Whamcloud - gitweb
LU-11158 mdt: grow lvb buffer to hold layout
[fs/lustre-release.git] / lustre / mdt / mdt_handler.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.gnu.org/licenses/gpl-2.0.html
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2010, 2017, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  * Lustre is a trademark of Sun Microsystems, Inc.
31  *
32  * lustre/mdt/mdt_handler.c
33  *
34  * Lustre Metadata Target (mdt) request handler
35  *
36  * Author: Peter Braam <braam@clusterfs.com>
37  * Author: Andreas Dilger <adilger@clusterfs.com>
38  * Author: Phil Schwan <phil@clusterfs.com>
39  * Author: Mike Shaver <shaver@clusterfs.com>
40  * Author: Nikita Danilov <nikita@clusterfs.com>
41  * Author: Huang Hua <huanghua@clusterfs.com>
42  * Author: Yury Umanets <umka@clusterfs.com>
43  */
44
45 #define DEBUG_SUBSYSTEM S_MDS
46
47 #include <linux/module.h>
48 #include <linux/pagemap.h>
49
50 #include <dt_object.h>
51 #include <lustre_acl.h>
52 #include <lustre_export.h>
53 #include <uapi/linux/lustre/lustre_ioctl.h>
54 #include <lustre_lfsck.h>
55 #include <lustre_log.h>
56 #include <lustre_nodemap.h>
57 #include <lustre_mds.h>
58 #include <uapi/linux/lustre/lustre_param.h>
59 #include <lustre_quota.h>
60 #include <lustre_swab.h>
61 #include <obd.h>
62 #include <obd_support.h>
63 #include <lustre_barrier.h>
64 #include <obd_cksum.h>
65 #include <llog_swab.h>
66
67 #include "mdt_internal.h"
68
69 static unsigned int max_mod_rpcs_per_client = 8;
70 module_param(max_mod_rpcs_per_client, uint, 0644);
71 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
72
73 mdl_mode_t mdt_mdl_lock_modes[] = {
74         [LCK_MINMODE] = MDL_MINMODE,
75         [LCK_EX]      = MDL_EX,
76         [LCK_PW]      = MDL_PW,
77         [LCK_PR]      = MDL_PR,
78         [LCK_CW]      = MDL_CW,
79         [LCK_CR]      = MDL_CR,
80         [LCK_NL]      = MDL_NL,
81         [LCK_GROUP]   = MDL_GROUP
82 };
83
84 enum ldlm_mode mdt_dlm_lock_modes[] = {
85         [MDL_MINMODE]   = LCK_MINMODE,
86         [MDL_EX]        = LCK_EX,
87         [MDL_PW]        = LCK_PW,
88         [MDL_PR]        = LCK_PR,
89         [MDL_CW]        = LCK_CW,
90         [MDL_CR]        = LCK_CR,
91         [MDL_NL]        = LCK_NL,
92         [MDL_GROUP]     = LCK_GROUP
93 };
94
95 static struct mdt_device *mdt_dev(struct lu_device *d);
96
97 static const struct lu_object_operations mdt_obj_ops;
98
99 /* Slab for MDT object allocation */
100 static struct kmem_cache *mdt_object_kmem;
101
102 /* For HSM restore handles */
103 struct kmem_cache *mdt_hsm_cdt_kmem;
104
105 /* For HSM request handles */
106 struct kmem_cache *mdt_hsm_car_kmem;
107
108 static struct lu_kmem_descr mdt_caches[] = {
109         {
110                 .ckd_cache = &mdt_object_kmem,
111                 .ckd_name  = "mdt_obj",
112                 .ckd_size  = sizeof(struct mdt_object)
113         },
114         {
115                 .ckd_cache      = &mdt_hsm_cdt_kmem,
116                 .ckd_name       = "mdt_cdt_restore_handle",
117                 .ckd_size       = sizeof(struct cdt_restore_handle)
118         },
119         {
120                 .ckd_cache      = &mdt_hsm_car_kmem,
121                 .ckd_name       = "mdt_cdt_agent_req",
122                 .ckd_size       = sizeof(struct cdt_agent_req)
123         },
124         {
125                 .ckd_cache = NULL
126         }
127 };
128
129 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
130 {
131         if (!rep)
132                 return 0;
133         return rep->lock_policy_res1 & op_flag;
134 }
135
136 void mdt_clear_disposition(struct mdt_thread_info *info,
137                            struct ldlm_reply *rep, __u64 op_flag)
138 {
139         if (info) {
140                 info->mti_opdata &= ~op_flag;
141                 tgt_opdata_clear(info->mti_env, op_flag);
142         }
143         if (rep)
144                 rep->lock_policy_res1 &= ~op_flag;
145 }
146
147 void mdt_set_disposition(struct mdt_thread_info *info,
148                          struct ldlm_reply *rep, __u64 op_flag)
149 {
150         if (info) {
151                 info->mti_opdata |= op_flag;
152                 tgt_opdata_set(info->mti_env, op_flag);
153         }
154         if (rep)
155                 rep->lock_policy_res1 |= op_flag;
156 }
157
158 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
159 {
160         lh->mlh_pdo_hash = 0;
161         lh->mlh_reg_mode = lm;
162         lh->mlh_rreg_mode = lm;
163         lh->mlh_type = MDT_REG_LOCK;
164 }
165
166 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
167                        const struct lu_name *lname)
168 {
169         lh->mlh_reg_mode = lock_mode;
170         lh->mlh_pdo_mode = LCK_MINMODE;
171         lh->mlh_rreg_mode = lock_mode;
172         lh->mlh_type = MDT_PDO_LOCK;
173
174         if (lu_name_is_valid(lname)) {
175                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
176                                                      lname->ln_namelen);
177                 /* XXX Workaround for LU-2856
178                  *
179                  * Zero is a valid return value of full_name_hash, but
180                  * several users of mlh_pdo_hash assume a non-zero
181                  * hash value. We therefore map zero onto an
182                  * arbitrary, but consistent value (1) to avoid
183                  * problems further down the road. */
184                 if (unlikely(lh->mlh_pdo_hash == 0))
185                         lh->mlh_pdo_hash = 1;
186         } else {
187                 lh->mlh_pdo_hash = 0;
188         }
189 }
190
191 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
192                               struct mdt_lock_handle *lh)
193 {
194         mdl_mode_t mode;
195         ENTRY;
196
197         /*
198          * Any dir access needs couple of locks:
199          *
200          * 1) on part of dir we gonna take lookup/modify;
201          *
202          * 2) on whole dir to protect it from concurrent splitting and/or to
203          * flush client's cache for readdir().
204          *
205          * so, for a given mode and object this routine decides what lock mode
206          * to use for lock #2:
207          *
208          * 1) if caller's gonna lookup in dir then we need to protect dir from
209          * being splitted only - LCK_CR
210          *
211          * 2) if caller's gonna modify dir then we need to protect dir from
212          * being splitted and to flush cache - LCK_CW
213          *
214          * 3) if caller's gonna modify dir and that dir seems ready for
215          * splitting then we need to protect it from any type of access
216          * (lookup/modify/split) - LCK_EX --bzzz
217          */
218
219         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
220         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
221
222         /*
223          * Ask underlaying level its opinion about preferable PDO lock mode
224          * having access type passed as regular lock mode:
225          *
226          * - MDL_MINMODE means that lower layer does not want to specify lock
227          * mode;
228          *
229          * - MDL_NL means that no PDO lock should be taken. This is used in some
230          * cases. Say, for non-splittable directories no need to use PDO locks
231          * at all.
232          */
233         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
234                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
235
236         if (mode != MDL_MINMODE) {
237                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
238         } else {
239                 /*
240                  * Lower layer does not want to specify locking mode. We do it
241                  * our selves. No special protection is needed, just flush
242                  * client's cache on modification and allow concurrent
243                  * mondification.
244                  */
245                 switch (lh->mlh_reg_mode) {
246                 case LCK_EX:
247                         lh->mlh_pdo_mode = LCK_EX;
248                         break;
249                 case LCK_PR:
250                         lh->mlh_pdo_mode = LCK_CR;
251                         break;
252                 case LCK_PW:
253                         lh->mlh_pdo_mode = LCK_CW;
254                         break;
255                 default:
256                         CERROR("Not expected lock type (0x%x)\n",
257                                (int)lh->mlh_reg_mode);
258                         LBUG();
259                 }
260         }
261
262         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
263         EXIT;
264 }
265
266 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
267                               struct lu_fid *fid)
268 {
269         struct mdt_device *mdt = info->mti_mdt;
270         struct lu_name *lname = &info->mti_name;
271         char *filename = info->mti_filename;
272         struct mdt_object *parent;
273         u32 mode;
274         int rc = 0;
275
276         LASSERT(!info->mti_cross_ref);
277
278         /*
279          * We may want to allow this to mount a completely separate
280          * fileset from the MDT in the future, but keeping it to
281          * ROOT/ only for now avoid potential security issues.
282          */
283         *fid = mdt->mdt_md_root_fid;
284
285         while (rc == 0 && fileset != NULL && *fileset != '\0') {
286                 const char *s1 = fileset;
287                 const char *s2;
288
289                 while (*++s1 == '/')
290                         ;
291                 s2 = s1;
292                 while (*s2 != '/' && *s2 != '\0')
293                         s2++;
294
295                 if (s2 == s1)
296                         break;
297
298                 fileset = s2;
299
300                 lname->ln_namelen = s2 - s1;
301                 if (lname->ln_namelen > NAME_MAX) {
302                         rc = -EINVAL;
303                         break;
304                 }
305
306                 /* reject .. as a path component */
307                 if (lname->ln_namelen == 2 &&
308                     strncmp(s1, "..", 2) == 0) {
309                         rc = -EINVAL;
310                         break;
311                 }
312
313                 strncpy(filename, s1, lname->ln_namelen);
314                 filename[lname->ln_namelen] = '\0';
315                 lname->ln_name = filename;
316
317                 parent = mdt_object_find(info->mti_env, mdt, fid);
318                 if (IS_ERR(parent)) {
319                         rc = PTR_ERR(parent);
320                         break;
321                 }
322                 /* Only got the fid of this obj by name */
323                 fid_zero(fid);
324                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
325                                 fid, &info->mti_spec);
326                 mdt_object_put(info->mti_env, parent);
327         }
328         if (!rc) {
329                 parent = mdt_object_find(info->mti_env, mdt, fid);
330                 if (IS_ERR(parent))
331                         rc = PTR_ERR(parent);
332                 else {
333                         mode = lu_object_attr(&parent->mot_obj);
334                         mdt_object_put(info->mti_env, parent);
335                         if (!S_ISDIR(mode))
336                                 rc = -ENOTDIR;
337                 }
338         }
339
340         return rc;
341 }
342
343 static int mdt_get_root(struct tgt_session_info *tsi)
344 {
345         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
346         struct mdt_device       *mdt = info->mti_mdt;
347         struct mdt_body         *repbody;
348         char                    *fileset = NULL, *buffer = NULL;
349         int                      rc;
350         struct obd_export       *exp = info->mti_exp;
351         char                    *nodemap_fileset;
352
353         ENTRY;
354
355         rc = mdt_check_ucred(info);
356         if (rc)
357                 GOTO(out, rc = err_serious(rc));
358
359         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
360                 GOTO(out, rc = err_serious(-ENOMEM));
361
362         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
363         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
364                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
365                 if (fileset == NULL)
366                         GOTO(out, rc = err_serious(-EFAULT));
367         }
368
369         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
370         if (nodemap_fileset && nodemap_fileset[0]) {
371                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
372                 if (fileset) {
373                         /* consider fileset from client as a sub-fileset
374                          * of the nodemap one */
375                         OBD_ALLOC(buffer, PATH_MAX + 1);
376                         if (buffer == NULL)
377                                 GOTO(out, rc = err_serious(-ENOMEM));
378                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
379                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
380                                 GOTO(out, rc = err_serious(-EINVAL));
381                         fileset = buffer;
382                 } else {
383                         /* enforce fileset as specified in the nodemap */
384                         fileset = nodemap_fileset;
385                 }
386         }
387
388         if (fileset) {
389                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
390                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
391                 if (rc < 0)
392                         GOTO(out, rc = err_serious(rc));
393         } else {
394                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
395         }
396         repbody->mbo_valid |= OBD_MD_FLID;
397
398         EXIT;
399 out:
400         mdt_thread_info_fini(info);
401         if (buffer)
402                 OBD_FREE(buffer, PATH_MAX+1);
403         return rc;
404 }
405
406 static int mdt_statfs(struct tgt_session_info *tsi)
407 {
408         struct ptlrpc_request *req = tgt_ses_req(tsi);
409         struct mdt_thread_info *info = tsi2mdt_info(tsi);
410         struct mdt_device *mdt = info->mti_mdt;
411         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
412         struct md_device *next = mdt->mdt_child;
413         struct ptlrpc_service_part *svcpt;
414         struct obd_statfs *osfs;
415         struct mdt_body *reqbody = NULL;
416         struct mdt_statfs_cache *msf;
417         int rc;
418
419         ENTRY;
420
421         svcpt = req->rq_rqbd->rqbd_svcpt;
422
423         /* This will trigger a watchdog timeout */
424         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
425                          (MDT_SERVICE_WATCHDOG_FACTOR *
426                           at_get(&svcpt->scp_at_estimate)) + 1);
427
428         rc = mdt_check_ucred(info);
429         if (rc)
430                 GOTO(out, rc = err_serious(rc));
431
432         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
433                 GOTO(out, rc = err_serious(-ENOMEM));
434
435         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
436         if (!osfs)
437                 GOTO(out, rc = -EPROTO);
438
439         if (mdt_is_sum_statfs_client(req->rq_export) &&
440                 lustre_packed_msg_size(req->rq_reqmsg) ==
441                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
442                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
443                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
444                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
445         }
446
447         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
448                 msf = &mdt->mdt_sum_osfs;
449         else
450                 msf = &mdt->mdt_osfs;
451
452         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
453                         /** statfs data is too old, get up-to-date one */
454                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
455                                 rc = next->md_ops->mdo_statfs(info->mti_env,
456                                                               next, osfs);
457                         else
458                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
459                                                osfs);
460                         if (rc)
461                                 GOTO(out, rc);
462                         spin_lock(&mdt->mdt_lock);
463                         msf->msf_osfs = *osfs;
464                         msf->msf_age = ktime_get_seconds();
465                         spin_unlock(&mdt->mdt_lock);
466         } else {
467                         /** use cached statfs data */
468                         spin_lock(&mdt->mdt_lock);
469                         *osfs = msf->msf_osfs;
470                         spin_unlock(&mdt->mdt_lock);
471         }
472
473         /* at least try to account for cached pages.  its still racy and
474          * might be under-reporting if clients haven't announced their
475          * caches with brw recently */
476         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
477                " pending %llu free %llu avail %llu\n",
478                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
479                tgd->tgd_tot_pending,
480                osfs->os_bfree << tgd->tgd_blockbits,
481                osfs->os_bavail << tgd->tgd_blockbits);
482
483         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
484                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
485                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
486
487         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
488         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
489                "%llu objects: %llu free; state %x\n",
490                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
491                osfs->os_files, osfs->os_ffree, osfs->os_state);
492
493         if (!exp_grant_param_supp(tsi->tsi_exp) &&
494             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
495                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
496                  * should not see a block size > page size, otherwise
497                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
498                  * block size which is the biggest block size known to work
499                  * with all client's page size. */
500                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
501                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
502                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
503                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
504         }
505         if (rc == 0)
506                 mdt_counter_incr(req, LPROC_MDT_STATFS);
507 out:
508         mdt_thread_info_fini(info);
509         RETURN(rc);
510 }
511
512 /**
513  * Pack size attributes into the reply.
514  */
515 int mdt_pack_size2body(struct mdt_thread_info *info,
516                         const struct lu_fid *fid, struct lustre_handle *lh)
517 {
518         struct mdt_body *b;
519         struct md_attr *ma = &info->mti_attr;
520         int dom_stripe;
521         bool dom_lock = false;
522
523         ENTRY;
524
525         LASSERT(ma->ma_attr.la_valid & LA_MODE);
526
527         if (!S_ISREG(ma->ma_attr.la_mode) ||
528             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
529                 RETURN(-ENODATA);
530
531         dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
532         /* no DoM stripe, no size in reply */
533         if (dom_stripe == LMM_NO_DOM)
534                 RETURN(-ENOENT);
535
536         if (lustre_handle_is_used(lh)) {
537                 struct ldlm_lock *lock;
538
539                 lock = ldlm_handle2lock(lh);
540                 if (lock != NULL) {
541                         dom_lock = ldlm_has_dom(lock);
542                         LDLM_LOCK_PUT(lock);
543                 }
544         }
545
546         /* no DoM lock, no size in reply */
547         if (!dom_lock)
548                 RETURN(0);
549
550         /* Either DoM lock exists or LMM has only DoM stripe then
551          * return size on body. */
552         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
553
554         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
555         RETURN(0);
556 }
557
558 #ifdef CONFIG_FS_POSIX_ACL
559 /*
560  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
561  *
562  * \param       info    thread info object
563  * \param       repbody reply to pack ACLs into
564  * \param       o       mdt object of file to examine
565  * \param       nodemap nodemap of client to reply to
566  * \retval      0       success
567  * \retval      -errno  error getting or parsing ACL from disk
568  */
569 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
570                       struct mdt_object *o, struct lu_nodemap *nodemap)
571 {
572         const struct lu_env     *env = info->mti_env;
573         struct md_object        *next = mdt_object_child(o);
574         struct lu_buf           *buf = &info->mti_buf;
575         struct mdt_device       *mdt = info->mti_mdt;
576         struct req_capsule *pill = info->mti_pill;
577         int rc;
578
579         ENTRY;
580
581         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
582         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
583         if (buf->lb_len == 0)
584                 RETURN(0);
585
586 again:
587         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
588         if (rc < 0) {
589                 if (rc == -ENODATA) {
590                         repbody->mbo_aclsize = 0;
591                         repbody->mbo_valid |= OBD_MD_FLACL;
592                         rc = 0;
593                 } else if (rc == -EOPNOTSUPP) {
594                         rc = 0;
595                 } else {
596                         if (rc == -ERANGE &&
597                             exp_connect_large_acl(info->mti_exp) &&
598                             buf->lb_buf != info->mti_big_acl) {
599                                 if (info->mti_big_acl == NULL) {
600                                         OBD_ALLOC_LARGE(info->mti_big_acl,
601                                                         mdt->mdt_max_ea_size);
602                                         if (info->mti_big_acl == NULL) {
603                                                 CERROR("%s: unable to grow "
604                                                        DFID" ACL buffer\n",
605                                                        mdt_obd_name(mdt),
606                                                        PFID(mdt_object_fid(o)));
607                                                 RETURN(-ENOMEM);
608                                         }
609
610                                         info->mti_big_aclsize =
611                                                         mdt->mdt_max_ea_size;
612                                 }
613
614                                 CDEBUG(D_INODE, "%s: grow the "DFID
615                                        " ACL buffer to size %d\n",
616                                        mdt_obd_name(mdt),
617                                        PFID(mdt_object_fid(o)),
618                                        mdt->mdt_max_ea_size);
619
620                                 buf->lb_buf = info->mti_big_acl;
621                                 buf->lb_len = info->mti_big_aclsize;
622
623                                 goto again;
624                         }
625
626                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
627                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
628                 }
629         } else {
630                 int client;
631                 int server;
632                 int acl_buflen;
633                 int lmm_buflen = 0;
634                 int lmmsize = 0;
635
636                 acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
637                 if (acl_buflen >= rc)
638                         goto map;
639
640                 /* If LOV/LMA EA is small, we can reuse part of their buffer */
641                 client = ptlrpc_req_get_repsize(pill->rc_req);
642                 server = lustre_packed_msg_size(pill->rc_req->rq_repmsg);
643                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
644                         lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD,
645                                                           RCL_SERVER);
646                         lmmsize = repbody->mbo_eadatasize;
647                 }
648
649                 if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) {
650                         CDEBUG(D_INODE, "%s: client prepared buffer size %d "
651                                "is not big enough with the ACL size %d (%d)\n",
652                                mdt_obd_name(mdt), client, rc,
653                                server - acl_buflen - lmm_buflen + rc + lmmsize);
654                         repbody->mbo_aclsize = 0;
655                         repbody->mbo_valid &= ~OBD_MD_FLACL;
656                         RETURN(-ERANGE);
657                 }
658
659 map:
660                 if (buf->lb_buf == info->mti_big_acl)
661                         info->mti_big_acl_used = 1;
662
663                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
664                                      rc, NODEMAP_FS_TO_CLIENT);
665                 /* if all ACLs mapped out, rc is still >= 0 */
666                 if (rc < 0) {
667                         CERROR("%s: nodemap_map_acl unable to parse "DFID
668                                " ACL: rc = %d\n", mdt_obd_name(mdt),
669                                PFID(mdt_object_fid(o)), rc);
670                         repbody->mbo_aclsize = 0;
671                         repbody->mbo_valid &= ~OBD_MD_FLACL;
672                 } else {
673                         repbody->mbo_aclsize = rc;
674                         repbody->mbo_valid |= OBD_MD_FLACL;
675                         rc = 0;
676                 }
677         }
678
679         RETURN(rc);
680 }
681 #endif
682
683 /* XXX Look into layout in MDT layer. */
684 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
685 {
686         struct lov_comp_md_v1   *comp_v1;
687         struct lov_mds_md       *v1;
688         int                      i;
689
690         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
691                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
692
693                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
694                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
695                                 comp_v1->lcm_entries[i].lcme_offset);
696                         /* We don't support partial release for now */
697                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
698                                 return false;
699                 }
700                 return true;
701         } else {
702                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
703                         true : false;
704         }
705 }
706
707 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
708                         const struct lu_attr *attr, const struct lu_fid *fid)
709 {
710         struct md_attr *ma = &info->mti_attr;
711         struct obd_export *exp = info->mti_exp;
712         struct lu_nodemap *nodemap = NULL;
713
714         LASSERT(ma->ma_valid & MA_INODE);
715
716         if (attr->la_valid & LA_ATIME) {
717                 b->mbo_atime = attr->la_atime;
718                 b->mbo_valid |= OBD_MD_FLATIME;
719         }
720         if (attr->la_valid & LA_MTIME) {
721                 b->mbo_mtime = attr->la_mtime;
722                 b->mbo_valid |= OBD_MD_FLMTIME;
723         }
724         if (attr->la_valid & LA_CTIME) {
725                 b->mbo_ctime = attr->la_ctime;
726                 b->mbo_valid |= OBD_MD_FLCTIME;
727         }
728         if (attr->la_valid & LA_FLAGS) {
729                 b->mbo_flags = attr->la_flags;
730                 b->mbo_valid |= OBD_MD_FLFLAGS;
731         }
732         if (attr->la_valid & LA_NLINK) {
733                 b->mbo_nlink = attr->la_nlink;
734                 b->mbo_valid |= OBD_MD_FLNLINK;
735         }
736         if (attr->la_valid & (LA_UID|LA_GID)) {
737                 nodemap = nodemap_get_from_exp(exp);
738                 if (IS_ERR(nodemap))
739                         goto out;
740         }
741         if (attr->la_valid & LA_UID) {
742                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
743                                             NODEMAP_FS_TO_CLIENT,
744                                             attr->la_uid);
745                 b->mbo_valid |= OBD_MD_FLUID;
746         }
747         if (attr->la_valid & LA_GID) {
748                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
749                                             NODEMAP_FS_TO_CLIENT,
750                                             attr->la_gid);
751                 b->mbo_valid |= OBD_MD_FLGID;
752         }
753
754         if (attr->la_valid & LA_PROJID) {
755                 /* TODO, nodemap for project id */
756                 b->mbo_projid = attr->la_projid;
757                 b->mbo_valid |= OBD_MD_FLPROJID;
758         }
759
760         b->mbo_mode = attr->la_mode;
761         if (attr->la_valid & LA_MODE)
762                 b->mbo_valid |= OBD_MD_FLMODE;
763         if (attr->la_valid & LA_TYPE)
764                 b->mbo_valid |= OBD_MD_FLTYPE;
765
766         if (fid != NULL) {
767                 b->mbo_fid1 = *fid;
768                 b->mbo_valid |= OBD_MD_FLID;
769                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
770                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
771         }
772
773         if (!(attr->la_valid & LA_TYPE))
774                 return;
775
776         b->mbo_rdev   = attr->la_rdev;
777         b->mbo_size   = attr->la_size;
778         b->mbo_blocks = attr->la_blocks;
779
780         if (!S_ISREG(attr->la_mode)) {
781                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
782         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
783                 /* means no objects are allocated on osts. */
784                 LASSERT(!(ma->ma_valid & MA_LOV));
785                 /* just ignore blocks occupied by extend attributes on MDS */
786                 b->mbo_blocks = 0;
787                 /* if no object is allocated on osts, the size on mds is valid.
788                  * b=22272 */
789                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
790         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
791                 if (mdt_hsm_is_released(ma->ma_lmm)) {
792                         /* A released file stores its size on MDS. */
793                         /* But return 1 block for released file, unless tools
794                          * like tar will consider it fully sparse. (LU-3864)
795                          */
796                         if (unlikely(b->mbo_size == 0))
797                                 b->mbo_blocks = 0;
798                         else
799                                 b->mbo_blocks = 1;
800                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
801                 } else if (info->mti_som_valid) { /* som is valid */
802                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
803                 }
804         }
805
806         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
807                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
808                        PFID(fid), (unsigned long long)b->mbo_size);
809
810 out:
811         if (!IS_ERR_OR_NULL(nodemap))
812                 nodemap_putref(nodemap);
813 }
814
815 static inline int mdt_body_has_lov(const struct lu_attr *la,
816                                    const struct mdt_body *body)
817 {
818         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
819                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
820 }
821
822 void mdt_client_compatibility(struct mdt_thread_info *info)
823 {
824         struct mdt_body       *body;
825         struct ptlrpc_request *req = mdt_info_req(info);
826         struct obd_export     *exp = req->rq_export;
827         struct md_attr        *ma = &info->mti_attr;
828         struct lu_attr        *la = &ma->ma_attr;
829         ENTRY;
830
831         if (exp_connect_layout(exp))
832                 /* the client can deal with 16-bit lmm_stripe_count */
833                 RETURN_EXIT;
834
835         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
836
837         if (!mdt_body_has_lov(la, body))
838                 RETURN_EXIT;
839
840         /* now we have a reply with a lov for a client not compatible with the
841          * layout lock so we have to clean the layout generation number */
842         if (S_ISREG(la->la_mode))
843                 ma->ma_lmm->lmm_layout_gen = 0;
844         EXIT;
845 }
846
847 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
848                                    struct mdt_object *o)
849 {
850         const struct lu_env *env = info->mti_env;
851         int rc, rc2;
852
853         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
854                           XATTR_NAME_LOV);
855
856         if (rc == -ENODATA)
857                 rc = 0;
858
859         if (rc < 0)
860                 goto out;
861
862         /* Is it a directory? Let's check for the LMV as well */
863         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
864                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
865                                    XATTR_NAME_LMV);
866
867                 if (rc2 == -ENODATA)
868                         rc2 = mo_xattr_get(env, mdt_object_child(o),
869                                            &LU_BUF_NULL,
870                                            XATTR_NAME_DEFAULT_LMV);
871
872                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
873                         rc = rc2;
874         }
875
876 out:
877         return rc;
878 }
879
880 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
881                       const char *name)
882 {
883         const struct lu_env *env = info->mti_env;
884         int rc;
885         ENTRY;
886
887         LASSERT(info->mti_big_lmm_used == 0);
888         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
889         if (rc < 0)
890                 RETURN(rc);
891
892         /* big_lmm may need to be grown */
893         if (info->mti_big_lmmsize < rc) {
894                 int size = size_roundup_power2(rc);
895
896                 if (info->mti_big_lmmsize > 0) {
897                         /* free old buffer */
898                         LASSERT(info->mti_big_lmm);
899                         OBD_FREE_LARGE(info->mti_big_lmm,
900                                        info->mti_big_lmmsize);
901                         info->mti_big_lmm = NULL;
902                         info->mti_big_lmmsize = 0;
903                 }
904
905                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
906                 if (info->mti_big_lmm == NULL)
907                         RETURN(-ENOMEM);
908                 info->mti_big_lmmsize = size;
909         }
910         LASSERT(info->mti_big_lmmsize >= rc);
911
912         info->mti_buf.lb_buf = info->mti_big_lmm;
913         info->mti_buf.lb_len = info->mti_big_lmmsize;
914         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
915
916         RETURN(rc);
917 }
918
919 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
920                    struct md_attr *ma, const char *name)
921 {
922         struct md_object *next = mdt_object_child(o);
923         struct lu_buf    *buf = &info->mti_buf;
924         int rc;
925
926         if (strcmp(name, XATTR_NAME_LOV) == 0) {
927                 buf->lb_buf = ma->ma_lmm;
928                 buf->lb_len = ma->ma_lmm_size;
929                 LASSERT(!(ma->ma_valid & MA_LOV));
930         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
931                 buf->lb_buf = ma->ma_lmv;
932                 buf->lb_len = ma->ma_lmv_size;
933                 LASSERT(!(ma->ma_valid & MA_LMV));
934         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
935                 buf->lb_buf = ma->ma_lmv;
936                 buf->lb_len = ma->ma_lmv_size;
937                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
938         } else {
939                 return -EINVAL;
940         }
941
942         LASSERT(buf->lb_buf);
943
944         rc = mo_xattr_get(info->mti_env, next, buf, name);
945         if (rc > 0) {
946
947 got:
948                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
949                         if (info->mti_big_lmm_used)
950                                 ma->ma_lmm = info->mti_big_lmm;
951
952                         /* NOT return LOV EA with hole to old client. */
953                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
954                                      LOV_PATTERN_F_HOLE) &&
955                             !(exp_connect_flags(info->mti_exp) &
956                               OBD_CONNECT_LFSCK)) {
957                                 return -EIO;
958                         } else {
959                                 ma->ma_lmm_size = rc;
960                                 ma->ma_valid |= MA_LOV;
961                         }
962                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
963                         if (info->mti_big_lmm_used)
964                                 ma->ma_lmv = info->mti_big_lmm;
965
966                         ma->ma_lmv_size = rc;
967                         ma->ma_valid |= MA_LMV;
968                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
969                         ma->ma_lmv_size = rc;
970                         ma->ma_valid |= MA_LMV_DEF;
971                 }
972
973                 /* Update mdt_max_mdsize so all clients will be aware that */
974                 if (info->mti_mdt->mdt_max_mdsize < rc)
975                         info->mti_mdt->mdt_max_mdsize = rc;
976
977                 rc = 0;
978         } else if (rc == -ENODATA) {
979                 /* no LOV EA */
980                 rc = 0;
981         } else if (rc == -ERANGE) {
982                 /* Default LMV has fixed size, so it must be able to fit
983                  * in the original buffer */
984                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
985                         return rc;
986                 rc = mdt_big_xattr_get(info, o, name);
987                 if (rc > 0) {
988                         info->mti_big_lmm_used = 1;
989                         goto got;
990                 }
991         }
992
993         return rc;
994 }
995
996 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
997                       struct lu_fid *pfid)
998 {
999         struct lu_buf           *buf = &info->mti_buf;
1000         struct link_ea_header   *leh;
1001         struct link_ea_entry    *lee;
1002         int                      rc;
1003         ENTRY;
1004
1005         buf->lb_buf = info->mti_big_lmm;
1006         buf->lb_len = info->mti_big_lmmsize;
1007         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1008                           buf, XATTR_NAME_LINK);
1009         /* ignore errors, MA_PFID won't be set and it is
1010          * up to the caller to treat this as an error */
1011         if (rc == -ERANGE || buf->lb_len == 0) {
1012                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1013                 buf->lb_buf = info->mti_big_lmm;
1014                 buf->lb_len = info->mti_big_lmmsize;
1015         }
1016
1017         if (rc < 0)
1018                 RETURN(rc);
1019         if (rc < sizeof(*leh)) {
1020                 CERROR("short LinkEA on "DFID": rc = %d\n",
1021                        PFID(mdt_object_fid(o)), rc);
1022                 RETURN(-ENODATA);
1023         }
1024
1025         leh = (struct link_ea_header *) buf->lb_buf;
1026         lee = (struct link_ea_entry *)(leh + 1);
1027         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1028                 leh->leh_magic = LINK_EA_MAGIC;
1029                 leh->leh_reccount = __swab32(leh->leh_reccount);
1030                 leh->leh_len = __swab64(leh->leh_len);
1031         }
1032         if (leh->leh_magic != LINK_EA_MAGIC)
1033                 RETURN(-EINVAL);
1034         if (leh->leh_reccount == 0)
1035                 RETURN(-ENODATA);
1036
1037         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1038         fid_be_to_cpu(pfid, pfid);
1039
1040         RETURN(0);
1041 }
1042
1043 int mdt_attr_get_complex(struct mdt_thread_info *info,
1044                          struct mdt_object *o, struct md_attr *ma)
1045 {
1046         const struct lu_env *env = info->mti_env;
1047         struct md_object    *next = mdt_object_child(o);
1048         struct lu_buf       *buf = &info->mti_buf;
1049         int                  need = ma->ma_need;
1050         int                  rc = 0, rc2;
1051         u32                  mode;
1052         ENTRY;
1053
1054         ma->ma_valid = 0;
1055
1056         if (mdt_object_exists(o) == 0)
1057                 GOTO(out, rc = -ENOENT);
1058         mode = lu_object_attr(&next->mo_lu);
1059
1060         if (need & MA_INODE) {
1061                 ma->ma_need = MA_INODE;
1062                 rc = mo_attr_get(env, next, ma);
1063                 if (rc)
1064                         GOTO(out, rc);
1065
1066                 if (S_ISREG(mode))
1067                         (void) mdt_get_som(info, o, ma);
1068                 ma->ma_valid |= MA_INODE;
1069         }
1070
1071         if (need & MA_PFID) {
1072                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1073                 if (rc == 0)
1074                         ma->ma_valid |= MA_PFID;
1075                 /* ignore this error, parent fid is not mandatory */
1076                 rc = 0;
1077         }
1078
1079         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1080                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1081                 if (rc)
1082                         GOTO(out, rc);
1083         }
1084
1085         if (need & MA_LMV && S_ISDIR(mode)) {
1086                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1087                 if (rc != 0)
1088                         GOTO(out, rc);
1089         }
1090
1091         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1092                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1093                 if (rc != 0)
1094                         GOTO(out, rc);
1095         }
1096
1097         /*
1098          * In the handle of MA_INODE, we may already get the SOM attr.
1099          */
1100         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1101                 rc = mdt_get_som(info, o, ma);
1102                 if (rc != 0)
1103                         GOTO(out, rc);
1104         }
1105
1106         if (need & MA_HSM && S_ISREG(mode)) {
1107                 buf->lb_buf = info->mti_xattr_buf;
1108                 buf->lb_len = sizeof(info->mti_xattr_buf);
1109                 CLASSERT(sizeof(struct hsm_attrs) <=
1110                          sizeof(info->mti_xattr_buf));
1111                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1112                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1113                 if (rc2 == 0)
1114                         ma->ma_valid |= MA_HSM;
1115                 else if (rc2 < 0 && rc2 != -ENODATA)
1116                         GOTO(out, rc = rc2);
1117         }
1118
1119 #ifdef CONFIG_FS_POSIX_ACL
1120         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1121                 buf->lb_buf = ma->ma_acl;
1122                 buf->lb_len = ma->ma_acl_size;
1123                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1124                 if (rc2 > 0) {
1125                         ma->ma_acl_size = rc2;
1126                         ma->ma_valid |= MA_ACL_DEF;
1127                 } else if (rc2 == -ENODATA) {
1128                         /* no ACLs */
1129                         ma->ma_acl_size = 0;
1130                 } else
1131                         GOTO(out, rc = rc2);
1132         }
1133 #endif
1134 out:
1135         ma->ma_need = need;
1136         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1137                rc, ma->ma_valid, ma->ma_lmm);
1138         RETURN(rc);
1139 }
1140
1141 static int mdt_getattr_internal(struct mdt_thread_info *info,
1142                                 struct mdt_object *o, int ma_need)
1143 {
1144         struct md_object        *next = mdt_object_child(o);
1145         const struct mdt_body   *reqbody = info->mti_body;
1146         struct ptlrpc_request   *req = mdt_info_req(info);
1147         struct md_attr          *ma = &info->mti_attr;
1148         struct lu_attr          *la = &ma->ma_attr;
1149         struct req_capsule      *pill = info->mti_pill;
1150         const struct lu_env     *env = info->mti_env;
1151         struct mdt_body         *repbody;
1152         struct lu_buf           *buffer = &info->mti_buf;
1153         struct obd_export       *exp = info->mti_exp;
1154         int                      rc;
1155         ENTRY;
1156
1157         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1158                 RETURN(err_serious(-ENOMEM));
1159
1160         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1161
1162         ma->ma_valid = 0;
1163
1164         if (mdt_object_remote(o)) {
1165                 /* This object is located on remote node.*/
1166                 /* Return -ENOTSUPP for old client */
1167                 if (!mdt_is_dne_client(req->rq_export))
1168                         GOTO(out, rc = -ENOTSUPP);
1169
1170                 repbody->mbo_fid1 = *mdt_object_fid(o);
1171                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1172                 GOTO(out, rc = 0);
1173         }
1174
1175         if (reqbody->mbo_eadatasize > 0) {
1176                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1177                 if (buffer->lb_buf == NULL)
1178                         GOTO(out, rc = -EPROTO);
1179                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1180                                                       RCL_SERVER);
1181         } else {
1182                 buffer->lb_buf = NULL;
1183                 buffer->lb_len = 0;
1184                 ma_need &= ~(MA_LOV | MA_LMV);
1185                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1186                        mdt_obd_name(info->mti_mdt),
1187                        req->rq_export->exp_client_uuid.uuid);
1188         }
1189
1190         /* If it is dir object and client require MEA, then we got MEA */
1191         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1192             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1193                 /* Assumption: MDT_MD size is enough for lmv size. */
1194                 ma->ma_lmv = buffer->lb_buf;
1195                 ma->ma_lmv_size = buffer->lb_len;
1196                 ma->ma_need = MA_INODE;
1197                 if (ma->ma_lmv_size > 0) {
1198                         if (reqbody->mbo_valid & OBD_MD_MEA)
1199                                 ma->ma_need |= MA_LMV;
1200                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1201                                 ma->ma_need |= MA_LMV_DEF;
1202                 }
1203         } else {
1204                 ma->ma_lmm = buffer->lb_buf;
1205                 ma->ma_lmm_size = buffer->lb_len;
1206                 ma->ma_need = MA_INODE | MA_HSM;
1207                 if (ma->ma_lmm_size > 0)
1208                         ma->ma_need |= MA_LOV;
1209         }
1210
1211         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1212             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1213             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1214                 /* get default stripe info for this dir. */
1215                 ma->ma_need |= MA_LOV_DEF;
1216         }
1217         ma->ma_need |= ma_need;
1218
1219         rc = mdt_attr_get_complex(info, o, ma);
1220         if (unlikely(rc)) {
1221                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1222                        "%s: getattr error for "DFID": rc = %d\n",
1223                        mdt_obd_name(info->mti_mdt),
1224                        PFID(mdt_object_fid(o)), rc);
1225                 RETURN(rc);
1226         }
1227
1228         /* if file is released, check if a restore is running */
1229         if (ma->ma_valid & MA_HSM) {
1230                 repbody->mbo_valid |= OBD_MD_TSTATE;
1231                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1232                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1233                         repbody->mbo_t_state = MS_RESTORE;
1234         }
1235
1236         if (likely(ma->ma_valid & MA_INODE))
1237                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1238         else
1239                 RETURN(-EFAULT);
1240
1241         if (mdt_body_has_lov(la, reqbody)) {
1242                 if (ma->ma_valid & MA_LOV) {
1243                         LASSERT(ma->ma_lmm_size);
1244                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1245                         if (S_ISDIR(la->la_mode))
1246                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1247                         else
1248                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1249                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1250                 }
1251                 if (ma->ma_valid & MA_LMV) {
1252                         /* Return -ENOTSUPP for old client */
1253                         if (!mdt_is_striped_client(req->rq_export))
1254                                 RETURN(-ENOTSUPP);
1255
1256                         LASSERT(S_ISDIR(la->la_mode));
1257                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1258                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1259                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1260                 }
1261                 if (ma->ma_valid & MA_LMV_DEF) {
1262                         /* Return -ENOTSUPP for old client */
1263                         if (!mdt_is_striped_client(req->rq_export))
1264                                 RETURN(-ENOTSUPP);
1265                         LASSERT(S_ISDIR(la->la_mode));
1266                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1267                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1268                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1269                                                OBD_MD_DEFAULT_MEA);
1270                 }
1271         } else if (S_ISLNK(la->la_mode) &&
1272                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1273                 buffer->lb_buf = ma->ma_lmm;
1274                 /* eadatasize from client includes NULL-terminator, so
1275                  * there is no need to read it */
1276                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1277                 rc = mo_readlink(env, next, buffer);
1278                 if (unlikely(rc <= 0)) {
1279                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1280                                mdt_obd_name(info->mti_mdt),
1281                                PFID(mdt_object_fid(o)), rc);
1282                         rc = -EFAULT;
1283                 } else {
1284                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1285
1286                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1287                                 rc -= 2;
1288                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1289                         /* we need to report back size with NULL-terminator
1290                          * because client expects that */
1291                         repbody->mbo_eadatasize = rc + 1;
1292                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1293                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1294                                        "on "DFID ", expected %d\n",
1295                                        mdt_obd_name(info->mti_mdt),
1296                                        rc, PFID(mdt_object_fid(o)),
1297                                        reqbody->mbo_eadatasize - 1);
1298                         /* NULL terminate */
1299                         ((char *)ma->ma_lmm)[rc] = 0;
1300
1301                         /* If the total CDEBUG() size is larger than a page, it
1302                          * will print a warning to the console, avoid this by
1303                          * printing just the last part of the symlink. */
1304                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1305                                print_limit < rc ? "..." : "", print_limit,
1306                                (char *)ma->ma_lmm + rc - print_limit, rc);
1307                         rc = 0;
1308                 }
1309         }
1310
1311         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1312                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1313                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1314                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1315                        repbody->mbo_max_mdsize);
1316         }
1317
1318 #ifdef CONFIG_FS_POSIX_ACL
1319         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1320                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1321                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1322                 if (IS_ERR(nodemap))
1323                         RETURN(PTR_ERR(nodemap));
1324
1325                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1326                 nodemap_putref(nodemap);
1327         }
1328 #endif
1329
1330 out:
1331         if (rc == 0)
1332                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1333
1334         RETURN(rc);
1335 }
1336
1337 static int mdt_getattr(struct tgt_session_info *tsi)
1338 {
1339         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1340         struct mdt_object       *obj = info->mti_object;
1341         struct req_capsule      *pill = info->mti_pill;
1342         struct mdt_body         *reqbody;
1343         struct mdt_body         *repbody;
1344         int rc, rc2;
1345         ENTRY;
1346
1347         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1348         LASSERT(reqbody);
1349         LASSERT(obj != NULL);
1350         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1351
1352         /* Special case for Data-on-MDT files to get data version */
1353         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1354                 rc = mdt_data_version_get(tsi);
1355                 GOTO(out, rc);
1356         }
1357
1358         /* Unlike intent case where we need to pre-fill out buffers early on
1359          * in intent policy for ldlm reasons, here we can have a much better
1360          * guess at EA size by just reading it from disk.
1361          * Exceptions are readdir and (missing) directory striping */
1362         /* Readlink */
1363         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1364                 /* No easy way to know how long is the symlink, but it cannot
1365                  * be more than PATH_MAX, so we allocate +1 */
1366                 rc = PATH_MAX + 1;
1367         /* A special case for fs ROOT: getattr there might fetch
1368          * default EA for entire fs, not just for this dir!
1369          */
1370         } else if (lu_fid_eq(mdt_object_fid(obj),
1371                              &info->mti_mdt->mdt_md_root_fid) &&
1372                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1373                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1374                                                                  MDS_GETATTR)) {
1375                 /* Should the default strping be bigger, mdt_fix_reply
1376                  * will reallocate */
1377                 rc = DEF_REP_MD_SIZE;
1378         } else {
1379                 /* Read the actual EA size from disk */
1380                 rc = mdt_attr_get_eabuf_size(info, obj);
1381         }
1382
1383         if (rc < 0)
1384                 GOTO(out, rc = err_serious(rc));
1385
1386         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1387
1388         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1389          * by default. If the target object has more ACL entries, then
1390          * enlarge the buffer when necessary. */
1391         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1392                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1393
1394         rc = req_capsule_server_pack(pill);
1395         if (unlikely(rc != 0))
1396                 GOTO(out, rc = err_serious(rc));
1397
1398         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1399         LASSERT(repbody != NULL);
1400         repbody->mbo_eadatasize = 0;
1401         repbody->mbo_aclsize = 0;
1402
1403         rc = mdt_check_ucred(info);
1404         if (unlikely(rc))
1405                 GOTO(out_shrink, rc);
1406
1407         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1408
1409         rc = mdt_getattr_internal(info, obj, 0);
1410         EXIT;
1411 out_shrink:
1412         mdt_client_compatibility(info);
1413         rc2 = mdt_fix_reply(info);
1414         if (rc == 0)
1415                 rc = rc2;
1416 out:
1417         mdt_thread_info_fini(info);
1418         return rc;
1419 }
1420
1421 /**
1422  * Handler of layout intent RPC requiring the layout modification
1423  *
1424  * \param[in] info      thread environment
1425  * \param[in] obj       object
1426  * \param[in] layout    layout change descriptor
1427  *
1428  * \retval 0    on success
1429  * \retval < 0  error code
1430  */
1431 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1432                       struct md_layout_change *layout)
1433 {
1434         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1435         int rc;
1436         ENTRY;
1437
1438         if (!mdt_object_exists(obj))
1439                 GOTO(out, rc = -ENOENT);
1440
1441         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1442                 GOTO(out, rc = -EINVAL);
1443
1444         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1445                            MAY_WRITE);
1446         if (rc)
1447                 GOTO(out, rc);
1448
1449         /* take layout lock to prepare layout change */
1450         mdt_lock_reg_init(lh, LCK_EX);
1451         rc = mdt_object_lock(info, obj, lh, MDS_INODELOCK_LAYOUT);
1452         if (rc)
1453                 GOTO(out, rc);
1454
1455         mutex_lock(&obj->mot_som_mutex);
1456         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1457         mutex_unlock(&obj->mot_som_mutex);
1458         mdt_object_unlock(info, obj, lh, 1);
1459 out:
1460         RETURN(rc);
1461 }
1462
1463 /**
1464  * Exchange MOF_LOV_CREATED flags between two objects after a
1465  * layout swap. No assumption is made on whether o1 or o2 have
1466  * created objects or not.
1467  *
1468  * \param[in,out] o1    First swap layout object
1469  * \param[in,out] o2    Second swap layout object
1470  */
1471 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1472 {
1473         unsigned int o1_lov_created = o1->mot_lov_created;
1474
1475         mutex_lock(&o1->mot_lov_mutex);
1476         mutex_lock(&o2->mot_lov_mutex);
1477
1478         o1->mot_lov_created = o2->mot_lov_created;
1479         o2->mot_lov_created = o1_lov_created;
1480
1481         mutex_unlock(&o2->mot_lov_mutex);
1482         mutex_unlock(&o1->mot_lov_mutex);
1483 }
1484
1485 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1486 {
1487         struct mdt_thread_info  *info;
1488         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1489         struct obd_export       *exp = req->rq_export;
1490         struct mdt_object       *o1, *o2, *o;
1491         struct mdt_lock_handle  *lh1, *lh2;
1492         struct mdc_swap_layouts *msl;
1493         int                      rc;
1494         ENTRY;
1495
1496         /* client does not support layout lock, so layout swaping
1497          * is disabled.
1498          * FIXME: there is a problem for old clients which don't support
1499          * layout lock yet. If those clients have already opened the file
1500          * they won't be notified at all so that old layout may still be
1501          * used to do IO. This can be fixed after file release is landed by
1502          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1503         if (!exp_connect_layout(exp))
1504                 RETURN(-EOPNOTSUPP);
1505
1506         info = tsi2mdt_info(tsi);
1507
1508         if (info->mti_dlm_req != NULL)
1509                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1510
1511         o1 = info->mti_object;
1512         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1513                                 &info->mti_body->mbo_fid2);
1514         if (IS_ERR(o))
1515                 GOTO(out, rc = PTR_ERR(o));
1516
1517         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1518                 GOTO(put, rc = -ENOENT);
1519
1520         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1521         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1522                 GOTO(put, rc);
1523
1524         if (rc < 0)
1525                 swap(o1, o2);
1526
1527         /* permission check. Make sure the calling process having permission
1528          * to write both files. */
1529         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1530                                 MAY_WRITE);
1531         if (rc < 0)
1532                 GOTO(put, rc);
1533
1534         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1535                                 MAY_WRITE);
1536         if (rc < 0)
1537                 GOTO(put, rc);
1538
1539         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1540         if (msl == NULL)
1541                 GOTO(put, rc = -EPROTO);
1542
1543         lh1 = &info->mti_lh[MDT_LH_NEW];
1544         mdt_lock_reg_init(lh1, LCK_EX);
1545         lh2 = &info->mti_lh[MDT_LH_OLD];
1546         mdt_lock_reg_init(lh2, LCK_EX);
1547
1548         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1549                              MDS_INODELOCK_XATTR);
1550         if (rc < 0)
1551                 GOTO(put, rc);
1552
1553         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1554                              MDS_INODELOCK_XATTR);
1555         if (rc < 0)
1556                 GOTO(unlock1, rc);
1557
1558         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1559                              mdt_object_child(o2), msl->msl_flags);
1560         if (rc < 0)
1561                 GOTO(unlock2, rc);
1562
1563         mdt_swap_lov_flag(o1, o2);
1564
1565 unlock2:
1566         mdt_object_unlock(info, o2, lh2, rc);
1567 unlock1:
1568         mdt_object_unlock(info, o1, lh1, rc);
1569 put:
1570         mdt_object_put(info->mti_env, o);
1571 out:
1572         mdt_thread_info_fini(info);
1573         RETURN(rc);
1574 }
1575
1576 static int mdt_raw_lookup(struct mdt_thread_info *info,
1577                           struct mdt_object *parent,
1578                           const struct lu_name *lname,
1579                           struct ldlm_reply *ldlm_rep)
1580 {
1581         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1582         int              rc;
1583         ENTRY;
1584
1585         LASSERT(!info->mti_cross_ref);
1586
1587         /* Only got the fid of this obj by name */
1588         fid_zero(child_fid);
1589         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1590                         lname, child_fid, &info->mti_spec);
1591         if (rc == 0) {
1592                 struct mdt_body *repbody;
1593
1594                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1595                 repbody->mbo_fid1 = *child_fid;
1596                 repbody->mbo_valid = OBD_MD_FLID;
1597                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1598         } else if (rc == -ENOENT) {
1599                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1600         }
1601
1602         RETURN(rc);
1603 }
1604
1605 /*
1606  * UPDATE lock should be taken against parent, and be released before exit;
1607  * child_bits lock should be taken against child, and be returned back:
1608  *            (1)normal request should release the child lock;
1609  *            (2)intent request will grant the lock to client.
1610  */
1611 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1612                                  struct mdt_lock_handle *lhc,
1613                                  __u64 child_bits,
1614                                  struct ldlm_reply *ldlm_rep)
1615 {
1616         struct ptlrpc_request  *req = mdt_info_req(info);
1617         struct mdt_body        *reqbody = NULL;
1618         struct mdt_object      *parent = info->mti_object;
1619         struct mdt_object      *child;
1620         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1621         struct lu_name         *lname = NULL;
1622         struct mdt_lock_handle *lhp = NULL;
1623         struct ldlm_lock       *lock;
1624         __u64 try_bits = 0;
1625         bool is_resent;
1626         int ma_need = 0;
1627         int rc;
1628
1629         ENTRY;
1630
1631         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1632         LASSERT(ergo(is_resent,
1633                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1634
1635         if (parent == NULL)
1636                 RETURN(-ENOENT);
1637
1638         if (info->mti_cross_ref) {
1639                 /* Only getattr on the child. Parent is on another node. */
1640                 mdt_set_disposition(info, ldlm_rep,
1641                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1642                 child = parent;
1643                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1644                        "ldlm_rep = %p\n",
1645                        PFID(mdt_object_fid(child)), ldlm_rep);
1646
1647                 rc = mdt_check_resent_lock(info, child, lhc);
1648                 if (rc < 0) {
1649                         RETURN(rc);
1650                 } else if (rc > 0) {
1651                         mdt_lock_handle_init(lhc);
1652                         mdt_lock_reg_init(lhc, LCK_PR);
1653
1654                         /*
1655                          * Object's name is on another MDS, no lookup or layout
1656                          * lock is needed here but update lock is.
1657                          */
1658                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1659                                         MDS_INODELOCK_LAYOUT);
1660                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1661
1662                         rc = mdt_object_lock(info, child, lhc, child_bits);
1663                         if (rc < 0)
1664                                 RETURN(rc);
1665                 }
1666
1667                 /* Finally, we can get attr for child. */
1668                 if (!mdt_object_exists(child)) {
1669                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1670                                         &child->mot_obj,
1671                                         "remote object doesn't exist.");
1672                         mdt_object_unlock(info, child, lhc, 1);
1673                         RETURN(-ENOENT);
1674                 }
1675
1676                 rc = mdt_getattr_internal(info, child, 0);
1677                 if (unlikely(rc != 0))
1678                         mdt_object_unlock(info, child, lhc, 1);
1679
1680                 RETURN(rc);
1681         }
1682
1683         lname = &info->mti_name;
1684         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1685
1686         if (lu_name_is_valid(lname)) {
1687                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1688                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1689                        PNAME(lname), ldlm_rep);
1690         } else {
1691                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1692                 if (unlikely(reqbody == NULL))
1693                         RETURN(err_serious(-EPROTO));
1694
1695                 *child_fid = reqbody->mbo_fid2;
1696
1697                 if (unlikely(!fid_is_sane(child_fid)))
1698                         RETURN(err_serious(-EINVAL));
1699
1700                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1701                        "ldlm_rep = %p\n",
1702                        PFID(mdt_object_fid(parent)),
1703                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1704         }
1705
1706         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1707
1708         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1709                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1710                                 &parent->mot_obj,
1711                                 "Parent doesn't exist!");
1712                 RETURN(-ESTALE);
1713         }
1714
1715         if (mdt_object_remote(parent)) {
1716                 CERROR("%s: parent "DFID" is on remote target\n",
1717                        mdt_obd_name(info->mti_mdt),
1718                        PFID(mdt_object_fid(parent)));
1719                 RETURN(-EIO);
1720         }
1721
1722         if (lu_name_is_valid(lname)) {
1723                 /* Always allow to lookup ".." */
1724                 if (unlikely(lname->ln_namelen == 2 &&
1725                              lname->ln_name[0] == '.' &&
1726                              lname->ln_name[1] == '.'))
1727                         info->mti_spec.sp_permitted = 1;
1728
1729                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1730                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1731
1732                         RETURN(rc);
1733                 }
1734
1735                 /* step 1: lock parent only if parent is a directory */
1736                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1737                         lhp = &info->mti_lh[MDT_LH_PARENT];
1738                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1739                         rc = mdt_object_lock(info, parent, lhp,
1740                                              MDS_INODELOCK_UPDATE);
1741                         if (unlikely(rc != 0))
1742                                 RETURN(rc);
1743                 }
1744
1745                 /* step 2: lookup child's fid by name */
1746                 fid_zero(child_fid);
1747                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1748                                 child_fid, &info->mti_spec);
1749                 if (rc == -ENOENT)
1750                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1751
1752                 if (rc != 0)
1753                         GOTO(out_parent, rc);
1754         }
1755
1756         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1757
1758         /*
1759          *step 3: find the child object by fid & lock it.
1760          *        regardless if it is local or remote.
1761          *
1762          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1763          *      set parent dir fid the same as child fid in getattr by fid case
1764          *      we should not lu_object_find() the object again, could lead
1765          *      to hung if there is a concurrent unlink destroyed the object.
1766          */
1767         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1768                 mdt_object_get(info->mti_env, parent);
1769                 child = parent;
1770         } else {
1771                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1772                                         child_fid);
1773         }
1774
1775         if (unlikely(IS_ERR(child)))
1776                 GOTO(out_parent, rc = PTR_ERR(child));
1777
1778         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1779         if (!mdt_object_exists(child)) {
1780                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1781                                 &child->mot_obj,
1782                                 "Object doesn't exist!");
1783                 GOTO(out_child, rc = -ENOENT);
1784         }
1785
1786         rc = mdt_check_resent_lock(info, child, lhc);
1787         if (rc < 0) {
1788                 GOTO(out_child, rc);
1789         } else if (rc > 0) {
1790                 mdt_lock_handle_init(lhc);
1791                 mdt_lock_reg_init(lhc, LCK_PR);
1792
1793                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1794                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1795                         struct md_attr *ma = &info->mti_attr;
1796
1797                         ma->ma_valid = 0;
1798                         ma->ma_need = MA_INODE;
1799                         rc = mdt_attr_get_complex(info, child, ma);
1800                         if (unlikely(rc != 0))
1801                                 GOTO(out_child, rc);
1802
1803                         /* If the file has not been changed for some time, we
1804                          * return not only a LOOKUP lock, but also an UPDATE
1805                          * lock and this might save us RPC on later STAT. For
1806                          * directories, it also let negative dentry cache start
1807                          * working for this dir. */
1808                         if (ma->ma_valid & MA_INODE &&
1809                             ma->ma_attr.la_valid & LA_CTIME &&
1810                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1811                                 ma->ma_attr.la_ctime < ktime_get_real_seconds())
1812                                 child_bits |= MDS_INODELOCK_UPDATE;
1813                 }
1814
1815                 /* layout lock must be granted in a best-effort way
1816                  * for IT operations */
1817                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1818                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1819                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1820                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1821                             exp_connect_layout(info->mti_exp)) {
1822                                 /* try to grant layout lock for regular file. */
1823                                 try_bits = MDS_INODELOCK_LAYOUT;
1824                         }
1825                         /* Acquire DOM lock in advance for data-on-mdt file */
1826                         if (child != parent)
1827                                 try_bits |= MDS_INODELOCK_DOM;
1828                 }
1829
1830                 if (try_bits != 0) {
1831                         /* try layout lock, it may fail to be granted due to
1832                          * contention at LOOKUP or UPDATE */
1833                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1834                                                  try_bits, false);
1835                         if (child_bits & MDS_INODELOCK_LAYOUT)
1836                                 ma_need |= MA_LOV;
1837                 } else {
1838                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1839                          * client will enqueue the lock to the remote MDT */
1840                         if (mdt_object_remote(child))
1841                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1842                         rc = mdt_object_lock(info, child, lhc, child_bits);
1843                 }
1844                 if (unlikely(rc != 0))
1845                         GOTO(out_child, rc);
1846         }
1847
1848         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1849
1850         /* finally, we can get attr for child. */
1851         rc = mdt_getattr_internal(info, child, ma_need);
1852         if (unlikely(rc != 0)) {
1853                 mdt_object_unlock(info, child, lhc, 1);
1854         } else if (lock) {
1855                 /* Debugging code. */
1856                 LDLM_DEBUG(lock, "Returning lock to client");
1857                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1858                                          &lock->l_resource->lr_name),
1859                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1860                          PLDLMRES(lock->l_resource),
1861                          PFID(mdt_object_fid(child)));
1862
1863                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1864                     mdt_object_exists(child) && !mdt_object_remote(child) &&
1865                     child != parent) {
1866                         LDLM_LOCK_PUT(lock);
1867                         mdt_object_put(info->mti_env, child);
1868                         /* NB: call the mdt_pack_size2body always after
1869                          * mdt_object_put(), that is why this special
1870                          * exit path is used. */
1871                         rc = mdt_pack_size2body(info, child_fid,
1872                                                 &lhc->mlh_reg_lh);
1873                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1874                                 /* DOM lock was taken in advance but this is
1875                                  * not DoM file. Drop the lock. */
1876                                 lock_res_and_lock(lock);
1877                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1878                                 unlock_res_and_lock(lock);
1879                         }
1880
1881                         GOTO(out_parent, rc = 0);
1882                 }
1883         }
1884         if (lock)
1885                 LDLM_LOCK_PUT(lock);
1886
1887         EXIT;
1888 out_child:
1889         mdt_object_put(info->mti_env, child);
1890 out_parent:
1891         if (lhp)
1892                 mdt_object_unlock(info, parent, lhp, 1);
1893         return rc;
1894 }
1895
1896 /* normal handler: should release the child lock */
1897 static int mdt_getattr_name(struct tgt_session_info *tsi)
1898 {
1899         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1900         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1901         struct mdt_body        *reqbody;
1902         struct mdt_body        *repbody;
1903         int rc, rc2;
1904         ENTRY;
1905
1906         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1907         LASSERT(reqbody != NULL);
1908         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1909         LASSERT(repbody != NULL);
1910
1911         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1912         repbody->mbo_eadatasize = 0;
1913         repbody->mbo_aclsize = 0;
1914
1915         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1916         if (unlikely(rc))
1917                 GOTO(out_shrink, rc);
1918
1919         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1920         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1921                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1922                 lhc->mlh_reg_lh.cookie = 0;
1923         }
1924         mdt_exit_ucred(info);
1925         EXIT;
1926 out_shrink:
1927         mdt_client_compatibility(info);
1928         rc2 = mdt_fix_reply(info);
1929         if (rc == 0)
1930                 rc = rc2;
1931         mdt_thread_info_fini(info);
1932         return rc;
1933 }
1934
1935 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1936                          void *karg, void __user *uarg);
1937
1938 static int mdt_set_info(struct tgt_session_info *tsi)
1939 {
1940         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1941         char                    *key;
1942         void                    *val;
1943         int                      keylen, vallen, rc = 0;
1944
1945         ENTRY;
1946
1947         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1948         if (key == NULL) {
1949                 DEBUG_REQ(D_HA, req, "no set_info key");
1950                 RETURN(err_serious(-EFAULT));
1951         }
1952
1953         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1954                                       RCL_CLIENT);
1955
1956         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1957         if (val == NULL) {
1958                 DEBUG_REQ(D_HA, req, "no set_info val");
1959                 RETURN(err_serious(-EFAULT));
1960         }
1961
1962         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1963                                       RCL_CLIENT);
1964
1965         /* Swab any part of val you need to here */
1966         if (KEY_IS(KEY_READ_ONLY)) {
1967                 spin_lock(&req->rq_export->exp_lock);
1968                 if (*(__u32 *)val)
1969                         *exp_connect_flags_ptr(req->rq_export) |=
1970                                 OBD_CONNECT_RDONLY;
1971                 else
1972                         *exp_connect_flags_ptr(req->rq_export) &=
1973                                 ~OBD_CONNECT_RDONLY;
1974                 spin_unlock(&req->rq_export->exp_lock);
1975         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1976                 struct changelog_setinfo *cs = val;
1977
1978                 if (vallen != sizeof(*cs)) {
1979                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1980                                tgt_name(tsi->tsi_tgt), vallen);
1981                         RETURN(-EINVAL);
1982                 }
1983                 if (ptlrpc_req_need_swab(req)) {
1984                         __swab64s(&cs->cs_recno);
1985                         __swab32s(&cs->cs_id);
1986                 }
1987
1988                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1989                                    vallen, val, NULL);
1990         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1991                 if (vallen > 0)
1992                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1993         } else {
1994                 RETURN(-EINVAL);
1995         }
1996         RETURN(rc);
1997 }
1998
1999 static int mdt_readpage(struct tgt_session_info *tsi)
2000 {
2001         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2002         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2003         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2004         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2005         struct mdt_body         *repbody;
2006         int                      rc;
2007         int                      i;
2008
2009         ENTRY;
2010
2011         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2012                 RETURN(err_serious(-ENOMEM));
2013
2014         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2015         if (repbody == NULL || reqbody == NULL)
2016                 RETURN(err_serious(-EFAULT));
2017
2018         /*
2019          * prepare @rdpg before calling lower layers and transfer itself. Here
2020          * reqbody->size contains offset of where to start to read and
2021          * reqbody->nlink contains number bytes to read.
2022          */
2023         rdpg->rp_hash = reqbody->mbo_size;
2024         if (rdpg->rp_hash != reqbody->mbo_size) {
2025                 CERROR("Invalid hash: %#llx != %#llx\n",
2026                        rdpg->rp_hash, reqbody->mbo_size);
2027                 RETURN(-EFAULT);
2028         }
2029
2030         rdpg->rp_attrs = reqbody->mbo_mode;
2031         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2032                 rdpg->rp_attrs |= LUDA_64BITHASH;
2033         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2034                                 exp_max_brw_size(tsi->tsi_exp));
2035         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2036                           PAGE_SHIFT;
2037         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2038         if (rdpg->rp_pages == NULL)
2039                 RETURN(-ENOMEM);
2040
2041         for (i = 0; i < rdpg->rp_npages; ++i) {
2042                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2043                 if (rdpg->rp_pages[i] == NULL)
2044                         GOTO(free_rdpg, rc = -ENOMEM);
2045         }
2046
2047         /* call lower layers to fill allocated pages with directory data */
2048         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2049         if (rc < 0)
2050                 GOTO(free_rdpg, rc);
2051
2052         /* send pages to client */
2053         rc = tgt_sendpage(tsi, rdpg, rc);
2054
2055         EXIT;
2056 free_rdpg:
2057
2058         for (i = 0; i < rdpg->rp_npages; i++)
2059                 if (rdpg->rp_pages[i] != NULL)
2060                         __free_page(rdpg->rp_pages[i]);
2061         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2062
2063         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2064                 RETURN(0);
2065
2066         return rc;
2067 }
2068
2069 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2070 {
2071         struct lu_ucred *uc = mdt_ucred_check(info);
2072         struct lu_attr *attr = &info->mti_attr.ma_attr;
2073
2074         if (uc == NULL)
2075                 return -EINVAL;
2076
2077         if (op != REINT_SETATTR) {
2078                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2079                         attr->la_uid = uc->uc_fsuid;
2080                 /* for S_ISGID, inherit gid from his parent, such work will be
2081                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2082                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2083                         attr->la_gid = uc->uc_fsgid;
2084         }
2085
2086         return 0;
2087 }
2088
2089 static int mdt_reint_internal(struct mdt_thread_info *info,
2090                               struct mdt_lock_handle *lhc,
2091                               __u32 op)
2092 {
2093         struct req_capsule      *pill = info->mti_pill;
2094         struct mdt_body         *repbody;
2095         int                      rc = 0, rc2;
2096
2097         ENTRY;
2098
2099         rc = mdt_reint_unpack(info, op);
2100         if (rc != 0) {
2101                 CERROR("Can't unpack reint, rc %d\n", rc);
2102                 RETURN(err_serious(rc));
2103         }
2104
2105         /* for replay (no_create) lmm is not needed, client has it already */
2106         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2107                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2108                                      DEF_REP_MD_SIZE);
2109
2110         /* llog cookies are always 0, the field is kept for compatibility */
2111         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2112                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2113
2114         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2115          * by default. If the target object has more ACL entries, then
2116          * enlarge the buffer when necessary. */
2117         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2118                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2119                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2120
2121         rc = req_capsule_server_pack(pill);
2122         if (rc != 0) {
2123                 CERROR("Can't pack response, rc %d\n", rc);
2124                 RETURN(err_serious(rc));
2125         }
2126
2127         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2128                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2129                 LASSERT(repbody);
2130                 repbody->mbo_eadatasize = 0;
2131                 repbody->mbo_aclsize = 0;
2132         }
2133
2134         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2135
2136         /* for replay no cookkie / lmm need, because client have this already */
2137         if (info->mti_spec.no_create)
2138                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2139                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2140
2141         rc = mdt_init_ucred_reint(info);
2142         if (rc)
2143                 GOTO(out_shrink, rc);
2144
2145         rc = mdt_fix_attr_ucred(info, op);
2146         if (rc != 0)
2147                 GOTO(out_ucred, rc = err_serious(rc));
2148
2149         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2150         if (rc < 0) {
2151                 GOTO(out_ucred, rc);
2152         } else if (rc == 1) {
2153                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
2154                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2155                 GOTO(out_ucred, rc);
2156         }
2157         rc = mdt_reint_rec(info, lhc);
2158         EXIT;
2159 out_ucred:
2160         mdt_exit_ucred(info);
2161 out_shrink:
2162         mdt_client_compatibility(info);
2163
2164         rc2 = mdt_fix_reply(info);
2165         if (rc == 0)
2166                 rc = rc2;
2167
2168         /*
2169          * Data-on-MDT optimization - read data along with OPEN and return it
2170          * in reply. Do that only if we have both DOM and LAYOUT locks.
2171          */
2172         if (rc == 0 && op == REINT_OPEN &&
2173             info->mti_attr.ma_lmm != NULL &&
2174             mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
2175                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2176                                           &lhc->mlh_reg_lh);
2177         }
2178
2179         return rc;
2180 }
2181
2182 static long mdt_reint_opcode(struct ptlrpc_request *req,
2183                              const struct req_format **fmt)
2184 {
2185         struct mdt_device       *mdt;
2186         struct mdt_rec_reint    *rec;
2187         long                     opc;
2188
2189         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2190         if (rec != NULL) {
2191                 opc = rec->rr_opcode;
2192                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2193                 if (opc < REINT_MAX && fmt[opc] != NULL)
2194                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2195                 else {
2196                         mdt = mdt_exp2dev(req->rq_export);
2197                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2198                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2199                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2200                         opc = err_serious(-EFAULT);
2201                 }
2202         } else {
2203                 opc = err_serious(-EFAULT);
2204         }
2205         return opc;
2206 }
2207
2208 static int mdt_reint(struct tgt_session_info *tsi)
2209 {
2210         long opc;
2211         int  rc;
2212         static const struct req_format *reint_fmts[REINT_MAX] = {
2213                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2214                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2215                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2216                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2217                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2218                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2219                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2220                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2221                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2222                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2223         };
2224
2225         ENTRY;
2226
2227         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2228         if (opc >= 0) {
2229                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2230                 /*
2231                  * No lock possible here from client to pass it to reint code
2232                  * path.
2233                  */
2234                 rc = mdt_reint_internal(info, NULL, opc);
2235                 mdt_thread_info_fini(info);
2236         } else {
2237                 rc = opc;
2238         }
2239
2240         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2241         RETURN(rc);
2242 }
2243
2244 /* this should sync the whole device */
2245 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2246 {
2247         struct dt_device *dt = mdt->mdt_bottom;
2248         int rc;
2249         ENTRY;
2250
2251         rc = dt->dd_ops->dt_sync(env, dt);
2252         RETURN(rc);
2253 }
2254
2255 /* this should sync this object */
2256 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2257                            struct mdt_object *mo)
2258 {
2259         int rc;
2260
2261         ENTRY;
2262
2263         if (!mdt_object_exists(mo)) {
2264                 CWARN("%s: non existing object "DFID": rc = %d\n",
2265                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2266                       -ESTALE);
2267                 RETURN(-ESTALE);
2268         }
2269
2270         rc = mo_object_sync(env, mdt_object_child(mo));
2271
2272         RETURN(rc);
2273 }
2274
2275 static int mdt_sync(struct tgt_session_info *tsi)
2276 {
2277         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2278         struct req_capsule      *pill = tsi->tsi_pill;
2279         struct mdt_body         *body;
2280         int                      rc;
2281
2282         ENTRY;
2283
2284         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2285                 RETURN(err_serious(-ENOMEM));
2286
2287         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2288                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2289         } else {
2290                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2291
2292                 /* sync an object */
2293                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2294                                      info->mti_object);
2295                 if (rc == 0) {
2296                         const struct lu_fid *fid;
2297                         struct lu_attr *la = &info->mti_attr.ma_attr;
2298
2299                         info->mti_attr.ma_need = MA_INODE;
2300                         info->mti_attr.ma_valid = 0;
2301                         rc = mdt_attr_get_complex(info, info->mti_object,
2302                                                   &info->mti_attr);
2303                         if (rc == 0) {
2304                                 body = req_capsule_server_get(pill,
2305                                                               &RMF_MDT_BODY);
2306                                 fid = mdt_object_fid(info->mti_object);
2307                                 mdt_pack_attr2body(info, body, la, fid);
2308                         }
2309                 }
2310                 mdt_thread_info_fini(info);
2311         }
2312         if (rc == 0)
2313                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2314
2315         RETURN(rc);
2316 }
2317
2318 static int mdt_data_sync(struct tgt_session_info *tsi)
2319 {
2320         struct mdt_thread_info *info;
2321         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2322         struct ost_body *body = tsi->tsi_ost_body;
2323         struct ost_body *repbody;
2324         struct mdt_object *mo = NULL;
2325         struct md_attr *ma;
2326         int rc = 0;
2327
2328         ENTRY;
2329
2330         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2331
2332         /* if no fid is specified then do nothing,
2333          * device sync is done via MDS_SYNC */
2334         if (fid_is_zero(&tsi->tsi_fid))
2335                 RETURN(0);
2336
2337         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2338         if (IS_ERR(mo))
2339                 RETURN(PTR_ERR(mo));
2340
2341         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2342         if (rc)
2343                 GOTO(put, rc);
2344
2345         repbody->oa.o_oi = body->oa.o_oi;
2346         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2347
2348         info = tsi2mdt_info(tsi);
2349         ma = &info->mti_attr;
2350         ma->ma_need = MA_INODE;
2351         ma->ma_valid = 0;
2352         rc = mdt_attr_get_complex(info, mo, ma);
2353         if (rc == 0)
2354                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2355         else
2356                 rc = 0;
2357         mdt_thread_info_fini(info);
2358
2359         EXIT;
2360 put:
2361         if (mo != NULL)
2362                 mdt_object_put(tsi->tsi_env, mo);
2363         return rc;
2364 }
2365
2366 /*
2367  * Handle quota control requests to consult current usage/limit, but also
2368  * to configure quota enforcement
2369  */
2370 static int mdt_quotactl(struct tgt_session_info *tsi)
2371 {
2372         struct obd_export       *exp  = tsi->tsi_exp;
2373         struct req_capsule      *pill = tsi->tsi_pill;
2374         struct obd_quotactl     *oqctl, *repoqc;
2375         int                      id, rc;
2376         struct mdt_device       *mdt = mdt_exp2dev(exp);
2377         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2378         struct lu_nodemap       *nodemap;
2379         ENTRY;
2380
2381         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2382         if (oqctl == NULL)
2383                 RETURN(err_serious(-EPROTO));
2384
2385         rc = req_capsule_server_pack(pill);
2386         if (rc)
2387                 RETURN(err_serious(rc));
2388
2389         nodemap = nodemap_get_from_exp(exp);
2390         if (IS_ERR(nodemap))
2391                 RETURN(PTR_ERR(nodemap));
2392
2393         switch (oqctl->qc_cmd) {
2394                 /* master quotactl */
2395         case Q_SETINFO:
2396         case Q_SETQUOTA:
2397         case LUSTRE_Q_SETDEFAULT:
2398                 if (!nodemap_can_setquota(nodemap))
2399                         GOTO(out_nodemap, rc = -EPERM);
2400         case Q_GETINFO:
2401         case Q_GETQUOTA:
2402         case LUSTRE_Q_GETDEFAULT:
2403                 if (qmt == NULL)
2404                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2405                 /* slave quotactl */
2406         case Q_GETOINFO:
2407         case Q_GETOQUOTA:
2408                 break;
2409         default:
2410                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2411                 GOTO(out_nodemap, rc = -EFAULT);
2412         }
2413
2414         id = oqctl->qc_id;
2415         switch (oqctl->qc_type) {
2416         case USRQUOTA:
2417                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2418                                     NODEMAP_CLIENT_TO_FS, id);
2419                 break;
2420         case GRPQUOTA:
2421                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2422                                     NODEMAP_CLIENT_TO_FS, id);
2423                 break;
2424         case PRJQUOTA:
2425                 /* todo: check/map project id */
2426                 id = oqctl->qc_id;
2427                 break;
2428         default:
2429                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2430         }
2431         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2432         if (repoqc == NULL)
2433                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2434
2435         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2436                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2437
2438         if (oqctl->qc_id != id)
2439                 swap(oqctl->qc_id, id);
2440
2441         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2442                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2443                         RETURN(-EINPROGRESS);
2444         }
2445
2446         switch (oqctl->qc_cmd) {
2447
2448         case Q_GETINFO:
2449         case Q_SETINFO:
2450         case Q_SETQUOTA:
2451         case Q_GETQUOTA:
2452         case LUSTRE_Q_SETDEFAULT:
2453         case LUSTRE_Q_GETDEFAULT:
2454                 /* forward quotactl request to QMT */
2455                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2456                 break;
2457
2458         case Q_GETOINFO:
2459         case Q_GETOQUOTA:
2460                 /* slave quotactl */
2461                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2462                                    oqctl);
2463                 break;
2464
2465         default:
2466                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2467                 GOTO(out_nodemap, rc = -EFAULT);
2468         }
2469
2470         if (oqctl->qc_id != id)
2471                 swap(oqctl->qc_id, id);
2472
2473         *repoqc = *oqctl;
2474
2475         EXIT;
2476
2477 out_nodemap:
2478         nodemap_putref(nodemap);
2479
2480         return rc;
2481 }
2482
2483 /** clone llog ctxt from child (mdd)
2484  * This allows remote llog (replicator) access.
2485  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2486  * context was originally set up, or we can handle them directly.
2487  * I choose the latter, but that means I need any llog
2488  * contexts set up by child to be accessable by the mdt.  So we clone the
2489  * context into our context list here.
2490  */
2491 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2492                                int idx)
2493 {
2494         struct md_device  *next = mdt->mdt_child;
2495         struct llog_ctxt *ctxt;
2496         int rc;
2497
2498         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2499                 return 0;
2500
2501         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2502         if (rc || ctxt == NULL) {
2503                 return 0;
2504         }
2505
2506         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2507         if (rc)
2508                 CERROR("Can't set mdt ctxt %d\n", rc);
2509
2510         return rc;
2511 }
2512
2513 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2514                                  struct mdt_device *mdt, int idx)
2515 {
2516         struct llog_ctxt *ctxt;
2517
2518         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2519         if (ctxt == NULL)
2520                 return 0;
2521         /* Put once for the get we just did, and once for the clone */
2522         llog_ctxt_put(ctxt);
2523         llog_ctxt_put(ctxt);
2524         return 0;
2525 }
2526
2527 /*
2528  * sec context handlers
2529  */
2530 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2531 {
2532         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2533
2534         return 0;
2535 }
2536
2537 /*
2538  * quota request handlers
2539  */
2540 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2541 {
2542         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2543         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2544         int                      rc;
2545         ENTRY;
2546
2547         if (qmt == NULL)
2548                 RETURN(err_serious(-EOPNOTSUPP));
2549
2550         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2551         RETURN(rc);
2552 }
2553
2554 struct mdt_object *mdt_object_new(const struct lu_env *env,
2555                                   struct mdt_device *d,
2556                                   const struct lu_fid *f)
2557 {
2558         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2559         struct lu_object *o;
2560         struct mdt_object *m;
2561         ENTRY;
2562
2563         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2564         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2565         if (unlikely(IS_ERR(o)))
2566                 m = (struct mdt_object *)o;
2567         else
2568                 m = mdt_obj(o);
2569         RETURN(m);
2570 }
2571
2572 struct mdt_object *mdt_object_find(const struct lu_env *env,
2573                                    struct mdt_device *d,
2574                                    const struct lu_fid *f)
2575 {
2576         struct lu_object *o;
2577         struct mdt_object *m;
2578         ENTRY;
2579
2580         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2581         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2582         if (unlikely(IS_ERR(o)))
2583                 m = (struct mdt_object *)o;
2584         else
2585                 m = mdt_obj(o);
2586
2587         RETURN(m);
2588 }
2589
2590 /**
2591  * Asyncronous commit for mdt device.
2592  *
2593  * Pass asynchonous commit call down the MDS stack.
2594  *
2595  * \param env environment
2596  * \param mdt the mdt device
2597  */
2598 static void mdt_device_commit_async(const struct lu_env *env,
2599                                     struct mdt_device *mdt)
2600 {
2601         struct dt_device *dt = mdt->mdt_bottom;
2602         int rc;
2603         ENTRY;
2604
2605         rc = dt->dd_ops->dt_commit_async(env, dt);
2606         if (unlikely(rc != 0))
2607                 CWARN("%s: async commit start failed: rc = %d\n",
2608                       mdt_obd_name(mdt), rc);
2609         atomic_inc(&mdt->mdt_async_commit_count);
2610         EXIT;
2611 }
2612
2613 /**
2614  * Mark the lock as "synchonous".
2615  *
2616  * Mark the lock to deffer transaction commit to the unlock time.
2617  *
2618  * \param lock the lock to mark as "synchonous"
2619  *
2620  * \see mdt_is_lock_sync
2621  * \see mdt_save_lock
2622  */
2623 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2624 {
2625         lock->l_ast_data = (void*)1;
2626 }
2627
2628 /**
2629  * Check whehter the lock "synchonous" or not.
2630  *
2631  * \param lock the lock to check
2632  * \retval 1 the lock is "synchonous"
2633  * \retval 0 the lock isn't "synchronous"
2634  *
2635  * \see mdt_set_lock_sync
2636  * \see mdt_save_lock
2637  */
2638 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2639 {
2640         return lock->l_ast_data != NULL;
2641 }
2642
2643 /**
2644  * Blocking AST for mdt locks.
2645  *
2646  * Starts transaction commit if in case of COS lock conflict or
2647  * deffers such a commit to the mdt_save_lock.
2648  *
2649  * \param lock the lock which blocks a request or cancelling lock
2650  * \param desc unused
2651  * \param data unused
2652  * \param flag indicates whether this cancelling or blocking callback
2653  * \retval 0
2654  * \see ldlm_blocking_ast_nocheck
2655  */
2656 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2657                      void *data, int flag)
2658 {
2659         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2660         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2661         bool commit_async = false;
2662         int rc;
2663         ENTRY;
2664
2665         if (flag == LDLM_CB_CANCELING)
2666                 RETURN(0);
2667
2668         lock_res_and_lock(lock);
2669         if (lock->l_blocking_ast != mdt_blocking_ast) {
2670                 unlock_res_and_lock(lock);
2671                 RETURN(0);
2672         }
2673         /* There is no lock conflict if l_blocking_lock == NULL,
2674          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2675          * when the last reference to a local lock was released */
2676         if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2677             lock->l_blocking_lock != NULL) {
2678                 if (mdt_cos_is_enabled(mdt)) {
2679                         if (lock->l_client_cookie !=
2680                             lock->l_blocking_lock->l_client_cookie)
2681                                 mdt_set_lock_sync(lock);
2682                 } else if (mdt_slc_is_enabled(mdt) &&
2683                            ldlm_is_cos_incompat(lock->l_blocking_lock)) {
2684                         mdt_set_lock_sync(lock);
2685                         /*
2686                          * we may do extra commit here, but there is a small
2687                          * window to miss a commit: lock was unlocked (saved),
2688                          * then a conflict lock queued and we come here, but
2689                          * REP-ACK not received, so lock was not converted to
2690                          * COS mode yet.
2691                          * Fortunately this window is quite small, so the
2692                          * extra commit should be rare (not to say distributed
2693                          * operation is rare too).
2694                          */
2695                         commit_async = true;
2696                 }
2697         } else if (lock->l_req_mode == LCK_COS &&
2698                    lock->l_blocking_lock != NULL) {
2699                 commit_async = true;
2700         }
2701
2702         rc = ldlm_blocking_ast_nocheck(lock);
2703
2704         if (commit_async) {
2705                 struct lu_env env;
2706
2707                 rc = lu_env_init(&env, LCT_LOCAL);
2708                 if (unlikely(rc != 0))
2709                         CWARN("%s: lu_env initialization failed, cannot "
2710                               "start asynchronous commit: rc = %d\n",
2711                               obd->obd_name, rc);
2712                 else
2713                         mdt_device_commit_async(&env, mdt);
2714                 lu_env_fini(&env);
2715         }
2716         RETURN(rc);
2717 }
2718
2719 /*
2720  * Blocking AST for cross-MDT lock
2721  *
2722  * Discard lock from uncommitted_slc_locks and cancel it.
2723  *
2724  * \param lock  the lock which blocks a request or cancelling lock
2725  * \param desc  unused
2726  * \param data  unused
2727  * \param flag  indicates whether this cancelling or blocking callback
2728  * \retval      0 on success
2729  * \retval      negative number on error
2730  */
2731 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2732                             void *data, int flag)
2733 {
2734         int rc = 0;
2735         ENTRY;
2736
2737         switch (flag) {
2738         case LDLM_CB_BLOCKING: {
2739                 struct lustre_handle lockh;
2740
2741                 ldlm_lock2handle(lock, &lockh);
2742                 rc = ldlm_cli_cancel(&lockh,
2743                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2744                 if (rc < 0) {
2745                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2746                         RETURN(rc);
2747                 }
2748                 break;
2749         }
2750         case LDLM_CB_CANCELING: {
2751                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2752                 struct mdt_device *mdt =
2753                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2754
2755                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2756
2757                 /* discard slc lock here so that it can be cleaned anytime,
2758                  * especially for cleanup_resource() */
2759                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2760
2761                 /* once we cache lock, l_ast_data is set to mdt_object */
2762                 if (lock->l_ast_data != NULL) {
2763                         struct mdt_object *mo = lock->l_ast_data;
2764                         struct lu_env env;
2765
2766                         rc = lu_env_init(&env, LCT_MD_THREAD);
2767                         if (unlikely(rc != 0)) {
2768                                 CWARN("%s: lu_env initialization failed, object"
2769                                       "%p "DFID" is leaked!\n",
2770                                       obd->obd_name, mo,
2771                                       PFID(mdt_object_fid(mo)));
2772                                 RETURN(rc);
2773                         }
2774
2775                         if (lock->l_policy_data.l_inodebits.bits &
2776                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2777                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2778                                 mo->mot_cache_attr = 0;
2779                         }
2780                         mdt_object_put(&env, mo);
2781                         lu_env_fini(&env);
2782                 }
2783                 break;
2784         }
2785         default:
2786                 LBUG();
2787         }
2788
2789         RETURN(rc);
2790 }
2791
2792 int mdt_check_resent_lock(struct mdt_thread_info *info,
2793                           struct mdt_object *mo,
2794                           struct mdt_lock_handle *lhc)
2795 {
2796         /* the lock might already be gotten in ldlm_handle_enqueue() */
2797         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2798                 struct ptlrpc_request *req = mdt_info_req(info);
2799                 struct ldlm_lock      *lock;
2800
2801                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2802                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2803                 if (lock == NULL) {
2804                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2805                          * a resend case, however, it could be already destroyed
2806                          * due to client eviction or a raced cancel RPC. */
2807                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2808                                           lhc->mlh_reg_lh.cookie);
2809                         RETURN(-ESTALE);
2810                 }
2811
2812                 if (!fid_res_name_eq(mdt_object_fid(mo),
2813                                      &lock->l_resource->lr_name)) {
2814                         CWARN("%s: Although resent, but still not "
2815                               "get child lock:"DFID"\n",
2816                               info->mti_exp->exp_obd->obd_name,
2817                               PFID(mdt_object_fid(mo)));
2818                         LDLM_LOCK_PUT(lock);
2819                         RETURN(-EPROTO);
2820                 }
2821                 LDLM_LOCK_PUT(lock);
2822                 return 0;
2823         }
2824         return 1;
2825 }
2826
2827 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
2828                                struct mdt_object *o, const struct lu_fid *fid,
2829                                struct lustre_handle *lh, enum ldlm_mode mode,
2830                                __u64 *ibits, __u64 trybits, bool cache)
2831 {
2832         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
2833         union ldlm_policy_data *policy = &mti->mti_policy;
2834         struct ldlm_res_id *res_id = &mti->mti_res_id;
2835         int rc = 0;
2836         ENTRY;
2837
2838         LASSERT(mdt_object_remote(o));
2839
2840         fid_build_reg_res_name(fid, res_id);
2841
2842         memset(einfo, 0, sizeof(*einfo));
2843         einfo->ei_type = LDLM_IBITS;
2844         einfo->ei_mode = mode;
2845         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2846         einfo->ei_cb_cp = ldlm_completion_ast;
2847         einfo->ei_enq_slave = 0;
2848         einfo->ei_res_id = res_id;
2849
2850         if (cache) {
2851                 /*
2852                  * if we cache lock, couple lock with mdt_object, so that object
2853                  * can be easily found in lock ASTs.
2854                  */
2855                 mdt_object_get(mti->mti_env, o);
2856                 einfo->ei_cbdata = o;
2857         }
2858
2859         memset(policy, 0, sizeof(*policy));
2860         policy->l_inodebits.bits = *ibits;
2861         policy->l_inodebits.try_bits = trybits;
2862
2863         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2864                             policy);
2865         if (rc < 0 && cache)
2866                 mdt_object_put(mti->mti_env, o);
2867
2868         /* Return successfully acquired bits to a caller */
2869         if (rc == 0) {
2870                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
2871
2872                 LASSERT(lock);
2873                 *ibits = lock->l_policy_data.l_inodebits.bits;
2874                 LDLM_LOCK_PUT(lock);
2875         }
2876         RETURN(rc);
2877 }
2878
2879 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2880                            const struct lu_fid *fid, struct lustre_handle *lh,
2881                            enum ldlm_mode mode, __u64 ibits, bool cache)
2882 {
2883         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
2884                                           cache);
2885 }
2886
2887 static int mdt_object_local_lock(struct mdt_thread_info *info,
2888                                  struct mdt_object *o,
2889                                  struct mdt_lock_handle *lh, __u64 *ibits,
2890                                  __u64 trybits, bool cos_incompat)
2891 {
2892         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2893         union ldlm_policy_data *policy = &info->mti_policy;
2894         struct ldlm_res_id *res_id = &info->mti_res_id;
2895         __u64 dlmflags = 0, *cookie = NULL;
2896         int rc;
2897         ENTRY;
2898
2899         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2900         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2901         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2902         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2903
2904         if (cos_incompat) {
2905                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2906                         lh->mlh_reg_mode == LCK_EX);
2907                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2908         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2909                 dlmflags |= LDLM_FL_COS_ENABLED;
2910         }
2911
2912         /* Only enqueue LOOKUP lock for remote object */
2913         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
2914
2915         if (lh->mlh_type == MDT_PDO_LOCK) {
2916                 /* check for exists after object is locked */
2917                 if (mdt_object_exists(o) == 0) {
2918                         /* Non-existent object shouldn't have PDO lock */
2919                         RETURN(-ESTALE);
2920                 } else {
2921                         /* Non-dir object shouldn't have PDO lock */
2922                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2923                                 RETURN(-ENOTDIR);
2924                 }
2925         }
2926
2927         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2928         dlmflags |= LDLM_FL_ATOMIC_CB;
2929
2930         if (info->mti_exp)
2931                 cookie = &info->mti_exp->exp_handle.h_cookie;
2932
2933         /*
2934          * Take PDO lock on whole directory and build correct @res_id for lock
2935          * on part of directory.
2936          */
2937         if (lh->mlh_pdo_hash != 0) {
2938                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2939                 mdt_lock_pdo_mode(info, o, lh);
2940                 if (lh->mlh_pdo_mode != LCK_NL) {
2941                         /*
2942                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2943                          * is never going to be sent to client and we do not
2944                          * want it slowed down due to possible cancels.
2945                          */
2946                         policy->l_inodebits.bits =
2947                                 *ibits & MDS_INODELOCK_UPDATE;
2948                         policy->l_inodebits.try_bits =
2949                                 trybits & MDS_INODELOCK_UPDATE;
2950                         /* at least one of them should be set */
2951                         LASSERT(policy->l_inodebits.bits |
2952                                 policy->l_inodebits.try_bits);
2953                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
2954                                           lh->mlh_pdo_mode, policy, res_id,
2955                                           dlmflags, cookie);
2956                         if (unlikely(rc != 0))
2957                                 GOTO(out_unlock, rc);
2958                 }
2959
2960                 /*
2961                  * Finish res_id initializing by name hash marking part of
2962                  * directory which is taking modification.
2963                  */
2964                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2965         }
2966
2967         policy->l_inodebits.bits = *ibits;
2968         policy->l_inodebits.try_bits = trybits;
2969
2970         /*
2971          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2972          * going to be sent to client. If it is - mdt_intent_policy() path will
2973          * fix it up and turn FL_LOCAL flag off.
2974          */
2975         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
2976                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2977                           cookie);
2978 out_unlock:
2979         if (rc != 0)
2980                 mdt_object_unlock(info, o, lh, 1);
2981         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2982                    lh->mlh_pdo_hash != 0 &&
2983                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2984                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2985
2986         /* Return successfully acquired bits to a caller */
2987         if (rc == 0) {
2988                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2989
2990                 LASSERT(lock);
2991                 *ibits = lock->l_policy_data.l_inodebits.bits;
2992                 LDLM_LOCK_PUT(lock);
2993         }
2994         RETURN(rc);
2995 }
2996
2997 static int
2998 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2999                          struct mdt_lock_handle *lh, __u64 *ibits,
3000                          __u64 trybits, bool cos_incompat)
3001 {
3002         struct mdt_lock_handle *local_lh = NULL;
3003         int rc;
3004         ENTRY;
3005
3006         if (!mdt_object_remote(o)) {
3007                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3008                                            cos_incompat);
3009                 RETURN(rc);
3010         }
3011
3012         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3013         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3014                     MDS_INODELOCK_XATTR);
3015
3016         /* Only enqueue LOOKUP lock for remote object */
3017         if (*ibits & MDS_INODELOCK_LOOKUP) {
3018                 __u64 local = MDS_INODELOCK_LOOKUP;
3019
3020                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3021                                            cos_incompat);
3022                 if (rc != ELDLM_OK)
3023                         RETURN(rc);
3024
3025                 local_lh = lh;
3026         }
3027
3028         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3029                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3030                  * enqueue RPC can only request 1 lock, to avoid extra
3031                  * RPC, so it will instead enqueue EX lock for remote
3032                  * object anyway XXX*/
3033                 if (lh->mlh_type == MDT_PDO_LOCK &&
3034                     lh->mlh_pdo_hash != 0) {
3035                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
3036                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
3037                                PFID(mdt_object_fid(o)));
3038                         lh->mlh_pdo_hash = 0;
3039                         lh->mlh_rreg_mode = LCK_EX;
3040                         lh->mlh_type = MDT_REG_LOCK;
3041                 }
3042
3043                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3044                                                 &lh->mlh_rreg_lh,
3045                                                 lh->mlh_rreg_mode,
3046                                                 ibits, trybits, false);
3047                 if (rc != ELDLM_OK) {
3048                         if (local_lh != NULL)
3049                                 mdt_object_unlock(info, o, local_lh, rc);
3050                         RETURN(rc);
3051                 }
3052         }
3053
3054         RETURN(0);
3055 }
3056
3057 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3058                     struct mdt_lock_handle *lh, __u64 ibits)
3059 {
3060         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3061 }
3062
3063 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3064                           struct mdt_lock_handle *lh, __u64 ibits,
3065                           bool cos_incompat)
3066 {
3067         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3068         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3069                                         cos_incompat);
3070 }
3071
3072 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3073                         struct mdt_lock_handle *lh, __u64 *ibits,
3074                         __u64 trybits, bool cos_incompat)
3075 {
3076         bool trylock_only = *ibits == 0;
3077         int rc;
3078
3079         LASSERT(!(*ibits & trybits));
3080         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3081                                       cos_incompat);
3082         if (rc && trylock_only) { /* clear error for try ibits lock only */
3083                 LASSERT(*ibits == 0);
3084                 rc = 0;
3085         }
3086         return rc;
3087 }
3088
3089 /**
3090  * Save a lock within request object.
3091  *
3092  * Keep the lock referenced until whether client ACK or transaction
3093  * commit happens or release the lock immediately depending on input
3094  * parameters. If COS is ON, a write lock is converted to COS lock
3095  * before saving.
3096  *
3097  * \param info thead info object
3098  * \param h lock handle
3099  * \param mode lock mode
3100  * \param decref force immediate lock releasing
3101  */
3102 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3103                    enum ldlm_mode mode, int decref)
3104 {
3105         ENTRY;
3106
3107         if (lustre_handle_is_used(h)) {
3108                 if (decref || !info->mti_has_trans ||
3109                     !(mode & (LCK_PW | LCK_EX))) {
3110                         mdt_fid_unlock(h, mode);
3111                 } else {
3112                         struct mdt_device *mdt = info->mti_mdt;
3113                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3114                         struct ptlrpc_request *req = mdt_info_req(info);
3115                         bool cos = mdt_cos_is_enabled(mdt);
3116                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3117
3118                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3119                                  h->cookie);
3120
3121                         /* there is no request if mdt_object_unlock() is called
3122                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3123                         if (likely(req != NULL)) {
3124                                 LDLM_DEBUG(lock, "save lock request %p reply "
3125                                         "state %p transno %lld\n", req,
3126                                         req->rq_reply_state, req->rq_transno);
3127                                 if (cos) {
3128                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3129                                         mode = LCK_COS;
3130                                 }
3131                                 if (req->rq_export->exp_disconnected)
3132                                         mdt_fid_unlock(h, mode);
3133                                 else
3134                                         ptlrpc_save_lock(req, h, mode, cos,
3135                                                          convert_lock);
3136                         } else {
3137                                 mdt_fid_unlock(h, mode);
3138                         }
3139                         if (mdt_is_lock_sync(lock)) {
3140                                 CDEBUG(D_HA, "found sync-lock,"
3141                                        " async commit started\n");
3142                                 mdt_device_commit_async(info->mti_env,
3143                                                         mdt);
3144                         }
3145                         LDLM_LOCK_PUT(lock);
3146                 }
3147                 h->cookie = 0ull;
3148         }
3149
3150         EXIT;
3151 }
3152
3153 /**
<