Whamcloud - gitweb
3c5ebd50d75d0d83eccffffe226ea80224d0f584
[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  *
31  * lustre/mdt/mdt_handler.c
32  *
33  * Lustre Metadata Target (mdt) request handler
34  *
35  * Author: Peter Braam <braam@clusterfs.com>
36  * Author: Andreas Dilger <adilger@clusterfs.com>
37  * Author: Phil Schwan <phil@clusterfs.com>
38  * Author: Mike Shaver <shaver@clusterfs.com>
39  * Author: Nikita Danilov <nikita@clusterfs.com>
40  * Author: Huang Hua <huanghua@clusterfs.com>
41  * Author: Yury Umanets <umka@clusterfs.com>
42  */
43
44 #define DEBUG_SUBSYSTEM S_MDS
45
46 #include <linux/module.h>
47 #include <linux/pagemap.h>
48
49 #include <dt_object.h>
50 #include <lustre_acl.h>
51 #include <lustre_export.h>
52 #include <uapi/linux/lustre/lustre_ioctl.h>
53 #include <lustre_lfsck.h>
54 #include <lustre_log.h>
55 #include <lustre_nodemap.h>
56 #include <lustre_mds.h>
57 #include <uapi/linux/lustre/lustre_param.h>
58 #include <lustre_quota.h>
59 #include <lustre_swab.h>
60 #include <lustre_lmv.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 #include <lustre_crypto.h>
67
68 #include "mdt_internal.h"
69
70 static unsigned int max_mod_rpcs_per_client = 8;
71 module_param(max_mod_rpcs_per_client, uint, 0644);
72 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
73
74 mdl_mode_t mdt_mdl_lock_modes[] = {
75         [LCK_MINMODE] = MDL_MINMODE,
76         [LCK_EX]      = MDL_EX,
77         [LCK_PW]      = MDL_PW,
78         [LCK_PR]      = MDL_PR,
79         [LCK_CW]      = MDL_CW,
80         [LCK_CR]      = MDL_CR,
81         [LCK_NL]      = MDL_NL,
82         [LCK_GROUP]   = MDL_GROUP
83 };
84
85 enum ldlm_mode mdt_dlm_lock_modes[] = {
86         [MDL_MINMODE]   = LCK_MINMODE,
87         [MDL_EX]        = LCK_EX,
88         [MDL_PW]        = LCK_PW,
89         [MDL_PR]        = LCK_PR,
90         [MDL_CW]        = LCK_CW,
91         [MDL_CR]        = LCK_CR,
92         [MDL_NL]        = LCK_NL,
93         [MDL_GROUP]     = LCK_GROUP
94 };
95
96 static struct mdt_device *mdt_dev(struct lu_device *d);
97
98 static const struct lu_object_operations mdt_obj_ops;
99
100 /* Slab for MDT object allocation */
101 static struct kmem_cache *mdt_object_kmem;
102
103 /* For HSM restore handles */
104 struct kmem_cache *mdt_hsm_cdt_kmem;
105
106 /* For HSM request handles */
107 struct kmem_cache *mdt_hsm_car_kmem;
108
109 static struct lu_kmem_descr mdt_caches[] = {
110         {
111                 .ckd_cache = &mdt_object_kmem,
112                 .ckd_name  = "mdt_obj",
113                 .ckd_size  = sizeof(struct mdt_object)
114         },
115         {
116                 .ckd_cache      = &mdt_hsm_cdt_kmem,
117                 .ckd_name       = "mdt_cdt_restore_handle",
118                 .ckd_size       = sizeof(struct cdt_restore_handle)
119         },
120         {
121                 .ckd_cache      = &mdt_hsm_car_kmem,
122                 .ckd_name       = "mdt_cdt_agent_req",
123                 .ckd_size       = sizeof(struct cdt_agent_req)
124         },
125         {
126                 .ckd_cache = NULL
127         }
128 };
129
130 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
131 {
132         if (!rep)
133                 return 0;
134         return rep->lock_policy_res1 & op_flag;
135 }
136
137 void mdt_clear_disposition(struct mdt_thread_info *info,
138                            struct ldlm_reply *rep, __u64 op_flag)
139 {
140         if (info) {
141                 info->mti_opdata &= ~op_flag;
142                 tgt_opdata_clear(info->mti_env, op_flag);
143         }
144         if (rep)
145                 rep->lock_policy_res1 &= ~op_flag;
146 }
147
148 void mdt_set_disposition(struct mdt_thread_info *info,
149                          struct ldlm_reply *rep, __u64 op_flag)
150 {
151         if (info) {
152                 info->mti_opdata |= op_flag;
153                 tgt_opdata_set(info->mti_env, op_flag);
154         }
155         if (rep)
156                 rep->lock_policy_res1 |= op_flag;
157 }
158
159 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
160 {
161         lh->mlh_pdo_hash = 0;
162         lh->mlh_reg_mode = lm;
163         lh->mlh_rreg_mode = lm;
164         lh->mlh_type = MDT_REG_LOCK;
165 }
166
167 void mdt_lh_reg_init(struct mdt_lock_handle *lh, struct ldlm_lock *lock)
168 {
169         mdt_lock_reg_init(lh, lock->l_req_mode);
170         if (lock->l_req_mode == LCK_GROUP)
171                 lh->mlh_gid = lock->l_policy_data.l_inodebits.li_gid;
172 }
173
174 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
175                        const struct lu_name *lname)
176 {
177         lh->mlh_reg_mode = lock_mode;
178         lh->mlh_pdo_mode = LCK_MINMODE;
179         lh->mlh_rreg_mode = lock_mode;
180         lh->mlh_type = MDT_PDO_LOCK;
181
182         if (lu_name_is_valid(lname)) {
183                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
184                                                      lname->ln_namelen);
185                 /* XXX Workaround for LU-2856
186                  *
187                  * Zero is a valid return value of full_name_hash, but
188                  * several users of mlh_pdo_hash assume a non-zero
189                  * hash value. We therefore map zero onto an
190                  * arbitrary, but consistent value (1) to avoid
191                  * problems further down the road. */
192                 if (unlikely(lh->mlh_pdo_hash == 0))
193                         lh->mlh_pdo_hash = 1;
194         } else {
195                 lh->mlh_pdo_hash = 0;
196         }
197 }
198
199 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
200                               struct mdt_lock_handle *lh)
201 {
202         mdl_mode_t mode;
203         ENTRY;
204
205         /*
206          * Any dir access needs couple of locks:
207          *
208          * 1) on part of dir we gonna take lookup/modify;
209          *
210          * 2) on whole dir to protect it from concurrent splitting and/or to
211          * flush client's cache for readdir().
212          *
213          * so, for a given mode and object this routine decides what lock mode
214          * to use for lock #2:
215          *
216          * 1) if caller's gonna lookup in dir then we need to protect dir from
217          * being splitted only - LCK_CR
218          *
219          * 2) if caller's gonna modify dir then we need to protect dir from
220          * being splitted and to flush cache - LCK_CW
221          *
222          * 3) if caller's gonna modify dir and that dir seems ready for
223          * splitting then we need to protect it from any type of access
224          * (lookup/modify/split) - LCK_EX --bzzz
225          */
226
227         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
228         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
229
230         /*
231          * Ask underlaying level its opinion about preferable PDO lock mode
232          * having access type passed as regular lock mode:
233          *
234          * - MDL_MINMODE means that lower layer does not want to specify lock
235          * mode;
236          *
237          * - MDL_NL means that no PDO lock should be taken. This is used in some
238          * cases. Say, for non-splittable directories no need to use PDO locks
239          * at all.
240          */
241         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
242                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
243
244         if (mode != MDL_MINMODE) {
245                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
246         } else {
247                 /*
248                  * Lower layer does not want to specify locking mode. We do it
249                  * our selves. No special protection is needed, just flush
250                  * client's cache on modification and allow concurrent
251                  * mondification.
252                  */
253                 switch (lh->mlh_reg_mode) {
254                 case LCK_EX:
255                         lh->mlh_pdo_mode = LCK_EX;
256                         break;
257                 case LCK_PR:
258                         lh->mlh_pdo_mode = LCK_CR;
259                         break;
260                 case LCK_PW:
261                         lh->mlh_pdo_mode = LCK_CW;
262                         break;
263                 default:
264                         CERROR("Not expected lock type (0x%x)\n",
265                                (int)lh->mlh_reg_mode);
266                         LBUG();
267                 }
268         }
269
270         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
271         EXIT;
272 }
273
274 /**
275  * Check whether \a o is directory stripe object.
276  *
277  * \param[in]  info     thread environment
278  * \param[in]  o        MDT object
279  *
280  * \retval 1    is directory stripe.
281  * \retval 0    isn't directory stripe.
282  * \retval < 1  error code
283  */
284 static int mdt_is_dir_stripe(struct mdt_thread_info *info,
285                                 struct mdt_object *o)
286 {
287         struct md_attr *ma = &info->mti_attr;
288         struct lmv_mds_md_v1 *lmv;
289         int rc;
290
291         rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
292         if (rc < 0)
293                 return rc;
294
295         if (!(ma->ma_valid & MA_LMV))
296                 return 0;
297
298         lmv = &ma->ma_lmv->lmv_md_v1;
299
300         if (!lmv_is_sane2(lmv))
301                 return -EBADF;
302
303         if (le32_to_cpu(lmv->lmv_magic) == LMV_MAGIC_STRIPE)
304                 return 1;
305
306         return 0;
307 }
308
309 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
310                               struct lu_fid *fid)
311 {
312         struct mdt_device *mdt = info->mti_mdt;
313         struct lu_name *lname = &info->mti_name;
314         const char *start = fileset;
315         char *filename = info->mti_filename;
316         struct mdt_object *parent;
317         u32 mode;
318         int rc = 0;
319
320         LASSERT(!info->mti_cross_ref);
321
322         /*
323          * We may want to allow this to mount a completely separate
324          * fileset from the MDT in the future, but keeping it to
325          * ROOT/ only for now avoid potential security issues.
326          */
327         *fid = mdt->mdt_md_root_fid;
328
329         while (rc == 0 && start != NULL && *start != '\0') {
330                 const char *s1 = start;
331                 const char *s2;
332
333                 while (*++s1 == '/')
334                         ;
335                 s2 = s1;
336                 while (*s2 != '/' && *s2 != '\0')
337                         s2++;
338
339                 if (s2 == s1)
340                         break;
341
342                 start = s2;
343
344                 lname->ln_namelen = s2 - s1;
345                 if (lname->ln_namelen > NAME_MAX) {
346                         rc = -EINVAL;
347                         break;
348                 }
349
350                 /* reject .. as a path component */
351                 if (lname->ln_namelen == 2 &&
352                     strncmp(s1, "..", 2) == 0) {
353                         rc = -EINVAL;
354                         break;
355                 }
356
357                 strncpy(filename, s1, lname->ln_namelen);
358                 filename[lname->ln_namelen] = '\0';
359                 lname->ln_name = filename;
360
361                 parent = mdt_object_find(info->mti_env, mdt, fid);
362                 if (IS_ERR(parent)) {
363                         rc = PTR_ERR(parent);
364                         break;
365                 }
366                 /* Only got the fid of this obj by name */
367                 fid_zero(fid);
368                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
369                                 fid, &info->mti_spec);
370                 mdt_object_put(info->mti_env, parent);
371         }
372         if (!rc) {
373                 parent = mdt_object_find(info->mti_env, mdt, fid);
374                 if (IS_ERR(parent))
375                         rc = PTR_ERR(parent);
376                 else {
377                         mode = lu_object_attr(&parent->mot_obj);
378                         if (!S_ISDIR(mode)) {
379                                 rc = -ENOTDIR;
380                         } else if (mdt_is_remote_object(info, parent, parent)) {
381                                 if (!mdt->mdt_enable_remote_subdir_mount) {
382                                         rc = -EREMOTE;
383                                         LCONSOLE_WARN("%s: subdir mount '%s' refused because 'enable_remote_subdir_mount=0': rc = %d\n",
384                                                       mdt_obd_name(mdt),
385                                                       fileset, rc);
386                                 } else {
387                                         LCONSOLE_INFO("%s: subdir mount '%s' is remote and may be slow\n",
388                                                       mdt_obd_name(mdt),
389                                                       fileset);
390                                 }
391                         }
392                         mdt_object_put(info->mti_env, parent);
393                 }
394         }
395
396         return rc;
397 }
398
399 static int mdt_get_root(struct tgt_session_info *tsi)
400 {
401         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
402         struct mdt_device       *mdt = info->mti_mdt;
403         struct mdt_body         *repbody;
404         char                    *fileset = NULL, *buffer = NULL;
405         int                      rc;
406         struct obd_export       *exp = info->mti_exp;
407         char                    *nodemap_fileset;
408
409         ENTRY;
410
411         rc = mdt_check_ucred(info);
412         if (rc)
413                 GOTO(out, rc = err_serious(rc));
414
415         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
416                 GOTO(out, rc = err_serious(-ENOMEM));
417
418         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
419         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
420                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
421                 if (fileset == NULL)
422                         GOTO(out, rc = err_serious(-EFAULT));
423         }
424
425         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
426         if (nodemap_fileset && nodemap_fileset[0]) {
427                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
428                 if (fileset) {
429                         /* consider fileset from client as a sub-fileset
430                          * of the nodemap one */
431                         OBD_ALLOC(buffer, PATH_MAX + 1);
432                         if (buffer == NULL)
433                                 GOTO(out, rc = err_serious(-ENOMEM));
434                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
435                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
436                                 GOTO(out, rc = err_serious(-EINVAL));
437                         fileset = buffer;
438                 } else {
439                         /* enforce fileset as specified in the nodemap */
440                         fileset = nodemap_fileset;
441                 }
442         }
443
444         if (fileset) {
445                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
446                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
447                 if (rc < 0)
448                         GOTO(out, rc = err_serious(rc));
449         } else {
450                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
451         }
452         repbody->mbo_valid |= OBD_MD_FLID;
453
454         EXIT;
455 out:
456         mdt_thread_info_fini(info);
457         if (buffer)
458                 OBD_FREE(buffer, PATH_MAX+1);
459         return rc;
460 }
461
462 static int mdt_statfs(struct tgt_session_info *tsi)
463 {
464         struct ptlrpc_request *req = tgt_ses_req(tsi);
465         struct mdt_thread_info *info = tsi2mdt_info(tsi);
466         struct mdt_device *mdt = info->mti_mdt;
467         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
468         struct md_device *next = mdt->mdt_child;
469         struct ptlrpc_service_part *svcpt;
470         struct obd_statfs *osfs;
471         struct mdt_body *reqbody = NULL;
472         struct mdt_statfs_cache *msf;
473         ktime_t kstart = ktime_get();
474         int current_blockbits;
475         int rc;
476
477         ENTRY;
478
479         svcpt = req->rq_rqbd->rqbd_svcpt;
480
481         /* This will trigger a watchdog timeout */
482         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
483                          (MDT_SERVICE_WATCHDOG_FACTOR *
484                           at_get(&svcpt->scp_at_estimate)) + 1);
485
486         rc = mdt_check_ucred(info);
487         if (rc)
488                 GOTO(out, rc = err_serious(rc));
489
490         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
491                 GOTO(out, rc = err_serious(-ENOMEM));
492
493         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
494         if (!osfs)
495                 GOTO(out, rc = -EPROTO);
496
497         if (mdt_is_sum_statfs_client(req->rq_export) &&
498                 lustre_packed_msg_size(req->rq_reqmsg) ==
499                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
500                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
501                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
502                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
503         }
504
505         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
506                 msf = &mdt->mdt_sum_osfs;
507         else
508                 msf = &mdt->mdt_osfs;
509
510         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
511                         /** statfs data is too old, get up-to-date one */
512                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
513                                 rc = next->md_ops->mdo_statfs(info->mti_env,
514                                                               next, osfs);
515                         else
516                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
517                                                osfs);
518                         if (rc)
519                                 GOTO(out, rc);
520                         spin_lock(&mdt->mdt_lock);
521                         msf->msf_osfs = *osfs;
522                         msf->msf_age = ktime_get_seconds();
523                         spin_unlock(&mdt->mdt_lock);
524         } else {
525                         /** use cached statfs data */
526                         spin_lock(&mdt->mdt_lock);
527                         *osfs = msf->msf_osfs;
528                         spin_unlock(&mdt->mdt_lock);
529         }
530
531         /* tgd_blockbit is recordsize bits set during mkfs.
532          * This once set does not change. However, 'zfs set'
533          * can be used to change the MDT blocksize. Instead
534          * of using cached value of 'tgd_blockbit' always
535          * calculate the blocksize bits which may have
536          * changed.
537          */
538         current_blockbits = fls64(osfs->os_bsize) - 1;
539
540         /* at least try to account for cached pages.  its still racy and
541          * might be under-reporting if clients haven't announced their
542          * caches with brw recently */
543         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
544                " pending %llu free %llu avail %llu\n",
545                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
546                tgd->tgd_tot_pending,
547                osfs->os_bfree << current_blockbits,
548                osfs->os_bavail << current_blockbits);
549
550         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
551                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
552                                    osfs->os_bsize - 1) >> current_blockbits));
553
554         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
555         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
556                "%llu objects: %llu free; state %x\n",
557                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
558                osfs->os_files, osfs->os_ffree, osfs->os_state);
559
560         if (!exp_grant_param_supp(tsi->tsi_exp) &&
561             current_blockbits > COMPAT_BSIZE_SHIFT) {
562                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
563                  * should not see a block size > page size, otherwise
564                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
565                  * block size which is the biggest block size known to work
566                  * with all client's page size. */
567                 osfs->os_blocks <<= current_blockbits - COMPAT_BSIZE_SHIFT;
568                 osfs->os_bfree  <<= current_blockbits - COMPAT_BSIZE_SHIFT;
569                 osfs->os_bavail <<= current_blockbits - COMPAT_BSIZE_SHIFT;
570                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
571         }
572         if (rc == 0)
573                 mdt_counter_incr(req, LPROC_MDT_STATFS,
574                                  ktime_us_delta(ktime_get(), kstart));
575 out:
576         mdt_thread_info_fini(info);
577         RETURN(rc);
578 }
579
580 __u32 mdt_lmm_dom_entry_check(struct lov_mds_md *lmm, int *is_dom_only)
581 {
582         struct lov_comp_md_v1 *comp_v1;
583         struct lov_mds_md *v1;
584         __u32 off;
585         __u32 dom_stripesize = 0;
586         int i;
587         bool has_ost_stripes = false;
588
589         ENTRY;
590
591         if (is_dom_only)
592                 *is_dom_only = 0;
593
594         if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
595                 RETURN(0);
596
597         comp_v1 = (struct lov_comp_md_v1 *)lmm;
598         off = le32_to_cpu(comp_v1->lcm_entries[0].lcme_offset);
599         v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
600
601         /* Fast check for DoM entry with no mirroring, should be the first */
602         if (le16_to_cpu(comp_v1->lcm_mirror_count) == 0 &&
603             lov_pattern(le32_to_cpu(v1->lmm_pattern)) != LOV_PATTERN_MDT)
604                 RETURN(0);
605
606         /* check all entries otherwise */
607         for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
608                 struct lov_comp_md_entry_v1 *lcme;
609
610                 lcme = &comp_v1->lcm_entries[i];
611                 if (!(le32_to_cpu(lcme->lcme_flags) & LCME_FL_INIT))
612                         continue;
613
614                 off = le32_to_cpu(lcme->lcme_offset);
615                 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
616
617                 if (lov_pattern(le32_to_cpu(v1->lmm_pattern)) ==
618                     LOV_PATTERN_MDT)
619                         dom_stripesize = le32_to_cpu(v1->lmm_stripe_size);
620                 else
621                         has_ost_stripes = true;
622
623                 if (dom_stripesize && has_ost_stripes)
624                         RETURN(dom_stripesize);
625         }
626         /* DoM-only case exits here */
627         if (is_dom_only && dom_stripesize)
628                 *is_dom_only = 1;
629         RETURN(dom_stripesize);
630 }
631
632 /**
633  * Pack size attributes into the reply.
634  */
635 int mdt_pack_size2body(struct mdt_thread_info *info,
636                         const struct lu_fid *fid, struct lustre_handle *lh)
637 {
638         struct mdt_body *b;
639         struct md_attr *ma = &info->mti_attr;
640         __u32 dom_stripe;
641         bool dom_lock = false;
642
643         ENTRY;
644
645         LASSERT(ma->ma_attr.la_valid & LA_MODE);
646
647         if (!S_ISREG(ma->ma_attr.la_mode) ||
648             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
649                 RETURN(-ENODATA);
650
651         dom_stripe = mdt_lmm_dom_stripesize(ma->ma_lmm);
652         /* no DoM stripe, no size in reply */
653         if (!dom_stripe)
654                 RETURN(-ENOENT);
655
656         if (lustre_handle_is_used(lh)) {
657                 struct ldlm_lock *lock;
658
659                 lock = ldlm_handle2lock(lh);
660                 if (lock != NULL) {
661                         dom_lock = ldlm_has_dom(lock);
662                         LDLM_LOCK_PUT(lock);
663                 }
664         }
665
666         /* no DoM lock, no size in reply */
667         if (!dom_lock)
668                 RETURN(0);
669
670         /* Either DoM lock exists or LMM has only DoM stripe then
671          * return size on body. */
672         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
673
674         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
675         RETURN(0);
676 }
677
678 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
679 /*
680  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
681  *
682  * \param       info    thread info object
683  * \param       repbody reply to pack ACLs into
684  * \param       o       mdt object of file to examine
685  * \param       nodemap nodemap of client to reply to
686  * \retval      0       success
687  * \retval      -errno  error getting or parsing ACL from disk
688  */
689 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
690                       struct mdt_object *o, struct lu_nodemap *nodemap)
691 {
692         const struct lu_env     *env = info->mti_env;
693         struct md_object        *next = mdt_object_child(o);
694         struct lu_buf           *buf = &info->mti_buf;
695         struct mdt_device       *mdt = info->mti_mdt;
696         struct req_capsule *pill = info->mti_pill;
697         int rc;
698
699         ENTRY;
700
701         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
702         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
703         if (buf->lb_len == 0)
704                 RETURN(0);
705
706         LASSERT(!info->mti_big_acl_used);
707 again:
708         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
709         if (rc < 0) {
710                 if (rc == -ENODATA) {
711                         repbody->mbo_aclsize = 0;
712                         repbody->mbo_valid |= OBD_MD_FLACL;
713                         rc = 0;
714                 } else if (rc == -EOPNOTSUPP) {
715                         rc = 0;
716                 } else if (rc == -ERANGE) {
717                         if (exp_connect_large_acl(info->mti_exp) &&
718                             !info->mti_big_acl_used) {
719                                 if (info->mti_big_acl == NULL) {
720                                         info->mti_big_aclsize =
721                                                         min_t(unsigned int,
722                                                               mdt->mdt_max_ea_size,
723                                                               XATTR_SIZE_MAX);
724                                         OBD_ALLOC_LARGE(info->mti_big_acl,
725                                                         info->mti_big_aclsize);
726                                         if (info->mti_big_acl == NULL) {
727                                                 info->mti_big_aclsize = 0;
728                                                 CERROR("%s: unable to grow "
729                                                        DFID" ACL buffer\n",
730                                                        mdt_obd_name(mdt),
731                                                        PFID(mdt_object_fid(o)));
732                                                 RETURN(-ENOMEM);
733                                         }
734                                 }
735
736                                 CDEBUG(D_INODE, "%s: grow the "DFID
737                                        " ACL buffer to size %d\n",
738                                        mdt_obd_name(mdt),
739                                        PFID(mdt_object_fid(o)),
740                                        info->mti_big_aclsize);
741
742                                 buf->lb_buf = info->mti_big_acl;
743                                 buf->lb_len = info->mti_big_aclsize;
744                                 info->mti_big_acl_used = 1;
745                                 goto again;
746                         }
747                         /* FS has ACL bigger that our limits */
748                         CDEBUG(D_INODE, "%s: "DFID" ACL can't fit into %d\n",
749                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)),
750                                info->mti_big_aclsize);
751                         rc = -E2BIG;
752                 } else {
753                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
754                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
755                 }
756         } else {
757                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
758                                      rc, NODEMAP_FS_TO_CLIENT);
759                 /* if all ACLs mapped out, rc is still >= 0 */
760                 if (rc < 0) {
761                         CERROR("%s: nodemap_map_acl unable to parse "DFID
762                                " ACL: rc = %d\n", mdt_obd_name(mdt),
763                                PFID(mdt_object_fid(o)), rc);
764                         repbody->mbo_aclsize = 0;
765                         repbody->mbo_valid &= ~OBD_MD_FLACL;
766                 } else {
767                         repbody->mbo_aclsize = rc;
768                         repbody->mbo_valid |= OBD_MD_FLACL;
769                         rc = 0;
770                 }
771         }
772
773         RETURN(rc);
774 }
775 #endif
776
777 /* XXX Look into layout in MDT layer. */
778 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
779 {
780         struct lov_comp_md_v1   *comp_v1;
781         struct lov_mds_md       *v1;
782         int                      i;
783
784         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
785                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
786
787                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
788                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
789                                 comp_v1->lcm_entries[i].lcme_offset);
790                         /* We don't support partial release for now */
791                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
792                                 return false;
793                 }
794                 return true;
795         } else {
796                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
797                         true : false;
798         }
799 }
800
801 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
802                         const struct lu_attr *attr, const struct lu_fid *fid)
803 {
804         struct mdt_device *mdt = info->mti_mdt;
805         struct obd_export *exp = info->mti_exp;
806         struct md_attr *ma = &info->mti_attr;
807         struct lu_nodemap *nodemap = NULL;
808
809         LASSERT(ma->ma_valid & MA_INODE);
810
811         if (attr->la_valid & LA_ATIME) {
812                 b->mbo_atime = attr->la_atime;
813                 b->mbo_valid |= OBD_MD_FLATIME;
814         }
815         if (attr->la_valid & LA_MTIME) {
816                 b->mbo_mtime = attr->la_mtime;
817                 b->mbo_valid |= OBD_MD_FLMTIME;
818         }
819         if (attr->la_valid & LA_CTIME) {
820                 b->mbo_ctime = attr->la_ctime;
821                 b->mbo_valid |= OBD_MD_FLCTIME;
822         }
823         if (attr->la_valid & LA_BTIME) {
824                 b->mbo_btime = attr->la_btime;
825                 b->mbo_valid |= OBD_MD_FLBTIME;
826         }
827         if (attr->la_valid & LA_FLAGS) {
828                 b->mbo_flags = attr->la_flags;
829                 b->mbo_valid |= OBD_MD_FLFLAGS;
830         }
831         if (attr->la_valid & LA_NLINK) {
832                 b->mbo_nlink = attr->la_nlink;
833                 b->mbo_valid |= OBD_MD_FLNLINK;
834         }
835         if (attr->la_valid & (LA_UID|LA_GID|LA_PROJID)) {
836                 nodemap = nodemap_get_from_exp(exp);
837                 if (IS_ERR(nodemap))
838                         goto out;
839         }
840         if (attr->la_valid & LA_UID) {
841                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
842                                             NODEMAP_FS_TO_CLIENT,
843                                             attr->la_uid);
844                 b->mbo_valid |= OBD_MD_FLUID;
845         }
846         if (attr->la_valid & LA_GID) {
847                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
848                                             NODEMAP_FS_TO_CLIENT,
849                                             attr->la_gid);
850                 b->mbo_valid |= OBD_MD_FLGID;
851         }
852
853         if (attr->la_valid & LA_PROJID) {
854                 b->mbo_projid = nodemap_map_id(nodemap, NODEMAP_PROJID,
855                                                NODEMAP_FS_TO_CLIENT,
856                                                attr->la_projid);
857                 b->mbo_valid |= OBD_MD_FLPROJID;
858         }
859
860         b->mbo_mode = attr->la_mode;
861         if (attr->la_valid & LA_MODE)
862                 b->mbo_valid |= OBD_MD_FLMODE;
863         if (attr->la_valid & LA_TYPE)
864                 b->mbo_valid |= OBD_MD_FLTYPE;
865
866         if (fid != NULL) {
867                 b->mbo_fid1 = *fid;
868                 b->mbo_valid |= OBD_MD_FLID;
869                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
870                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
871         }
872
873         if (!(attr->la_valid & LA_TYPE))
874                 return;
875
876         b->mbo_rdev   = attr->la_rdev;
877         b->mbo_size   = attr->la_size;
878         b->mbo_blocks = attr->la_blocks;
879
880         if (!S_ISREG(attr->la_mode)) {
881                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
882         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
883                 /* means no objects are allocated on osts. */
884                 LASSERT(!(ma->ma_valid & MA_LOV));
885                 /* just ignore blocks occupied by extend attributes on MDS */
886                 b->mbo_blocks = 0;
887                 /* if no object is allocated on osts, the size on mds is valid.
888                  * b=22272 */
889                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
890         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
891                 if (mdt_hsm_is_released(ma->ma_lmm)) {
892                         /* A released file stores its size on MDS. */
893                         /* But return 1 block for released file, unless tools
894                          * like tar will consider it fully sparse. (LU-3864)
895                          */
896                         if (unlikely(b->mbo_size == 0))
897                                 b->mbo_blocks = 0;
898                         else
899                                 b->mbo_blocks = 1;
900                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
901                 } else if (info->mti_som_strict &&
902                            mdt->mdt_opts.mo_enable_strict_som) {
903                         /* use SOM for size*/
904                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
905                 } else if (ma->ma_valid & MA_SOM) { /* lsom is valid */
906                         b->mbo_valid |= OBD_MD_FLLAZYSIZE | OBD_MD_FLLAZYBLOCKS;
907                         b->mbo_size = ma->ma_som.ms_size;
908                         b->mbo_blocks = ma->ma_som.ms_blocks;
909                 }
910         }
911
912         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE ||
913                             b->mbo_valid & OBD_MD_FLLAZYSIZE))
914                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
915                        PFID(fid), (unsigned long long)b->mbo_size);
916
917 out:
918         if (!IS_ERR_OR_NULL(nodemap))
919                 nodemap_putref(nodemap);
920 }
921
922 static inline int mdt_body_has_lov(const struct lu_attr *la,
923                                    const struct mdt_body *body)
924 {
925         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
926                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
927 }
928
929 void mdt_client_compatibility(struct mdt_thread_info *info)
930 {
931         struct mdt_body       *body;
932         struct ptlrpc_request *req = mdt_info_req(info);
933         struct obd_export     *exp = req->rq_export;
934         struct md_attr        *ma = &info->mti_attr;
935         struct lu_attr        *la = &ma->ma_attr;
936         ENTRY;
937
938         if (exp_connect_layout(exp))
939                 /* the client can deal with 16-bit lmm_stripe_count */
940                 RETURN_EXIT;
941
942         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
943
944         if (!mdt_body_has_lov(la, body))
945                 RETURN_EXIT;
946
947         /* now we have a reply with a lov for a client not compatible with the
948          * layout lock so we have to clean the layout generation number */
949         if (S_ISREG(la->la_mode))
950                 ma->ma_lmm->lmm_layout_gen = 0;
951         EXIT;
952 }
953
954 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
955                                    struct mdt_object *o)
956 {
957         const struct lu_env *env = info->mti_env;
958         int rc, rc2;
959
960         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
961                           XATTR_NAME_LOV);
962
963         if (rc == -ENODATA)
964                 rc = 0;
965
966         if (rc < 0)
967                 goto out;
968
969         /* Is it a directory? Let's check for the LMV as well */
970         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
971                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
972                                    XATTR_NAME_LMV);
973
974                 if (rc2 == -ENODATA)
975                         rc2 = mo_xattr_get(env, mdt_object_child(o),
976                                            &LU_BUF_NULL,
977                                            XATTR_NAME_DEFAULT_LMV);
978
979                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
980                         rc = rc2;
981         }
982
983 out:
984         return rc;
985 }
986
987 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
988                       const char *name)
989 {
990         const struct lu_env *env = info->mti_env;
991         int rc;
992         ENTRY;
993
994         LASSERT(info->mti_big_lmm_used == 0);
995         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
996         if (rc < 0)
997                 RETURN(rc);
998
999         /* big_lmm may need to be grown */
1000         if (info->mti_big_lmmsize < rc) {
1001                 int size = size_roundup_power2(rc);
1002
1003                 if (info->mti_big_lmmsize > 0) {
1004                         /* free old buffer */
1005                         LASSERT(info->mti_big_lmm);
1006                         OBD_FREE_LARGE(info->mti_big_lmm,
1007                                        info->mti_big_lmmsize);
1008                         info->mti_big_lmm = NULL;
1009                         info->mti_big_lmmsize = 0;
1010                 }
1011
1012                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
1013                 if (info->mti_big_lmm == NULL)
1014                         RETURN(-ENOMEM);
1015                 info->mti_big_lmmsize = size;
1016         }
1017         LASSERT(info->mti_big_lmmsize >= rc);
1018
1019         info->mti_buf.lb_buf = info->mti_big_lmm;
1020         info->mti_buf.lb_len = info->mti_big_lmmsize;
1021         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
1022
1023         RETURN(rc);
1024 }
1025
1026 int __mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
1027                      struct md_attr *ma, const char *name)
1028 {
1029         struct md_object *next = mdt_object_child(o);
1030         struct lu_buf    *buf = &info->mti_buf;
1031         int rc;
1032
1033         if (strcmp(name, XATTR_NAME_LOV) == 0) {
1034                 buf->lb_buf = ma->ma_lmm;
1035                 buf->lb_len = ma->ma_lmm_size;
1036                 LASSERT(!(ma->ma_valid & MA_LOV));
1037         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1038                 buf->lb_buf = ma->ma_lmv;
1039                 buf->lb_len = ma->ma_lmv_size;
1040                 LASSERT(!(ma->ma_valid & MA_LMV));
1041         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1042                 buf->lb_buf = ma->ma_default_lmv;
1043                 buf->lb_len = ma->ma_default_lmv_size;
1044                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
1045         } else {
1046                 return -EINVAL;
1047         }
1048
1049         LASSERT(buf->lb_buf);
1050
1051         rc = mo_xattr_get(info->mti_env, next, buf, name);
1052         if (rc > 0) {
1053
1054 got:
1055                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
1056                         if (info->mti_big_lmm_used)
1057                                 ma->ma_lmm = info->mti_big_lmm;
1058
1059                         /* NOT return LOV EA with hole to old client. */
1060                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
1061                                      LOV_PATTERN_F_HOLE) &&
1062                             !(exp_connect_flags(info->mti_exp) &
1063                               OBD_CONNECT_LFSCK)) {
1064                                 return -EIO;
1065                         } else {
1066                                 ma->ma_lmm_size = rc;
1067                                 ma->ma_valid |= MA_LOV;
1068                         }
1069                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1070                         if (info->mti_big_lmm_used)
1071                                 ma->ma_lmv = info->mti_big_lmm;
1072
1073                         ma->ma_lmv_size = rc;
1074                         ma->ma_valid |= MA_LMV;
1075                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1076                         ma->ma_default_lmv_size = rc;
1077                         ma->ma_valid |= MA_LMV_DEF;
1078                 }
1079
1080                 /* Update mdt_max_mdsize so all clients will be aware that */
1081                 if (info->mti_mdt->mdt_max_mdsize < rc)
1082                         info->mti_mdt->mdt_max_mdsize = rc;
1083
1084                 rc = 0;
1085         } else if (rc == -ENODATA) {
1086                 /* no LOV EA */
1087                 rc = 0;
1088         } else if (rc == -ERANGE) {
1089                 /* Default LMV has fixed size, so it must be able to fit
1090                  * in the original buffer */
1091                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
1092                         return rc;
1093                 rc = mdt_big_xattr_get(info, o, name);
1094                 if (rc > 0) {
1095                         info->mti_big_lmm_used = 1;
1096                         goto got;
1097                 }
1098         }
1099
1100         return rc;
1101 }
1102
1103 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
1104                    struct md_attr *ma, const char *name)
1105 {
1106         int rc;
1107
1108         if (!info->mti_big_lmm) {
1109                 OBD_ALLOC(info->mti_big_lmm, PAGE_SIZE);
1110                 if (!info->mti_big_lmm)
1111                         return -ENOMEM;
1112                 info->mti_big_lmmsize = PAGE_SIZE;
1113         }
1114
1115         if (strcmp(name, XATTR_NAME_LOV) == 0) {
1116                 ma->ma_lmm = info->mti_big_lmm;
1117                 ma->ma_lmm_size = info->mti_big_lmmsize;
1118                 ma->ma_valid &= ~MA_LOV;
1119         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1120                 ma->ma_lmv = info->mti_big_lmm;
1121                 ma->ma_lmv_size = info->mti_big_lmmsize;
1122                 ma->ma_valid &= ~MA_LMV;
1123         } else {
1124                 LBUG();
1125         }
1126
1127         LASSERT(!info->mti_big_lmm_used);
1128         rc = __mdt_stripe_get(info, o, ma, name);
1129         /* since big_lmm is always used here, clear 'used' flag to avoid
1130          * assertion in mdt_big_xattr_get().
1131          */
1132         info->mti_big_lmm_used = 0;
1133
1134         return rc;
1135 }
1136
1137 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
1138                       struct lu_fid *pfid)
1139 {
1140         struct lu_buf           *buf = &info->mti_buf;
1141         struct link_ea_header   *leh;
1142         struct link_ea_entry    *lee;
1143         int                      rc;
1144         ENTRY;
1145
1146         buf->lb_buf = info->mti_big_lmm;
1147         buf->lb_len = info->mti_big_lmmsize;
1148         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1149                           buf, XATTR_NAME_LINK);
1150         /* ignore errors, MA_PFID won't be set and it is
1151          * up to the caller to treat this as an error */
1152         if (rc == -ERANGE || buf->lb_len == 0) {
1153                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1154                 buf->lb_buf = info->mti_big_lmm;
1155                 buf->lb_len = info->mti_big_lmmsize;
1156         }
1157
1158         if (rc < 0)
1159                 RETURN(rc);
1160         if (rc < sizeof(*leh)) {
1161                 CERROR("short LinkEA on "DFID": rc = %d\n",
1162                        PFID(mdt_object_fid(o)), rc);
1163                 RETURN(-ENODATA);
1164         }
1165
1166         leh = (struct link_ea_header *) buf->lb_buf;
1167         lee = (struct link_ea_entry *)(leh + 1);
1168         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1169                 leh->leh_magic = LINK_EA_MAGIC;
1170                 leh->leh_reccount = __swab32(leh->leh_reccount);
1171                 leh->leh_len = __swab64(leh->leh_len);
1172         }
1173         if (leh->leh_magic != LINK_EA_MAGIC)
1174                 RETURN(-EINVAL);
1175         if (leh->leh_reccount == 0)
1176                 RETURN(-ENODATA);
1177
1178         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1179         fid_be_to_cpu(pfid, pfid);
1180
1181         RETURN(0);
1182 }
1183
1184 int mdt_attr_get_pfid_name(struct mdt_thread_info *info, struct mdt_object *o,
1185                            struct lu_fid *pfid, struct lu_name *lname)
1186 {
1187         struct lu_buf *buf = &info->mti_buf;
1188         struct link_ea_header *leh;
1189         struct link_ea_entry *lee;
1190         int reclen;
1191         int rc;
1192
1193         buf->lb_buf = info->mti_xattr_buf;
1194         buf->lb_len = sizeof(info->mti_xattr_buf);
1195         rc = mo_xattr_get(info->mti_env, mdt_object_child(o), buf,
1196                           XATTR_NAME_LINK);
1197         if (rc == -ERANGE) {
1198                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1199                 buf->lb_buf = info->mti_big_lmm;
1200                 buf->lb_len = info->mti_big_lmmsize;
1201         }
1202         if (rc < 0)
1203                 return rc;
1204
1205         if (rc < sizeof(*leh)) {
1206                 CERROR("short LinkEA on "DFID": rc = %d\n",
1207                        PFID(mdt_object_fid(o)), rc);
1208                 return -ENODATA;
1209         }
1210
1211         leh = (struct link_ea_header *)buf->lb_buf;
1212         lee = (struct link_ea_entry *)(leh + 1);
1213         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1214                 leh->leh_magic = LINK_EA_MAGIC;
1215                 leh->leh_reccount = __swab32(leh->leh_reccount);
1216                 leh->leh_len = __swab64(leh->leh_len);
1217         }
1218         if (leh->leh_magic != LINK_EA_MAGIC)
1219                 return -EINVAL;
1220
1221         if (leh->leh_reccount == 0)
1222                 return -ENODATA;
1223
1224         linkea_entry_unpack(lee, &reclen, lname, pfid);
1225
1226         return 0;
1227 }
1228
1229 int mdt_attr_get_complex(struct mdt_thread_info *info,
1230                          struct mdt_object *o, struct md_attr *ma)
1231 {
1232         const struct lu_env *env = info->mti_env;
1233         struct md_object    *next = mdt_object_child(o);
1234         struct lu_buf       *buf = &info->mti_buf;
1235         int                  need = ma->ma_need;
1236         int                  rc = 0, rc2;
1237         u32                  mode;
1238         ENTRY;
1239
1240         ma->ma_valid = 0;
1241
1242         if (mdt_object_exists(o) == 0)
1243                 GOTO(out, rc = -ENOENT);
1244         mode = lu_object_attr(&next->mo_lu);
1245
1246         if (need & MA_INODE) {
1247                 ma->ma_need = MA_INODE;
1248                 rc = mo_attr_get(env, next, ma);
1249                 if (rc)
1250                         GOTO(out, rc);
1251
1252                 if (S_ISREG(mode))
1253                         (void) mdt_get_som(info, o, ma);
1254                 ma->ma_valid |= MA_INODE;
1255         }
1256
1257         if (need & MA_PFID) {
1258                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1259                 if (rc == 0)
1260                         ma->ma_valid |= MA_PFID;
1261                 /* ignore this error, parent fid is not mandatory */
1262                 rc = 0;
1263         }
1264
1265         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1266                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1267                 if (rc)
1268                         GOTO(out, rc);
1269         }
1270
1271         if (need & MA_LMV && S_ISDIR(mode)) {
1272                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1273                 if (rc != 0)
1274                         GOTO(out, rc);
1275         }
1276
1277         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1278                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1279                 if (rc != 0)
1280                         GOTO(out, rc);
1281         }
1282
1283         /*
1284          * In the handle of MA_INODE, we may already get the SOM attr.
1285          */
1286         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1287                 rc = mdt_get_som(info, o, ma);
1288                 if (rc != 0)
1289                         GOTO(out, rc);
1290         }
1291
1292         if (need & MA_HSM && S_ISREG(mode)) {
1293                 buf->lb_buf = info->mti_xattr_buf;
1294                 buf->lb_len = sizeof(info->mti_xattr_buf);
1295                 BUILD_BUG_ON(sizeof(struct hsm_attrs) >
1296                              sizeof(info->mti_xattr_buf));
1297                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1298                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1299                 if (rc2 == 0)
1300                         ma->ma_valid |= MA_HSM;
1301                 else if (rc2 < 0 && rc2 != -ENODATA)
1302                         GOTO(out, rc = rc2);
1303         }
1304
1305 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1306         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1307                 buf->lb_buf = ma->ma_acl;
1308                 buf->lb_len = ma->ma_acl_size;
1309                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1310                 if (rc2 > 0) {
1311                         ma->ma_acl_size = rc2;
1312                         ma->ma_valid |= MA_ACL_DEF;
1313                 } else if (rc2 == -ENODATA) {
1314                         /* no ACLs */
1315                         ma->ma_acl_size = 0;
1316                 } else
1317                         GOTO(out, rc = rc2);
1318         }
1319 #endif
1320 out:
1321         ma->ma_need = need;
1322         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1323                rc, ma->ma_valid, ma->ma_lmm);
1324         RETURN(rc);
1325 }
1326
1327 static void mdt_preset_encctx_size(struct mdt_thread_info *info)
1328 {
1329         struct req_capsule *pill = info->mti_pill;
1330
1331         ENTRY;
1332         if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX,
1333                                   RCL_SERVER))
1334                 /* pre-set size in server part with max size */
1335                 req_capsule_set_size(pill, &RMF_FILE_ENCCTX,
1336                                      RCL_SERVER,
1337                                      info->mti_mdt->mdt_max_mdsize);
1338         EXIT;
1339 }
1340
1341 static int mdt_getattr_internal(struct mdt_thread_info *info,
1342                                 struct mdt_object *o, int ma_need)
1343 {
1344         struct mdt_device *mdt = info->mti_mdt;
1345         struct md_object *next = mdt_object_child(o);
1346         const struct mdt_body *reqbody = info->mti_body;
1347         struct ptlrpc_request *req = mdt_info_req(info);
1348         struct md_attr *ma = &info->mti_attr;
1349         struct lu_attr *la = &ma->ma_attr;
1350         struct req_capsule *pill = info->mti_pill;
1351         const struct lu_env *env = info->mti_env;
1352         struct mdt_body *repbody;
1353         struct lu_buf *buffer = &info->mti_buf;
1354         struct obd_export *exp = info->mti_exp;
1355         ktime_t kstart = ktime_get();
1356         int rc;
1357
1358         ENTRY;
1359
1360         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1361                 RETURN(err_serious(-ENOMEM));
1362
1363         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1364
1365         ma->ma_valid = 0;
1366
1367         if (mdt_object_remote(o)) {
1368                 /* This object is located on remote node.*/
1369                 /* Return -ENOTSUPP for old client */
1370                 if (!mdt_is_dne_client(req->rq_export))
1371                         GOTO(out, rc = -ENOTSUPP);
1372
1373                 repbody->mbo_fid1 = *mdt_object_fid(o);
1374                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1375                 GOTO(out, rc = 0);
1376         }
1377
1378         if (reqbody->mbo_eadatasize > 0) {
1379                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1380                 if (buffer->lb_buf == NULL)
1381                         GOTO(out, rc = -EPROTO);
1382                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1383                                                       RCL_SERVER);
1384         } else {
1385                 buffer->lb_buf = NULL;
1386                 buffer->lb_len = 0;
1387                 ma_need &= ~(MA_LOV | MA_LMV);
1388                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1389                        mdt_obd_name(info->mti_mdt),
1390                        req->rq_export->exp_client_uuid.uuid);
1391         }
1392
1393         /* from 2.12.58 intent_getattr pack default LMV in reply */
1394         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1395             ((reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) ==
1396                     (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) &&
1397             req_capsule_has_field(&req->rq_pill, &RMF_DEFAULT_MDT_MD,
1398                                   RCL_SERVER)) {
1399                 ma->ma_lmv = buffer->lb_buf;
1400                 ma->ma_lmv_size = buffer->lb_len;
1401                 ma->ma_default_lmv = req_capsule_server_get(pill,
1402                                                 &RMF_DEFAULT_MDT_MD);
1403                 ma->ma_default_lmv_size = req_capsule_get_size(pill,
1404                                                 &RMF_DEFAULT_MDT_MD,
1405                                                 RCL_SERVER);
1406                 ma->ma_need = MA_INODE;
1407                 if (ma->ma_lmv_size > 0)
1408                         ma->ma_need |= MA_LMV;
1409                 if (ma->ma_default_lmv_size > 0)
1410                         ma->ma_need |= MA_LMV_DEF;
1411         } else if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1412                    (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1413                 /* If it is dir and client require MEA, then we got MEA */
1414                 /* Assumption: MDT_MD size is enough for lmv size. */
1415                 ma->ma_lmv = buffer->lb_buf;
1416                 ma->ma_lmv_size = buffer->lb_len;
1417                 ma->ma_need = MA_INODE;
1418                 if (ma->ma_lmv_size > 0) {
1419                         if (reqbody->mbo_valid & OBD_MD_MEA) {
1420                                 ma->ma_need |= MA_LMV;
1421                         } else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA) {
1422                                 ma->ma_need |= MA_LMV_DEF;
1423                                 ma->ma_default_lmv = buffer->lb_buf;
1424                                 ma->ma_lmv = NULL;
1425                                 ma->ma_default_lmv_size = buffer->lb_len;
1426                                 ma->ma_lmv_size = 0;
1427                         }
1428                 }
1429         } else {
1430                 ma->ma_lmm = buffer->lb_buf;
1431                 ma->ma_lmm_size = buffer->lb_len;
1432                 ma->ma_need = MA_INODE | MA_HSM;
1433                 if (ma->ma_lmm_size > 0) {
1434                         ma->ma_need |= MA_LOV;
1435                         /* Older clients may crash if they getattr overstriped
1436                          * files
1437                          */
1438                         if (!exp_connect_overstriping(exp) &&
1439                             mdt_lmm_is_overstriping(ma->ma_lmm))
1440                                 RETURN(-EOPNOTSUPP);
1441                 }
1442         }
1443
1444         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1445             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1446             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1447                 /* get default stripe info for this dir. */
1448                 ma->ma_need |= MA_LOV_DEF;
1449         }
1450         ma->ma_need |= ma_need;
1451
1452         rc = mdt_attr_get_complex(info, o, ma);
1453         if (unlikely(rc)) {
1454                 CDEBUG_LIMIT(rc == -ENOENT ? D_OTHER : D_ERROR,
1455                              "%s: getattr error for "DFID": rc = %d\n",
1456                              mdt_obd_name(info->mti_mdt),
1457                              PFID(mdt_object_fid(o)), rc);
1458                 RETURN(rc);
1459         }
1460
1461         /* if file is released, check if a restore is running */
1462         if (ma->ma_valid & MA_HSM) {
1463                 repbody->mbo_valid |= OBD_MD_TSTATE;
1464                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1465                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1466                         repbody->mbo_t_state = MS_RESTORE;
1467         }
1468
1469         if (unlikely(!(ma->ma_valid & MA_INODE)))
1470                 RETURN(-EFAULT);
1471
1472         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1473
1474         if (mdt_body_has_lov(la, reqbody)) {
1475                 u32 stripe_count = 1;
1476                 bool fixed_layout = false;
1477
1478                 if (ma->ma_valid & MA_LOV) {
1479                         LASSERT(ma->ma_lmm_size);
1480                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1481                         if (S_ISDIR(la->la_mode))
1482                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1483                         else
1484                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1485                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1486                 }
1487                 if (ma->ma_valid & MA_LMV) {
1488                         struct lmv_mds_md_v1 *lmv = &ma->ma_lmv->lmv_md_v1;
1489                         u32 magic = le32_to_cpu(lmv->lmv_magic);
1490
1491                         /* Return -ENOTSUPP for old client */
1492                         if (!mdt_is_striped_client(req->rq_export))
1493                                 RETURN(-ENOTSUPP);
1494
1495                         LASSERT(S_ISDIR(la->la_mode));
1496                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1497                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1498                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1499
1500                         stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
1501                         fixed_layout = lmv_is_fixed(lmv);
1502                         if (magic == LMV_MAGIC_STRIPE && lmv_is_restriping(lmv))
1503                                 mdt_restripe_migrate_add(info, o);
1504                         else if (magic == LMV_MAGIC_V1 &&
1505                                  lmv_is_restriping(lmv))
1506                                 mdt_restripe_update_add(info, o);
1507                 }
1508                 if (ma->ma_valid & MA_LMV_DEF) {
1509                         /* Return -ENOTSUPP for old client */
1510                         if (!mdt_is_striped_client(req->rq_export))
1511                                 RETURN(-ENOTSUPP);
1512                         LASSERT(S_ISDIR(la->la_mode));
1513                         /*
1514                          * when ll_dir_getstripe() gets default LMV, it
1515                          * checks mbo_eadatasize.
1516                          */
1517                         if (!(ma->ma_valid & MA_LMV))
1518                                 repbody->mbo_eadatasize =
1519                                         ma->ma_default_lmv_size;
1520                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1521                                                OBD_MD_DEFAULT_MEA);
1522                 }
1523                 CDEBUG(D_VFSTRACE,
1524                        "dirent count %llu stripe count %u MDT count %d\n",
1525                        ma->ma_attr.la_dirent_count, stripe_count,
1526                        atomic_read(&mdt->mdt_mds_mds_conns) + 1);
1527                 if (ma->ma_attr.la_dirent_count != LU_DIRENT_COUNT_UNSET &&
1528                     ma->ma_attr.la_dirent_count >
1529                         mdt->mdt_restriper.mdr_dir_split_count &&
1530                     !fid_is_root(mdt_object_fid(o)) &&
1531                     mdt->mdt_enable_dir_auto_split &&
1532                     !o->mot_restriping &&
1533                     stripe_count < atomic_read(&mdt->mdt_mds_mds_conns) + 1 &&
1534                     !fixed_layout)
1535                         mdt_auto_split_add(info, o);
1536         } else if (S_ISLNK(la->la_mode) &&
1537                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1538                 buffer->lb_buf = ma->ma_lmm;
1539                 /* eadatasize from client includes NULL-terminator, so
1540                  * there is no need to read it */
1541                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1542                 rc = mo_readlink(env, next, buffer);
1543                 if (unlikely(rc <= 0)) {
1544                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1545                                mdt_obd_name(info->mti_mdt),
1546                                PFID(mdt_object_fid(o)), rc);
1547                         rc = -EFAULT;
1548                 } else {
1549                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1550
1551                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1552                                 rc -= 2;
1553                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1554                         /* we need to report back size with NULL-terminator
1555                          * because client expects that */
1556                         repbody->mbo_eadatasize = rc + 1;
1557                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1558                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1559                                        "on "DFID ", expected %d\n",
1560                                        mdt_obd_name(info->mti_mdt),
1561                                        rc, PFID(mdt_object_fid(o)),
1562                                        reqbody->mbo_eadatasize - 1);
1563                         /* NULL terminate */
1564                         ((char *)ma->ma_lmm)[rc] = 0;
1565
1566                         /* If the total CDEBUG() size is larger than a page, it
1567                          * will print a warning to the console, avoid this by
1568                          * printing just the last part of the symlink. */
1569                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1570                                print_limit < rc ? "..." : "", print_limit,
1571                                (char *)ma->ma_lmm + rc - print_limit, rc);
1572                         rc = 0;
1573                 }
1574         }
1575
1576         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1577                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1578                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1579                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1580                        repbody->mbo_max_mdsize);
1581         }
1582
1583 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1584         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1585                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1586                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1587                 if (IS_ERR(nodemap))
1588                         RETURN(PTR_ERR(nodemap));
1589
1590                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1591                 nodemap_putref(nodemap);
1592         }
1593 #endif
1594
1595 out:
1596         if (rc == 0)
1597                 mdt_counter_incr(req, LPROC_MDT_GETATTR,
1598                                  ktime_us_delta(ktime_get(), kstart));
1599
1600         RETURN(rc);
1601 }
1602
1603 static int mdt_getattr(struct tgt_session_info *tsi)
1604 {
1605         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1606         struct mdt_object       *obj = info->mti_object;
1607         struct req_capsule      *pill = info->mti_pill;
1608         struct mdt_body         *reqbody;
1609         struct mdt_body         *repbody;
1610         int rc, rc2;
1611         ENTRY;
1612
1613         if (unlikely(info->mti_object == NULL))
1614                 RETURN(-EPROTO);
1615
1616         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1617         LASSERT(reqbody);
1618         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1619
1620         /* Special case for Data-on-MDT files to get data version */
1621         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1622                 rc = mdt_data_version_get(tsi);
1623                 GOTO(out, rc);
1624         }
1625
1626         /* Unlike intent case where we need to pre-fill out buffers early on
1627          * in intent policy for ldlm reasons, here we can have a much better
1628          * guess at EA size by just reading it from disk.
1629          * Exceptions are readdir and (missing) directory striping */
1630         /* Readlink */
1631         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1632                 /* No easy way to know how long is the symlink, but it cannot
1633                  * be more than PATH_MAX, so we allocate +1 */
1634                 rc = PATH_MAX + 1;
1635         /* A special case for fs ROOT: getattr there might fetch
1636          * default EA for entire fs, not just for this dir!
1637          */
1638         } else if (lu_fid_eq(mdt_object_fid(obj),
1639                              &info->mti_mdt->mdt_md_root_fid) &&
1640                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1641                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1642                                                                  MDS_GETATTR)) {
1643                 /* Should the default strping be bigger, mdt_fix_reply
1644                  * will reallocate */
1645                 rc = DEF_REP_MD_SIZE;
1646         } else {
1647                 /* Read the actual EA size from disk */
1648                 rc = mdt_attr_get_eabuf_size(info, obj);
1649         }
1650
1651         if (rc < 0)
1652                 GOTO(out, rc = err_serious(rc));
1653
1654         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1655
1656         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1657          * by default. If the target object has more ACL entries, then
1658          * enlarge the buffer when necessary. */
1659         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1660                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1661         mdt_preset_encctx_size(info);
1662
1663         rc = req_capsule_server_pack(pill);
1664         if (unlikely(rc != 0))
1665                 GOTO(out, rc = err_serious(rc));
1666
1667         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1668         LASSERT(repbody != NULL);
1669         repbody->mbo_eadatasize = 0;
1670         repbody->mbo_aclsize = 0;
1671
1672         rc = mdt_check_ucred(info);
1673         if (unlikely(rc))
1674                 GOTO(out_shrink, rc);
1675
1676         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1677
1678         rc = mdt_getattr_internal(info, obj, 0);
1679         if (unlikely(rc))
1680                 GOTO(out_shrink, rc);
1681
1682         rc = mdt_pack_encctx_in_reply(info, obj);
1683         EXIT;
1684 out_shrink:
1685         mdt_client_compatibility(info);
1686         rc2 = mdt_fix_reply(info);
1687         if (rc == 0)
1688                 rc = rc2;
1689 out:
1690         mdt_thread_info_fini(info);
1691         return rc;
1692 }
1693
1694 /**
1695  * Handler of layout intent RPC requiring the layout modification
1696  *
1697  * \param[in]  info     thread environment
1698  * \param[in]  obj      object
1699  * \param[out] lhc      object ldlm lock handle
1700  * \param[in]  layout   layout change descriptor
1701  *
1702  * \retval 0    on success
1703  * \retval < 0  error code
1704  */
1705 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1706                       struct mdt_lock_handle *lhc,
1707                       struct md_layout_change *layout)
1708 {
1709         int rc;
1710
1711         ENTRY;
1712
1713         if (!mdt_object_exists(obj))
1714                 RETURN(-ENOENT);
1715
1716         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1717                 RETURN(-EINVAL);
1718
1719         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1720                            MAY_WRITE);
1721         if (rc)
1722                 RETURN(rc);
1723
1724         rc = mdt_check_resent_lock(info, obj, lhc);
1725         if (rc < 0)
1726                 RETURN(rc);
1727
1728         if (rc > 0) {
1729                 /* not resent */
1730                 __u64 lockpart = MDS_INODELOCK_LAYOUT;
1731
1732                 /* take layout lock to prepare layout change */
1733                 if (layout->mlc_opc == MD_LAYOUT_WRITE)
1734                         lockpart |= MDS_INODELOCK_UPDATE;
1735
1736                 mdt_lock_handle_init(lhc);
1737                 mdt_lock_reg_init(lhc, LCK_EX);
1738                 rc = mdt_reint_object_lock(info, obj, lhc, lockpart, false);
1739                 if (rc)
1740                         RETURN(rc);
1741         }
1742
1743         mutex_lock(&obj->mot_som_mutex);
1744         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1745         mutex_unlock(&obj->mot_som_mutex);
1746
1747         if (rc)
1748                 mdt_object_unlock(info, obj, lhc, 1);
1749
1750         RETURN(rc);
1751 }
1752
1753 /**
1754  * Exchange MOF_LOV_CREATED flags between two objects after a
1755  * layout swap. No assumption is made on whether o1 or o2 have
1756  * created objects or not.
1757  *
1758  * \param[in,out] o1    First swap layout object
1759  * \param[in,out] o2    Second swap layout object
1760  */
1761 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1762 {
1763         unsigned int o1_lov_created = o1->mot_lov_created;
1764
1765         mutex_lock(&o1->mot_lov_mutex);
1766         mutex_lock(&o2->mot_lov_mutex);
1767
1768         o1->mot_lov_created = o2->mot_lov_created;
1769         o2->mot_lov_created = o1_lov_created;
1770
1771         mutex_unlock(&o2->mot_lov_mutex);
1772         mutex_unlock(&o1->mot_lov_mutex);
1773 }
1774
1775 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1776 {
1777         struct mdt_thread_info  *info;
1778         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1779         struct obd_export       *exp = req->rq_export;
1780         struct mdt_object       *o1, *o2, *o;
1781         struct mdt_lock_handle  *lh1, *lh2;
1782         struct mdc_swap_layouts *msl;
1783         int                      rc;
1784         ENTRY;
1785
1786         /* client does not support layout lock, so layout swaping
1787          * is disabled.
1788          * FIXME: there is a problem for old clients which don't support
1789          * layout lock yet. If those clients have already opened the file
1790          * they won't be notified at all so that old layout may still be
1791          * used to do IO. This can be fixed after file release is landed by
1792          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1793         if (!exp_connect_layout(exp))
1794                 RETURN(-EOPNOTSUPP);
1795
1796         info = tsi2mdt_info(tsi);
1797         if (unlikely(info->mti_object == NULL))
1798                 RETURN(-EPROTO);
1799
1800         if (info->mti_dlm_req != NULL)
1801                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1802
1803         o1 = info->mti_object;
1804         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1805                                 &info->mti_body->mbo_fid2);
1806         if (IS_ERR(o))
1807                 GOTO(out, rc = PTR_ERR(o));
1808
1809         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1810                 GOTO(put, rc = -ENOENT);
1811
1812         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1813         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1814                 GOTO(put, rc);
1815
1816         if (rc < 0)
1817                 swap(o1, o2);
1818
1819         /* permission check. Make sure the calling process having permission
1820          * to write both files. */
1821         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1822                            MAY_WRITE);
1823         if (rc < 0)
1824                 GOTO(put, rc);
1825
1826         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1827                            MAY_WRITE);
1828         if (rc < 0)
1829                 GOTO(put, rc);
1830
1831         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1832         if (msl == NULL)
1833                 GOTO(put, rc = -EPROTO);
1834
1835         lh1 = &info->mti_lh[MDT_LH_NEW];
1836         mdt_lock_reg_init(lh1, LCK_EX);
1837         lh2 = &info->mti_lh[MDT_LH_OLD];
1838         mdt_lock_reg_init(lh2, LCK_EX);
1839
1840         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1841                              MDS_INODELOCK_XATTR);
1842         if (rc < 0)
1843                 GOTO(put, rc);
1844
1845         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1846                              MDS_INODELOCK_XATTR);
1847         if (rc < 0)
1848                 GOTO(unlock1, rc);
1849
1850         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1851                              mdt_object_child(o2), msl->msl_flags);
1852         if (rc < 0)
1853                 GOTO(unlock2, rc);
1854
1855         mdt_swap_lov_flag(o1, o2);
1856
1857 unlock2:
1858         mdt_object_unlock(info, o2, lh2, rc);
1859 unlock1:
1860         mdt_object_unlock(info, o1, lh1, rc);
1861 put:
1862         mdt_object_put(info->mti_env, o);
1863 out:
1864         mdt_thread_info_fini(info);
1865         RETURN(rc);
1866 }
1867
1868 static int mdt_raw_lookup(struct mdt_thread_info *info,
1869                           struct mdt_object *parent,
1870                           const struct lu_name *lname)
1871 {
1872         struct lu_fid *fid = &info->mti_tmp_fid1;
1873         struct mdt_body *repbody;
1874         bool is_dotdot = false;
1875         bool is_old_parent_stripe = false;
1876         bool is_new_parent_checked = false;
1877         int rc;
1878
1879         ENTRY;
1880
1881         LASSERT(!info->mti_cross_ref);
1882         /* Always allow to lookup ".." */
1883         if (lname->ln_namelen == 2 &&
1884             lname->ln_name[0] == '.' && lname->ln_name[1] == '.') {
1885                 info->mti_spec.sp_permitted = 1;
1886                 is_dotdot = true;
1887                 if (mdt_is_dir_stripe(info, parent) == 1)
1888                         is_old_parent_stripe = true;
1889         }
1890
1891         mdt_object_get(info->mti_env, parent);
1892 lookup:
1893         /* Only got the fid of this obj by name */
1894         fid_zero(fid);
1895         rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname, fid,
1896                         &info->mti_spec);
1897         mdt_object_put(info->mti_env, parent);
1898         if (rc)
1899                 RETURN(rc);
1900
1901         /* getattr_name("..") should return master object FID for striped dir */
1902         if (is_dotdot && (is_old_parent_stripe || !is_new_parent_checked)) {
1903                 parent = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1904                 if (IS_ERR(parent))
1905                         RETURN(PTR_ERR(parent));
1906
1907                 /* old client getattr_name("..") with stripe FID */
1908                 if (unlikely(is_old_parent_stripe)) {
1909                         is_old_parent_stripe = false;
1910                         goto lookup;
1911                 }
1912
1913                 /* ".." may be a stripe */
1914                 if (unlikely(mdt_is_dir_stripe(info, parent) == 1)) {
1915                         is_new_parent_checked = true;
1916                         goto lookup;
1917                 }
1918
1919                 mdt_object_put(info->mti_env, parent);
1920         }
1921
1922         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1923         repbody->mbo_fid1 = *fid;
1924         repbody->mbo_valid = OBD_MD_FLID;
1925
1926         RETURN(rc);
1927 }
1928
1929 /**
1930  * Find name matching hash
1931  *
1932  * We search \a child LinkEA for a name whose hash matches \a lname
1933  * (it contains an encoded hash).
1934  *
1935  * \param info mdt thread info
1936  * \param lname encoded hash to find
1937  * \param parent parent object
1938  * \param child object to search with LinkEA
1939  *
1940  * \retval 1 match found
1941  * \retval 0 no match found
1942  * \retval -ev negative errno upon error
1943  */
1944 int find_name_matching_hash(struct mdt_thread_info *info, struct lu_name *lname,
1945                             struct mdt_object *parent, struct mdt_object *child)
1946 {
1947         /* Here, lname is an encoded hash of on-disk name, and
1948          * client is doing access without encryption key.
1949          * So we need to get LinkEA, check parent fid is correct and
1950          * compare name hash with the one in the request.
1951          */
1952         struct lu_buf *buf = &info->mti_big_buf;
1953         struct lu_name name;
1954         struct lu_fid pfid;
1955         struct linkea_data ldata = { NULL };
1956         struct link_ea_header *leh;
1957         struct link_ea_entry *lee;
1958         struct lu_buf link = { 0 };
1959         char *hash;
1960         int reclen, count, rc;
1961
1962         ENTRY;
1963         if (lname->ln_namelen < LL_CRYPTO_BLOCK_SIZE)
1964                 RETURN(-EINVAL);
1965
1966         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
1967         if (!buf->lb_buf)
1968                 RETURN(-ENOMEM);
1969
1970         ldata.ld_buf = buf;
1971         rc = mdt_links_read(info, child, &ldata);
1972         if (rc < 0)
1973                 RETURN(rc);
1974
1975         hash = kmalloc(lname->ln_namelen, GFP_NOFS);
1976         if (!hash)
1977                 RETURN(-ENOMEM);
1978         rc = critical_decode(lname->ln_name, lname->ln_namelen, hash);
1979
1980         leh = buf->lb_buf;
1981         lee = (struct link_ea_entry *)(leh + 1);
1982         for (count = 0; count < leh->leh_reccount; count++) {
1983                 linkea_entry_unpack(lee, &reclen, &name, &pfid);
1984                 if (!parent || lu_fid_eq(&pfid, mdt_object_fid(parent))) {
1985                         lu_buf_check_and_alloc(&link, name.ln_namelen);
1986                         if (!link.lb_buf)
1987                                 GOTO(out_match, rc = -ENOMEM);
1988                         rc = critical_decode(name.ln_name, name.ln_namelen,
1989                                              link.lb_buf);
1990
1991                         if (memcmp(LLCRYPT_EXTRACT_DIGEST(link.lb_buf, rc),
1992                                    hash, LL_CRYPTO_BLOCK_SIZE) == 0) {
1993                                 *lname = name;
1994                                 break;
1995                         }
1996                 }
1997                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
1998         }
1999         if (count == leh->leh_reccount)
2000                 rc = 0;
2001         else
2002                 rc = 1;
2003
2004 out_match:
2005         lu_buf_free(&link);
2006         kfree(hash);
2007
2008         RETURN(rc);
2009 }
2010
2011 /*
2012  * UPDATE lock should be taken against parent, and be released before exit;
2013  * child_bits lock should be taken against child, and be returned back:
2014  *            (1)normal request should release the child lock;
2015  *            (2)intent request will grant the lock to client.
2016  */
2017 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
2018                                  struct mdt_lock_handle *lhc,
2019                                  __u64 child_bits,
2020                                  struct ldlm_reply *ldlm_rep)
2021 {
2022         struct ptlrpc_request *req = mdt_info_req(info);
2023         struct mdt_body *reqbody = NULL;
2024         struct mdt_object *parent = info->mti_object;
2025         struct mdt_object *child = NULL;
2026         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2027         struct lu_name *lname = NULL;
2028         struct mdt_lock_handle *lhp = NULL;
2029         struct ldlm_lock *lock;
2030         struct req_capsule *pill = info->mti_pill;
2031         __u64 try_bits = 0;
2032         bool is_resent;
2033         int ma_need = 0;
2034         int rc;
2035
2036         ENTRY;
2037
2038         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
2039         LASSERT(ergo(is_resent,
2040                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
2041
2042         if (parent == NULL)
2043                 RETURN(-ENOENT);
2044
2045         if (info->mti_cross_ref) {
2046                 /* Only getattr on the child. Parent is on another node. */
2047                 mdt_set_disposition(info, ldlm_rep,
2048                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
2049                 child = parent;
2050                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
2051                        "ldlm_rep = %p\n",
2052                        PFID(mdt_object_fid(child)), ldlm_rep);
2053
2054                 rc = mdt_check_resent_lock(info, child, lhc);
2055                 if (rc < 0) {
2056                         RETURN(rc);
2057                 } else if (rc > 0) {
2058                         mdt_lock_handle_init(lhc);
2059                         mdt_lock_reg_init(lhc, LCK_PR);
2060
2061                         /*
2062                          * Object's name entry is on another MDS, it will
2063                          * request PERM lock only because LOOKUP lock is owned
2064                          * by the MDS where name entry resides.
2065                          *
2066                          * TODO: it should try layout lock too. - Jinshan
2067                          */
2068                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
2069                                         MDS_INODELOCK_LAYOUT);
2070                         child_bits |= MDS_INODELOCK_PERM;
2071
2072                         rc = mdt_object_lock(info, child, lhc, child_bits);
2073                         if (rc < 0)
2074                                 RETURN(rc);
2075                 }
2076
2077                 /* Finally, we can get attr for child. */
2078                 if (!mdt_object_exists(child)) {
2079                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
2080                                         &child->mot_obj,
2081                                         "remote object doesn't exist.");
2082                         mdt_object_unlock(info, child, lhc, 1);
2083                         RETURN(-ENOENT);
2084                 }
2085
2086                 rc = mdt_getattr_internal(info, child, 0);
2087                 if (unlikely(rc != 0)) {
2088                         mdt_object_unlock(info, child, lhc, 1);
2089                         RETURN(rc);
2090                 }
2091
2092                 rc = mdt_pack_secctx_in_reply(info, child);
2093                 if (unlikely(rc)) {
2094                         mdt_object_unlock(info, child, lhc, 1);
2095                         RETURN(rc);
2096                 }
2097
2098                 rc = mdt_pack_encctx_in_reply(info, child);
2099                 if (unlikely(rc))
2100                         mdt_object_unlock(info, child, lhc, 1);
2101                 RETURN(rc);
2102         }
2103
2104         lname = &info->mti_name;
2105         mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
2106
2107         if (info->mti_body->mbo_valid & OBD_MD_NAMEHASH) {
2108                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
2109                 if (unlikely(reqbody == NULL))
2110                         RETURN(err_serious(-EPROTO));
2111
2112                 *child_fid = reqbody->mbo_fid2;
2113                 if (unlikely(!fid_is_sane(child_fid)))
2114                         RETURN(err_serious(-EINVAL));
2115
2116                 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
2117                         mdt_object_get(info->mti_env, parent);
2118                         child = parent;
2119                 } else {
2120                         child = mdt_object_find(info->mti_env, info->mti_mdt,
2121                                                 child_fid);
2122                         if (IS_ERR(child))
2123                                 RETURN(PTR_ERR(child));
2124                 }
2125
2126                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
2127                        "ldlm_rep = %p\n",
2128                        PFID(mdt_object_fid(parent)),
2129                        PFID(&reqbody->mbo_fid2), ldlm_rep);
2130         } else if (lu_name_is_valid(lname)) {
2131                 if (mdt_object_remote(parent)) {
2132                         CERROR("%s: parent "DFID" is on remote target\n",
2133                                mdt_obd_name(info->mti_mdt),
2134                                PFID(mdt_object_fid(parent)));
2135                         RETURN(-EPROTO);
2136                 }
2137
2138                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
2139                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
2140                        PNAME(lname), ldlm_rep);
2141         } else {
2142                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
2143                 if (unlikely(reqbody == NULL))
2144                         RETURN(err_serious(-EPROTO));
2145
2146                 *child_fid = reqbody->mbo_fid2;
2147                 if (unlikely(!fid_is_sane(child_fid)))
2148                         RETURN(err_serious(-EINVAL));
2149
2150                 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
2151                         mdt_object_get(info->mti_env, parent);
2152                         child = parent;
2153                 } else {
2154                         child = mdt_object_find(info->mti_env, info->mti_mdt,
2155                                                 child_fid);
2156                         if (IS_ERR(child))
2157                                 RETURN(PTR_ERR(child));
2158                 }
2159
2160                 if (mdt_object_remote(child)) {
2161                         CERROR("%s: child "DFID" is on remote target\n",
2162                                mdt_obd_name(info->mti_mdt),
2163                                PFID(mdt_object_fid(child)));
2164                         GOTO(out_child, rc = -EPROTO);
2165                 }
2166
2167                 /* don't fetch LOOKUP lock if it's remote object */
2168                 rc = mdt_is_remote_object(info, parent, child);
2169                 if (rc < 0)
2170                         GOTO(out_child, rc);
2171                 if (rc)
2172                         child_bits &= ~MDS_INODELOCK_LOOKUP;
2173
2174                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
2175                        "ldlm_rep = %p\n",
2176                        PFID(mdt_object_fid(parent)),
2177                        PFID(&reqbody->mbo_fid2), ldlm_rep);
2178         }
2179
2180         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
2181
2182         if (unlikely(!mdt_object_exists(parent)) &&
2183             !(info->mti_body->mbo_valid & OBD_MD_NAMEHASH) &&
2184             lu_name_is_valid(lname)) {
2185                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2186                                 &parent->mot_obj,
2187                                 "Parent doesn't exist!");
2188                 GOTO(out_child, rc = -ESTALE);
2189         }
2190
2191         if (!child && is_resent) {
2192                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2193                 if (lock == NULL) {
2194                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2195                          * a resend case, however, it could be already destroyed
2196                          * due to client eviction or a raced cancel RPC.
2197                          */
2198                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2199                                           lhc->mlh_reg_lh.cookie);
2200                         RETURN(-ESTALE);
2201                 }
2202                 fid_extract_from_res_name(child_fid,
2203                                           &lock->l_resource->lr_name);
2204                 LDLM_LOCK_PUT(lock);
2205                 child = mdt_object_find(info->mti_env, info->mti_mdt,
2206                                         child_fid);
2207                 if (IS_ERR(child))
2208                         RETURN(PTR_ERR(child));
2209         } else if (!(info->mti_body->mbo_valid & OBD_MD_NAMEHASH) &&
2210             lu_name_is_valid(lname)) {
2211                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
2212                         rc = mdt_raw_lookup(info, parent, lname);
2213
2214                         RETURN(rc);
2215                 }
2216
2217                 /* step 1: lock parent only if parent is a directory */
2218                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
2219                         lhp = &info->mti_lh[MDT_LH_PARENT];
2220                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
2221                         rc = mdt_object_lock(info, parent, lhp,
2222                                              MDS_INODELOCK_UPDATE);
2223                         if (unlikely(rc != 0))
2224                                 RETURN(rc);
2225                 }
2226
2227                 /* step 2: lookup child's fid by name */
2228                 fid_zero(child_fid);
2229                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
2230                                 child_fid, &info->mti_spec);
2231                 if (rc == -ENOENT)
2232                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
2233
2234                 if (rc != 0)
2235                         GOTO(unlock_parent, rc);
2236
2237                 child = mdt_object_find(info->mti_env, info->mti_mdt,
2238                                         child_fid);
2239                 if (unlikely(IS_ERR(child)))
2240                         GOTO(unlock_parent, rc = PTR_ERR(child));
2241         }
2242
2243         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
2244
2245         /* step 3: lock child regardless if it is local or remote. */
2246         LASSERT(child);
2247
2248         if (info->mti_body->mbo_valid & OBD_MD_NAMEHASH) {
2249                 /* Here, lname is an encoded hash of on-disk name, and
2250                  * client is doing access without encryption key.
2251                  * So we need to compare name hash with the one in the request.
2252                  */
2253                 if (!find_name_matching_hash(info, lname, parent,
2254                                              child)) {
2255                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
2256                         mdt_clear_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
2257                         GOTO(out_child, rc = -ENOENT);
2258                 }
2259         }
2260
2261         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
2262         if (!mdt_object_exists(child)) {
2263                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2264                                 &child->mot_obj,
2265                                 "Object doesn't exist!");
2266                 GOTO(out_child, rc = -ENOENT);
2267         }
2268
2269         rc = mdt_check_resent_lock(info, child, lhc);
2270         if (rc < 0) {
2271                 GOTO(out_child, rc);
2272         } else if (rc > 0) {
2273                 mdt_lock_handle_init(lhc);
2274                 mdt_lock_reg_init(lhc, LCK_PR);
2275
2276                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
2277                     !mdt_object_remote(child)) {
2278                         struct md_attr *ma = &info->mti_attr;
2279
2280                         ma->ma_valid = 0;
2281                         ma->ma_need = MA_INODE;
2282                         rc = mdt_attr_get_complex(info, child, ma);
2283                         if (unlikely(rc != 0))
2284                                 GOTO(out_child, rc);
2285
2286                         /* If the file has not been changed for some time, we
2287                          * return not only a LOOKUP lock, but also an UPDATE
2288                          * lock and this might save us RPC on later STAT. For
2289                          * directories, it also let negative dentry cache start
2290                          * working for this dir. */
2291                         if (ma->ma_valid & MA_INODE &&
2292                             ma->ma_attr.la_valid & LA_CTIME &&
2293                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
2294                             ma->ma_attr.la_ctime < ktime_get_real_seconds())
2295                                 child_bits |= MDS_INODELOCK_UPDATE;
2296                 }
2297
2298                 /* layout lock must be granted in a best-effort way
2299                  * for IT operations */
2300                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
2301                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2302                     !mdt_object_remote(child) && ldlm_rep != NULL) {
2303                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
2304                             exp_connect_layout(info->mti_exp)) {
2305                                 /* try to grant layout lock for regular file. */
2306                                 try_bits = MDS_INODELOCK_LAYOUT;
2307                         }
2308                         /* Acquire DOM lock in advance for data-on-mdt file */
2309                         if (child != parent)
2310                                 try_bits |= MDS_INODELOCK_DOM;
2311                 }
2312
2313                 if (try_bits != 0) {
2314                         /* try layout lock, it may fail to be granted due to
2315                          * contention at LOOKUP or UPDATE */
2316                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
2317                                                  try_bits, false);
2318                         if (child_bits & MDS_INODELOCK_LAYOUT)
2319                                 ma_need |= MA_LOV;
2320                 } else {
2321                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
2322                          * client will enqueue the lock to the remote MDT */
2323                         if (mdt_object_remote(child))
2324                                 child_bits &= ~MDS_INODELOCK_UPDATE;
2325                         rc = mdt_object_lock(info, child, lhc, child_bits);
2326                 }
2327                 if (unlikely(rc != 0))
2328                         GOTO(out_child, rc);
2329         }
2330
2331         /* finally, we can get attr for child. */
2332         rc = mdt_getattr_internal(info, child, ma_need);
2333         if (unlikely(rc != 0)) {
2334                 if (!is_resent)
2335                         mdt_object_unlock(info, child, lhc, 1);
2336                 GOTO(out_child, rc);
2337         }
2338
2339         rc = mdt_pack_secctx_in_reply(info, child);
2340         if (unlikely(rc)) {
2341                 if (!is_resent)
2342                         mdt_object_unlock(info, child, lhc, 1);
2343                 GOTO(out_child, rc);
2344         }
2345
2346         rc = mdt_pack_encctx_in_reply(info, child);
2347         if (unlikely(rc)) {
2348                 if (!is_resent)
2349                         mdt_object_unlock(info, child, lhc, 1);
2350                 GOTO(out_child, rc);
2351         }
2352
2353         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2354         if (lock) {
2355                 /* Debugging code. */
2356                 LDLM_DEBUG(lock, "Returning lock to client");
2357                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
2358                                          &lock->l_resource->lr_name),
2359                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
2360                          PLDLMRES(lock->l_resource),
2361                          PFID(mdt_object_fid(child)));
2362
2363                 if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_PTLRPC_ENQ_RESEND))) {
2364                         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2365                                 OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_ENQ_RESEND,
2366                                                  req->rq_deadline -
2367                                                  req->rq_arrival_time.tv_sec +
2368                                                  cfs_fail_val ?: 3);
2369                         /* Put the lock to the waiting list and force the cancel */
2370                         ldlm_set_ast_sent(lock);
2371                 }
2372
2373                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2374                     !mdt_object_remote(child) && child != parent) {
2375                         mdt_object_put(info->mti_env, child);
2376                         rc = mdt_pack_size2body(info, child_fid,
2377                                                 &lhc->mlh_reg_lh);
2378                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
2379                                 /* DOM lock was taken in advance but this is
2380                                  * not DoM file. Drop the lock.
2381                                  */
2382                                 lock_res_and_lock(lock);
2383                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
2384                                 unlock_res_and_lock(lock);
2385                         }
2386                         LDLM_LOCK_PUT(lock);
2387                         GOTO(unlock_parent, rc = 0);
2388                 }
2389                 LDLM_LOCK_PUT(lock);
2390         }
2391
2392         EXIT;
2393 out_child:
2394         if (child)
2395                 mdt_object_put(info->mti_env, child);
2396 unlock_parent:
2397         if (lhp)
2398                 mdt_object_unlock(info, parent, lhp, 1);
2399         if (rc == -ENOENT) {
2400                 /* return -ENOKEY instead of -ENOENT to encryption-unaware
2401                  * client if trying to access an encrypted file
2402                  */
2403                 int rc2 = mdt_check_enc(info, parent);
2404
2405                 if (rc2)
2406                         rc = rc2;
2407         }
2408         return rc;
2409 }
2410
2411 /* normal handler: should release the child lock */
2412 static int mdt_getattr_name(struct tgt_session_info *tsi)
2413 {
2414         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
2415         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
2416         struct mdt_body *reqbody;
2417         struct mdt_body *repbody;
2418         int rc, rc2;
2419
2420         ENTRY;
2421
2422         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2423         LASSERT(reqbody != NULL);
2424         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2425         LASSERT(repbody != NULL);
2426
2427         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
2428         repbody->mbo_eadatasize = 0;
2429         repbody->mbo_aclsize = 0;
2430
2431         rc = mdt_init_ucred(info, reqbody);
2432         if (unlikely(rc))
2433                 GOTO(out_shrink, rc);
2434
2435         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
2436         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
2437                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
2438                 lhc->mlh_reg_lh.cookie = 0;
2439         }
2440         mdt_exit_ucred(info);
2441         EXIT;
2442 out_shrink:
2443         mdt_client_compatibility(info);
2444         rc2 = mdt_fix_reply(info);
2445         if (rc == 0)
2446                 rc = rc2;
2447         mdt_thread_info_fini(info);
2448         return rc;
2449 }
2450
2451 static int mdt_rmfid_unlink(struct mdt_thread_info *info,
2452                             const struct lu_fid *pfid,
2453                             const struct lu_name *name,
2454                             struct mdt_object *obj, s64 ctime)
2455 {
2456         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2457         struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
2458         struct mdt_device *mdt = info->mti_mdt;
2459         struct md_attr *ma = &info->mti_attr;
2460         struct mdt_lock_handle *parent_lh;
2461         struct mdt_lock_handle *child_lh;
2462         struct mdt_object *pobj;
2463         bool cos_incompat = false;
2464         int rc;
2465         ENTRY;
2466
2467         pobj = mdt_object_find(info->mti_env, mdt, pfid);
2468         if (IS_ERR(pobj))
2469                 GOTO(out, rc = PTR_ERR(pobj));
2470
2471         parent_lh = &info->mti_lh[MDT_LH_PARENT];
2472         mdt_lock_pdo_init(parent_lh, LCK_PW, name);
2473         rc = mdt_object_lock(info, pobj, parent_lh, MDS_INODELOCK_UPDATE);
2474         if (rc != 0)
2475                 GOTO(put_parent, rc);
2476
2477         if (mdt_object_remote(pobj))
2478                 cos_incompat = true;
2479
2480         rc = mdo_lookup(info->mti_env, mdt_object_child(pobj),
2481                         name, child_fid, &info->mti_spec);
2482         if (rc != 0)
2483                 GOTO(unlock_parent, rc);
2484
2485         if (!lu_fid_eq(child_fid, mdt_object_fid(obj)))
2486                 GOTO(unlock_parent, rc = -EREMCHG);
2487
2488         child_lh = &info->mti_lh[MDT_LH_CHILD];
2489         mdt_lock_reg_init(child_lh, LCK_EX);
2490         rc = mdt_reint_striped_lock(info, obj, child_lh,
2491                                     MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
2492                                     einfo, cos_incompat);
2493         if (rc != 0)
2494                 GOTO(unlock_parent, rc);
2495
2496         if (atomic_read(&obj->mot_open_count)) {
2497                 CDEBUG(D_OTHER, "object "DFID" open, skip\n",
2498                        PFID(mdt_object_fid(obj)));
2499                 GOTO(unlock_child, rc = -EBUSY);
2500         }
2501
2502         ma->ma_need = 0;
2503         ma->ma_valid = MA_INODE;
2504         ma->ma_attr.la_valid = LA_CTIME;
2505         ma->ma_attr.la_ctime = ctime;
2506
2507         mutex_lock(&obj->mot_lov_mutex);
2508
2509         rc = mdo_unlink(info->mti_env, mdt_object_child(pobj),
2510                         mdt_object_child(obj), name, ma, 0);
2511
2512         mutex_unlock(&obj->mot_lov_mutex);
2513
2514 unlock_child:
2515         mdt_reint_striped_unlock(info, obj, child_lh, einfo, 1);
2516 unlock_parent:
2517         mdt_object_unlock(info, pobj, parent_lh, 1);
2518 put_parent:
2519         mdt_object_put(info->mti_env, pobj);
2520 out:
2521         RETURN(rc);
2522 }
2523
2524 static int mdt_rmfid_check_permission(struct mdt_thread_info *info,
2525                                         struct mdt_object *obj)
2526 {
2527         struct lu_ucred *uc = lu_ucred(info->mti_env);
2528         struct md_attr *ma = &info->mti_attr;
2529         struct lu_attr *la = &ma->ma_attr;
2530         int rc = 0;
2531         ENTRY;
2532
2533         ma->ma_need = MA_INODE;
2534         rc = mo_attr_get(info->mti_env, mdt_object_child(obj), ma);
2535         if (rc)
2536                 GOTO(out, rc);
2537
2538         if (la->la_flags & LUSTRE_IMMUTABLE_FL)
2539                         rc = -EACCES;
2540
2541         if (cap_raised(uc->uc_cap, CAP_DAC_OVERRIDE))
2542                 RETURN(0);
2543         if (uc->uc_fsuid == la->la_uid) {
2544                 if ((la->la_mode & S_IWUSR) == 0)
2545                         rc = -EACCES;
2546         } else if (uc->uc_fsgid == la->la_gid) {
2547                 if ((la->la_mode & S_IWGRP) == 0)
2548                         rc = -EACCES;
2549         } else if ((la->la_mode & S_IWOTH) == 0) {
2550                         rc = -EACCES;
2551         }
2552
2553 out:
2554         RETURN(rc);
2555 }
2556
2557 static int mdt_rmfid_one(struct mdt_thread_info *info, struct lu_fid *fid,
2558                          s64 ctime)
2559 {
2560         struct mdt_device *mdt = info->mti_mdt;
2561         struct mdt_object *obj = NULL;
2562         struct linkea_data ldata = { NULL };
2563         struct lu_buf *buf = &info->mti_big_buf;
2564         struct lu_name *name = &info->mti_name;
2565         struct lu_fid *pfid = &info->mti_tmp_fid1;
2566         struct link_ea_header *leh;
2567         struct link_ea_entry *lee;
2568         int reclen, count, rc = 0;
2569         ENTRY;
2570
2571         if (!fid_is_sane(fid))
2572                 GOTO(out, rc = -EINVAL);
2573
2574         if (!fid_is_namespace_visible(fid))
2575                 GOTO(out, rc = -EINVAL);
2576
2577         obj = mdt_object_find(info->mti_env, mdt, fid);
2578         if (IS_ERR(obj))
2579                 GOTO(out, rc = PTR_ERR(obj));
2580
2581         if (mdt_object_remote(obj))
2582                 GOTO(out, rc = -EREMOTE);
2583         if (!mdt_object_exists(obj) || lu_object_is_dying(&obj->mot_header))
2584                 GOTO(out, rc = -ENOENT);
2585
2586         rc = mdt_rmfid_check_permission(info, obj);
2587         if (rc)
2588                 GOTO(out, rc);
2589
2590         /* take LinkEA */
2591         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2592         if (!buf->lb_buf)
2593                 GOTO(out, rc = -ENOMEM);
2594
2595         ldata.ld_buf = buf;
2596         rc = mdt_links_read(info, obj, &ldata);
2597         if (rc)
2598                 GOTO(out, rc);
2599
2600         leh = buf->lb_buf;
2601         lee = (struct link_ea_entry *)(leh + 1);
2602         for (count = 0; count < leh->leh_reccount; count++) {
2603                 /* remove every hardlink */
2604                 linkea_entry_unpack(lee, &reclen, name, pfid);
2605                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2606                 rc = mdt_rmfid_unlink(info, pfid, name, obj, ctime);
2607                 if (rc)
2608                         break;
2609         }
2610
2611 out:
2612         if (obj && !IS_ERR(obj))
2613                 mdt_object_put(info->mti_env, obj);
2614         if (info->mti_big_buf.lb_buf)
2615                 lu_buf_free(&info->mti_big_buf);
2616
2617         RETURN(rc);
2618 }
2619
2620 static int mdt_rmfid(struct tgt_session_info *tsi)
2621 {
2622         struct mdt_thread_info *mti = tsi2mdt_info(tsi);
2623         struct mdt_body *reqbody;
2624         struct lu_fid *fids, *rfids;
2625         int bufsize, rc;
2626         __u32 *rcs;
2627         int i, nr;
2628         ENTRY;
2629
2630         reqbody = req_capsule_client_get(tsi->tsi_pill, &RMF_MDT_BODY);
2631         if (reqbody == NULL)
2632                 RETURN(-EPROTO);
2633         bufsize = req_capsule_get_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2634                                        RCL_CLIENT);
2635         nr = bufsize / sizeof(struct lu_fid);
2636         if (nr * sizeof(struct lu_fid) != bufsize)
2637                 RETURN(-EINVAL);
2638         req_capsule_set_size(tsi->tsi_pill, &RMF_RCS,
2639                              RCL_SERVER, nr * sizeof(__u32));
2640         req_capsule_set_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2641                              RCL_SERVER, nr * sizeof(struct lu_fid));
2642         rc = req_capsule_server_pack(tsi->tsi_pill);
2643         if (rc)
2644                 GOTO(out, rc = err_serious(rc));
2645         fids = req_capsule_client_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2646         if (fids == NULL)
2647                 RETURN(-EPROTO);
2648         rcs = req_capsule_server_get(tsi->tsi_pill, &RMF_RCS);
2649         LASSERT(rcs);
2650         rfids = req_capsule_server_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2651         LASSERT(rfids);
2652
2653         mdt_init_ucred(mti, reqbody);
2654         for (i = 0; i < nr; i++) {
2655                 rfids[i] = fids[i];
2656                 rcs[i] = mdt_rmfid_one(mti, fids + i, reqbody->mbo_ctime);
2657         }
2658         mdt_exit_ucred(mti);
2659
2660 out:
2661         RETURN(rc);
2662 }
2663
2664 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2665                          void *karg, void __user *uarg);
2666
2667 int mdt_io_set_info(struct tgt_session_info *tsi)
2668 {
2669         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2670         struct ost_body         *body = NULL, *repbody;
2671         void                    *key, *val = NULL;
2672         int                      keylen, vallen, rc = 0;
2673         bool                     is_grant_shrink;
2674
2675         ENTRY;
2676
2677         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2678         if (key == NULL) {
2679                 DEBUG_REQ(D_HA, req, "no set_info key");
2680                 RETURN(err_serious(-EFAULT));
2681         }
2682         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2683                                       RCL_CLIENT);
2684
2685         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2686         if (val == NULL) {
2687                 DEBUG_REQ(D_HA, req, "no set_info val");
2688                 RETURN(err_serious(-EFAULT));
2689         }
2690         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2691                                       RCL_CLIENT);
2692
2693         is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
2694         if (is_grant_shrink)
2695                 /* In this case the value is actually an RMF_OST_BODY, so we
2696                  * transmutate the type of this PTLRPC */
2697                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
2698
2699         rc = req_capsule_server_pack(tsi->tsi_pill);
2700         if (rc < 0)
2701                 RETURN(rc);
2702
2703         if (is_grant_shrink) {
2704                 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
2705
2706                 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2707                 *repbody = *body;
2708
2709                 /** handle grant shrink, similar to a read request */
2710                 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
2711                                        &repbody->oa);
2712         } else {
2713                 CERROR("%s: Unsupported key %s\n",
2714                        tgt_name(tsi->tsi_tgt), (char *)key);
2715                 rc = -EOPNOTSUPP;
2716         }
2717
2718         RETURN(rc);
2719 }
2720
2721
2722 static int mdt_set_info(struct tgt_session_info *tsi)
2723 {
2724         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2725         char                    *key;
2726         void                    *val;
2727         int                      keylen, vallen, rc = 0;
2728
2729         ENTRY;
2730
2731         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2732         if (key == NULL) {
2733                 DEBUG_REQ(D_HA, req, "no set_info key");
2734                 RETURN(err_serious(-EFAULT));
2735         }
2736
2737         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2738                                       RCL_CLIENT);
2739
2740         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2741         if (val == NULL) {
2742                 DEBUG_REQ(D_HA, req, "no set_info val");
2743                 RETURN(err_serious(-EFAULT));
2744         }
2745
2746         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2747                                       RCL_CLIENT);
2748
2749         /* Swab any part of val you need to here */
2750         if (KEY_IS(KEY_READ_ONLY)) {
2751                 spin_lock(&req->rq_export->exp_lock);
2752                 if (*(__u32 *)val)
2753                         *exp_connect_flags_ptr(req->rq_export) |=
2754                                 OBD_CONNECT_RDONLY;
2755                 else
2756                         *exp_connect_flags_ptr(req->rq_export) &=
2757                                 ~OBD_CONNECT_RDONLY;
2758                 spin_unlock(&req->rq_export->exp_lock);
2759         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2760                 struct changelog_setinfo *cs = val;
2761
2762                 if (vallen != sizeof(*cs)) {
2763                         CERROR("%s: bad changelog_clear setinfo size %d\n",
2764                                tgt_name(tsi->tsi_tgt), vallen);
2765                         RETURN(-EINVAL);
2766                 }
2767                 if (req_capsule_req_need_swab(&req->rq_pill)) {
2768                         __swab64s(&cs->cs_recno);
2769                         __swab32s(&cs->cs_id);
2770                 }
2771
2772                 if (!mdt_is_rootadmin(tsi2mdt_info(tsi)))
2773                         RETURN(-EACCES);
2774                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2775                                    vallen, val, NULL);
2776         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2777                 if (vallen > 0)
2778                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2779         } else {
2780                 RETURN(-EINVAL);
2781         }
2782         RETURN(rc);
2783 }
2784
2785 static int mdt_readpage(struct tgt_session_info *tsi)
2786 {
2787         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2788         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2789         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2790         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2791         struct mdt_body         *repbody;
2792         int                      rc;
2793         int                      i;
2794
2795         ENTRY;
2796
2797         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2798                 RETURN(err_serious(-ENOMEM));
2799
2800         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2801         if (repbody == NULL || reqbody == NULL)
2802                 RETURN(err_serious(-EFAULT));
2803
2804         /*
2805          * prepare @rdpg before calling lower layers and transfer itself. Here
2806          * reqbody->size contains offset of where to start to read and
2807          * reqbody->nlink contains number bytes to read.
2808          */
2809         rdpg->rp_hash = reqbody->mbo_size;
2810         if (rdpg->rp_hash != reqbody->mbo_size) {
2811                 CERROR("Invalid hash: %#llx != %#llx\n",
2812                        rdpg->rp_hash, reqbody->mbo_size);
2813                 RETURN(-EFAULT);
2814         }
2815
2816         rdpg->rp_attrs = reqbody->mbo_mode;
2817         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2818                 rdpg->rp_attrs |= LUDA_64BITHASH;
2819         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2820                                 exp_max_brw_size(tsi->tsi_exp));
2821         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2822                           PAGE_SHIFT;
2823         OBD_ALLOC_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2824         if (rdpg->rp_pages == NULL)
2825                 RETURN(-ENOMEM);
2826
2827         for (i = 0; i < rdpg->rp_npages; ++i) {
2828                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2829                 if (rdpg->rp_pages[i] == NULL)
2830                         GOTO(free_rdpg, rc = -ENOMEM);
2831         }
2832
2833         /* call lower layers to fill allocated pages with directory data */
2834         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2835         if (rc < 0)
2836                 GOTO(free_rdpg, rc);
2837
2838         /* send pages to client */
2839         rc = tgt_sendpage(tsi, rdpg, rc);
2840
2841         EXIT;
2842 free_rdpg:
2843
2844         for (i = 0; i < rdpg->rp_npages; i++)
2845                 if (rdpg->rp_pages[i] != NULL)
2846                         __free_page(rdpg->rp_pages[i]);
2847         OBD_FREE_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2848
2849         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2850                 RETURN(0);
2851
2852         return rc;
2853 }
2854
2855 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2856 {
2857         struct lu_ucred *uc = mdt_ucred_check(info);
2858         struct lu_attr *attr = &info->mti_attr.ma_attr;
2859
2860         if (uc == NULL)
2861                 return -EINVAL;
2862
2863         if (op != REINT_SETATTR) {
2864                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2865                         attr->la_uid = uc->uc_fsuid;
2866                 /* for S_ISGID, inherit gid from his parent, such work will be
2867                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2868                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2869                         attr->la_gid = uc->uc_fsgid;
2870         }
2871
2872         return 0;
2873 }
2874
2875 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2876 {
2877         return op == REINT_OPEN &&
2878              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2879 }
2880
2881 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2882 {
2883         struct req_capsule *pill = info->mti_pill;
2884
2885         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2886                                   RCL_SERVER) &&
2887             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2888                                   RCL_CLIENT)) {
2889                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2890                                          RCL_CLIENT) != 0)
2891                         /* pre-set size in server part with max size */
2892                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2893                                              RCL_SERVER,
2894                                              OBD_MAX_DEFAULT_EA_SIZE);
2895                 else
2896                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2897                                              RCL_SERVER, 0);
2898         }
2899 }
2900
2901 static int mdt_reint_internal(struct mdt_thread_info *info,
2902                               struct mdt_lock_handle *lhc,
2903                               __u32 op)
2904 {
2905         struct req_capsule      *pill = info->mti_pill;
2906         struct mdt_body         *repbody;
2907         int                      rc = 0, rc2;
2908
2909         ENTRY;
2910
2911         rc = mdt_reint_unpack(info, op);
2912         if (rc != 0) {
2913                 CERROR("Can't unpack reint, rc %d\n", rc);
2914                 RETURN(err_serious(rc));
2915         }
2916
2917
2918         /* check if the file system is set to readonly. O_RDONLY open
2919          * is still allowed even the file system is set to readonly mode */
2920         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2921                 RETURN(err_serious(-EROFS));
2922
2923         /* for replay (no_create) lmm is not needed, client has it already */
2924         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2925                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2926                                      DEF_REP_MD_SIZE);
2927
2928         /* llog cookies are always 0, the field is kept for compatibility */
2929         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2930                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2931
2932         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2933          * by default. If the target object has more ACL entries, then
2934          * enlarge the buffer when necessary. */
2935         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2936                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2937                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2938
2939         mdt_preset_secctx_size(info);
2940         mdt_preset_encctx_size(info);
2941
2942         rc = req_capsule_server_pack(pill);
2943         if (rc != 0) {
2944                 CERROR("Can't pack response, rc %d\n", rc);
2945                 RETURN(err_serious(rc));
2946         }
2947
2948         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2949                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2950                 LASSERT(repbody);
2951                 repbody->mbo_eadatasize = 0;
2952                 repbody->mbo_aclsize = 0;
2953         }
2954
2955         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2956
2957         /* for replay no cookkie / lmm need, because client have this already */
2958         if (info->mti_spec.no_create)
2959                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2960                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2961
2962         rc = mdt_init_ucred_reint(info);
2963         if (rc)
2964                 GOTO(out_shrink, rc);
2965
2966         rc = mdt_fix_attr_ucred(info, op);
2967         if (rc != 0)
2968                 GOTO(out_ucred, rc = err_serious(rc));
2969
2970         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2971         if (rc < 0) {
2972                 GOTO(out_ucred, rc);
2973         } else if (rc == 1) {
2974                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt");
2975                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2976                 GOTO(out_ucred, rc);
2977         }
2978         rc = mdt_reint_rec(info, lhc);
2979         EXIT;
2980 out_ucred:
2981         mdt_exit_ucred(info);
2982 out_shrink:
2983         mdt_client_compatibility(info);
2984
2985         rc2 = mdt_fix_reply(info);
2986         if (rc == 0)
2987                 rc = rc2;
2988
2989         /*
2990          * Data-on-MDT optimization - read data along with OPEN and return it
2991          * in reply when possible.
2992          */
2993         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req))
2994                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2995                                           &lhc->mlh_reg_lh);
2996
2997         return rc;
2998 }
2999
3000 static long mdt_reint_opcode(struct ptlrpc_request *req,
3001                              const struct req_format **fmt)
3002 {
3003         struct mdt_device       *mdt;
3004         struct mdt_rec_reint    *rec;
3005         long                     opc;
3006
3007         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
3008         if (rec != NULL) {
3009                 opc = rec->rr_opcode;
3010                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
3011                 if (opc < REINT_MAX && fmt[opc] != NULL)
3012                         req_capsule_extend(&req->rq_pill, fmt[opc]);
3013                 else {
3014                         mdt = mdt_exp2dev(req->rq_export);
3015                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
3016                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
3017                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
3018                         opc = err_serious(-EFAULT);
3019                 }
3020         } else {
3021                 opc = err_serious(-EFAULT);
3022         }
3023         return opc;
3024 }
3025
3026 static int mdt_reint(struct tgt_session_info *tsi)
3027 {
3028         long opc;
3029         int  rc;
3030         static const struct req_format *reint_fmts[REINT_MAX] = {
3031                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
3032                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
3033                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
3034                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
3035                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
3036                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
3037                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
3038                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
3039                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
3040                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
3041         };
3042
3043         ENTRY;
3044
3045         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
3046         if (opc >= 0) {
3047                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3048                 /*
3049                  * No lock possible here from client to pass it to reint code
3050                  * path.
3051                  */
3052                 rc = mdt_reint_internal(info, NULL, opc);
3053                 mdt_thread_info_fini(info);
3054         } else {
3055                 rc = opc;
3056         }
3057
3058         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
3059         RETURN(rc);
3060 }
3061
3062 /* this should sync the whole device */
3063 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
3064 {
3065         struct dt_device *dt = mdt->mdt_bottom;
3066         int rc;
3067         ENTRY;
3068
3069         rc = dt->dd_ops->dt_sync(env, dt);
3070         RETURN(rc);
3071 }
3072
3073 /* this should sync this object */
3074 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
3075                            struct mdt_object *mo)
3076 {
3077         int rc = 0;
3078
3079         ENTRY;
3080
3081         if (!mdt_object_exists(mo)) {
3082                 CWARN("%s: non existing object "DFID": rc = %d\n",
3083                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
3084                       -ESTALE);
3085                 RETURN(-ESTALE);
3086         }
3087
3088         if (S_ISREG(lu_object_attr(&mo->mot_obj))) {
3089                 struct lu_target *tgt = tgt_ses_info(env)->tsi_tgt;
3090                 dt_obj_version_t version;
3091
3092                 version = dt_version_get(env, mdt_obj2dt(mo));
3093                 if (version > tgt->lut_obd->obd_last_committed)
3094                         rc = mo_object_sync(env, mdt_object_child(mo));
3095         } else {
3096                 rc = mo_object_sync(env, mdt_object_child(mo));
3097         }
3098
3099         RETURN(rc);
3100 }
3101
3102 static int mdt_sync(struct tgt_session_info *tsi)
3103 {
3104         struct ptlrpc_request   *req = tgt_ses_req(tsi);
3105         struct req_capsule      *pill = tsi->tsi_pill;
3106         struct mdt_body         *body;
3107         ktime_t                  kstart = ktime_get();
3108         int                      rc;
3109
3110         ENTRY;
3111
3112         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
3113                 RETURN(err_serious(-ENOMEM));
3114
3115         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
3116                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
3117         } else {
3118                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3119
3120                 if (unlikely(info->mti_object == NULL))
3121                         RETURN(-EPROTO);
3122
3123                 /* sync an object */
3124                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
3125                                      info->mti_object);
3126                 if (rc == 0) {
3127                         const struct lu_fid *fid;
3128                         struct lu_attr *la = &info->mti_attr.ma_attr;
3129
3130                         info->mti_attr.ma_need = MA_INODE;
3131                         info->mti_attr.ma_valid = 0;
3132                         rc = mdt_attr_get_complex(info, info->mti_object,
3133                                                   &info->mti_attr);
3134                         if (rc == 0) {
3135                                 body = req_capsule_server_get(pill,
3136                                                               &RMF_MDT_BODY);
3137                                 fid = mdt_object_fid(info->mti_object);
3138                                 mdt_pack_attr2body(info, body, la, fid);
3139                         }
3140                 }
3141                 mdt_thread_info_fini(info);
3142         }
3143         if (rc == 0)
3144                 mdt_counter_incr(req, LPROC_MDT_SYNC,
3145                                  ktime_us_delta(ktime_get(), kstart));
3146
3147         RETURN(rc);
3148 }
3149
3150 static int mdt_data_sync(struct tgt_session_info *tsi)
3151 {
3152         struct mdt_thread_info *info;
3153         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
3154         struct ost_body *body = tsi->tsi_ost_body;
3155         struct ost_body *repbody;
3156         struct mdt_object *mo = NULL;
3157         struct md_attr *ma;
3158         int rc = 0;
3159
3160         ENTRY;
3161
3162         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
3163
3164         /* if no fid is specified then do nothing,
3165          * device sync is done via MDS_SYNC */
3166         if (fid_is_zero(&tsi->tsi_fid))
3167                 RETURN(0);
3168
3169         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
3170         if (IS_ERR(mo))
3171                 RETURN(PTR_ERR(mo));
3172
3173         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
3174         if (rc)
3175                 GOTO(put, rc);
3176
3177         repbody->oa.o_oi = body->oa.o_oi;
3178         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
3179
3180         info = tsi2mdt_info(tsi);
3181         ma = &info->mti_attr;
3182         ma->ma_need = MA_INODE;
3183         ma->ma_valid = 0;
3184         rc = mdt_attr_get_complex(info, mo, ma);
3185         if (rc == 0)
3186                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
3187         else
3188                 rc = 0;
3189         mdt_thread_info_fini(info);
3190
3191         EXIT;
3192 put:
3193         if (mo != NULL)
3194                 mdt_object_put(tsi->tsi_env, mo);
3195         return rc;
3196 }
3197
3198 /*
3199  * Handle quota control requests to consult current usage/limit, but also
3200  * to configure quota enforcement
3201  */
3202 static int mdt_quotactl(struct tgt_session_info *tsi)
3203 {
3204         struct obd_export *exp  = tsi->tsi_exp;
3205         struct req_capsule *pill = tsi->tsi_pill;
3206         struct obd_quotactl *oqctl, *repoqc;
3207         int id, rc;
3208         struct mdt_device *mdt = mdt_exp2dev(exp);
3209         struct lu_device *qmt = mdt->mdt_qmt_dev;
3210         struct lu_nodemap *nodemap;
3211         ENTRY;
3212
3213         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
3214         if (!oqctl)
3215                 RETURN(err_serious(-EPROTO));
3216
3217         rc = req_capsule_server_pack(pill);
3218         if (rc)
3219                 RETURN(err_serious(rc));
3220
3221         nodemap = nodemap_get_from_exp(exp);
3222         if (IS_ERR(nodemap))
3223                 RETURN(PTR_ERR(nodemap));
3224
3225         switch (oqctl->qc_cmd) {
3226                 /* master quotactl */
3227         case Q_SETINFO:
3228         case Q_SETQUOTA:
3229         case LUSTRE_Q_SETDEFAULT:
3230         case LUSTRE_Q_SETQUOTAPOOL:
3231         case LUSTRE_Q_SETINFOPOOL:
3232         case LUSTRE_Q_SETDEFAULT_POOL:
3233         case LUSTRE_Q_DELETEQID:
3234                 if (!nodemap_can_setquota(nodemap, oqctl->qc_type,
3235                                           oqctl->qc_id))
3236                         GOTO(out_nodemap, rc = -EPERM);
3237                 fallthrough;
3238         case Q_GETINFO:
3239         case Q_GETQUOTA:
3240         case LUSTRE_Q_GETDEFAULT:
3241         case LUSTRE_Q_GETQUOTAPOOL:
3242         case LUSTRE_Q_GETINFOPOOL:
3243         case LUSTRE_Q_GETDEFAULT_POOL:
3244                 if (qmt == NULL)
3245                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
3246                 /* slave quotactl */
3247                 fallthrough;
3248         case Q_GETOINFO:
3249         case Q_GETOQUOTA:
3250                 break;
3251         default:
3252                 rc = -EFAULT;
3253                 CERROR("%s: unsupported quotactl command %d: rc = %d\n",
3254                        mdt_obd_name(mdt), oqctl->qc_cmd, rc);
3255                 GOTO(out_nodemap, rc);
3256         }
3257
3258         id = oqctl->qc_id;
3259         switch (oqctl->qc_type) {
3260         case USRQUOTA:
3261                 id = nodemap_map_id(nodemap, NODEMAP_UID,
3262                                     NODEMAP_CLIENT_TO_FS, id);
3263                 break;
3264         case GRPQUOTA:
3265                 id = nodemap_map_id(nodemap, NODEMAP_GID,
3266                                     NODEMAP_CLIENT_TO_FS, id);
3267                 break;
3268         case PRJQUOTA:
3269                 id = nodemap_map_id(nodemap, NODEMAP_PROJID,
3270                                     NODEMAP_CLIENT_TO_FS, id);
3271                 break;
3272         default:
3273                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
3274         }
3275         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
3276         if (repoqc == NULL)
3277                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
3278
3279         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
3280                 barrier_exit(tsi->tsi_tgt->lut_bottom);
3281
3282         if (oqctl->qc_id != id)
3283                 swap(oqctl->qc_id, id);
3284
3285         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
3286                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
3287                         RETURN(-EINPROGRESS);
3288         }
3289
3290         switch (oqctl->qc_cmd) {
3291
3292         case Q_GETINFO:
3293         case Q_SETINFO:
3294         case Q_SETQUOTA:
3295         case Q_GETQUOTA:
3296         case LUSTRE_Q_SETDEFAULT:
3297         case LUSTRE_Q_GETDEFAULT:
3298         case LUSTRE_Q_SETQUOTAPOOL:
3299         case LUSTRE_Q_GETQUOTAPOOL:
3300         case LUSTRE_Q_SETINFOPOOL:
3301         case LUSTRE_Q_GETINFOPOOL:
3302         case LUSTRE_Q_SETDEFAULT_POOL:
3303         case LUSTRE_Q_GETDEFAULT_POOL:
3304         case LUSTRE_Q_DELETEQID:
3305                 /* forward quotactl request to QMT */
3306                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
3307                 break;
3308
3309         case Q_GETOINFO:
3310         case Q_GETOQUOTA:
3311                 /* slave quotactl */
3312                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
3313                                    oqctl);
3314                 break;
3315
3316         default:
3317                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
3318                 GOTO(out_nodemap, rc = -EFAULT);
3319         }
3320
3321         if (oqctl->qc_id != id)
3322                 swap(oqctl->qc_id, id);
3323
3324         QCTL_COPY(repoqc, oqctl);
3325         EXIT;
3326
3327 out_nodemap:
3328         nodemap_putref(nodemap);
3329
3330         return rc;
3331 }
3332
3333 /** clone llog ctxt from child (mdd)
3334  * This allows remote llog (replicator) access.
3335  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
3336  * context was originally set up, or we can handle them directly.
3337  * I choose the latter, but that means I need any llog
3338  * contexts set up by child to be accessable by the mdt.  So we clone the
3339  * context into our context list here.
3340  */
3341 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
3342                                int idx)
3343 {
3344         struct md_device  *next = mdt->mdt_child;
3345         struct llog_ctxt *ctxt;
3346         int rc;
3347
3348         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
3349                 return 0;
3350
3351         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
3352         if (rc || ctxt == NULL) {
3353                 return 0;
3354         }
3355
3356         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
3357         if (rc)
3358                 CERROR("Can't set mdt ctxt %d\n", rc);
3359
3360         return rc;
3361 }
3362
3363 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
3364                                  struct mdt_device *mdt, int idx)
3365 {
3366         struct llog_ctxt *ctxt;
3367
3368         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
3369         if (ctxt == NULL)
3370                 return 0;
3371         /* Put once for the get we just did, and once for the clone */
3372         llog_ctxt_put(ctxt);
3373         llog_ctxt_put(ctxt);
3374         return 0;
3375 }
3376
3377 /*
3378  * sec context handlers
3379  */
3380 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
3381 {
3382         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
3383
3384         return 0;
3385 }
3386
3387 /*
3388  * quota request handlers
3389  */
3390 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
3391 {
3392         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
3393         struct lu_device        *qmt = mdt->mdt_qmt_dev;
3394         int                      rc;
3395         ENTRY;
3396
3397         if (qmt == NULL)
3398                 RETURN(err_serious(-EOPNOTSUPP));
3399
3400         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
3401         RETURN(rc);
3402 }
3403
3404 struct mdt_object *mdt_object_new(const struct lu_env *env,
3405                                   struct mdt_device *d,
3406                                   const struct lu_fid *f)
3407 {
3408         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
3409         struct lu_object *o;
3410         struct mdt_object *m;
3411         ENTRY;
3412
3413         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
3414         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
3415         if (unlikely(IS_ERR(o)))
3416                 m = (struct mdt_object *)o;
3417         else
3418                 m = mdt_obj(o);
3419         RETURN(m);
3420 }
3421
3422 struct mdt_object *mdt_object_find(const struct lu_env *env,
3423                                    struct mdt_device *d,
3424                                    const struct lu_fid *f)
3425 {
3426         struct lu_object *o;
3427         struct mdt_object *m;
3428         ENTRY;
3429
3430         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
3431         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
3432         if (unlikely(IS_ERR(o)))
3433                 m = (struct mdt_object *)o;
3434         else
3435                 m = mdt_obj(o);
3436
3437         RETURN(m);
3438 }
3439
3440 /**
3441  * Asyncronous commit for mdt device.
3442  *
3443  * Pass asynchonous commit call down the MDS stack.
3444  *
3445  * \param env environment
3446  * \param mdt the mdt device
3447  */
3448 static void mdt_device_commit_async(const struct lu_env *env,
3449                                     struct mdt_device *mdt)
3450 {
3451         struct dt_device *dt = mdt->mdt_bottom;
3452         int rc;
3453         ENTRY;
3454
3455         rc = dt->dd_ops->dt_commit_async(env, dt);
3456         if (unlikely(rc != 0))
3457                 CWARN("%s: async commit start failed: rc = %d\n",
3458                       mdt_obd_name(mdt), rc);
3459         atomic_inc(&mdt->mdt_async_commit_count);
3460         EXIT;
3461 }
3462
3463 /**
3464  * Mark the lock as "synchonous".
3465  *
3466  * Mark the lock to deffer transaction commit to the unlock time.
3467  *
3468  * \param lock the lock to mark as "synchonous"
3469  *
3470  * \see mdt_is_lock_sync
3471  * \see mdt_save_lock
3472  */
3473 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
3474 {
3475         lock->l_ast_data = (void*)1;
3476 }
3477
3478 /**
3479  * Check whehter the lock "synchonous" or not.
3480  *
3481  * \param lock the lock to check
3482  * \retval 1 the lock is "synchonous"
3483  * \retval 0 the lock isn't "synchronous"
3484  *
3485  * \see mdt_set_lock_sync
3486  * \see mdt_save_lock
3487  */
3488 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
3489 {
3490         return lock->l_ast_data != NULL;
3491 }
3492
3493 /**
3494  * Blocking AST for mdt locks.
3495  *
3496  * Starts transaction commit if in case of COS lock conflict or
3497  * deffers such a commit to the mdt_save_lock.
3498  *
3499  * \param lock the lock which blocks a request or cancelling lock
3500  * \param desc unused
3501  * \param data unused
3502  * \param flag indicates whether this cancelling or blocking callback
3503  * \retval 0
3504  * \see ldlm_blocking_ast_nocheck
3505  */
3506 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3507                      void *data, int flag)
3508 {
3509         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3510         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
3511         struct ldlm_cb_set_arg *arg = data;
3512         bool commit_async = false;
3513         int rc;
3514         ENTRY;
3515
3516         if (flag == LDLM_CB_CANCELING)
3517                 RETURN(0);
3518
3519         lock_res_and_lock(lock);
3520         if (lock->l_blocking_ast != mdt_blocking_ast) {
3521                 unlock_res_and_lock(lock);
3522                 RETURN(0);
3523         }
3524
3525         /* A blocking ast may be sent from ldlm_lock_decref_internal
3526          * when the last reference to a local lock was released and
3527          * during blocking event from ldlm_work_bl_ast_lock().
3528          * The 'data' parameter is l_ast_data in the first case and
3529          * callback arguments in the second one. Distinguish them by that.
3530          */
3531         if (!data || data == lock->l_ast_data || !arg->bl_desc)
3532                 goto skip_cos_checks;
3533
3534         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
3535                 if (mdt_cos_is_enabled(mdt)) {
3536                         if (!arg->bl_desc->bl_same_client)
3537                                 mdt_set_lock_sync(lock);
3538                 } else if (mdt_slc_is_enabled(mdt) &&
3539                            arg->bl_desc->bl_cos_incompat) {
3540                         mdt_set_lock_sync(lock);
3541                         /*
3542                          * we may do extra commit here, but there is a small
3543                          * window to miss a commit: lock was unlocked (saved),
3544                          * then a conflict lock queued and we come here, but
3545                          * REP-ACK not received, so lock was not converted to
3546                          * COS mode yet.
3547                          * Fortunately this window is quite small, so the
3548                          * extra commit should be rare (not to say distributed
3549                          * operation is rare too).
3550                          */
3551                         commit_async = true;
3552                 }
3553         } else if (lock->l_req_mode == LCK_COS) {
3554                 commit_async = true;
3555         }
3556
3557 skip_cos_checks:
3558         rc = ldlm_blocking_ast_nocheck(lock);
3559
3560         if (commit_async) {
3561                 struct lu_env env;
3562
3563                 rc = lu_env_init(&env, LCT_LOCAL);
3564                 if (unlikely(rc != 0))
3565                         CWARN("%s: lu_env initialization failed, cannot "
3566                               "start asynchronous commit: rc = %d\n",
3567                               obd->obd_name, rc);
3568                 else
3569                         mdt_device_commit_async(&env, mdt);
3570                 lu_env_fini(&env);
3571         }
3572         RETURN(rc);
3573 }
3574
3575 /*
3576  * Blocking AST for cross-MDT lock
3577  *
3578  * Discard lock from uncommitted_slc_locks and cancel it.
3579  *
3580  * \param lock  the lock which blocks a request or cancelling lock
3581  * \param desc  unused
3582  * \param data  unused
3583  * \param flag  indicates whether this cancelling or blocking callback
3584  * \retval      0 on success
3585  * \retval      negative number on error
3586  */
3587 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3588                             void *data, int flag)
3589 {
3590         int rc = 0;
3591         ENTRY;
3592
3593         switch (flag) {
3594         case LDLM_CB_BLOCKING: {
3595                 struct lustre_handle lockh;
3596
3597                 ldlm_lock2handle(lock, &lockh);
3598                 rc = ldlm_cli_cancel(&lockh,
3599                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
3600                 if (rc < 0) {
3601                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
3602                         RETURN(rc);
3603                 }
3604                 break;
3605         }
3606         case LDLM_CB_CANCELING: {
3607                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3608                 struct mdt_device *mdt =
3609                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
3610
3611                 LDLM_DEBUG(lock, "Revoke remote lock\n");
3612
3613                 /* discard slc lock here so that it can be cleaned anytime,
3614                  * especially for cleanup_resource() */
3615                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
3616
3617                 /* once we cache lock, l_ast_data is set to mdt_object */
3618                 if (lock->l_ast_data != NULL) {
3619                         struct mdt_object *mo = lock->l_ast_data;
3620                         struct lu_env env;
3621
3622                         rc = lu_env_init(&env, LCT_MD_THREAD);
3623                         if (unlikely(rc != 0)) {
3624                                 CWARN("%s: lu_env initialization failed, object %p "DFID" is leaked!: rc = %d\n",
3625                                       obd->obd_name, mo,
3626                                       PFID(mdt_object_fid(mo)), rc);
3627                                 RETURN(rc);
3628                         }
3629
3630                         if (lock->l_policy_data.l_inodebits.bits &
3631                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
3632                                 rc = mo_invalidate(&env, mdt_object_child(mo));
3633                                 mo->mot_cache_attr = 0;
3634                         }
3635                         mdt_object_put(&env, mo);
3636                         lu_env_fini(&env);
3637                 }
3638                 break;
3639         }
3640         default:
3641                 LBUG();
3642         }
3643
3644         RETURN(rc);
3645 }
3646
3647 int mdt_check_resent_lock(struct mdt_thread_info *info,
3648                           struct mdt_object *mo,
3649                           struct mdt_lock_handle *lhc)
3650 {
3651         /* the lock might already be gotten in ldlm_handle_enqueue() */
3652         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
3653                 struct ptlrpc_request *req = mdt_info_req(info);
3654                 struct ldlm_lock      *lock;
3655
3656                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
3657   &