         /* check if the file system is set to readonly. O_RDONLY open
2418          * is still allowed even the file system is set to readonly mode */
2419         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2420                 RETURN(err_serious(-EROFS));
2421
2422         /* for replay (no_create) lmm is not needed, client has it already */
2423         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2424                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2425                                      DEF_REP_MD_SIZE);
2426
2427         /* llog cookies are always 0, the field is kept for compatibility */
2428         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2429                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2430
2431         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2432          * by default. If the target object has more ACL entries, then
2433          * enlarge the buffer when necessary. */
2434         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2435                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2436                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2437
2438         mdt_preset_secctx_size(info);
2439
2440         rc = req_capsule_server_pack(pill);
2441         if (rc != 0) {
2442                 CERROR("Can't pack response, rc %d\n", rc);
2443                 RETURN(err_serious(rc));
2444         }
2445
2446         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2447                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2448                 LASSERT(repbody);
2449                 repbody->mbo_eadatasize = 0;
2450                 repbody->mbo_aclsize = 0;
2451         }
2452
2453         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2454
2455         /* for replay no cookkie / lmm need, because client have this already */
2456         if (info->mti_spec.no_create)
2457                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2458                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2459
2460         rc = mdt_init_ucred_reint(info);
2461         if (rc)
2462                 GOTO(out_shrink, rc);
2463
2464         rc = mdt_fix_attr_ucred(info, op);
2465         if (rc != 0)
2466                 GOTO(out_ucred, rc = err_serious(rc));
2467
2468         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2469         if (rc < 0) {
2470                 GOTO(out_ucred, rc);
2471         } else if (rc == 1) {
2472                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt");
2473                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2474                 GOTO(out_ucred, rc);
2475         }
2476         rc = mdt_reint_rec(info, lhc);
2477         EXIT;
2478 out_ucred:
2479         mdt_exit_ucred(info);
2480 out_shrink:
2481         mdt_client_compatibility(info);
2482
2483         rc2 = mdt_fix_reply(info);
2484         if (rc == 0)
2485                 rc = rc2;
2486
2487         /*
2488          * Data-on-MDT optimization - read data along with OPEN and return it
2489          * in reply. Do that only if we have both DOM and LAYOUT locks.
2490          */
2491         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req) &&
2492             info->mti_attr.ma_lmm != NULL &&
2493             mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
2494                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2495                                           &lhc->mlh_reg_lh);
2496         }
2497
2498         return rc;
2499 }
2500
2501 static long mdt_reint_opcode(struct ptlrpc_request *req,
2502                              const struct req_format **fmt)
2503 {
2504         struct mdt_device       *mdt;
2505         struct mdt_rec_reint    *rec;
2506         long                     opc;
2507
2508         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2509         if (rec != NULL) {
2510                 opc = rec->rr_opcode;
2511                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2512                 if (opc < REINT_MAX && fmt[opc] != NULL)
2513                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2514                 else {
2515                         mdt = mdt_exp2dev(req->rq_export);
2516                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2517                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2518                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2519                         opc = err_serious(-EFAULT);
2520                 }
2521         } else {
2522                 opc = err_serious(-EFAULT);
2523         }
2524         return opc;
2525 }
2526
2527 static int mdt_reint(struct tgt_session_info *tsi)
2528 {
2529         long opc;
2530         int  rc;
2531         static const struct req_format *reint_fmts[REINT_MAX] = {
2532                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2533                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2534                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2535                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2536                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2537                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2538                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2539                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2540                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2541                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2542         };
2543
2544         ENTRY;
2545
2546         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2547         if (opc >= 0) {
2548                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2549                 /*
2550                  * No lock possible here from client to pass it to reint code
2551                  * path.
2552                  */
2553                 rc = mdt_reint_internal(info, NULL, opc);
2554                 mdt_thread_info_fini(info);
2555         } else {
2556                 rc = opc;
2557         }
2558
2559         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2560         RETURN(rc);
2561 }
2562
2563 /* this should sync the whole device */
2564 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2565 {
2566         struct dt_device *dt = mdt->mdt_bottom;
2567         int rc;
2568         ENTRY;
2569
2570         rc = dt->dd_ops->dt_sync(env, dt);
2571         RETURN(rc);
2572 }
2573
2574 /* this should sync this object */
2575 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2576                            struct mdt_object *mo)
2577 {
2578         int rc = 0;
2579
2580         ENTRY;
2581
2582         if (!mdt_object_exists(mo)) {
2583                 CWARN("%s: non existing object "DFID": rc = %d\n",
2584                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2585                       -ESTALE);
2586                 RETURN(-ESTALE);
2587         }
2588
2589         if (S_ISREG(lu_object_attr(&mo->mot_obj))) {
2590                 struct lu_target *tgt = tgt_ses_info(env)->tsi_tgt;
2591                 dt_obj_version_t version;
2592
2593                 version = dt_version_get(env, mdt_obj2dt(mo));
2594                 if (version > tgt->lut_obd->obd_last_committed)
2595                         rc = mo_object_sync(env, mdt_object_child(mo));
2596         } else {
2597                 rc = mo_object_sync(env, mdt_object_child(mo));
2598         }
2599
2600         RETURN(rc);
2601 }
2602
2603 static int mdt_sync(struct tgt_session_info *tsi)
2604 {
2605         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2606         struct req_capsule      *pill = tsi->tsi_pill;
2607         struct mdt_body         *body;
2608         int                      rc;
2609
2610         ENTRY;
2611
2612         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2613                 RETURN(err_serious(-ENOMEM));
2614
2615         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2616                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2617         } else {
2618                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2619
2620                 if (unlikely(info->mti_object == NULL))
2621                         RETURN(-EPROTO);
2622
2623                 /* sync an object */
2624                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2625                                      info->mti_object);
2626                 if (rc == 0) {
2627                         const struct lu_fid *fid;
2628                         struct lu_attr *la = &info->mti_attr.ma_attr;
2629
2630                         info->mti_attr.ma_need = MA_INODE;
2631                         info->mti_attr.ma_valid = 0;
2632                         rc = mdt_attr_get_complex(info, info->mti_object,
2633                                                   &info->mti_attr);
2634                         if (rc == 0) {
2635                                 body = req_capsule_server_get(pill,
2636                                                               &RMF_MDT_BODY);
2637                                 fid = mdt_object_fid(info->mti_object);
2638                                 mdt_pack_attr2body(info, body, la, fid);
2639                         }
2640                 }
2641                 mdt_thread_info_fini(info);
2642         }
2643         if (rc == 0)
2644                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2645
2646         RETURN(rc);
2647 }
2648
2649 static int mdt_data_sync(struct tgt_session_info *tsi)
2650 {
2651         struct mdt_thread_info *info;
2652         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2653         struct ost_body *body = tsi->tsi_ost_body;
2654         struct ost_body *repbody;
2655         struct mdt_object *mo = NULL;
2656         struct md_attr *ma;
2657         int rc = 0;
2658
2659         ENTRY;
2660
2661         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2662
2663         /* if no fid is specified then do nothing,
2664          * device sync is done via MDS_SYNC */
2665         if (fid_is_zero(&tsi->tsi_fid))
2666                 RETURN(0);
2667
2668         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2669         if (IS_ERR(mo))
2670                 RETURN(PTR_ERR(mo));
2671
2672         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2673         if (rc)
2674                 GOTO(put, rc);
2675
2676         repbody->oa.o_oi = body->oa.o_oi;
2677         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2678
2679         info = tsi2mdt_info(tsi);
2680         ma = &info->mti_attr;
2681         ma->ma_need = MA_INODE;
2682         ma->ma_valid = 0;
2683         rc = mdt_attr_get_complex(info, mo, ma);
2684         if (rc == 0)
2685                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2686         else
2687                 rc = 0;
2688         mdt_thread_info_fini(info);
2689
2690         EXIT;
2691 put:
2692         if (mo != NULL)
2693                 mdt_object_put(tsi->tsi_env, mo);
2694         return rc;
2695 }
2696
2697 /*
2698  * Handle quota control requests to consult current usage/limit, but also
2699  * to configure quota enforcement
2700  */
2701 static int mdt_quotactl(struct tgt_session_info *tsi)
2702 {
2703         struct obd_export       *exp  = tsi->tsi_exp;
2704         struct req_capsule      *pill = tsi->tsi_pill;
2705         struct obd_quotactl     *oqctl, *repoqc;
2706         int                      id, rc;
2707         struct mdt_device       *mdt = mdt_exp2dev(exp);
2708         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2709         struct lu_nodemap       *nodemap;
2710         ENTRY;
2711
2712         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2713         if (oqctl == NULL)
2714                 RETURN(err_serious(-EPROTO));
2715
2716         rc = req_capsule_server_pack(pill);
2717         if (rc)
2718                 RETURN(err_serious(rc));
2719
2720         nodemap = nodemap_get_from_exp(exp);
2721         if (IS_ERR(nodemap))
2722                 RETURN(PTR_ERR(nodemap));
2723
2724         switch (oqctl->qc_cmd) {
2725                 /* master quotactl */
2726         case Q_SETINFO:
2727         case Q_SETQUOTA:
2728         case LUSTRE_Q_SETDEFAULT:
2729                 if (!nodemap_can_setquota(nodemap))
2730                         GOTO(out_nodemap, rc = -EPERM);
2731                 /* fallthrough */
2732         case Q_GETINFO:
2733         case Q_GETQUOTA:
2734         case LUSTRE_Q_GETDEFAULT:
2735                 if (qmt == NULL)
2736                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2737                 /* slave quotactl */
2738                 /* fallthrough */
2739         case Q_GETOINFO:
2740         case Q_GETOQUOTA:
2741                 break;
2742         default:
2743                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2744                 GOTO(out_nodemap, rc = -EFAULT);
2745         }
2746
2747         id = oqctl->qc_id;
2748         switch (oqctl->qc_type) {
2749         case USRQUOTA:
2750                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2751                                     NODEMAP_CLIENT_TO_FS, id);
2752                 break;
2753         case GRPQUOTA:
2754                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2755                                     NODEMAP_CLIENT_TO_FS, id);
2756                 break;
2757         case PRJQUOTA:
2758                 /* todo: check/map project id */
2759                 id = oqctl->qc_id;
2760                 break;
2761         default:
2762                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2763         }
2764         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2765         if (repoqc == NULL)
2766                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2767
2768         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2769                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2770
2771         if (oqctl->qc_id != id)
2772                 swap(oqctl->qc_id, id);
2773
2774         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2775                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2776                         RETURN(-EINPROGRESS);
2777         }
2778
2779         switch (oqctl->qc_cmd) {
2780
2781         case Q_GETINFO:
2782         case Q_SETINFO:
2783         case Q_SETQUOTA:
2784         case Q_GETQUOTA:
2785         case LUSTRE_Q_SETDEFAULT:
2786         case LUSTRE_Q_GETDEFAULT:
2787                 /* forward quotactl request to QMT */
2788                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2789                 break;
2790
2791         case Q_GETOINFO:
2792         case Q_GETOQUOTA:
2793                 /* slave quotactl */
2794                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2795                                    oqctl);
2796                 break;
2797
2798         default:
2799                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2800                 GOTO(out_nodemap, rc = -EFAULT);
2801         }
2802
2803         if (oqctl->qc_id != id)
2804                 swap(oqctl->qc_id, id);
2805
2806         *repoqc = *oqctl;
2807
2808         EXIT;
2809
2810 out_nodemap:
2811         nodemap_putref(nodemap);
2812
2813         return rc;
2814 }
2815
2816 /** clone llog ctxt from child (mdd)
2817  * This allows remote llog (replicator) access.
2818  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2819  * context was originally set up, or we can handle them directly.
2820  * I choose the latter, but that means I need any llog
2821  * contexts set up by child to be accessable by the mdt.  So we clone the
2822  * context into our context list here.
2823  */
2824 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2825                                int idx)
2826 {
2827         struct md_device  *next = mdt->mdt_child;
2828         struct llog_ctxt *ctxt;
2829         int rc;
2830
2831         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2832                 return 0;
2833
2834         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2835         if (rc || ctxt == NULL) {
2836                 return 0;
2837         }
2838
2839         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2840         if (rc)
2841                 CERROR("Can't set mdt ctxt %d\n", rc);
2842
2843         return rc;
2844 }
2845
2846 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2847                                  struct mdt_device *mdt, int idx)
2848 {
2849         struct llog_ctxt *ctxt;
2850
2851         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2852         if (ctxt == NULL)
2853                 return 0;
2854         /* Put once for the get we just did, and once for the clone */
2855         llog_ctxt_put(ctxt);
2856         llog_ctxt_put(ctxt);
2857         return 0;
2858 }
2859
2860 /*
2861  * sec context handlers
2862  */
2863 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2864 {
2865         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2866
2867         return 0;
2868 }
2869
2870 /*
2871  * quota request handlers
2872  */
2873 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2874 {
2875         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2876         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2877         int                      rc;
2878         ENTRY;
2879
2880         if (qmt == NULL)
2881                 RETURN(err_serious(-EOPNOTSUPP));
2882
2883         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2884         RETURN(rc);
2885 }
2886
2887 struct mdt_object *mdt_object_new(const struct lu_env *env,
2888                                   struct mdt_device *d,
2889                                   const struct lu_fid *f)
2890 {
2891         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2892         struct lu_object *o;
2893         struct mdt_object *m;
2894         ENTRY;
2895
2896         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2897         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2898         if (unlikely(IS_ERR(o)))
2899                 m = (struct mdt_object *)o;
2900         else
2901                 m = mdt_obj(o);
2902         RETURN(m);
2903 }
2904
2905 struct mdt_object *mdt_object_find(const struct lu_env *env,
2906                                    struct mdt_device *d,
2907                                    const struct lu_fid *f)
2908 {
2909         struct lu_object *o;
2910         struct mdt_object *m;
2911         ENTRY;
2912
2913         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2914         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2915         if (unlikely(IS_ERR(o)))
2916                 m = (struct mdt_object *)o;
2917         else
2918                 m = mdt_obj(o);
2919
2920         RETURN(m);
2921 }
2922
2923 /**
2924  * Asyncronous commit for mdt device.
2925  *
2926  * Pass asynchonous commit call down the MDS stack.
2927  *
2928  * \param env environment
2929  * \param mdt the mdt device
2930  */
2931 static void mdt_device_commit_async(const struct lu_env *env,
2932                                     struct mdt_device *mdt)
2933 {
2934         struct dt_device *dt = mdt->mdt_bottom;
2935         int rc;
2936         ENTRY;
2937
2938         rc = dt->dd_ops->dt_commit_async(env, dt);
2939         if (unlikely(rc != 0))
2940                 CWARN("%s: async commit start failed: rc = %d\n",
2941                       mdt_obd_name(mdt), rc);
2942         atomic_inc(&mdt->mdt_async_commit_count);
2943         EXIT;
2944 }
2945
2946 /**
2947  * Mark the lock as "synchonous".
2948  *
2949  * Mark the lock to deffer transaction commit to the unlock time.
2950  *
2951  * \param lock the lock to mark as "synchonous"
2952  *
2953  * \see mdt_is_lock_sync
2954  * \see mdt_save_lock
2955  */
2956 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2957 {
2958         lock->l_ast_data = (void*)1;
2959 }
2960
2961 /**
2962  * Check whehter the lock "synchonous" or not.
2963  *
2964  * \param lock the lock to check
2965  * \retval 1 the lock is "synchonous"
2966  * \retval 0 the lock isn't "synchronous"
2967  *
2968  * \see mdt_set_lock_sync
2969  * \see mdt_save_lock
2970  */
2971 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2972 {
2973         return lock->l_ast_data != NULL;
2974 }
2975
2976 /**
2977  * Blocking AST for mdt locks.
2978  *
2979  * Starts transaction commit if in case of COS lock conflict or
2980  * deffers such a commit to the mdt_save_lock.
2981  *
2982  * \param lock the lock which blocks a request or cancelling lock
2983  * \param desc unused
2984  * \param data unused
2985  * \param flag indicates whether this cancelling or blocking callback
2986  * \retval 0
2987  * \see ldlm_blocking_ast_nocheck
2988  */
2989 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2990                      void *data, int flag)
2991 {
2992         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2993         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2994         struct ldlm_cb_set_arg *arg = data;
2995         bool commit_async = false;
2996         int rc;
2997         ENTRY;
2998
2999         if (flag == LDLM_CB_CANCELING)
3000                 RETURN(0);
3001
3002         lock_res_and_lock(lock);
3003         if (lock->l_blocking_ast != mdt_blocking_ast) {
3004                 unlock_res_and_lock(lock);
3005                 RETURN(0);
3006         }
3007
3008         /* A blocking ast may be sent from ldlm_lock_decref_internal
3009          * when the last reference to a local lock was released and
3010          * during blocking event from ldlm_work_bl_ast_lock().
3011          * The 'data' parameter is l_ast_data in the first case and
3012          * callback arguments in the second one. Distinguish them by that.
3013          */
3014         if (!data || data == lock->l_ast_data || !arg->bl_desc)
3015                 goto skip_cos_checks;
3016
3017         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
3018                 if (mdt_cos_is_enabled(mdt)) {
3019                         if (!arg->bl_desc->bl_same_client)
3020                                 mdt_set_lock_sync(lock);
3021                 } else if (mdt_slc_is_enabled(mdt) &&
3022                            arg->bl_desc->bl_cos_incompat) {
3023                         mdt_set_lock_sync(lock);
3024                         /*
3025                          * we may do extra commit here, but there is a small
3026                          * window to miss a commit: lock was unlocked (saved),
3027                          * then a conflict lock queued and we come here, but
3028                          * REP-ACK not received, so lock was not converted to
3029                          * COS mode yet.
3030                          * Fortunately this window is quite small, so the
3031                          * extra commit should be rare (not to say distributed
3032                          * operation is rare too).
3033                          */
3034                         commit_async = true;
3035                 }
3036         } else if (lock->l_req_mode == LCK_COS) {
3037                 commit_async = true;
3038         }
3039
3040 skip_cos_checks:
3041         rc = ldlm_blocking_ast_nocheck(lock);
3042
3043         if (commit_async) {
3044                 struct lu_env env;
3045
3046                 rc = lu_env_init(&env, LCT_LOCAL);
3047                 if (unlikely(rc != 0))
3048                         CWARN("%s: lu_env initialization failed, cannot "
3049                               "start asynchronous commit: rc = %d\n",
3050                               obd->obd_name, rc);
3051                 else
3052                         mdt_device_commit_async(&env, mdt);
3053                 lu_env_fini(&env);
3054         }
3055         RETURN(rc);
3056 }
3057
3058 /*
3059  * Blocking AST for cross-MDT lock
3060  *
3061  * Discard lock from uncommitted_slc_locks and cancel it.
3062  *
3063  * \param lock  the lock which blocks a request or cancelling lock
3064  * \param desc  unused
3065  * \param data  unused
3066  * \param flag  indicates whether this cancelling or blocking callback
3067  * \retval      0 on success
3068  * \retval      negative number on error
3069  */
3070 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3071                             void *data, int flag)
3072 {
3073         int rc = 0;
3074         ENTRY;
3075
3076         switch (flag) {
3077         case LDLM_CB_BLOCKING: {
3078                 struct lustre_handle lockh;
3079
3080                 ldlm_lock2handle(lock, &lockh);
3081                 rc = ldlm_cli_cancel(&lockh,
3082                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
3083                 if (rc < 0) {
3084                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
3085                         RETURN(rc);
3086                 }
3087                 break;
3088         }
3089         case LDLM_CB_CANCELING: {
3090                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3091                 struct mdt_device *mdt =
3092                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
3093
3094                 LDLM_DEBUG(lock, "Revoke remote lock\n");
3095
3096                 /* discard slc lock here so that it can be cleaned anytime,
3097                  * especially for cleanup_resource() */
3098                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
3099
3100                 /* once we cache lock, l_ast_data is set to mdt_object */
3101                 if (lock->l_ast_data != NULL) {
3102                         struct mdt_object *mo = lock->l_ast_data;
3103                         struct lu_env env;
3104
3105                         rc = lu_env_init(&env, LCT_MD_THREAD);
3106                         if (unlikely(rc != 0)) {
3107                                 CWARN("%s: lu_env initialization failed, object"
3108                                       "%p "DFID" is leaked!\n",
3109                                       obd->obd_name, mo,
3110                                       PFID(mdt_object_fid(mo)));
3111                                 RETURN(rc);
3112                         }
3113
3114                         if (lock->l_policy_data.l_inodebits.bits &
3115                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
3116                                 rc = mo_invalidate(&env, mdt_object_child(mo));
3117                                 mo->mot_cache_attr = 0;
3118                         }
3119                         mdt_object_put(&env, mo);
3120                         lu_env_fini(&env);
3121                 }
3122                 break;
3123         }
3124         default:
3125                 LBUG();
3126         }
3127
3128         RETURN(rc);
3129 }
3130
3131 int mdt_check_resent_lock(struct mdt_thread_info *info,
3132                           struct mdt_object *mo,
3133                           struct mdt_lock_handle *lhc)
3134 {
3135         /* the lock might already be gotten in ldlm_handle_enqueue() */
3136         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
3137                 struct ptlrpc_request *req = mdt_info_req(info);
3138                 struct ldlm_lock      *lock;
3139
3140                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
3141                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
3142                 if (lock == NULL) {
3143                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
3144                          * a resend case, however, it could be already destroyed
3145                          * due to client eviction or a raced cancel RPC. */
3146                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
3147                                           lhc->mlh_reg_lh.cookie);
3148                         RETURN(-ESTALE);
3149                 }
3150
3151                 if (!fid_res_name_eq(mdt_object_fid(mo),
3152                                      &lock->l_resource->lr_name)) {
3153                         CWARN("%s: Although resent, but still not "
3154                               "get child lock:"DFID"\n",
3155                               info->mti_exp->exp_obd->obd_name,
3156                               PFID(mdt_object_fid(mo)));
3157                         LDLM_LOCK_PUT(lock);
3158                         RETURN(-EPROTO);
3159                 }
3160                 LDLM_LOCK_PUT(lock);
3161                 return 0;
3162         }
3163         return 1;
3164 }
3165
3166 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
3167 {