Whamcloud - gitweb
LU-10318 dom: support DATA_VERSION IO type
[fs/lustre-release.git] / lustre / mdt / mdt_handler.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.gnu.org/licenses/gpl-2.0.html
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2010, 2017, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  * Lustre is a trademark of Sun Microsystems, Inc.
31  *
32  * lustre/mdt/mdt_handler.c
33  *
34  * Lustre Metadata Target (mdt) request handler
35  *
36  * Author: Peter Braam <braam@clusterfs.com>
37  * Author: Andreas Dilger <adilger@clusterfs.com>
38  * Author: Phil Schwan <phil@clusterfs.com>
39  * Author: Mike Shaver <shaver@clusterfs.com>
40  * Author: Nikita Danilov <nikita@clusterfs.com>
41  * Author: Huang Hua <huanghua@clusterfs.com>
42  * Author: Yury Umanets <umka@clusterfs.com>
43  */
44
45 #define DEBUG_SUBSYSTEM S_MDS
46
47 #include <linux/module.h>
48 #include <linux/pagemap.h>
49
50 #include <dt_object.h>
51 #include <lustre_acl.h>
52 #include <lustre_export.h>
53 #include <uapi/linux/lustre/lustre_ioctl.h>
54 #include <lustre_lfsck.h>
55 #include <lustre_log.h>
56 #include <lustre_nodemap.h>
57 #include <lustre_mds.h>
58 #include <uapi/linux/lustre/lustre_param.h>
59 #include <lustre_quota.h>
60 #include <lustre_swab.h>
61 #include <obd.h>
62 #include <obd_support.h>
63 #include <lustre_barrier.h>
64 #include <obd_cksum.h>
65 #include <llog_swab.h>
66
67 #include "mdt_internal.h"
68
69
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 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
98
99 static const struct lu_object_operations mdt_obj_ops;
100
101 /* Slab for MDT object allocation */
102 static struct kmem_cache *mdt_object_kmem;
103
104 /* For HSM restore handles */
105 struct kmem_cache *mdt_hsm_cdt_kmem;
106
107 /* For HSM request handles */
108 struct kmem_cache *mdt_hsm_car_kmem;
109
110 static struct lu_kmem_descr mdt_caches[] = {
111         {
112                 .ckd_cache = &mdt_object_kmem,
113                 .ckd_name  = "mdt_obj",
114                 .ckd_size  = sizeof(struct mdt_object)
115         },
116         {
117                 .ckd_cache      = &mdt_hsm_cdt_kmem,
118                 .ckd_name       = "mdt_cdt_restore_handle",
119                 .ckd_size       = sizeof(struct cdt_restore_handle)
120         },
121         {
122                 .ckd_cache      = &mdt_hsm_car_kmem,
123                 .ckd_name       = "mdt_cdt_agent_req",
124                 .ckd_size       = sizeof(struct cdt_agent_req)
125         },
126         {
127                 .ckd_cache = NULL
128         }
129 };
130
131 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
132 {
133         if (!rep)
134                 return 0;
135         return rep->lock_policy_res1 & op_flag;
136 }
137
138 void mdt_clear_disposition(struct mdt_thread_info *info,
139                            struct ldlm_reply *rep, __u64 op_flag)
140 {
141         if (info) {
142                 info->mti_opdata &= ~op_flag;
143                 tgt_opdata_clear(info->mti_env, op_flag);
144         }
145         if (rep)
146                 rep->lock_policy_res1 &= ~op_flag;
147 }
148
149 void mdt_set_disposition(struct mdt_thread_info *info,
150                          struct ldlm_reply *rep, __u64 op_flag)
151 {
152         if (info) {
153                 info->mti_opdata |= op_flag;
154                 tgt_opdata_set(info->mti_env, op_flag);
155         }
156         if (rep)
157                 rep->lock_policy_res1 |= op_flag;
158 }
159
160 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
161 {
162         lh->mlh_pdo_hash = 0;
163         lh->mlh_reg_mode = lm;
164         lh->mlh_rreg_mode = lm;
165         lh->mlh_type = MDT_REG_LOCK;
166 }
167
168 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
169                        const struct lu_name *lname)
170 {
171         lh->mlh_reg_mode = lock_mode;
172         lh->mlh_pdo_mode = LCK_MINMODE;
173         lh->mlh_rreg_mode = lock_mode;
174         lh->mlh_type = MDT_PDO_LOCK;
175
176         if (lu_name_is_valid(lname)) {
177                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
178                                                      lname->ln_namelen);
179                 /* XXX Workaround for LU-2856
180                  *
181                  * Zero is a valid return value of full_name_hash, but
182                  * several users of mlh_pdo_hash assume a non-zero
183                  * hash value. We therefore map zero onto an
184                  * arbitrary, but consistent value (1) to avoid
185                  * problems further down the road. */
186                 if (unlikely(lh->mlh_pdo_hash == 0))
187                         lh->mlh_pdo_hash = 1;
188         } else {
189                 lh->mlh_pdo_hash = 0;
190         }
191 }
192
193 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
194                               struct mdt_lock_handle *lh)
195 {
196         mdl_mode_t mode;
197         ENTRY;
198
199         /*
200          * Any dir access needs couple of locks:
201          *
202          * 1) on part of dir we gonna take lookup/modify;
203          *
204          * 2) on whole dir to protect it from concurrent splitting and/or to
205          * flush client's cache for readdir().
206          *
207          * so, for a given mode and object this routine decides what lock mode
208          * to use for lock #2:
209          *
210          * 1) if caller's gonna lookup in dir then we need to protect dir from
211          * being splitted only - LCK_CR
212          *
213          * 2) if caller's gonna modify dir then we need to protect dir from
214          * being splitted and to flush cache - LCK_CW
215          *
216          * 3) if caller's gonna modify dir and that dir seems ready for
217          * splitting then we need to protect it from any type of access
218          * (lookup/modify/split) - LCK_EX --bzzz
219          */
220
221         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
222         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
223
224         /*
225          * Ask underlaying level its opinion about preferable PDO lock mode
226          * having access type passed as regular lock mode:
227          *
228          * - MDL_MINMODE means that lower layer does not want to specify lock
229          * mode;
230          *
231          * - MDL_NL means that no PDO lock should be taken. This is used in some
232          * cases. Say, for non-splittable directories no need to use PDO locks
233          * at all.
234          */
235         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
236                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
237
238         if (mode != MDL_MINMODE) {
239                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
240         } else {
241                 /*
242                  * Lower layer does not want to specify locking mode. We do it
243                  * our selves. No special protection is needed, just flush
244                  * client's cache on modification and allow concurrent
245                  * mondification.
246                  */
247                 switch (lh->mlh_reg_mode) {
248                 case LCK_EX:
249                         lh->mlh_pdo_mode = LCK_EX;
250                         break;
251                 case LCK_PR:
252                         lh->mlh_pdo_mode = LCK_CR;
253                         break;
254                 case LCK_PW:
255                         lh->mlh_pdo_mode = LCK_CW;
256                         break;
257                 default:
258                         CERROR("Not expected lock type (0x%x)\n",
259                                (int)lh->mlh_reg_mode);
260                         LBUG();
261                 }
262         }
263
264         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
265         EXIT;
266 }
267
268 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
269                               struct lu_fid *fid)
270 {
271         struct mdt_device *mdt = info->mti_mdt;
272         struct lu_name *lname = &info->mti_name;
273         char *name = NULL;
274         struct mdt_object *parent;
275         u32 mode;
276         int rc = 0;
277
278         LASSERT(!info->mti_cross_ref);
279
280         OBD_ALLOC(name, NAME_MAX + 1);
281         if (name == NULL)
282                 return -ENOMEM;
283         lname->ln_name = name;
284
285         /*
286          * We may want to allow this to mount a completely separate
287          * fileset from the MDT in the future, but keeping it to
288          * ROOT/ only for now avoid potential security issues.
289          */
290         *fid = mdt->mdt_md_root_fid;
291
292         while (rc == 0 && fileset != NULL && *fileset != '\0') {
293                 const char *s1 = fileset;
294                 const char *s2;
295
296                 while (*++s1 == '/')
297                         ;
298                 s2 = s1;
299                 while (*s2 != '/' && *s2 != '\0')
300                         s2++;
301
302                 if (s2 == s1)
303                         break;
304
305                 fileset = s2;
306
307                 lname->ln_namelen = s2 - s1;
308                 if (lname->ln_namelen > NAME_MAX) {
309                         rc = -EINVAL;
310                         break;
311                 }
312
313                 /* reject .. as a path component */
314                 if (lname->ln_namelen == 2 &&
315                     strncmp(s1, "..", 2) == 0) {
316                         rc = -EINVAL;
317                         break;
318                 }
319
320                 strncpy(name, s1, lname->ln_namelen);
321                 name[lname->ln_namelen] = '\0';
322
323                 parent = mdt_object_find(info->mti_env, mdt, fid);
324                 if (IS_ERR(parent)) {
325                         rc = PTR_ERR(parent);
326                         break;
327                 }
328                 /* Only got the fid of this obj by name */
329                 fid_zero(fid);
330                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
331                                 fid, &info->mti_spec);
332                 mdt_object_put(info->mti_env, parent);
333         }
334         if (!rc) {
335                 parent = mdt_object_find(info->mti_env, mdt, fid);
336                 if (IS_ERR(parent))
337                         rc = PTR_ERR(parent);
338                 else {
339                         mode = lu_object_attr(&parent->mot_obj);
340                         mdt_object_put(info->mti_env, parent);
341                         if (!S_ISDIR(mode))
342                                 rc = -ENOTDIR;
343                 }
344         }
345
346         OBD_FREE(name, NAME_MAX + 1);
347
348         return rc;
349 }
350
351 static int mdt_get_root(struct tgt_session_info *tsi)
352 {
353         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
354         struct mdt_device       *mdt = info->mti_mdt;
355         struct mdt_body         *repbody;
356         char                    *fileset = NULL, *buffer = NULL;
357         int                      rc;
358         struct obd_export       *exp = info->mti_exp;
359         char                    *nodemap_fileset;
360
361         ENTRY;
362
363         rc = mdt_check_ucred(info);
364         if (rc)
365                 GOTO(out, rc = err_serious(rc));
366
367         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
368                 GOTO(out, rc = err_serious(-ENOMEM));
369
370         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
371         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
372                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
373                 if (fileset == NULL)
374                         GOTO(out, rc = err_serious(-EFAULT));
375         }
376
377         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
378         if (nodemap_fileset && nodemap_fileset[0]) {
379                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
380                 if (fileset) {
381                         /* consider fileset from client as a sub-fileset
382                          * of the nodemap one */
383                         OBD_ALLOC(buffer, PATH_MAX + 1);
384                         if (buffer == NULL)
385                                 GOTO(out, rc = err_serious(-ENOMEM));
386                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
387                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
388                                 GOTO(out, rc = err_serious(-EINVAL));
389                         fileset = buffer;
390                 } else {
391                         /* enforce fileset as specified in the nodemap */
392                         fileset = nodemap_fileset;
393                 }
394         }
395
396         if (fileset) {
397                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
398                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
399                 if (rc < 0)
400                         GOTO(out, rc = err_serious(rc));
401         } else {
402                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
403         }
404         repbody->mbo_valid |= OBD_MD_FLID;
405
406         EXIT;
407 out:
408         mdt_thread_info_fini(info);
409         if (buffer)
410                 OBD_FREE(buffer, PATH_MAX+1);
411         return rc;
412 }
413
414 static int mdt_statfs(struct tgt_session_info *tsi)
415 {
416         struct ptlrpc_request           *req = tgt_ses_req(tsi);
417         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
418         struct mdt_device               *mdt = info->mti_mdt;
419         struct tg_grants_data           *tgd = &mdt->mdt_lut.lut_tgd;
420         struct ptlrpc_service_part      *svcpt;
421         struct obd_statfs               *osfs;
422         int                             rc;
423
424         ENTRY;
425
426         svcpt = req->rq_rqbd->rqbd_svcpt;
427
428         /* This will trigger a watchdog timeout */
429         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
430                          (MDT_SERVICE_WATCHDOG_FACTOR *
431                           at_get(&svcpt->scp_at_estimate)) + 1);
432
433         rc = mdt_check_ucred(info);
434         if (rc)
435                 GOTO(out, rc = err_serious(rc));
436
437         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
438                 GOTO(out, rc = err_serious(-ENOMEM));
439
440         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
441         if (!osfs)
442                 GOTO(out, rc = -EPROTO);
443
444         rc = tgt_statfs_internal(tsi->tsi_env, &mdt->mdt_lut, osfs,
445                                  ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
446                                  NULL);
447         if (unlikely(rc))
448                 GOTO(out, rc);
449
450         /* at least try to account for cached pages.  its still racy and
451          * might be under-reporting if clients haven't announced their
452          * caches with brw recently */
453         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
454                " pending %llu free %llu avail %llu\n",
455                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
456                tgd->tgd_tot_pending,
457                osfs->os_bfree << tgd->tgd_blockbits,
458                osfs->os_bavail << tgd->tgd_blockbits);
459
460         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
461                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
462                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
463
464         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
465         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
466                "%llu objects: %llu free; state %x\n",
467                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
468                osfs->os_files, osfs->os_ffree, osfs->os_state);
469
470         if (!exp_grant_param_supp(tsi->tsi_exp) &&
471             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
472                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
473                  * should not see a block size > page size, otherwise
474                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
475                  * block size which is the biggest block size known to work
476                  * with all client's page size. */
477                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
478                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
479                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
480                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
481         }
482         if (rc == 0)
483                 mdt_counter_incr(req, LPROC_MDT_STATFS);
484 out:
485         mdt_thread_info_fini(info);
486         RETURN(rc);
487 }
488
489 /**
490  * Pack size attributes into the reply.
491  */
492 int mdt_pack_size2body(struct mdt_thread_info *info,
493                         const struct lu_fid *fid, bool dom_lock)
494 {
495         struct mdt_body *b;
496         struct md_attr *ma = &info->mti_attr;
497         int dom_stripe;
498
499         ENTRY;
500
501         LASSERT(ma->ma_attr.la_valid & LA_MODE);
502
503         if (!S_ISREG(ma->ma_attr.la_mode) ||
504             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
505                 RETURN(-ENODATA);
506
507         dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
508         /* no DoM stripe, no size in reply */
509         if (dom_stripe == LMM_NO_DOM)
510                 RETURN(-ENOENT);
511
512         /* no DoM lock, no size in reply */
513         if (!dom_lock)
514                 RETURN(0);
515
516         /* Either DoM lock exists or LMM has only DoM stripe then
517          * return size on body. */
518         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
519
520         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
521         RETURN(0);
522 }
523
524 #ifdef CONFIG_FS_POSIX_ACL
525 /*
526  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
527  *
528  * \param       info    thread info object
529  * \param       repbody reply to pack ACLs into
530  * \param       o       mdt object of file to examine
531  * \param       nodemap nodemap of client to reply to
532  * \retval      0       success
533  * \retval      -errno  error getting or parsing ACL from disk
534  */
535 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
536                       struct mdt_object *o, struct lu_nodemap *nodemap)
537 {
538         const struct lu_env     *env = info->mti_env;
539         struct md_object        *next = mdt_object_child(o);
540         struct lu_buf           *buf = &info->mti_buf;
541         struct mdt_device       *mdt = info->mti_mdt;
542         struct req_capsule *pill = info->mti_pill;
543         int rc;
544
545         ENTRY;
546
547         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
548         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
549         if (buf->lb_len == 0)
550                 RETURN(0);
551
552 again:
553         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
554         if (rc < 0) {
555                 if (rc == -ENODATA) {
556                         repbody->mbo_aclsize = 0;
557                         repbody->mbo_valid |= OBD_MD_FLACL;
558                         rc = 0;
559                 } else if (rc == -EOPNOTSUPP) {
560                         rc = 0;
561                 } else {
562                         if (rc == -ERANGE &&
563                             exp_connect_large_acl(info->mti_exp) &&
564                             buf->lb_buf != info->mti_big_acl) {
565                                 if (info->mti_big_acl == NULL) {
566                                         OBD_ALLOC_LARGE(info->mti_big_acl,
567                                                         mdt->mdt_max_ea_size);
568                                         if (info->mti_big_acl == NULL) {
569                                                 CERROR("%s: unable to grow "
570                                                        DFID" ACL buffer\n",
571                                                        mdt_obd_name(mdt),
572                                                        PFID(mdt_object_fid(o)));
573                                                 RETURN(-ENOMEM);
574                                         }
575
576                                         info->mti_big_aclsize =
577                                                         mdt->mdt_max_ea_size;
578                                 }
579
580                                 CDEBUG(D_INODE, "%s: grow the "DFID
581                                        " ACL buffer to size %d\n",
582                                        mdt_obd_name(mdt),
583                                        PFID(mdt_object_fid(o)),
584                                        mdt->mdt_max_ea_size);
585
586                                 buf->lb_buf = info->mti_big_acl;
587                                 buf->lb_len = info->mti_big_aclsize;
588
589                                 goto again;
590                         }
591
592                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
593                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
594                 }
595         } else {
596                 int client;
597                 int server;
598                 int acl_buflen;
599                 int lmm_buflen = 0;
600                 int lmmsize = 0;
601
602                 acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
603                 if (acl_buflen >= rc)
604                         goto map;
605
606                 /* If LOV/LMA EA is small, we can reuse part of their buffer */
607                 client = ptlrpc_req_get_repsize(pill->rc_req);
608                 server = lustre_packed_msg_size(pill->rc_req->rq_repmsg);
609                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
610                         lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD,
611                                                           RCL_SERVER);
612                         lmmsize = repbody->mbo_eadatasize;
613                 }
614
615                 if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) {
616                         CDEBUG(D_INODE, "%s: client prepared buffer size %d "
617                                "is not big enough with the ACL size %d (%d)\n",
618                                mdt_obd_name(mdt), client, rc,
619                                server - acl_buflen - lmm_buflen + rc + lmmsize);
620                         repbody->mbo_aclsize = 0;
621                         repbody->mbo_valid &= ~OBD_MD_FLACL;
622                         RETURN(-ERANGE);
623                 }
624
625 map:
626                 if (buf->lb_buf == info->mti_big_acl)
627                         info->mti_big_acl_used = 1;
628
629                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
630                                      rc, NODEMAP_FS_TO_CLIENT);
631                 /* if all ACLs mapped out, rc is still >= 0 */
632                 if (rc < 0) {
633                         CERROR("%s: nodemap_map_acl unable to parse "DFID
634                                " ACL: rc = %d\n", mdt_obd_name(mdt),
635                                PFID(mdt_object_fid(o)), rc);
636                         repbody->mbo_aclsize = 0;
637                         repbody->mbo_valid &= ~OBD_MD_FLACL;
638                 } else {
639                         repbody->mbo_aclsize = rc;
640                         repbody->mbo_valid |= OBD_MD_FLACL;
641                         rc = 0;
642                 }
643         }
644
645         RETURN(rc);
646 }
647 #endif
648
649 /* XXX Look into layout in MDT layer. */
650 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
651 {
652         struct lov_comp_md_v1   *comp_v1;
653         struct lov_mds_md       *v1;
654         int                      i;
655
656         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
657                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
658
659                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
660                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
661                                 comp_v1->lcm_entries[i].lcme_offset);
662                         /* We don't support partial release for now */
663                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
664                                 return false;
665                 }
666                 return true;
667         } else {
668                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
669                         true : false;
670         }
671 }
672
673 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
674                         const struct lu_attr *attr, const struct lu_fid *fid)
675 {
676         struct md_attr *ma = &info->mti_attr;
677         struct obd_export *exp = info->mti_exp;
678         struct lu_nodemap *nodemap = NULL;
679
680         LASSERT(ma->ma_valid & MA_INODE);
681
682         if (attr->la_valid & LA_ATIME) {
683                 b->mbo_atime = attr->la_atime;
684                 b->mbo_valid |= OBD_MD_FLATIME;
685         }
686         if (attr->la_valid & LA_MTIME) {
687                 b->mbo_mtime = attr->la_mtime;
688                 b->mbo_valid |= OBD_MD_FLMTIME;
689         }
690         if (attr->la_valid & LA_CTIME) {
691                 b->mbo_ctime = attr->la_ctime;
692                 b->mbo_valid |= OBD_MD_FLCTIME;
693         }
694         if (attr->la_valid & LA_FLAGS) {
695                 b->mbo_flags = attr->la_flags;
696                 b->mbo_valid |= OBD_MD_FLFLAGS;
697         }
698         if (attr->la_valid & LA_NLINK) {
699                 b->mbo_nlink = attr->la_nlink;
700                 b->mbo_valid |= OBD_MD_FLNLINK;
701         }
702         if (attr->la_valid & (LA_UID|LA_GID)) {
703                 nodemap = nodemap_get_from_exp(exp);
704                 if (IS_ERR(nodemap))
705                         goto out;
706         }
707         if (attr->la_valid & LA_UID) {
708                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
709                                             NODEMAP_FS_TO_CLIENT,
710                                             attr->la_uid);
711                 b->mbo_valid |= OBD_MD_FLUID;
712         }
713         if (attr->la_valid & LA_GID) {
714                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
715                                             NODEMAP_FS_TO_CLIENT,
716                                             attr->la_gid);
717                 b->mbo_valid |= OBD_MD_FLGID;
718         }
719
720         if (attr->la_valid & LA_PROJID) {
721                 /* TODO, nodemap for project id */
722                 b->mbo_projid = attr->la_projid;
723                 b->mbo_valid |= OBD_MD_FLPROJID;
724         }
725
726         b->mbo_mode = attr->la_mode;
727         if (attr->la_valid & LA_MODE)
728                 b->mbo_valid |= OBD_MD_FLMODE;
729         if (attr->la_valid & LA_TYPE)
730                 b->mbo_valid |= OBD_MD_FLTYPE;
731
732         if (fid != NULL) {
733                 b->mbo_fid1 = *fid;
734                 b->mbo_valid |= OBD_MD_FLID;
735                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
736                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
737         }
738
739         if (!(attr->la_valid & LA_TYPE))
740                 return;
741
742         b->mbo_rdev   = attr->la_rdev;
743         b->mbo_size   = attr->la_size;
744         b->mbo_blocks = attr->la_blocks;
745
746         if (!S_ISREG(attr->la_mode)) {
747                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
748         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
749                 /* means no objects are allocated on osts. */
750                 LASSERT(!(ma->ma_valid & MA_LOV));
751                 /* just ignore blocks occupied by extend attributes on MDS */
752                 b->mbo_blocks = 0;
753                 /* if no object is allocated on osts, the size on mds is valid.
754                  * b=22272 */
755                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
756         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
757                 if (mdt_hsm_is_released(ma->ma_lmm)) {
758                         /* A released file stores its size on MDS. */
759                         /* But return 1 block for released file, unless tools
760                          * like tar will consider it fully sparse. (LU-3864)
761                          */
762                         if (unlikely(b->mbo_size == 0))
763                                 b->mbo_blocks = 0;
764                         else
765                                 b->mbo_blocks = 1;
766                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
767                 } else if (info->mti_som_valid) { /* som is valid */
768                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
769                 }
770         }
771
772         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
773                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
774                        PFID(fid), (unsigned long long)b->mbo_size);
775
776 out:
777         if (!IS_ERR_OR_NULL(nodemap))
778                 nodemap_putref(nodemap);
779 }
780
781 static inline int mdt_body_has_lov(const struct lu_attr *la,
782                                    const struct mdt_body *body)
783 {
784         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
785                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
786 }
787
788 void mdt_client_compatibility(struct mdt_thread_info *info)
789 {
790         struct mdt_body       *body;
791         struct ptlrpc_request *req = mdt_info_req(info);
792         struct obd_export     *exp = req->rq_export;
793         struct md_attr        *ma = &info->mti_attr;
794         struct lu_attr        *la = &ma->ma_attr;
795         ENTRY;
796
797         if (exp_connect_layout(exp))
798                 /* the client can deal with 16-bit lmm_stripe_count */
799                 RETURN_EXIT;
800
801         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
802
803         if (!mdt_body_has_lov(la, body))
804                 RETURN_EXIT;
805
806         /* now we have a reply with a lov for a client not compatible with the
807          * layout lock so we have to clean the layout generation number */
808         if (S_ISREG(la->la_mode))
809                 ma->ma_lmm->lmm_layout_gen = 0;
810         EXIT;
811 }
812
813 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
814                                    struct mdt_object *o)
815 {
816         const struct lu_env *env = info->mti_env;
817         int rc, rc2;
818
819         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
820                           XATTR_NAME_LOV);
821
822         if (rc == -ENODATA)
823                 rc = 0;
824
825         if (rc < 0)
826                 goto out;
827
828         /* Is it a directory? Let's check for the LMV as well */
829         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
830                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
831                                    XATTR_NAME_LMV);
832
833                 if (rc2 == -ENODATA)
834                         rc2 = mo_xattr_get(env, mdt_object_child(o),
835                                            &LU_BUF_NULL,
836                                            XATTR_NAME_DEFAULT_LMV);
837
838                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
839                         rc = rc2;
840         }
841
842 out:
843         return rc;
844 }
845
846 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
847                       const char *name)
848 {
849         const struct lu_env *env = info->mti_env;
850         int rc;
851         ENTRY;
852
853         LASSERT(info->mti_big_lmm_used == 0);
854         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
855         if (rc < 0)
856                 RETURN(rc);
857
858         /* big_lmm may need to be grown */
859         if (info->mti_big_lmmsize < rc) {
860                 int size = size_roundup_power2(rc);
861
862                 if (info->mti_big_lmmsize > 0) {
863                         /* free old buffer */
864                         LASSERT(info->mti_big_lmm);
865                         OBD_FREE_LARGE(info->mti_big_lmm,
866                                        info->mti_big_lmmsize);
867                         info->mti_big_lmm = NULL;
868                         info->mti_big_lmmsize = 0;
869                 }
870
871                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
872                 if (info->mti_big_lmm == NULL)
873                         RETURN(-ENOMEM);
874                 info->mti_big_lmmsize = size;
875         }
876         LASSERT(info->mti_big_lmmsize >= rc);
877
878         info->mti_buf.lb_buf = info->mti_big_lmm;
879         info->mti_buf.lb_len = info->mti_big_lmmsize;
880         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
881
882         RETURN(rc);
883 }
884
885 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
886                    struct md_attr *ma, const char *name)
887 {
888         struct md_object *next = mdt_object_child(o);
889         struct lu_buf    *buf = &info->mti_buf;
890         int rc;
891
892         if (strcmp(name, XATTR_NAME_LOV) == 0) {
893                 buf->lb_buf = ma->ma_lmm;
894                 buf->lb_len = ma->ma_lmm_size;
895                 LASSERT(!(ma->ma_valid & MA_LOV));
896         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
897                 buf->lb_buf = ma->ma_lmv;
898                 buf->lb_len = ma->ma_lmv_size;
899                 LASSERT(!(ma->ma_valid & MA_LMV));
900         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
901                 buf->lb_buf = ma->ma_lmv;
902                 buf->lb_len = ma->ma_lmv_size;
903                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
904         } else {
905                 return -EINVAL;
906         }
907
908         rc = mo_xattr_get(info->mti_env, next, buf, name);
909         if (rc > 0) {
910
911 got:
912                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
913                         if (info->mti_big_lmm_used)
914                                 ma->ma_lmm = info->mti_big_lmm;
915
916                         /* NOT return LOV EA with hole to old client. */
917                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
918                                      LOV_PATTERN_F_HOLE) &&
919                             !(exp_connect_flags(info->mti_exp) &
920                               OBD_CONNECT_LFSCK)) {
921                                 return -EIO;
922                         } else {
923                                 ma->ma_lmm_size = rc;
924                                 ma->ma_valid |= MA_LOV;
925                         }
926                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
927                         if (info->mti_big_lmm_used)
928                                 ma->ma_lmv = info->mti_big_lmm;
929
930                         ma->ma_lmv_size = rc;
931                         ma->ma_valid |= MA_LMV;
932                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
933                         ma->ma_lmv_size = rc;
934                         ma->ma_valid |= MA_LMV_DEF;
935                 }
936
937                 /* Update mdt_max_mdsize so all clients will be aware that */
938                 if (info->mti_mdt->mdt_max_mdsize < rc)
939                         info->mti_mdt->mdt_max_mdsize = rc;
940
941                 rc = 0;
942         } else if (rc == -ENODATA) {
943                 /* no LOV EA */
944                 rc = 0;
945         } else if (rc == -ERANGE) {
946                 /* Default LMV has fixed size, so it must be able to fit
947                  * in the original buffer */
948                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
949                         return rc;
950                 rc = mdt_big_xattr_get(info, o, name);
951                 if (rc > 0) {
952                         info->mti_big_lmm_used = 1;
953                         goto got;
954                 }
955         }
956
957         return rc;
958 }
959
960 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
961                              struct mdt_object *o, struct lu_fid *pfid)
962 {
963         struct lu_buf           *buf = &info->mti_buf;
964         struct link_ea_header   *leh;
965         struct link_ea_entry    *lee;
966         int                      rc;
967         ENTRY;
968
969         buf->lb_buf = info->mti_big_lmm;
970         buf->lb_len = info->mti_big_lmmsize;
971         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
972                           buf, XATTR_NAME_LINK);
973         /* ignore errors, MA_PFID won't be set and it is
974          * up to the caller to treat this as an error */
975         if (rc == -ERANGE || buf->lb_len == 0) {
976                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
977                 buf->lb_buf = info->mti_big_lmm;
978                 buf->lb_len = info->mti_big_lmmsize;
979         }
980
981         if (rc < 0)
982                 RETURN(rc);
983         if (rc < sizeof(*leh)) {
984                 CERROR("short LinkEA on "DFID": rc = %d\n",
985                        PFID(mdt_object_fid(o)), rc);
986                 RETURN(-ENODATA);
987         }
988
989         leh = (struct link_ea_header *) buf->lb_buf;
990         lee = (struct link_ea_entry *)(leh + 1);
991         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
992                 leh->leh_magic = LINK_EA_MAGIC;
993                 leh->leh_reccount = __swab32(leh->leh_reccount);
994                 leh->leh_len = __swab64(leh->leh_len);
995         }
996         if (leh->leh_magic != LINK_EA_MAGIC)
997                 RETURN(-EINVAL);
998         if (leh->leh_reccount == 0)
999                 RETURN(-ENODATA);
1000
1001         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1002         fid_be_to_cpu(pfid, pfid);
1003
1004         RETURN(0);
1005 }
1006
1007 int mdt_attr_get_complex(struct mdt_thread_info *info,
1008                          struct mdt_object *o, struct md_attr *ma)
1009 {
1010         const struct lu_env *env = info->mti_env;
1011         struct md_object    *next = mdt_object_child(o);
1012         struct lu_buf       *buf = &info->mti_buf;
1013         int                  need = ma->ma_need;
1014         int                  rc = 0, rc2;
1015         u32                  mode;
1016         ENTRY;
1017
1018         ma->ma_valid = 0;
1019
1020         if (mdt_object_exists(o) == 0)
1021                 GOTO(out, rc = -ENOENT);
1022         mode = lu_object_attr(&next->mo_lu);
1023
1024         if (need & MA_INODE) {
1025                 ma->ma_need = MA_INODE;
1026                 rc = mo_attr_get(env, next, ma);
1027                 if (rc)
1028                         GOTO(out, rc);
1029
1030                 if (S_ISREG(mode))
1031                         (void) mdt_get_som(info, o, &ma->ma_attr);
1032                 ma->ma_valid |= MA_INODE;
1033         }
1034
1035         if (need & MA_PFID) {
1036                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1037                 if (rc == 0)
1038                         ma->ma_valid |= MA_PFID;
1039                 /* ignore this error, parent fid is not mandatory */
1040                 rc = 0;
1041         }
1042
1043         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1044                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1045                 if (rc)
1046                         GOTO(out, rc);
1047         }
1048
1049         if (need & MA_LMV && S_ISDIR(mode)) {
1050                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1051                 if (rc != 0)
1052                         GOTO(out, rc);
1053         }
1054
1055         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1056                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1057                 if (rc != 0)
1058                         GOTO(out, rc);
1059         }
1060
1061         if (need & MA_HSM && S_ISREG(mode)) {
1062                 buf->lb_buf = info->mti_xattr_buf;
1063                 buf->lb_len = sizeof(info->mti_xattr_buf);
1064                 CLASSERT(sizeof(struct hsm_attrs) <=
1065                          sizeof(info->mti_xattr_buf));
1066                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1067                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1068                 if (rc2 == 0)
1069                         ma->ma_valid |= MA_HSM;
1070                 else if (rc2 < 0 && rc2 != -ENODATA)
1071                         GOTO(out, rc = rc2);
1072         }
1073
1074 #ifdef CONFIG_FS_POSIX_ACL
1075         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1076                 buf->lb_buf = ma->ma_acl;
1077                 buf->lb_len = ma->ma_acl_size;
1078                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1079                 if (rc2 > 0) {
1080                         ma->ma_acl_size = rc2;
1081                         ma->ma_valid |= MA_ACL_DEF;
1082                 } else if (rc2 == -ENODATA) {
1083                         /* no ACLs */
1084                         ma->ma_acl_size = 0;
1085                 } else
1086                         GOTO(out, rc = rc2);
1087         }
1088 #endif
1089 out:
1090         ma->ma_need = need;
1091         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1092                rc, ma->ma_valid, ma->ma_lmm);
1093         RETURN(rc);
1094 }
1095
1096 static int mdt_getattr_internal(struct mdt_thread_info *info,
1097                                 struct mdt_object *o, int ma_need)
1098 {
1099         struct md_object        *next = mdt_object_child(o);
1100         const struct mdt_body   *reqbody = info->mti_body;
1101         struct ptlrpc_request   *req = mdt_info_req(info);
1102         struct md_attr          *ma = &info->mti_attr;
1103         struct lu_attr          *la = &ma->ma_attr;
1104         struct req_capsule      *pill = info->mti_pill;
1105         const struct lu_env     *env = info->mti_env;
1106         struct mdt_body         *repbody;
1107         struct lu_buf           *buffer = &info->mti_buf;
1108         struct obd_export       *exp = info->mti_exp;
1109         int                      rc;
1110         ENTRY;
1111
1112         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1113                 RETURN(err_serious(-ENOMEM));
1114
1115         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1116
1117         ma->ma_valid = 0;
1118
1119         if (mdt_object_remote(o)) {
1120                 /* This object is located on remote node.*/
1121                 /* Return -ENOTSUPP for old client */
1122                 if (!mdt_is_dne_client(req->rq_export))
1123                         GOTO(out, rc = -ENOTSUPP);
1124
1125                 repbody->mbo_fid1 = *mdt_object_fid(o);
1126                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1127                 GOTO(out, rc = 0);
1128         }
1129
1130         if (reqbody->mbo_eadatasize > 0) {
1131                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1132                 if (buffer->lb_buf == NULL)
1133                         GOTO(out, rc = -EPROTO);
1134                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1135                                                       RCL_SERVER);
1136         } else {
1137                 buffer->lb_buf = NULL;
1138                 buffer->lb_len = 0;
1139                 ma_need &= ~(MA_LOV | MA_LMV);
1140                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1141                        mdt_obd_name(info->mti_mdt),
1142                        req->rq_export->exp_client_uuid.uuid);
1143         }
1144
1145         /* If it is dir object and client require MEA, then we got MEA */
1146         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1147             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1148                 /* Assumption: MDT_MD size is enough for lmv size. */
1149                 ma->ma_lmv = buffer->lb_buf;
1150                 ma->ma_lmv_size = buffer->lb_len;
1151                 ma->ma_need = MA_INODE;
1152                 if (ma->ma_lmv_size > 0) {
1153                         if (reqbody->mbo_valid & OBD_MD_MEA)
1154                                 ma->ma_need |= MA_LMV;
1155                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1156                                 ma->ma_need |= MA_LMV_DEF;
1157                 }
1158         } else {
1159                 ma->ma_lmm = buffer->lb_buf;
1160                 ma->ma_lmm_size = buffer->lb_len;
1161                 ma->ma_need = MA_INODE | MA_HSM;
1162                 if (ma->ma_lmm_size > 0)
1163                         ma->ma_need |= MA_LOV;
1164         }
1165
1166         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1167             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1168             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1169                 /* get default stripe info for this dir. */
1170                 ma->ma_need |= MA_LOV_DEF;
1171         }
1172         ma->ma_need |= ma_need;
1173
1174         rc = mdt_attr_get_complex(info, o, ma);
1175         if (unlikely(rc)) {
1176                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1177                        "%s: getattr error for "DFID": rc = %d\n",
1178                        mdt_obd_name(info->mti_mdt),
1179                        PFID(mdt_object_fid(o)), rc);
1180                 RETURN(rc);
1181         }
1182
1183         /* if file is released, check if a restore is running */
1184         if (ma->ma_valid & MA_HSM) {
1185                 repbody->mbo_valid |= OBD_MD_TSTATE;
1186                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1187                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1188                         repbody->mbo_t_state = MS_RESTORE;
1189         }
1190
1191         if (likely(ma->ma_valid & MA_INODE))
1192                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1193         else
1194                 RETURN(-EFAULT);
1195
1196         if (mdt_body_has_lov(la, reqbody)) {
1197                 if (ma->ma_valid & MA_LOV) {
1198                         LASSERT(ma->ma_lmm_size);
1199                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1200                         if (S_ISDIR(la->la_mode))
1201                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1202                         else
1203                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1204                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1205                 }
1206                 if (ma->ma_valid & MA_LMV) {
1207                         /* Return -ENOTSUPP for old client */
1208                         if (!mdt_is_striped_client(req->rq_export))
1209                                 RETURN(-ENOTSUPP);
1210
1211                         LASSERT(S_ISDIR(la->la_mode));
1212                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1213                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1214                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1215                 }
1216                 if (ma->ma_valid & MA_LMV_DEF) {
1217                         /* Return -ENOTSUPP for old client */
1218                         if (!mdt_is_striped_client(req->rq_export))
1219                                 RETURN(-ENOTSUPP);
1220                         LASSERT(S_ISDIR(la->la_mode));
1221                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1222                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1223                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1224                                                OBD_MD_DEFAULT_MEA);
1225                 }
1226         } else if (S_ISLNK(la->la_mode) &&
1227                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1228                 buffer->lb_buf = ma->ma_lmm;
1229                 /* eadatasize from client includes NULL-terminator, so
1230                  * there is no need to read it */
1231                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1232                 rc = mo_readlink(env, next, buffer);
1233                 if (unlikely(rc <= 0)) {
1234                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1235                                mdt_obd_name(info->mti_mdt),
1236                                PFID(mdt_object_fid(o)), rc);
1237                         rc = -EFAULT;
1238                 } else {
1239                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1240
1241                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1242                                 rc -= 2;
1243                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1244                         /* we need to report back size with NULL-terminator
1245                          * because client expects that */
1246                         repbody->mbo_eadatasize = rc + 1;
1247                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1248                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1249                                        "on "DFID ", expected %d\n",
1250                                        mdt_obd_name(info->mti_mdt),
1251                                        rc, PFID(mdt_object_fid(o)),
1252                                        reqbody->mbo_eadatasize - 1);
1253                         /* NULL terminate */
1254                         ((char *)ma->ma_lmm)[rc] = 0;
1255
1256                         /* If the total CDEBUG() size is larger than a page, it
1257                          * will print a warning to the console, avoid this by
1258                          * printing just the last part of the symlink. */
1259                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1260                                print_limit < rc ? "..." : "", print_limit,
1261                                (char *)ma->ma_lmm + rc - print_limit, rc);
1262                         rc = 0;
1263                 }
1264         }
1265
1266         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1267                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1268                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1269                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1270                        repbody->mbo_max_mdsize);
1271         }
1272
1273 #ifdef CONFIG_FS_POSIX_ACL
1274         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1275                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1276                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1277                 if (IS_ERR(nodemap))
1278                         RETURN(PTR_ERR(nodemap));
1279
1280                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1281                 nodemap_putref(nodemap);
1282         }
1283 #endif
1284
1285 out:
1286         if (rc == 0)
1287                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1288
1289         RETURN(rc);
1290 }
1291
1292 static int mdt_getattr(struct tgt_session_info *tsi)
1293 {
1294         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1295         struct mdt_object       *obj = info->mti_object;
1296         struct req_capsule      *pill = info->mti_pill;
1297         struct mdt_body         *reqbody;
1298         struct mdt_body         *repbody;
1299         int rc, rc2;
1300         ENTRY;
1301
1302         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1303         LASSERT(reqbody);
1304         LASSERT(obj != NULL);
1305         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1306
1307         /* Special case for Data-on-MDT files to get data version */
1308         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1309                 rc = mdt_data_version_get(tsi);
1310                 GOTO(out, rc);
1311         }
1312
1313         /* Unlike intent case where we need to pre-fill out buffers early on
1314          * in intent policy for ldlm reasons, here we can have a much better
1315          * guess at EA size by just reading it from disk.
1316          * Exceptions are readdir and (missing) directory striping */
1317         /* Readlink */
1318         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1319                 /* No easy way to know how long is the symlink, but it cannot
1320                  * be more than PATH_MAX, so we allocate +1 */
1321                 rc = PATH_MAX + 1;
1322         /* A special case for fs ROOT: getattr there might fetch
1323          * default EA for entire fs, not just for this dir!
1324          */
1325         } else if (lu_fid_eq(mdt_object_fid(obj),
1326                              &info->mti_mdt->mdt_md_root_fid) &&
1327                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1328                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1329                                                                  MDS_GETATTR)) {
1330                 /* Should the default strping be bigger, mdt_fix_reply
1331                  * will reallocate */
1332                 rc = DEF_REP_MD_SIZE;
1333         } else {
1334                 /* Read the actual EA size from disk */
1335                 rc = mdt_attr_get_eabuf_size(info, obj);
1336         }
1337
1338         if (rc < 0)
1339                 GOTO(out, rc = err_serious(rc));
1340
1341         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1342
1343         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1344          * by default. If the target object has more ACL entries, then
1345          * enlarge the buffer when necessary. */
1346         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1347                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1348
1349         rc = req_capsule_server_pack(pill);
1350         if (unlikely(rc != 0))
1351                 GOTO(out, rc = err_serious(rc));
1352
1353         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1354         LASSERT(repbody != NULL);
1355         repbody->mbo_eadatasize = 0;
1356         repbody->mbo_aclsize = 0;
1357
1358         rc = mdt_check_ucred(info);
1359         if (unlikely(rc))
1360                 GOTO(out_shrink, rc);
1361
1362         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1363
1364         rc = mdt_getattr_internal(info, obj, 0);
1365         EXIT;
1366 out_shrink:
1367         mdt_client_compatibility(info);
1368         rc2 = mdt_fix_reply(info);
1369         if (rc == 0)
1370                 rc = rc2;
1371 out:
1372         mdt_thread_info_fini(info);
1373         return rc;
1374 }
1375
1376 /**
1377  * Handler of layout intent RPC requiring the layout modification
1378  *
1379  * \param[in] info      thread environment
1380  * \param[in] obj       object
1381  * \param[in] layout    layout change descriptor
1382  *
1383  * \retval 0    on success
1384  * \retval < 0  error code
1385  */
1386 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1387                       struct md_layout_change *layout)
1388 {
1389         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1390         int rc;
1391         ENTRY;
1392
1393         if (!mdt_object_exists(obj))
1394                 GOTO(out, rc = -ENOENT);
1395
1396         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1397                 GOTO(out, rc = -EINVAL);
1398
1399         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1400                            MAY_WRITE);
1401         if (rc)
1402                 GOTO(out, rc);
1403
1404         /* take layout lock to prepare layout change */
1405         mdt_lock_reg_init(lh, LCK_EX);
1406         rc = mdt_object_lock(info, obj, lh, MDS_INODELOCK_LAYOUT);
1407         if (rc)
1408                 GOTO(out, rc);
1409
1410         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1411
1412         mdt_object_unlock(info, obj, lh, 1);
1413 out:
1414         RETURN(rc);
1415 }
1416
1417 /**
1418  * Exchange MOF_LOV_CREATED flags between two objects after a
1419  * layout swap. No assumption is made on whether o1 or o2 have
1420  * created objects or not.
1421  *
1422  * \param[in,out] o1    First swap layout object
1423  * \param[in,out] o2    Second swap layout object
1424  */
1425 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1426 {
1427         unsigned int o1_lov_created = o1->mot_lov_created;
1428
1429         mutex_lock(&o1->mot_lov_mutex);
1430         mutex_lock(&o2->mot_lov_mutex);
1431
1432         o1->mot_lov_created = o2->mot_lov_created;
1433         o2->mot_lov_created = o1_lov_created;
1434
1435         mutex_unlock(&o2->mot_lov_mutex);
1436         mutex_unlock(&o1->mot_lov_mutex);
1437 }
1438
1439 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1440 {
1441         struct mdt_thread_info  *info;
1442         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1443         struct obd_export       *exp = req->rq_export;
1444         struct mdt_object       *o1, *o2, *o;
1445         struct mdt_lock_handle  *lh1, *lh2;
1446         struct mdc_swap_layouts *msl;
1447         int                      rc;
1448         ENTRY;
1449
1450         /* client does not support layout lock, so layout swaping
1451          * is disabled.
1452          * FIXME: there is a problem for old clients which don't support
1453          * layout lock yet. If those clients have already opened the file
1454          * they won't be notified at all so that old layout may still be
1455          * used to do IO. This can be fixed after file release is landed by
1456          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1457         if (!exp_connect_layout(exp))
1458                 RETURN(-EOPNOTSUPP);
1459
1460         info = tsi2mdt_info(tsi);
1461
1462         if (info->mti_dlm_req != NULL)
1463                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1464
1465         o1 = info->mti_object;
1466         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1467                                 &info->mti_body->mbo_fid2);
1468         if (IS_ERR(o))
1469                 GOTO(out, rc = PTR_ERR(o));
1470
1471         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1472                 GOTO(put, rc = -ENOENT);
1473
1474         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1475         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1476                 GOTO(put, rc);
1477
1478         if (rc < 0)
1479                 swap(o1, o2);
1480
1481         /* permission check. Make sure the calling process having permission
1482          * to write both files. */
1483         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1484                                 MAY_WRITE);
1485         if (rc < 0)
1486                 GOTO(put, rc);
1487
1488         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1489                                 MAY_WRITE);
1490         if (rc < 0)
1491                 GOTO(put, rc);
1492
1493         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1494         if (msl == NULL)
1495                 GOTO(put, rc = -EPROTO);
1496
1497         lh1 = &info->mti_lh[MDT_LH_NEW];
1498         mdt_lock_reg_init(lh1, LCK_EX);
1499         lh2 = &info->mti_lh[MDT_LH_OLD];
1500         mdt_lock_reg_init(lh2, LCK_EX);
1501
1502         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1503                              MDS_INODELOCK_XATTR);
1504         if (rc < 0)
1505                 GOTO(put, rc);
1506
1507         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1508                              MDS_INODELOCK_XATTR);
1509         if (rc < 0)
1510                 GOTO(unlock1, rc);
1511
1512         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1513                              mdt_object_child(o2), msl->msl_flags);
1514         if (rc < 0)
1515                 GOTO(unlock2, rc);
1516
1517         mdt_swap_lov_flag(o1, o2);
1518
1519 unlock2:
1520         mdt_object_unlock(info, o2, lh2, rc);
1521 unlock1:
1522         mdt_object_unlock(info, o1, lh1, rc);
1523 put:
1524         mdt_object_put(info->mti_env, o);
1525 out:
1526         mdt_thread_info_fini(info);
1527         RETURN(rc);
1528 }
1529
1530 static int mdt_raw_lookup(struct mdt_thread_info *info,
1531                           struct mdt_object *parent,
1532                           const struct lu_name *lname,
1533                           struct ldlm_reply *ldlm_rep)
1534 {
1535         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1536         int              rc;
1537         ENTRY;
1538
1539         LASSERT(!info->mti_cross_ref);
1540
1541         /* Only got the fid of this obj by name */
1542         fid_zero(child_fid);
1543         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1544                         lname, child_fid, &info->mti_spec);
1545         if (rc == 0) {
1546                 struct mdt_body *repbody;
1547
1548                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1549                 repbody->mbo_fid1 = *child_fid;
1550                 repbody->mbo_valid = OBD_MD_FLID;
1551                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1552         } else if (rc == -ENOENT) {
1553                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1554         }
1555
1556         RETURN(rc);
1557 }
1558
1559 /*
1560  * UPDATE lock should be taken against parent, and be released before exit;
1561  * child_bits lock should be taken against child, and be returned back:
1562  *            (1)normal request should release the child lock;
1563  *            (2)intent request will grant the lock to client.
1564  */
1565 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1566                                  struct mdt_lock_handle *lhc,
1567                                  __u64 child_bits,
1568                                  struct ldlm_reply *ldlm_rep)
1569 {
1570         struct ptlrpc_request  *req = mdt_info_req(info);
1571         struct mdt_body        *reqbody = NULL;
1572         struct mdt_object      *parent = info->mti_object;
1573         struct mdt_object      *child;
1574         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1575         struct lu_name         *lname = NULL;
1576         struct mdt_lock_handle *lhp = NULL;
1577         struct ldlm_lock       *lock;
1578         __u64 try_bits = 0;
1579         bool is_resent;
1580         int ma_need = 0;
1581         int rc;
1582
1583         ENTRY;
1584
1585         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1586         LASSERT(ergo(is_resent,
1587                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1588
1589         if (parent == NULL)
1590                 RETURN(-ENOENT);
1591
1592         if (info->mti_cross_ref) {
1593                 /* Only getattr on the child. Parent is on another node. */
1594                 mdt_set_disposition(info, ldlm_rep,
1595                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1596                 child = parent;
1597                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1598                        "ldlm_rep = %p\n",
1599                        PFID(mdt_object_fid(child)), ldlm_rep);
1600
1601                 rc = mdt_check_resent_lock(info, child, lhc);
1602                 if (rc < 0) {
1603                         RETURN(rc);
1604                 } else if (rc > 0) {
1605                         mdt_lock_handle_init(lhc);
1606                         mdt_lock_reg_init(lhc, LCK_PR);
1607
1608                         /*
1609                          * Object's name is on another MDS, no lookup or layout
1610                          * lock is needed here but update lock is.
1611                          */
1612                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1613                                         MDS_INODELOCK_LAYOUT);
1614                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1615
1616                         rc = mdt_object_lock(info, child, lhc, child_bits);
1617                         if (rc < 0)
1618                                 RETURN(rc);
1619                 }
1620
1621                 /* Finally, we can get attr for child. */
1622                 if (!mdt_object_exists(child)) {
1623                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1624                                         &child->mot_obj,
1625                                         "remote object doesn't exist.");
1626                         mdt_object_unlock(info, child, lhc, 1);
1627                         RETURN(-ENOENT);
1628                 }
1629
1630                 rc = mdt_getattr_internal(info, child, 0);
1631                 if (unlikely(rc != 0))
1632                         mdt_object_unlock(info, child, lhc, 1);
1633
1634                 RETURN(rc);
1635         }
1636
1637         lname = &info->mti_name;
1638         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1639
1640         if (lu_name_is_valid(lname)) {
1641                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1642                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1643                        PNAME(lname), ldlm_rep);
1644         } else {
1645                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1646                 if (unlikely(reqbody == NULL))
1647                         RETURN(err_serious(-EPROTO));
1648
1649                 *child_fid = reqbody->mbo_fid2;
1650
1651                 if (unlikely(!fid_is_sane(child_fid)))
1652                         RETURN(err_serious(-EINVAL));
1653
1654                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1655                        "ldlm_rep = %p\n",
1656                        PFID(mdt_object_fid(parent)),
1657                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1658         }
1659
1660         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1661
1662         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1663                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1664                                 &parent->mot_obj,
1665                                 "Parent doesn't exist!");
1666                 RETURN(-ESTALE);
1667         }
1668
1669         if (mdt_object_remote(parent)) {
1670                 CERROR("%s: parent "DFID" is on remote target\n",
1671                        mdt_obd_name(info->mti_mdt),
1672                        PFID(mdt_object_fid(parent)));
1673                 RETURN(-EIO);
1674         }
1675
1676         if (lu_name_is_valid(lname)) {
1677                 /* Always allow to lookup ".." */
1678                 if (unlikely(lname->ln_namelen == 2 &&
1679                              lname->ln_name[0] == '.' &&
1680                              lname->ln_name[1] == '.'))
1681                         info->mti_spec.sp_permitted = 1;
1682
1683                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1684                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1685
1686                         RETURN(rc);
1687                 }
1688
1689                 /* step 1: lock parent only if parent is a directory */
1690                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1691                         lhp = &info->mti_lh[MDT_LH_PARENT];
1692                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1693                         rc = mdt_object_lock(info, parent, lhp,
1694                                              MDS_INODELOCK_UPDATE);
1695                         if (unlikely(rc != 0))
1696                                 RETURN(rc);
1697                 }
1698
1699                 /* step 2: lookup child's fid by name */
1700                 fid_zero(child_fid);
1701                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1702                                 child_fid, &info->mti_spec);
1703                 if (rc == -ENOENT)
1704                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1705
1706                 if (rc != 0)
1707                         GOTO(out_parent, rc);
1708         }
1709
1710         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1711
1712         /*
1713          *step 3: find the child object by fid & lock it.
1714          *        regardless if it is local or remote.
1715          *
1716          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1717          *      set parent dir fid the same as child fid in getattr by fid case
1718          *      we should not lu_object_find() the object again, could lead
1719          *      to hung if there is a concurrent unlink destroyed the object.
1720          */
1721         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1722                 mdt_object_get(info->mti_env, parent);
1723                 child = parent;
1724         } else {
1725                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1726                                         child_fid);
1727         }
1728
1729         if (unlikely(IS_ERR(child)))
1730                 GOTO(out_parent, rc = PTR_ERR(child));
1731
1732         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1733         if (!mdt_object_exists(child)) {
1734                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1735                                 &child->mot_obj,
1736                                 "Object doesn't exist!");
1737                 GOTO(out_child, rc = -ENOENT);
1738         }
1739
1740         rc = mdt_check_resent_lock(info, child, lhc);
1741         if (rc < 0) {
1742                 GOTO(out_child, rc);
1743         } else if (rc > 0) {
1744                 mdt_lock_handle_init(lhc);
1745                 mdt_lock_reg_init(lhc, LCK_PR);
1746
1747                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1748                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1749                         struct md_attr *ma = &info->mti_attr;
1750
1751                         ma->ma_valid = 0;
1752                         ma->ma_need = MA_INODE;
1753                         rc = mdt_attr_get_complex(info, child, ma);
1754                         if (unlikely(rc != 0))
1755                                 GOTO(out_child, rc);
1756
1757                         /* If the file has not been changed for some time, we
1758                          * return not only a LOOKUP lock, but also an UPDATE
1759                          * lock and this might save us RPC on later STAT. For
1760                          * directories, it also let negative dentry cache start
1761                          * working for this dir. */
1762                         if (ma->ma_valid & MA_INODE &&
1763                             ma->ma_attr.la_valid & LA_CTIME &&
1764                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1765                                 ma->ma_attr.la_ctime < ktime_get_real_seconds())
1766                                 child_bits |= MDS_INODELOCK_UPDATE;
1767                 }
1768
1769                 /* layout lock must be granted in a best-effort way
1770                  * for IT operations */
1771                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1772                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1773                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1774                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1775                             exp_connect_layout(info->mti_exp)) {
1776                                 /* try to grant layout lock for regular file. */
1777                                 try_bits = MDS_INODELOCK_LAYOUT;
1778                         }
1779                         /* Acquire DOM lock in advance for data-on-mdt file */
1780                         if (child != parent)
1781                                 try_bits |= MDS_INODELOCK_DOM;
1782                 }
1783
1784                 if (try_bits != 0) {
1785                         /* try layout lock, it may fail to be granted due to
1786                          * contention at LOOKUP or UPDATE */
1787                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1788                                                  try_bits, false);
1789                         if (child_bits & MDS_INODELOCK_LAYOUT)
1790                                 ma_need |= MA_LOV;
1791                 } else {
1792                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1793                          * client will enqueue the lock to the remote MDT */
1794                         if (mdt_object_remote(child))
1795                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1796                         rc = mdt_object_lock(info, child, lhc, child_bits);
1797                 }
1798                 if (unlikely(rc != 0))
1799                         GOTO(out_child, rc);
1800         }
1801
1802         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1803
1804         /* finally, we can get attr for child. */
1805         rc = mdt_getattr_internal(info, child, ma_need);
1806         if (unlikely(rc != 0)) {
1807                 mdt_object_unlock(info, child, lhc, 1);
1808         } else if (lock) {
1809                 /* Debugging code. */
1810                 LDLM_DEBUG(lock, "Returning lock to client");
1811                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1812                                          &lock->l_resource->lr_name),
1813                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1814                          PLDLMRES(lock->l_resource),
1815                          PFID(mdt_object_fid(child)));
1816
1817                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1818                     mdt_object_exists(child) && !mdt_object_remote(child) &&
1819                     child != parent) {
1820                         LDLM_LOCK_PUT(lock);
1821                         mdt_object_put(info->mti_env, child);
1822                         /* NB: call the mdt_pack_size2body always after
1823                          * mdt_object_put(), that is why this speacial
1824                          * exit path is used. */
1825                         rc = mdt_pack_size2body(info, child_fid,
1826                                                 child_bits & MDS_INODELOCK_DOM);
1827                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1828                                 /* DOM lock was taken in advance but this is
1829                                  * not DoM file. Drop the lock. */
1830                                 lock_res_and_lock(lock);
1831                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1832                                 unlock_res_and_lock(lock);
1833                         }
1834
1835                         GOTO(out_parent, rc = 0);
1836                 }
1837         }
1838         if (lock)
1839                 LDLM_LOCK_PUT(lock);
1840
1841         EXIT;
1842 out_child:
1843         mdt_object_put(info->mti_env, child);
1844 out_parent:
1845         if (lhp)
1846                 mdt_object_unlock(info, parent, lhp, 1);
1847         return rc;
1848 }
1849
1850 /* normal handler: should release the child lock */
1851 static int mdt_getattr_name(struct tgt_session_info *tsi)
1852 {
1853         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1854         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1855         struct mdt_body        *reqbody;
1856         struct mdt_body        *repbody;
1857         int rc, rc2;
1858         ENTRY;
1859
1860         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1861         LASSERT(reqbody != NULL);
1862         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1863         LASSERT(repbody != NULL);
1864
1865         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1866         repbody->mbo_eadatasize = 0;
1867         repbody->mbo_aclsize = 0;
1868
1869         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1870         if (unlikely(rc))
1871                 GOTO(out_shrink, rc);
1872
1873         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1874         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1875                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1876                 lhc->mlh_reg_lh.cookie = 0;
1877         }
1878         mdt_exit_ucred(info);
1879         EXIT;
1880 out_shrink:
1881         mdt_client_compatibility(info);
1882         rc2 = mdt_fix_reply(info);
1883         if (rc == 0)
1884                 rc = rc2;
1885         mdt_thread_info_fini(info);
1886         return rc;
1887 }
1888
1889 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1890                          void *karg, void __user *uarg);
1891
1892 static int mdt_set_info(struct tgt_session_info *tsi)
1893 {
1894         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1895         char                    *key;
1896         void                    *val;
1897         int                      keylen, vallen, rc = 0;
1898
1899         ENTRY;
1900
1901         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1902         if (key == NULL) {
1903                 DEBUG_REQ(D_HA, req, "no set_info key");
1904                 RETURN(err_serious(-EFAULT));
1905         }
1906
1907         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1908                                       RCL_CLIENT);
1909
1910         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1911         if (val == NULL) {
1912                 DEBUG_REQ(D_HA, req, "no set_info val");
1913                 RETURN(err_serious(-EFAULT));
1914         }
1915
1916         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1917                                       RCL_CLIENT);
1918
1919         /* Swab any part of val you need to here */
1920         if (KEY_IS(KEY_READ_ONLY)) {
1921                 spin_lock(&req->rq_export->exp_lock);
1922                 if (*(__u32 *)val)
1923                         *exp_connect_flags_ptr(req->rq_export) |=
1924                                 OBD_CONNECT_RDONLY;
1925                 else
1926                         *exp_connect_flags_ptr(req->rq_export) &=
1927                                 ~OBD_CONNECT_RDONLY;
1928                 spin_unlock(&req->rq_export->exp_lock);
1929         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1930                 struct changelog_setinfo *cs = val;
1931
1932                 if (vallen != sizeof(*cs)) {
1933                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1934                                tgt_name(tsi->tsi_tgt), vallen);
1935                         RETURN(-EINVAL);
1936                 }
1937                 if (ptlrpc_req_need_swab(req)) {
1938                         __swab64s(&cs->cs_recno);
1939                         __swab32s(&cs->cs_id);
1940                 }
1941
1942                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1943                                    vallen, val, NULL);
1944         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1945                 if (vallen > 0)
1946                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1947         } else {
1948                 RETURN(-EINVAL);
1949         }
1950         RETURN(rc);
1951 }
1952
1953 static int mdt_readpage(struct tgt_session_info *tsi)
1954 {
1955         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1956         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1957         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1958         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1959         struct mdt_body         *repbody;
1960         int                      rc;
1961         int                      i;
1962
1963         ENTRY;
1964
1965         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1966                 RETURN(err_serious(-ENOMEM));
1967
1968         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1969         if (repbody == NULL || reqbody == NULL)
1970                 RETURN(err_serious(-EFAULT));
1971
1972         /*
1973          * prepare @rdpg before calling lower layers and transfer itself. Here
1974          * reqbody->size contains offset of where to start to read and
1975          * reqbody->nlink contains number bytes to read.
1976          */
1977         rdpg->rp_hash = reqbody->mbo_size;
1978         if (rdpg->rp_hash != reqbody->mbo_size) {
1979                 CERROR("Invalid hash: %#llx != %#llx\n",
1980                        rdpg->rp_hash, reqbody->mbo_size);
1981                 RETURN(-EFAULT);
1982         }
1983
1984         rdpg->rp_attrs = reqbody->mbo_mode;
1985         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1986                 rdpg->rp_attrs |= LUDA_64BITHASH;
1987         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1988                                 exp_max_brw_size(tsi->tsi_exp));
1989         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
1990                           PAGE_SHIFT;
1991         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1992         if (rdpg->rp_pages == NULL)
1993                 RETURN(-ENOMEM);
1994
1995         for (i = 0; i < rdpg->rp_npages; ++i) {
1996                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
1997                 if (rdpg->rp_pages[i] == NULL)
1998                         GOTO(free_rdpg, rc = -ENOMEM);
1999         }
2000
2001         /* call lower layers to fill allocated pages with directory data */
2002         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2003         if (rc < 0)
2004                 GOTO(free_rdpg, rc);
2005
2006         /* send pages to client */
2007         rc = tgt_sendpage(tsi, rdpg, rc);
2008
2009         EXIT;
2010 free_rdpg:
2011
2012         for (i = 0; i < rdpg->rp_npages; i++)
2013                 if (rdpg->rp_pages[i] != NULL)
2014                         __free_page(rdpg->rp_pages[i]);
2015         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2016
2017         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2018                 RETURN(0);
2019
2020         return rc;
2021 }
2022
2023 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2024 {
2025         struct lu_ucred *uc = mdt_ucred_check(info);
2026         struct lu_attr *attr = &info->mti_attr.ma_attr;
2027
2028         if (uc == NULL)
2029                 return -EINVAL;
2030
2031         if (op != REINT_SETATTR) {
2032                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2033                         attr->la_uid = uc->uc_fsuid;
2034                 /* for S_ISGID, inherit gid from his parent, such work will be
2035                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2036                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2037                         attr->la_gid = uc->uc_fsgid;
2038         }
2039
2040         return 0;
2041 }
2042
2043 static int mdt_reint_internal(struct mdt_thread_info *info,
2044                               struct mdt_lock_handle *lhc,
2045                               __u32 op)
2046 {
2047         struct req_capsule      *pill = info->mti_pill;
2048         struct mdt_body         *repbody;
2049         int                      rc = 0, rc2;
2050
2051         ENTRY;
2052
2053         rc = mdt_reint_unpack(info, op);
2054         if (rc != 0) {
2055                 CERROR("Can't unpack reint, rc %d\n", rc);
2056                 RETURN(err_serious(rc));
2057         }
2058
2059         /* for replay (no_create) lmm is not needed, client has it already */
2060         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2061                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2062                                      DEF_REP_MD_SIZE);
2063
2064         /* llog cookies are always 0, the field is kept for compatibility */
2065         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2066                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2067
2068         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2069          * by default. If the target object has more ACL entries, then
2070          * enlarge the buffer when necessary. */
2071         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2072                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2073                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2074
2075         rc = req_capsule_server_pack(pill);
2076         if (rc != 0) {
2077                 CERROR("Can't pack response, rc %d\n", rc);
2078                 RETURN(err_serious(rc));
2079         }
2080
2081         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2082                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2083                 LASSERT(repbody);
2084                 repbody->mbo_eadatasize = 0;
2085                 repbody->mbo_aclsize = 0;
2086         }
2087
2088         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2089
2090         /* for replay no cookkie / lmm need, because client have this already */
2091         if (info->mti_spec.no_create)
2092                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2093                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2094
2095         rc = mdt_init_ucred_reint(info);
2096         if (rc)
2097                 GOTO(out_shrink, rc);
2098
2099         rc = mdt_fix_attr_ucred(info, op);
2100         if (rc != 0)
2101                 GOTO(out_ucred, rc = err_serious(rc));
2102
2103         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2104         if (rc < 0) {
2105                 GOTO(out_ucred, rc);
2106         } else if (rc == 1) {
2107                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
2108                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2109                 GOTO(out_ucred, rc);
2110         }
2111         rc = mdt_reint_rec(info, lhc);
2112         EXIT;
2113 out_ucred:
2114         mdt_exit_ucred(info);
2115 out_shrink:
2116         mdt_client_compatibility(info);
2117         rc2 = mdt_fix_reply(info);
2118         if (rc == 0)
2119                 rc = rc2;
2120         return rc;
2121 }
2122
2123 static long mdt_reint_opcode(struct ptlrpc_request *req,
2124                              const struct req_format **fmt)
2125 {
2126         struct mdt_device       *mdt;
2127         struct mdt_rec_reint    *rec;
2128         long                     opc;
2129
2130         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2131         if (rec != NULL) {
2132                 opc = rec->rr_opcode;
2133                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2134                 if (opc < REINT_MAX && fmt[opc] != NULL)
2135                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2136                 else {
2137                         mdt = mdt_exp2dev(req->rq_export);
2138                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2139                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2140                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2141                         opc = err_serious(-EFAULT);
2142                 }
2143         } else {
2144                 opc = err_serious(-EFAULT);
2145         }
2146         return opc;
2147 }
2148
2149 static int mdt_reint(struct tgt_session_info *tsi)
2150 {
2151         long opc;
2152         int  rc;
2153         static const struct req_format *reint_fmts[REINT_MAX] = {
2154                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2155                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2156                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2157                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2158                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2159                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2160                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2161                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2162                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME,
2163                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2164         };
2165
2166         ENTRY;
2167
2168         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2169         if (opc >= 0) {
2170                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2171                 /*
2172                  * No lock possible here from client to pass it to reint code
2173                  * path.
2174                  */
2175                 rc = mdt_reint_internal(info, NULL, opc);
2176                 mdt_thread_info_fini(info);
2177         } else {
2178                 rc = opc;
2179         }
2180
2181         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2182         RETURN(rc);
2183 }
2184
2185 /* this should sync the whole device */
2186 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2187 {
2188         struct dt_device *dt = mdt->mdt_bottom;
2189         int rc;
2190         ENTRY;
2191
2192         rc = dt->dd_ops->dt_sync(env, dt);
2193         RETURN(rc);
2194 }
2195
2196 /* this should sync this object */
2197 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2198                            struct mdt_object *mo)
2199 {
2200         int rc;
2201
2202         ENTRY;
2203
2204         if (!mdt_object_exists(mo)) {
2205                 CWARN("%s: non existing object "DFID": rc = %d\n",
2206                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2207                       -ESTALE);
2208                 RETURN(-ESTALE);
2209         }
2210
2211         rc = mo_object_sync(env, mdt_object_child(mo));
2212
2213         RETURN(rc);
2214 }
2215
2216 static int mdt_sync(struct tgt_session_info *tsi)
2217 {
2218         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2219         struct req_capsule      *pill = tsi->tsi_pill;
2220         struct mdt_body         *body;
2221         int                      rc;
2222
2223         ENTRY;
2224
2225         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2226                 RETURN(err_serious(-ENOMEM));
2227
2228         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2229                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2230         } else {
2231                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2232
2233                 /* sync an object */
2234                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2235                                      info->mti_object);
2236                 if (rc == 0) {
2237                         const struct lu_fid *fid;
2238                         struct lu_attr *la = &info->mti_attr.ma_attr;
2239
2240                         info->mti_attr.ma_need = MA_INODE;
2241                         info->mti_attr.ma_valid = 0;
2242                         rc = mdt_attr_get_complex(info, info->mti_object,
2243                                                   &info->mti_attr);
2244                         if (rc == 0) {
2245                                 body = req_capsule_server_get(pill,
2246                                                               &RMF_MDT_BODY);
2247                                 fid = mdt_object_fid(info->mti_object);
2248                                 mdt_pack_attr2body(info, body, la, fid);
2249                         }
2250                 }
2251                 mdt_thread_info_fini(info);
2252         }
2253         if (rc == 0)
2254                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2255
2256         RETURN(rc);
2257 }
2258
2259 static int mdt_data_sync(struct tgt_session_info *tsi)
2260 {
2261         struct mdt_thread_info *info;
2262         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2263         struct ost_body *body = tsi->tsi_ost_body;
2264         struct ost_body *repbody;
2265         struct mdt_object *mo = NULL;
2266         struct md_attr *ma;
2267         int rc = 0;
2268
2269         ENTRY;
2270
2271         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2272
2273         /* if no fid is specified then do nothing,
2274          * device sync is done via MDS_SYNC */
2275         if (fid_is_zero(&tsi->tsi_fid))
2276                 RETURN(0);
2277
2278         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2279         if (IS_ERR(mo))
2280                 RETURN(PTR_ERR(mo));
2281
2282         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2283         if (rc)
2284                 GOTO(put, rc);
2285
2286         repbody->oa.o_oi = body->oa.o_oi;
2287         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2288
2289         info = tsi2mdt_info(tsi);
2290         ma = &info->mti_attr;
2291         ma->ma_need = MA_INODE;
2292         ma->ma_valid = 0;
2293         rc = mdt_attr_get_complex(info, mo, ma);
2294         if (rc == 0)
2295                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2296         else
2297                 rc = 0;
2298         mdt_thread_info_fini(info);
2299
2300         EXIT;
2301 put:
2302         if (mo != NULL)
2303                 mdt_object_put(tsi->tsi_env, mo);
2304         return rc;
2305 }
2306
2307 /*
2308  * Handle quota control requests to consult current usage/limit, but also
2309  * to configure quota enforcement
2310  */
2311 static int mdt_quotactl(struct tgt_session_info *tsi)
2312 {
2313         struct obd_export       *exp  = tsi->tsi_exp;
2314         struct req_capsule      *pill = tsi->tsi_pill;
2315         struct obd_quotactl     *oqctl, *repoqc;
2316         int                      id, rc;
2317         struct mdt_device       *mdt = mdt_exp2dev(exp);
2318         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2319         struct lu_nodemap       *nodemap;
2320         ENTRY;
2321
2322         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2323         if (oqctl == NULL)
2324                 RETURN(err_serious(-EPROTO));
2325
2326         rc = req_capsule_server_pack(pill);
2327         if (rc)
2328                 RETURN(err_serious(rc));
2329
2330         nodemap = nodemap_get_from_exp(exp);
2331         if (IS_ERR(nodemap))
2332                 RETURN(PTR_ERR(nodemap));
2333
2334         switch (oqctl->qc_cmd) {
2335                 /* master quotactl */
2336         case Q_SETINFO:
2337         case Q_SETQUOTA:
2338                 if (!nodemap_can_setquota(nodemap))
2339                         GOTO(out_nodemap, rc = -EPERM);
2340         case Q_GETINFO:
2341         case Q_GETQUOTA:
2342                 if (qmt == NULL)
2343                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2344                 /* slave quotactl */
2345         case Q_GETOINFO:
2346         case Q_GETOQUOTA:
2347                 break;
2348         default:
2349                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2350                 GOTO(out_nodemap, rc = -EFAULT);
2351         }
2352
2353         id = oqctl->qc_id;
2354         switch (oqctl->qc_type) {
2355         case USRQUOTA:
2356                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2357                                     NODEMAP_CLIENT_TO_FS, id);
2358                 break;
2359         case GRPQUOTA:
2360                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2361                                     NODEMAP_CLIENT_TO_FS, id);
2362                 break;
2363         case PRJQUOTA:
2364                 /* todo: check/map project id */
2365                 id = oqctl->qc_id;
2366                 break;
2367         default:
2368                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2369         }
2370         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2371         if (repoqc == NULL)
2372                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2373
2374         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2375                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2376
2377         if (oqctl->qc_id != id)
2378                 swap(oqctl->qc_id, id);
2379
2380         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2381                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2382                         RETURN(-EINPROGRESS);
2383         }
2384
2385         switch (oqctl->qc_cmd) {
2386
2387         case Q_GETINFO:
2388         case Q_SETINFO:
2389         case Q_SETQUOTA:
2390         case Q_GETQUOTA:
2391                 /* forward quotactl request to QMT */
2392                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2393                 break;
2394
2395         case Q_GETOINFO:
2396         case Q_GETOQUOTA:
2397                 /* slave quotactl */
2398                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2399                                    oqctl);
2400                 break;
2401
2402         default:
2403                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2404                 GOTO(out_nodemap, rc = -EFAULT);
2405         }
2406
2407         if (oqctl->qc_id != id)
2408                 swap(oqctl->qc_id, id);
2409
2410         *repoqc = *oqctl;
2411
2412         EXIT;
2413
2414 out_nodemap:
2415         nodemap_putref(nodemap);
2416
2417         return rc;
2418 }
2419
2420 /** clone llog ctxt from child (mdd)
2421  * This allows remote llog (replicator) access.
2422  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2423  * context was originally set up, or we can handle them directly.
2424  * I choose the latter, but that means I need any llog
2425  * contexts set up by child to be accessable by the mdt.  So we clone the
2426  * context into our context list here.
2427  */
2428 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2429                                int idx)
2430 {
2431         struct md_device  *next = mdt->mdt_child;
2432         struct llog_ctxt *ctxt;
2433         int rc;
2434
2435         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2436                 return 0;
2437
2438         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2439         if (rc || ctxt == NULL) {
2440                 return 0;
2441         }
2442
2443         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2444         if (rc)
2445                 CERROR("Can't set mdt ctxt %d\n", rc);
2446
2447         return rc;
2448 }
2449
2450 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2451                                  struct mdt_device *mdt, int idx)
2452 {
2453         struct llog_ctxt *ctxt;
2454
2455         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2456         if (ctxt == NULL)
2457                 return 0;
2458         /* Put once for the get we just did, and once for the clone */
2459         llog_ctxt_put(ctxt);
2460         llog_ctxt_put(ctxt);
2461         return 0;
2462 }
2463
2464 /*
2465  * sec context handlers
2466  */
2467 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2468 {
2469         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2470
2471         return 0;
2472 }
2473
2474 /*
2475  * quota request handlers
2476  */
2477 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2478 {
2479         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2480         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2481         int                      rc;
2482         ENTRY;
2483
2484         if (qmt == NULL)
2485                 RETURN(err_serious(-EOPNOTSUPP));
2486
2487         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2488         RETURN(rc);
2489 }
2490
2491 struct mdt_object *mdt_object_new(const struct lu_env *env,
2492                                   struct mdt_device *d,
2493                                   const struct lu_fid *f)
2494 {
2495         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2496         struct lu_object *o;
2497         struct mdt_object *m;
2498         ENTRY;
2499
2500         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2501         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2502         if (unlikely(IS_ERR(o)))
2503                 m = (struct mdt_object *)o;
2504         else
2505                 m = mdt_obj(o);
2506         RETURN(m);
2507 }
2508
2509 struct mdt_object *mdt_object_find(const struct lu_env *env,
2510                                    struct mdt_device *d,
2511                                    const struct lu_fid *f)
2512 {
2513         struct lu_object *o;
2514         struct mdt_object *m;
2515         ENTRY;
2516
2517         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2518         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2519         if (unlikely(IS_ERR(o)))
2520                 m = (struct mdt_object *)o;
2521         else
2522                 m = mdt_obj(o);
2523
2524         RETURN(m);
2525 }
2526
2527 /**
2528  * Asyncronous commit for mdt device.
2529  *
2530  * Pass asynchonous commit call down the MDS stack.
2531  *
2532  * \param env environment
2533  * \param mdt the mdt device
2534  */
2535 static void mdt_device_commit_async(const struct lu_env *env,
2536                                     struct mdt_device *mdt)
2537 {
2538         struct dt_device *dt = mdt->mdt_bottom;
2539         int rc;
2540         ENTRY;
2541
2542         rc = dt->dd_ops->dt_commit_async(env, dt);
2543         if (unlikely(rc != 0))
2544                 CWARN("%s: async commit start failed: rc = %d\n",
2545                       mdt_obd_name(mdt), rc);
2546         atomic_inc(&mdt->mdt_async_commit_count);
2547         EXIT;
2548 }
2549
2550 /**
2551  * Mark the lock as "synchonous".
2552  *
2553  * Mark the lock to deffer transaction commit to the unlock time.
2554  *
2555  * \param lock the lock to mark as "synchonous"
2556  *
2557  * \see mdt_is_lock_sync
2558  * \see mdt_save_lock
2559  */
2560 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2561 {
2562         lock->l_ast_data = (void*)1;
2563 }
2564
2565 /**
2566  * Check whehter the lock "synchonous" or not.
2567  *
2568  * \param lock the lock to check
2569  * \retval 1 the lock is "synchonous"
2570  * \retval 0 the lock isn't "synchronous"
2571  *
2572  * \see mdt_set_lock_sync
2573  * \see mdt_save_lock
2574  */
2575 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2576 {
2577         return lock->l_ast_data != NULL;
2578 }
2579
2580 /**
2581  * Blocking AST for mdt locks.
2582  *
2583  * Starts transaction commit if in case of COS lock conflict or
2584  * deffers such a commit to the mdt_save_lock.
2585  *
2586  * \param lock the lock which blocks a request or cancelling lock
2587  * \param desc unused
2588  * \param data unused
2589  * \param flag indicates whether this cancelling or blocking callback
2590  * \retval 0
2591  * \see ldlm_blocking_ast_nocheck
2592  */
2593 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2594                      void *data, int flag)
2595 {
2596         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2597         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2598         bool commit_async = false;
2599         int rc;
2600         ENTRY;
2601
2602         if (flag == LDLM_CB_CANCELING)
2603                 RETURN(0);
2604
2605         lock_res_and_lock(lock);
2606         if (lock->l_blocking_ast != mdt_blocking_ast) {
2607                 unlock_res_and_lock(lock);
2608                 RETURN(0);
2609         }
2610         /* There is no lock conflict if l_blocking_lock == NULL,
2611          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2612          * when the last reference to a local lock was released */
2613         if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2614             lock->l_blocking_lock != NULL) {
2615                 if (mdt_cos_is_enabled(mdt)) {
2616                         if (lock->l_client_cookie !=
2617                             lock->l_blocking_lock->l_client_cookie)
2618                                 mdt_set_lock_sync(lock);
2619                 } else if (mdt_slc_is_enabled(mdt) &&
2620                            ldlm_is_cos_incompat(lock->l_blocking_lock)) {
2621                         mdt_set_lock_sync(lock);
2622                         /*
2623                          * we may do extra commit here, but there is a small
2624                          * window to miss a commit: lock was unlocked (saved),
2625                          * then a conflict lock queued and we come here, but
2626                          * REP-ACK not received, so lock was not converted to
2627                          * COS mode yet.
2628                          * Fortunately this window is quite small, so the
2629                          * extra commit should be rare (not to say distributed
2630                          * operation is rare too).
2631                          */
2632                         commit_async = true;
2633                 }
2634         } else if (lock->l_req_mode == LCK_COS &&
2635                    lock->l_blocking_lock != NULL) {
2636                 commit_async = true;
2637         }
2638
2639         rc = ldlm_blocking_ast_nocheck(lock);
2640
2641         if (commit_async) {
2642                 struct lu_env env;
2643
2644                 rc = lu_env_init(&env, LCT_LOCAL);
2645                 if (unlikely(rc != 0))
2646                         CWARN("%s: lu_env initialization failed, cannot "
2647                               "start asynchronous commit: rc = %d\n",
2648                               obd->obd_name, rc);
2649                 else
2650                         mdt_device_commit_async(&env, mdt);
2651                 lu_env_fini(&env);
2652         }
2653         RETURN(rc);
2654 }
2655
2656 /*
2657  * Blocking AST for cross-MDT lock
2658  *
2659  * Discard lock from uncommitted_slc_locks and cancel it.
2660  *
2661  * \param lock  the lock which blocks a request or cancelling lock
2662  * \param desc  unused
2663  * \param data  unused
2664  * \param flag  indicates whether this cancelling or blocking callback
2665  * \retval      0 on success
2666  * \retval      negative number on error
2667  */
2668 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2669                             void *data, int flag)
2670 {
2671         int rc = 0;
2672         ENTRY;
2673
2674         switch (flag) {
2675         case LDLM_CB_BLOCKING: {
2676                 struct lustre_handle lockh;
2677
2678                 ldlm_lock2handle(lock, &lockh);
2679                 rc = ldlm_cli_cancel(&lockh,
2680                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2681                 if (rc < 0) {
2682                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2683                         RETURN(rc);
2684                 }
2685                 break;
2686         }
2687         case LDLM_CB_CANCELING: {
2688                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2689                 struct mdt_device *mdt =
2690                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2691
2692                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2693
2694                 /* discard slc lock here so that it can be cleaned anytime,
2695                  * especially for cleanup_resource() */
2696                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2697
2698                 /* once we cache lock, l_ast_data is set to mdt_object */
2699                 if (lock->l_ast_data != NULL) {
2700                         struct mdt_object *mo = lock->l_ast_data;
2701                         struct lu_env env;
2702
2703                         rc = lu_env_init(&env, LCT_MD_THREAD);
2704                         if (unlikely(rc != 0)) {
2705                                 CWARN("%s: lu_env initialization failed, object"
2706                                       "%p "DFID" is leaked!\n",
2707                                       obd->obd_name, mo,
2708                                       PFID(mdt_object_fid(mo)));
2709                                 RETURN(rc);
2710                         }
2711
2712                         if (lock->l_policy_data.l_inodebits.bits &
2713                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2714                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2715                                 mo->mot_cache_attr = 0;
2716                         }
2717                         mdt_object_put(&env, mo);
2718                         lu_env_fini(&env);
2719                 }
2720                 break;
2721         }
2722         default:
2723                 LBUG();
2724         }
2725
2726         RETURN(rc);
2727 }
2728
2729 int mdt_check_resent_lock(struct mdt_thread_info *info,
2730                           struct mdt_object *mo,
2731                           struct mdt_lock_handle *lhc)
2732 {
2733         /* the lock might already be gotten in ldlm_handle_enqueue() */
2734         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2735                 struct ptlrpc_request *req = mdt_info_req(info);
2736                 struct ldlm_lock      *lock;
2737
2738                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2739                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2740                 if (lock == NULL) {
2741                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2742                          * a resend case, however, it could be already destroyed
2743                          * due to client eviction or a raced cancel RPC. */
2744                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2745                                           lhc->mlh_reg_lh.cookie);
2746                         RETURN(-ESTALE);
2747                 }
2748
2749                 if (!fid_res_name_eq(mdt_object_fid(mo),
2750                                      &lock->l_resource->lr_name)) {
2751                         CWARN("%s: Although resent, but still not "
2752                               "get child lock:"DFID"\n",
2753                               info->mti_exp->exp_obd->obd_name,
2754                               PFID(mdt_object_fid(mo)));
2755                         LDLM_LOCK_PUT(lock);
2756                         RETURN(-EPROTO);
2757                 }
2758                 LDLM_LOCK_PUT(lock);
2759                 return 0;
2760         }
2761         return 1;
2762 }
2763
2764 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
2765                                struct mdt_object *o, const struct lu_fid *fid,
2766                                struct lustre_handle *lh, enum ldlm_mode mode,
2767                                __u64 *ibits, __u64 trybits, bool cache)
2768 {
2769         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2770         union ldlm_policy_data *policy = &mti->mti_policy;
2771         struct ldlm_res_id *res_id = &mti->mti_res_id;
2772         int rc = 0;
2773         ENTRY;
2774
2775         LASSERT(mdt_object_remote(o));
2776
2777         fid_build_reg_res_name(fid, res_id);
2778
2779         memset(einfo, 0, sizeof(*einfo));
2780         einfo->ei_type = LDLM_IBITS;
2781         einfo->ei_mode = mode;
2782         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2783         einfo->ei_cb_cp = ldlm_completion_ast;
2784         einfo->ei_enq_slave = 0;
2785         einfo->ei_res_id = res_id;
2786
2787         if (cache) {
2788                 /*
2789                  * if we cache lock, couple lock with mdt_object, so that object
2790                  * can be easily found in lock ASTs.
2791                  */
2792                 mdt_object_get(mti->mti_env, o);
2793                 einfo->ei_cbdata = o;
2794         }
2795
2796
2797         memset(policy, 0, sizeof(*policy));
2798         policy->l_inodebits.bits = *ibits;
2799         policy->l_inodebits.try_bits = trybits;
2800
2801         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2802                             policy);
2803         if (rc < 0 && cache) {
2804                 mdt_object_put(mti->mti_env, o);
2805                 einfo->ei_cbdata = NULL;
2806         }
2807
2808         /* Return successfully acquired bits to a caller */
2809         if (rc == 0) {
2810                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
2811
2812                 LASSERT(lock);
2813                 *ibits = lock->l_policy_data.l_inodebits.bits;
2814                 LDLM_LOCK_PUT(lock);
2815         }
2816         RETURN(rc);
2817 }
2818
2819 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2820                            const struct lu_fid *fid, struct lustre_handle *lh,
2821                            enum ldlm_mode mode, __u64 ibits, bool cache)
2822 {
2823         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
2824                                           cache);
2825 }
2826
2827 static int mdt_object_local_lock(struct mdt_thread_info *info,
2828                                  struct mdt_object *o,
2829                                  struct mdt_lock_handle *lh, __u64 *ibits,
2830                                  __u64 trybits, bool cos_incompat)
2831 {
2832         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2833         union ldlm_policy_data *policy = &info->mti_policy;
2834         struct ldlm_res_id *res_id = &info->mti_res_id;
2835         __u64 dlmflags = 0;
2836         int rc;
2837         ENTRY;
2838
2839         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2840         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2841         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2842         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2843
2844         if (cos_incompat) {
2845                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2846                         lh->mlh_reg_mode == LCK_EX);
2847                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2848         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2849                 dlmflags |= LDLM_FL_COS_ENABLED;
2850         }
2851
2852         /* Only enqueue LOOKUP lock for remote object */
2853         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
2854
2855         if (lh->mlh_type == MDT_PDO_LOCK) {
2856                 /* check for exists after object is locked */
2857                 if (mdt_object_exists(o) == 0) {
2858                         /* Non-existent object shouldn't have PDO lock */
2859                         RETURN(-ESTALE);
2860                 } else {
2861                         /* Non-dir object shouldn't have PDO lock */
2862                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2863                                 RETURN(-ENOTDIR);
2864                 }
2865         }
2866
2867
2868         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2869         dlmflags |= LDLM_FL_ATOMIC_CB;
2870
2871         /*
2872          * Take PDO lock on whole directory and build correct @res_id for lock
2873          * on part of directory.
2874          */
2875         if (lh->mlh_pdo_hash != 0) {
2876                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2877                 mdt_lock_pdo_mode(info, o, lh);
2878                 if (lh->mlh_pdo_mode != LCK_NL) {
2879                         /*
2880                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2881                          * is never going to be sent to client and we do not
2882                          * want it slowed down due to possible cancels.
2883                          */
2884                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2885                         policy->l_inodebits.try_bits = 0;
2886                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2887                                           policy, res_id, dlmflags,
2888                                           info->mti_exp == NULL ? NULL :
2889                                           &info->mti_exp->exp_handle.h_cookie);
2890                         if (unlikely(rc != 0))
2891                                 GOTO(out_unlock, rc);
2892                 }
2893
2894                 /*
2895                  * Finish res_id initializing by name hash marking part of
2896                  * directory which is taking modification.
2897                  */
2898                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2899         }
2900
2901         policy->l_inodebits.bits = *ibits;
2902         policy->l_inodebits.try_bits = trybits;
2903
2904         /*
2905          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2906          * going to be sent to client. If it is - mdt_intent_policy() path will
2907          * fix it up and turn FL_LOCAL flag off.
2908          */
2909         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2910                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2911                           info->mti_exp == NULL ? NULL :
2912                           &info->mti_exp->exp_handle.h_cookie);
2913 out_unlock:
2914         if (rc != 0)
2915                 mdt_object_unlock(info, o, lh, 1);
2916         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2917                    lh->mlh_pdo_hash != 0 &&
2918                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2919                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2920
2921         /* Return successfully acquired bits to a caller */
2922         if (rc == 0) {
2923                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2924
2925                 LASSERT(lock);
2926                 *ibits = lock->l_policy_data.l_inodebits.bits;
2927                 LDLM_LOCK_PUT(lock);
2928         }
2929         RETURN(rc);
2930 }
2931
2932 static int
2933 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2934                          struct mdt_lock_handle *lh, __u64 *ibits,
2935                          __u64 trybits, bool cos_incompat)
2936 {
2937         struct mdt_lock_handle *local_lh = NULL;
2938         int rc;
2939         ENTRY;
2940
2941         if (!mdt_object_remote(o)) {
2942                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
2943                                            cos_incompat);
2944                 RETURN(rc);
2945         }
2946
2947         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2948         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2949                     MDS_INODELOCK_XATTR);
2950
2951         /* Only enqueue LOOKUP lock for remote object */
2952         if (*ibits & MDS_INODELOCK_LOOKUP) {
2953                 __u64 local = MDS_INODELOCK_LOOKUP;
2954
2955                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
2956                                            cos_incompat);
2957                 if (rc != ELDLM_OK)
2958                         RETURN(rc);
2959
2960                 local_lh = lh;
2961         }
2962
2963         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
2964                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2965                  * enqueue RPC can only request 1 lock, to avoid extra
2966                  * RPC, so it will instead enqueue EX lock for remote
2967                  * object anyway XXX*/
2968                 if (lh->mlh_type == MDT_PDO_LOCK &&
2969                     lh->mlh_pdo_hash != 0) {
2970                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2971                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2972                                PFID(mdt_object_fid(o)));
2973                         lh->mlh_pdo_hash = 0;
2974                         lh->mlh_rreg_mode = LCK_EX;
2975                         lh->mlh_type = MDT_REG_LOCK;
2976                 }
2977
2978                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
2979                                                 &lh->mlh_rreg_lh,
2980                                                 lh->mlh_rreg_mode,
2981                                                 ibits, trybits, false);
2982                 if (rc != ELDLM_OK) {
2983                         if (local_lh != NULL)
2984                                 mdt_object_unlock(info, o, local_lh, rc);
2985                         RETURN(rc);
2986                 }
2987         }
2988
2989         RETURN(0);
2990 }
2991
2992 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2993                     struct mdt_lock_handle *lh, __u64 ibits)
2994 {
2995         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
2996 }
2997
2998 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2999                           struct mdt_lock_handle *lh, __u64 ibits,
3000                           bool cos_incompat)
3001 {
3002         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3003         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3004                                         cos_incompat);
3005 }
3006
3007 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3008                         struct mdt_lock_handle *lh, __u64 *ibits,
3009                         __u64 trybits, bool cos_incompat)
3010 {
3011         bool trylock_only = *ibits == 0;
3012         int rc;
3013
3014         LASSERT(!(*ibits & trybits));
3015         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3016                                       cos_incompat);
3017         if (rc && trylock_only) { /* clear error for try ibits lock only */
3018                 LASSERT(*ibits == 0);
3019                 rc = 0;
3020         }
3021         return rc;
3022 }
3023
3024 /**
3025  * Save a lock within request object.
3026  *
3027  * Keep the lock referenced until whether client ACK or transaction
3028  * commit happens or release the lock immediately depending on input
3029  * parameters. If COS is ON, a write lock is converted to COS lock
3030  * before saving.
3031  *
3032  * \param info thead info object
3033  * \param h lock handle
3034  * \param mode lock mode
3035  * \param decref force immediate lock releasing
3036  */
3037 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3038                    enum ldlm_mode mode, int decref)
3039 {
3040         ENTRY;
3041
3042         if (lustre_handle_is_used(h)) {
3043                 if (decref || !info->mti_has_trans ||
3044                     !(mode & (LCK_PW | LCK_EX))) {
3045                         mdt_fid_unlock(h, mode);
3046                 } else {
3047                         struct mdt_device *mdt = info->mti_mdt;
3048                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3049                         struct ptlrpc_request *req = mdt_info_req(info);
3050                         bool cos = mdt_cos_is_enabled(mdt);
3051                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3052
3053                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3054                                  h->cookie);
3055
3056                         /* there is no request if mdt_object_unlock() is called
3057                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3058                         if (likely(req != NULL)) {
3059                                 LDLM_DEBUG(lock, "save lock request %p reply "
3060                                         "state %p transno %lld\n", req,
3061                                         req->rq_reply_state, req->rq_transno);
3062                                 if (cos) {
3063                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3064                                         mode = LCK_COS;
3065                                 }
3066                                 if (req->rq_export->exp_disconnected)
3067                                         mdt_fid_unlock(h, mode);
3068                                 else
3069                                         ptlrpc_save_lock(req, h, mode, cos,
3070                                                          convert_lock);
3071                         } else {
3072                                 mdt_fid_unlock(h, mode);
3073                         }
3074                         if (mdt_is_lock_sync(lock)) {
3075                                 CDEBUG(D_HA, "found sync-lock,"
3076                                        " async commit started\n");
3077                                 mdt_device_commit_async(info->mti_env,
3078                                                         mdt);
3079                         }
3080                         LDLM_LOCK_PUT(lock);
3081                 }
3082                 h->cookie = 0ull;
3083         }
3084
3085         EXIT;
3086 }
3087
3088 /**
3089  * Save cross-MDT lock in uncommitted_slc_locks
3090  *
3091  * Keep the lock referenced until transaction commit happens or release the lock
3092  * immediately depending on input parameters.
3093  *
3094  * \param info thead info object
3095  * \param h lock handle
3096  * \param mode lock mode
3097  * \param decref force immediate lock releasing
3098  */
3099 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3100                                  struct mdt_object *o, struct lustre_handle *h,
3101                                  enum ldlm_mode mode, int decref)
3102 {
3103         ENTRY;
3104
3105         if (lustre_handle_is_used(h)) {
3106                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3107
3108                 if (o != NULL &&
3109                     (lock->l_policy_data.l_inodebits.bits &
3110                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3111                         mo_invalidate(info->mti_env, mdt_object_child(o));
3112
3113                 if (decref || !info->mti_has_trans ||
3114                     !(mode & (LCK_PW | LCK_EX))) {
3115                         ldlm_lock_decref_and_cancel(h, mode);
3116                         LDLM_LOCK_PUT(lock);
3117                 } else {
3118                         struct ptlrpc_request *req = mdt_info_req(info);
3119
3120                         LASSERT(req != NULL);
3121                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3122                                           req->rq_transno);
3123                         ldlm_lock_decref(h, mode);
3124                 }
3125                 h->cookie = 0ull;
3126         }
3127
3128         EXIT;
3129 }
3130
3131 /**
3132  * Unlock mdt object.
3133  *
3134  * Immeditely release the regular lock and the PDO lock or save the
3135  * lock in request and keep them referenced until client ACK or
3136  * transaction commit.
3137  *
3138  * \param info thread info object
3139  * \param o mdt object
3140  * \param lh mdt lock handle referencing regular and PDO locks
3141  * \param decref force immediate lock releasing
3142  *
3143  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3144  */
3145 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3146                        struct mdt_lock_handle *lh, int decref)
3147 {
3148         ENTRY;
3149
3150         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3151         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3152         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3153                              decref);
3154
3155         EXIT;
3156 }
3157
3158 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3159                                         const struct lu_fid *f,
3160                                         struct mdt_lock_handle *lh,