Whamcloud - gitweb
LU-11564 tests: add version check sanity-hsm tests
[fs/lustre-release.git] / lustre / llite / llite_lib.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) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2011, 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/llite/llite_lib.c
33  *
34  * Lustre Light Super operations
35  */
36
37 #define DEBUG_SUBSYSTEM S_LLITE
38
39 #include <linux/module.h>
40 #include <linux/statfs.h>
41 #include <linux/time.h>
42 #include <linux/types.h>
43 #include <linux/version.h>
44 #include <linux/mm.h>
45 #include <linux/user_namespace.h>
46 #ifdef HAVE_UIDGID_HEADER
47 # include <linux/uidgid.h>
48 #endif
49 #include <linux/security.h>
50
51 #include <uapi/linux/lustre/lustre_ioctl.h>
52 #include <lustre_ha.h>
53 #include <lustre_dlm.h>
54 #include <lprocfs_status.h>
55 #include <lustre_disk.h>
56 #include <uapi/linux/lustre/lustre_param.h>
57 #include <lustre_log.h>
58 #include <cl_object.h>
59 #include <obd_cksum.h>
60 #include "llite_internal.h"
61
62 struct kmem_cache *ll_file_data_slab;
63
64 #ifndef log2
65 #define log2(n) ffz(~(n))
66 #endif
67
68 static struct ll_sb_info *ll_init_sbi(void)
69 {
70         struct ll_sb_info *sbi = NULL;
71         unsigned long pages;
72         unsigned long lru_page_max;
73         struct sysinfo si;
74         class_uuid_t uuid;
75         int i;
76         ENTRY;
77
78         OBD_ALLOC_PTR(sbi);
79         if (sbi == NULL)
80                 RETURN(NULL);
81
82         spin_lock_init(&sbi->ll_lock);
83         mutex_init(&sbi->ll_lco.lco_lock);
84         spin_lock_init(&sbi->ll_pp_extent_lock);
85         spin_lock_init(&sbi->ll_process_lock);
86         sbi->ll_rw_stats_on = 0;
87
88         si_meminfo(&si);
89         pages = si.totalram - si.totalhigh;
90         lru_page_max = pages / 2;
91
92         /* initialize ll_cache data */
93         sbi->ll_cache = cl_cache_init(lru_page_max);
94         if (sbi->ll_cache == NULL) {
95                 OBD_FREE(sbi, sizeof(*sbi));
96                 RETURN(NULL);
97         }
98
99         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
100                                            SBI_DEFAULT_READAHEAD_MAX);
101         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
102         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
103
104         ll_generate_random_uuid(uuid);
105         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
106         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
107
108         sbi->ll_flags |= LL_SBI_VERBOSE;
109 #ifdef ENABLE_CHECKSUM
110         sbi->ll_flags |= LL_SBI_CHECKSUM;
111 #endif
112
113 #ifdef HAVE_LRU_RESIZE_SUPPORT
114         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
115 #endif
116         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
117
118         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
119                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
120                                pp_r_hist.oh_lock);
121                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
122                                pp_w_hist.oh_lock);
123         }
124
125         /* metadata statahead is enabled by default */
126         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
127         sbi->ll_sa_max = LL_SA_RPC_DEF;
128         atomic_set(&sbi->ll_sa_total, 0);
129         atomic_set(&sbi->ll_sa_wrong, 0);
130         atomic_set(&sbi->ll_sa_running, 0);
131         atomic_set(&sbi->ll_agl_total, 0);
132         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
133         sbi->ll_flags |= LL_SBI_FAST_READ;
134         sbi->ll_flags |= LL_SBI_TINY_WRITE;
135
136         /* root squash */
137         sbi->ll_squash.rsi_uid = 0;
138         sbi->ll_squash.rsi_gid = 0;
139         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
140         init_rwsem(&sbi->ll_squash.rsi_sem);
141
142         RETURN(sbi);
143 }
144
145 static void ll_free_sbi(struct super_block *sb)
146 {
147         struct ll_sb_info *sbi = ll_s2sbi(sb);
148         ENTRY;
149
150         if (sbi != NULL) {
151                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
152                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
153                 if (sbi->ll_cache != NULL) {
154                         cl_cache_decref(sbi->ll_cache);
155                         sbi->ll_cache = NULL;
156                 }
157                 OBD_FREE(sbi, sizeof(*sbi));
158         }
159         EXIT;
160 }
161
162 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
163                                     struct vfsmount *mnt)
164 {
165         struct inode *root = NULL;
166         struct ll_sb_info *sbi = ll_s2sbi(sb);
167         struct obd_statfs *osfs = NULL;
168         struct ptlrpc_request *request = NULL;
169         struct obd_connect_data *data = NULL;
170         struct obd_uuid *uuid;
171         struct md_op_data *op_data;
172         struct lustre_md lmd;
173         u64 valid;
174         int size, err, checksum;
175
176         ENTRY;
177         sbi->ll_md_obd = class_name2obd(md);
178         if (!sbi->ll_md_obd) {
179                 CERROR("MD %s: not setup or attached\n", md);
180                 RETURN(-EINVAL);
181         }
182
183         OBD_ALLOC_PTR(data);
184         if (data == NULL)
185                 RETURN(-ENOMEM);
186
187         OBD_ALLOC_PTR(osfs);
188         if (osfs == NULL) {
189                 OBD_FREE_PTR(data);
190                 RETURN(-ENOMEM);
191         }
192
193         /* pass client page size via ocd_grant_blkbits, the server should report
194          * back its backend blocksize for grant calculation purpose */
195         data->ocd_grant_blkbits = PAGE_SHIFT;
196
197         /* indicate MDT features supported by this client */
198         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
199                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
200                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
201                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
202                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
203                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
204                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
205                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
206                                   OBD_CONNECT_64BITHASH |
207                                   OBD_CONNECT_EINPROGRESS |
208                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
209                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
210                                   OBD_CONNECT_MAX_EASIZE |
211                                   OBD_CONNECT_FLOCK_DEAD |
212                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
213                                   OBD_CONNECT_OPEN_BY_FID |
214                                   OBD_CONNECT_DIR_STRIPE |
215                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
216                                   OBD_CONNECT_SUBTREE |
217                                   OBD_CONNECT_MULTIMODRPCS |
218                                   OBD_CONNECT_GRANT_PARAM |
219                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
220
221         data->ocd_connect_flags2 = OBD_CONNECT2_FLR |
222                                    OBD_CONNECT2_LOCK_CONVERT |
223                                    OBD_CONNECT2_DIR_MIGRATE |
224                                    OBD_CONNECT2_SUM_STATFS;
225
226 #ifdef HAVE_LRU_RESIZE_SUPPORT
227         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
228                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
229 #endif
230 #ifdef CONFIG_FS_POSIX_ACL
231         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
232                                    OBD_CONNECT_LARGE_ACL;
233 #endif
234
235         data->ocd_cksum_types = obd_cksum_types_supported_client();
236
237         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
238                 /* flag mdc connection as lightweight, only used for test
239                  * purpose, use with care */
240                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
241
242         data->ocd_ibits_known = MDS_INODELOCK_FULL;
243         data->ocd_version = LUSTRE_VERSION_CODE;
244
245         if (sb->s_flags & MS_RDONLY)
246                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
247         if (sbi->ll_flags & LL_SBI_USER_XATTR)
248                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
249
250 #ifdef MS_NOSEC
251         /* Setting this indicates we correctly support S_NOSEC (See kernel
252          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
253          */
254         sb->s_flags |= MS_NOSEC;
255 #endif
256
257         if (sbi->ll_flags & LL_SBI_FLOCK)
258                 sbi->ll_fop = &ll_file_operations_flock;
259         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
260                 sbi->ll_fop = &ll_file_operations;
261         else
262                 sbi->ll_fop = &ll_file_operations_noflock;
263
264         /* always ping even if server suppress_pings */
265         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
266                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
267
268         obd_connect_set_secctx(data);
269
270         data->ocd_brw_size = MD_MAX_BRW_SIZE;
271
272         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
273                           &sbi->ll_sb_uuid, data, NULL);
274         if (err == -EBUSY) {
275                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
276                                    "recovery, of which this client is not a "
277                                    "part. Please wait for recovery to complete,"
278                                    " abort, or time out.\n", md);
279                 GOTO(out, err);
280         } else if (err) {
281                 CERROR("cannot connect to %s: rc = %d\n", md, err);
282                 GOTO(out, err);
283         }
284
285         sbi->ll_md_exp->exp_connect_data = *data;
286
287         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
288                            LUSTRE_SEQ_METADATA);
289         if (err) {
290                 CERROR("%s: Can't init metadata layer FID infrastructure, "
291                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
292                 GOTO(out_md, err);
293         }
294
295         /* For mount, we only need fs info from MDT0, and also in DNE, it
296          * can make sure the client can be mounted as long as MDT0 is
297          * avaible */
298         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
299                         ktime_get_seconds() -OBD_STATFS_CACHE_SECONDS,
300                         OBD_STATFS_FOR_MDT0);
301         if (err)
302                 GOTO(out_md_fid, err);
303
304         /* This needs to be after statfs to ensure connect has finished.
305          * Note that "data" does NOT contain the valid connect reply.
306          * If connecting to a 1.8 server there will be no LMV device, so
307          * we can access the MDC export directly and exp_connect_flags will
308          * be non-zero, but if accessing an upgraded 2.1 server it will
309          * have the correct flags filled in.
310          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
311         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
312         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
313             valid != CLIENT_CONNECT_MDT_REQD) {
314                 char *buf;
315
316                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
317                 obd_connect_flags2str(buf, PAGE_SIZE,
318                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
319                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
320                                    "feature(s) needed for correct operation "
321                                    "of this client (%s). Please upgrade "
322                                    "server or downgrade client.\n",
323                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
324                 OBD_FREE(buf, PAGE_SIZE);
325                 GOTO(out_md_fid, err = -EPROTO);
326         }
327
328         size = sizeof(*data);
329         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
330                            KEY_CONN_DATA,  &size, data);
331         if (err) {
332                 CERROR("%s: Get connect data failed: rc = %d\n",
333                        sbi->ll_md_exp->exp_obd->obd_name, err);
334                 GOTO(out_md_fid, err);
335         }
336
337         LASSERT(osfs->os_bsize);
338         sb->s_blocksize = osfs->os_bsize;
339         sb->s_blocksize_bits = log2(osfs->os_bsize);
340         sb->s_magic = LL_SUPER_MAGIC;
341         sb->s_maxbytes = MAX_LFS_FILESIZE;
342         sbi->ll_namelen = osfs->os_namelen;
343         sbi->ll_mnt.mnt = current->fs->root.mnt;
344
345         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
346             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
347                 LCONSOLE_INFO("Disabling user_xattr feature because "
348                               "it is not supported on the server\n");
349                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
350         }
351
352         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
353 #ifdef MS_POSIXACL
354                 sb->s_flags |= MS_POSIXACL;
355 #endif
356                 sbi->ll_flags |= LL_SBI_ACL;
357         } else {
358                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
359 #ifdef MS_POSIXACL
360                 sb->s_flags &= ~MS_POSIXACL;
361 #endif
362                 sbi->ll_flags &= ~LL_SBI_ACL;
363         }
364
365         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
366                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
367
368         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
369                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
370
371         if (obd_connect_has_secctx(data))
372                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
373
374         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
375                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
376                         LCONSOLE_INFO("%s: disabling xattr cache due to "
377                                       "unknown maximum xattr size.\n", dt);
378                 } else if (!sbi->ll_xattr_cache_set) {
379                         /* If xattr_cache is already set (no matter 0 or 1)
380                          * during processing llog, it won't be enabled here. */
381                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
382                         sbi->ll_xattr_cache_enabled = 1;
383                 }
384         }
385
386         sbi->ll_dt_obd = class_name2obd(dt);
387         if (!sbi->ll_dt_obd) {
388                 CERROR("DT %s: not setup or attached\n", dt);
389                 GOTO(out_md_fid, err = -ENODEV);
390         }
391
392         /* pass client page size via ocd_grant_blkbits, the server should report
393          * back its backend blocksize for grant calculation purpose */
394         data->ocd_grant_blkbits = PAGE_SHIFT;
395
396         /* indicate OST features supported by this client */
397         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
398                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
399                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
400                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
401                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
402                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
403                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
404                                   OBD_CONNECT_EINPROGRESS |
405                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
406                                   OBD_CONNECT_LAYOUTLOCK |
407                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
408                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
409                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
410
411 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
412  * can interoperate with an older version of lockahead which was released prior
413  * to landing in master. This support will be dropped when 2.13 development
414  * starts.  At the point, we should not just drop the connect flag (below), we
415  * should also remove the support in the code.
416  *
417  * Removing it means a few things:
418  * 1. Remove this section here
419  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
420  * 3. Remove function exp_connect_lockahead_old
421  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
422  * */
423 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
424         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
425 #endif
426
427         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
428
429         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
430                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
431
432         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
433          * disabled by default, because it can still be enabled on the
434          * fly via /proc. As a consequence, we still need to come to an
435          * agreement on the supported algorithms at connect time */
436         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
437
438         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
439                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
440         else
441                 data->ocd_cksum_types = obd_cksum_types_supported_client();
442
443 #ifdef HAVE_LRU_RESIZE_SUPPORT
444         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
445 #endif
446         /* always ping even if server suppress_pings */
447         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
448                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
449
450         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
451                "ocd_grant: %d\n", data->ocd_connect_flags,
452                data->ocd_version, data->ocd_grant);
453
454         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
455         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
456
457         data->ocd_brw_size = DT_MAX_BRW_SIZE;
458
459         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
460                           &sbi->ll_sb_uuid, data, NULL);
461         if (err == -EBUSY) {
462                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
463                                    "recovery, of which this client is not a "
464                                    "part.  Please wait for recovery to "
465                                    "complete, abort, or time out.\n", dt);
466                 GOTO(out_md, err);
467         } else if (err) {
468                 CERROR("%s: Cannot connect to %s: rc = %d\n",
469                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
470                 GOTO(out_md, err);
471         }
472
473         sbi->ll_dt_exp->exp_connect_data = *data;
474
475         /* Don't change value if it was specified in the config log */
476         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
477                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
478                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
479                               (data->ocd_brw_size >> PAGE_SHIFT));
480
481         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
482                            LUSTRE_SEQ_METADATA);
483         if (err) {
484                 CERROR("%s: Can't init data layer FID infrastructure, "
485                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
486                 GOTO(out_dt, err);
487         }
488
489         mutex_lock(&sbi->ll_lco.lco_lock);
490         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
491         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
492         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
493         mutex_unlock(&sbi->ll_lco.lco_lock);
494
495         fid_zero(&sbi->ll_root_fid);
496         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
497                            &sbi->ll_root_fid);
498         if (err) {
499                 CERROR("cannot mds_connect: rc = %d\n", err);
500                 GOTO(out_lock_cn_cb, err);
501         }
502         if (!fid_is_sane(&sbi->ll_root_fid)) {
503                 CERROR("%s: Invalid root fid "DFID" during mount\n",
504                        sbi->ll_md_exp->exp_obd->obd_name,
505                        PFID(&sbi->ll_root_fid));
506                 GOTO(out_lock_cn_cb, err = -EINVAL);
507         }
508         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
509
510         sb->s_op = &lustre_super_operations;
511 #ifdef HAVE_XATTR_HANDLER_FLAGS
512         sb->s_xattr = ll_xattr_handlers;
513 #endif
514 #if THREAD_SIZE >= 8192 /*b=17630*/
515         sb->s_export_op = &lustre_export_operations;
516 #endif
517
518         /* make root inode
519          * XXX: move this to after cbd setup? */
520         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
521         if (sbi->ll_flags & LL_SBI_ACL)
522                 valid |= OBD_MD_FLACL;
523
524         OBD_ALLOC_PTR(op_data);
525         if (op_data == NULL)
526                 GOTO(out_lock_cn_cb, err = -ENOMEM);
527
528         op_data->op_fid1 = sbi->ll_root_fid;
529         op_data->op_mode = 0;
530         op_data->op_valid = valid;
531
532         err = md_getattr(sbi->ll_md_exp, op_data, &request);
533
534         OBD_FREE_PTR(op_data);
535         if (err) {
536                 CERROR("%s: md_getattr failed for root: rc = %d\n",
537                        sbi->ll_md_exp->exp_obd->obd_name, err);
538                 GOTO(out_lock_cn_cb, err);
539         }
540
541         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
542                                sbi->ll_md_exp, &lmd);
543         if (err) {
544                 CERROR("failed to understand root inode md: rc = %d\n", err);
545                 ptlrpc_req_finished(request);
546                 GOTO(out_lock_cn_cb, err);
547         }
548
549         LASSERT(fid_is_sane(&sbi->ll_root_fid));
550         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
551                                             sbi->ll_flags & LL_SBI_32BIT_API),
552                        &lmd);
553         md_free_lustre_md(sbi->ll_md_exp, &lmd);
554         ptlrpc_req_finished(request);
555
556         if (IS_ERR(root)) {
557 #ifdef CONFIG_FS_POSIX_ACL
558                 if (lmd.posix_acl) {
559                         posix_acl_release(lmd.posix_acl);
560                         lmd.posix_acl = NULL;
561                 }
562 #endif
563                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
564                 root = NULL;
565                 CERROR("lustre_lite: bad iget4 for root\n");
566                 GOTO(out_root, err);
567         }
568
569         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
570         if (sbi->ll_checksum_set) {
571                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
572                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
573                                          sizeof(checksum), &checksum, NULL);
574                 if (err) {
575                         CERROR("%s: Set checksum failed: rc = %d\n",
576                                sbi->ll_dt_exp->exp_obd->obd_name, err);
577                         GOTO(out_root, err);
578                 }
579         }
580         cl_sb_init(sb);
581
582         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
583                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
584                                  sbi->ll_cache, NULL);
585         if (err) {
586                 CERROR("%s: Set cache_set failed: rc = %d\n",
587                        sbi->ll_dt_exp->exp_obd->obd_name, err);
588                 GOTO(out_root, err);
589         }
590
591         sb->s_root = d_make_root(root);
592         if (sb->s_root == NULL) {
593                 CERROR("%s: can't make root dentry\n",
594                         ll_get_fsname(sb, NULL, 0));
595                 GOTO(out_root, err = -ENOMEM);
596         }
597 #ifdef HAVE_DCACHE_LOCK
598         sb->s_root->d_op = &ll_d_ops;
599 #endif
600
601         sbi->ll_sdev_orig = sb->s_dev;
602
603         /* We set sb->s_dev equal on all lustre clients in order to support
604          * NFS export clustering.  NFSD requires that the FSID be the same
605          * on all clients. */
606         /* s_dev is also used in lt_compare() to compare two fs, but that is
607          * only a node-local comparison. */
608         uuid = obd_get_uuid(sbi->ll_md_exp);
609         if (uuid != NULL)
610                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
611
612         if (data != NULL)
613                 OBD_FREE_PTR(data);
614         if (osfs != NULL)
615                 OBD_FREE_PTR(osfs);
616
617         if (sbi->ll_dt_obd) {
618                 err = sysfs_create_link(&sbi->ll_kset.kobj,
619                                         &sbi->ll_dt_obd->obd_kset.kobj,
620                                         sbi->ll_dt_obd->obd_type->typ_name);
621                 if (err < 0) {
622                         CERROR("%s: could not register %s in llite: rc = %d\n",
623                                dt, ll_get_fsname(sb, NULL, 0), err);
624                         err = 0;
625                 }
626         }
627
628         if (sbi->ll_md_obd) {
629                 err = sysfs_create_link(&sbi->ll_kset.kobj,
630                                         &sbi->ll_md_obd->obd_kset.kobj,
631                                         sbi->ll_md_obd->obd_type->typ_name);
632                 if (err < 0) {
633                         CERROR("%s: could not register %s in llite: rc = %d\n",
634                                md, ll_get_fsname(sb, NULL, 0), err);
635                         err = 0;
636                 }
637         }
638
639         RETURN(err);
640 out_root:
641         if (root)
642                 iput(root);
643 out_lock_cn_cb:
644         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
645 out_dt:
646         obd_disconnect(sbi->ll_dt_exp);
647         sbi->ll_dt_exp = NULL;
648         sbi->ll_dt_obd = NULL;
649 out_md_fid:
650         obd_fid_fini(sbi->ll_md_exp->exp_obd);
651 out_md:
652         obd_disconnect(sbi->ll_md_exp);
653         sbi->ll_md_exp = NULL;
654         sbi->ll_md_obd = NULL;
655 out:
656         if (data != NULL)
657                 OBD_FREE_PTR(data);
658         if (osfs != NULL)
659                 OBD_FREE_PTR(osfs);
660         return err;
661 }
662
663 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
664 {
665         int size, rc;
666
667         size = sizeof(*lmmsize);
668         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
669                           KEY_MAX_EASIZE, &size, lmmsize);
670         if (rc != 0) {
671                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
672                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
673                 RETURN(rc);
674         }
675
676         size = sizeof(int);
677         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
678                           KEY_MAX_EASIZE, &size, lmmsize);
679         if (rc)
680                 CERROR("Get max mdsize error rc %d\n", rc);
681
682         RETURN(rc);
683 }
684
685 /**
686  * Get the value of the default_easize parameter.
687  *
688  * \see client_obd::cl_default_mds_easize
689  *
690  * \param[in] sbi       superblock info for this filesystem
691  * \param[out] lmmsize  pointer to storage location for value
692  *
693  * \retval 0            on success
694  * \retval negative     negated errno on failure
695  */
696 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
697 {
698         int size, rc;
699
700         size = sizeof(int);
701         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
702                          KEY_DEFAULT_EASIZE, &size, lmmsize);
703         if (rc)
704                 CERROR("Get default mdsize error rc %d\n", rc);
705
706         RETURN(rc);
707 }
708
709 /**
710  * Set the default_easize parameter to the given value.
711  *
712  * \see client_obd::cl_default_mds_easize
713  *
714  * \param[in] sbi       superblock info for this filesystem
715  * \param[in] lmmsize   the size to set
716  *
717  * \retval 0            on success
718  * \retval negative     negated errno on failure
719  */
720 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
721 {
722         int rc;
723
724         if (lmmsize < sizeof(struct lov_mds_md) ||
725             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
726                 return -EINVAL;
727
728         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
729                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
730                                 sizeof(int), &lmmsize, NULL);
731
732         RETURN(rc);
733 }
734
735 static void client_common_put_super(struct super_block *sb)
736 {
737         struct ll_sb_info *sbi = ll_s2sbi(sb);
738         ENTRY;
739
740         cl_sb_fini(sb);
741
742         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
743         obd_disconnect(sbi->ll_dt_exp);
744         sbi->ll_dt_exp = NULL;
745
746         ll_debugfs_unregister_super(sb);
747
748         obd_fid_fini(sbi->ll_md_exp->exp_obd);
749         obd_disconnect(sbi->ll_md_exp);
750         sbi->ll_md_exp = NULL;
751
752         EXIT;
753 }
754
755 void ll_kill_super(struct super_block *sb)
756 {
757         struct ll_sb_info *sbi;
758         ENTRY;
759
760         /* not init sb ?*/
761         if (!(sb->s_flags & MS_ACTIVE))
762                 return;
763
764         sbi = ll_s2sbi(sb);
765         /* we need restore s_dev from changed for clustred NFS before put_super
766          * because new kernels have cached s_dev and change sb->s_dev in
767          * put_super not affected real removing devices */
768         if (sbi) {
769                 sb->s_dev = sbi->ll_sdev_orig;
770                 sbi->ll_umounting = 1;
771
772                 /* wait running statahead threads to quit */
773                 while (atomic_read(&sbi->ll_sa_running) > 0) {
774                         set_current_state(TASK_UNINTERRUPTIBLE);
775                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
776                 }
777         }
778
779         EXIT;
780 }
781
782 static inline int ll_set_opt(const char *opt, char *data, int fl)
783 {
784         if (strncmp(opt, data, strlen(opt)) != 0)
785                 return 0;
786         else
787                 return fl;
788 }
789
790 /* non-client-specific mount options are parsed in lmd_parse */
791 static int ll_options(char *options, struct ll_sb_info *sbi)
792 {
793         int tmp;
794         char *s1 = options, *s2;
795         int *flags = &sbi->ll_flags;
796         ENTRY;
797
798         if (!options)
799                 RETURN(0);
800
801         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
802
803         while (*s1) {
804                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
805                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
806                 if (tmp) {
807                         *flags |= tmp;
808                         goto next;
809                 }
810                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
811                 if (tmp) {
812                         *flags |= tmp;
813                         goto next;
814                 }
815                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
816                 if (tmp) {
817                         *flags |= tmp;
818                         goto next;
819                 }
820                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
821                 if (tmp) {
822                         *flags &= ~tmp;
823                         goto next;
824                 }
825                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
826                 if (tmp) {
827                         *flags |= tmp;
828                         goto next;
829                 }
830                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
831                 if (tmp) {
832                         *flags &= ~tmp;
833                         goto next;
834                 }
835                 tmp = ll_set_opt("context", s1, 1);
836                 if (tmp)
837                         goto next;
838                 tmp = ll_set_opt("fscontext", s1, 1);
839                 if (tmp)
840                         goto next;
841                 tmp = ll_set_opt("defcontext", s1, 1);
842                 if (tmp)
843                         goto next;
844                 tmp = ll_set_opt("rootcontext", s1, 1);
845                 if (tmp)
846                         goto next;
847                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
848                 if (tmp) {
849                         *flags |= tmp;
850                         goto next;
851                 }
852                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
853                 if (tmp) {
854                         *flags &= ~tmp;
855                         goto next;
856                 }
857
858                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
859                 if (tmp) {
860                         *flags |= tmp;
861                         sbi->ll_checksum_set = 1;
862                         goto next;
863                 }
864                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
865                 if (tmp) {
866                         *flags &= ~tmp;
867                         sbi->ll_checksum_set = 1;
868                         goto next;
869                 }
870                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
871                 if (tmp) {
872                         *flags |= tmp;
873                         goto next;
874                 }
875                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
876                 if (tmp) {
877                         *flags &= ~tmp;
878                         goto next;
879                 }
880                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
881                 if (tmp) {
882                         *flags |= tmp;
883                         goto next;
884                 }
885                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
886                 if (tmp) {
887                         *flags &= ~tmp;
888                         goto next;
889                 }
890                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
891                 if (tmp) {
892                         *flags |= tmp;
893                         goto next;
894                 }
895                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
896                 if (tmp) {
897                         *flags |= tmp;
898                         goto next;
899                 }
900                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
901                 if (tmp) {
902                         *flags &= ~tmp;
903                         goto next;
904                 }
905                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
906                 if (tmp) {
907                         *flags |= tmp;
908                         goto next;
909                 }
910                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
911                                    s1);
912                 RETURN(-EINVAL);
913
914 next:
915                 /* Find next opt */
916                 s2 = strchr(s1, ',');
917                 if (s2 == NULL)
918                         break;
919                 s1 = s2 + 1;
920         }
921         RETURN(0);
922 }
923
924 void ll_lli_init(struct ll_inode_info *lli)
925 {
926         lli->lli_inode_magic = LLI_INODE_MAGIC;
927         lli->lli_flags = 0;
928         spin_lock_init(&lli->lli_lock);
929         lli->lli_posix_acl = NULL;
930         /* Do not set lli_fid, it has been initialized already. */
931         fid_zero(&lli->lli_pfid);
932         lli->lli_mds_read_och = NULL;
933         lli->lli_mds_write_och = NULL;
934         lli->lli_mds_exec_och = NULL;
935         lli->lli_open_fd_read_count = 0;
936         lli->lli_open_fd_write_count = 0;
937         lli->lli_open_fd_exec_count = 0;
938         mutex_init(&lli->lli_och_mutex);
939         spin_lock_init(&lli->lli_agl_lock);
940         spin_lock_init(&lli->lli_layout_lock);
941         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
942         lli->lli_clob = NULL;
943
944         init_rwsem(&lli->lli_xattrs_list_rwsem);
945         mutex_init(&lli->lli_xattrs_enq_lock);
946
947         LASSERT(lli->lli_vfs_inode.i_mode != 0);
948         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
949                 mutex_init(&lli->lli_readdir_mutex);
950                 lli->lli_opendir_key = NULL;
951                 lli->lli_sai = NULL;
952                 spin_lock_init(&lli->lli_sa_lock);
953                 lli->lli_opendir_pid = 0;
954                 lli->lli_sa_enabled = 0;
955                 lli->lli_def_stripe_offset = -1;
956         } else {
957                 mutex_init(&lli->lli_size_mutex);
958                 lli->lli_symlink_name = NULL;
959                 init_rwsem(&lli->lli_trunc_sem);
960                 range_lock_tree_init(&lli->lli_write_tree);
961                 init_rwsem(&lli->lli_glimpse_sem);
962                 lli->lli_glimpse_time = ktime_set(0, 0);
963                 INIT_LIST_HEAD(&lli->lli_agl_list);
964                 lli->lli_agl_index = 0;
965                 lli->lli_async_rc = 0;
966         }
967         mutex_init(&lli->lli_layout_mutex);
968         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
969 }
970
971 #define MAX_STRING_SIZE 128
972
973 #ifndef HAVE_SUPER_SETUP_BDI_NAME
974
975 #define LSI_BDI_INITIALIZED     0x00400000
976
977 #ifndef HAVE_BDI_CAP_MAP_COPY
978 # define BDI_CAP_MAP_COPY       0
979 #endif
980
981 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
982 {
983         struct  lustre_sb_info *lsi = s2lsi(sb);
984         char buf[MAX_STRING_SIZE];
985         va_list args;
986         int err;
987
988         err = bdi_init(&lsi->lsi_bdi);
989         if (err)
990                 return err;
991
992         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
993         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
994         lsi->lsi_bdi.name = "lustre";
995         va_start(args, fmt);
996         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
997         va_end(args);
998         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
999         va_end(args);
1000         if (!err)
1001                 sb->s_bdi = &lsi->lsi_bdi;
1002
1003         return err;
1004 }
1005 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1006
1007 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1008 {
1009         struct  lustre_profile *lprof = NULL;
1010         struct  lustre_sb_info *lsi = s2lsi(sb);
1011         struct  ll_sb_info *sbi = NULL;
1012         char    *dt = NULL, *md = NULL;
1013         char    *profilenm = get_profile_name(sb);
1014         struct config_llog_instance *cfg;
1015         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1016         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1017         char name[MAX_STRING_SIZE];
1018         int md_len = 0;
1019         int dt_len = 0;
1020         char *ptr;
1021         int len;
1022         int err;
1023
1024         ENTRY;
1025         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1026
1027         try_module_get(THIS_MODULE);
1028
1029         OBD_ALLOC_PTR(cfg);
1030         if (cfg == NULL)
1031                 GOTO(out_free, err = -ENOMEM);
1032
1033         /* client additional sb info */
1034         lsi->lsi_llsbi = sbi = ll_init_sbi();
1035         if (!sbi)
1036                 GOTO(out_free, err = -ENOMEM);
1037
1038         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1039         if (err)
1040                 GOTO(out_free, err);
1041
1042         err = super_setup_bdi_name(sb, "lustre-%p", sb);
1043         if (err)
1044                 GOTO(out_free, err);
1045
1046 #ifndef HAVE_DCACHE_LOCK
1047         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1048         sb->s_d_op = &ll_d_ops;
1049 #endif
1050         /* Get fsname */
1051         len = strlen(lsi->lsi_lmd->lmd_profile);
1052         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
1053         if (ptr && (strcmp(ptr, "-client") == 0))
1054                 len -= 7;
1055
1056         /* Mount info */
1057         snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
1058                  lsi->lsi_lmd->lmd_profile, sb);
1059
1060         /* Call ll_debugfs_register_super() before lustre_process_log()
1061          * so that "llite.*.*" params can be processed correctly.
1062          */
1063         err = ll_debugfs_register_super(sb, name);
1064         if (err < 0) {
1065                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1066                        ll_get_fsname(sb, NULL, 0), err);
1067                 err = 0;
1068         }
1069
1070         /* Generate a string unique to this super, in case some joker tries
1071            to mount the same fs at two mount points.
1072            Use the address of the super itself.*/
1073         cfg->cfg_instance = sb;
1074         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1075         cfg->cfg_callback = class_config_llog_handler;
1076         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1077         /* set up client obds */
1078         err = lustre_process_log(sb, profilenm, cfg);
1079         if (err < 0)
1080                 GOTO(out_proc, err);
1081
1082         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1083         lprof = class_get_profile(profilenm);
1084         if (lprof == NULL) {
1085                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1086                                    " read from the MGS.  Does that filesystem "
1087                                    "exist?\n", profilenm);
1088                 GOTO(out_proc, err = -EINVAL);
1089         }
1090         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1091                lprof->lp_md, lprof->lp_dt);
1092
1093         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1094         OBD_ALLOC(dt, dt_len);
1095         if (!dt)
1096                 GOTO(out_proc, err = -ENOMEM);
1097         snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1098
1099         md_len = strlen(lprof->lp_md) + instlen + 2;
1100         OBD_ALLOC(md, md_len);
1101         if (!md)
1102                 GOTO(out_proc, err = -ENOMEM);
1103         snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1104
1105         /* connections, registrations, sb setup */
1106         err = client_common_fill_super(sb, md, dt, mnt);
1107         if (err < 0)
1108                 GOTO(out_proc, err);
1109
1110         sbi->ll_client_common_fill_super_succeeded = 1;
1111
1112 out_proc:
1113         if (err < 0)
1114                 ll_debugfs_unregister_super(sb);
1115 out_free:
1116         if (md)
1117                 OBD_FREE(md, md_len);
1118         if (dt)
1119                 OBD_FREE(dt, dt_len);
1120         if (lprof != NULL)
1121                 class_put_profile(lprof);
1122         if (cfg)
1123                 OBD_FREE_PTR(cfg);
1124         if (err)
1125                 ll_put_super(sb);
1126         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1127                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1128         RETURN(err);
1129 } /* ll_fill_super */
1130
1131 void ll_put_super(struct super_block *sb)
1132 {
1133         struct config_llog_instance cfg, params_cfg;
1134         struct obd_device *obd;
1135         struct lustre_sb_info *lsi = s2lsi(sb);
1136         struct ll_sb_info *sbi = ll_s2sbi(sb);
1137         char *profilenm = get_profile_name(sb);
1138         long ccc_count;
1139         int next, force = 1, rc = 0;
1140         ENTRY;
1141
1142         if (!sbi)
1143                 GOTO(out_no_sbi, 0);
1144
1145         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1146
1147         cfg.cfg_instance = sb;
1148         lustre_end_log(sb, profilenm, &cfg);
1149
1150         params_cfg.cfg_instance = sb;
1151         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1152
1153         if (sbi->ll_md_exp) {
1154                 obd = class_exp2obd(sbi->ll_md_exp);
1155                 if (obd)
1156                         force = obd->obd_force;
1157         }
1158
1159         /* Wait for unstable pages to be committed to stable storage */
1160         if (force == 0) {
1161                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1162                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1163                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1164                         &lwi);
1165         }
1166
1167         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1168         if (force == 0 && rc != -EINTR)
1169                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1170
1171
1172         /* We need to set force before the lov_disconnect in
1173            lustre_common_put_super, since l_d cleans up osc's as well. */
1174         if (force) {
1175                 next = 0;
1176                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1177                                                      &next)) != NULL) {
1178                         obd->obd_force = force;
1179                 }
1180         }
1181
1182         if (sbi->ll_client_common_fill_super_succeeded) {
1183                 /* Only if client_common_fill_super succeeded */
1184                 client_common_put_super(sb);
1185         }
1186
1187         next = 0;
1188         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1189                 class_manual_cleanup(obd);
1190         }
1191
1192         if (sbi->ll_flags & LL_SBI_VERBOSE)
1193                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1194
1195         if (profilenm)
1196                 class_del_profile(profilenm);
1197
1198 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1199         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1200                 bdi_destroy(&lsi->lsi_bdi);
1201                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1202         }
1203 #endif
1204
1205         ll_free_sbi(sb);
1206         lsi->lsi_llsbi = NULL;
1207 out_no_sbi:
1208         lustre_common_put_super(sb);
1209
1210         cl_env_cache_purge(~0);
1211
1212         module_put(THIS_MODULE);
1213
1214         EXIT;
1215 } /* client_put_super */
1216
1217 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1218 {
1219         struct inode *inode = NULL;
1220
1221         /* NOTE: we depend on atomic igrab() -bzzz */
1222         lock_res_and_lock(lock);
1223         if (lock->l_resource->lr_lvb_inode) {
1224                 struct ll_inode_info * lli;
1225                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1226                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1227                         inode = igrab(lock->l_resource->lr_lvb_inode);
1228                 } else {
1229                         inode = lock->l_resource->lr_lvb_inode;
1230                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1231                                          D_WARNING, lock, "lr_lvb_inode %p is "
1232                                          "bogus: magic %08x",
1233                                          lock->l_resource->lr_lvb_inode,
1234                                          lli->lli_inode_magic);
1235                         inode = NULL;
1236                 }
1237         }
1238         unlock_res_and_lock(lock);
1239         return inode;
1240 }
1241
1242 void ll_dir_clear_lsm_md(struct inode *inode)
1243 {
1244         struct ll_inode_info *lli = ll_i2info(inode);
1245
1246         LASSERT(S_ISDIR(inode->i_mode));
1247
1248         if (lli->lli_lsm_md != NULL) {
1249                 lmv_free_memmd(lli->lli_lsm_md);
1250                 lli->lli_lsm_md = NULL;
1251         }
1252 }
1253
1254 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1255                                       const struct lu_fid *fid,
1256                                       struct lustre_md *md)
1257 {
1258         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1259         struct mdt_body         *body = md->body;
1260         struct inode            *inode;
1261         ino_t                   ino;
1262         ENTRY;
1263
1264         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1265         inode = iget_locked(sb, ino);
1266         if (inode == NULL) {
1267                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1268                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1269                 RETURN(ERR_PTR(-ENOENT));
1270         }
1271
1272         if (inode->i_state & I_NEW) {
1273                 struct ll_inode_info *lli = ll_i2info(inode);
1274                 struct lmv_stripe_md *lsm = md->lmv;
1275
1276                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1277                                 (body->mbo_mode & S_IFMT);
1278                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1279                          PFID(fid));
1280
1281                 LTIME_S(inode->i_mtime) = 0;
1282                 LTIME_S(inode->i_atime) = 0;
1283                 LTIME_S(inode->i_ctime) = 0;
1284                 inode->i_rdev = 0;
1285
1286 #ifdef HAVE_BACKING_DEV_INFO
1287                 /* initializing backing dev info. */
1288                 inode->i_mapping->backing_dev_info =
1289                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1290 #endif
1291                 inode->i_op = &ll_dir_inode_operations;
1292                 inode->i_fop = &ll_dir_operations;
1293                 lli->lli_fid = *fid;
1294                 ll_lli_init(lli);
1295
1296                 LASSERT(lsm != NULL);
1297                 /* master object FID */
1298                 lli->lli_pfid = body->mbo_fid1;
1299                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1300                        lli, PFID(fid), PFID(&lli->lli_pfid));
1301                 unlock_new_inode(inode);
1302         }
1303
1304         RETURN(inode);
1305 }
1306
1307 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1308 {
1309         struct lu_fid *fid;
1310         struct lmv_stripe_md *lsm = md->lmv;
1311         int i;
1312
1313         LASSERT(lsm != NULL);
1314         /* XXX sigh, this lsm_root initialization should be in
1315          * LMV layer, but it needs ll_iget right now, so we
1316          * put this here right now. */
1317         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1318                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1319                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1320                 /* Unfortunately ll_iget will call ll_update_inode,
1321                  * where the initialization of slave inode is slightly
1322                  * different, so it reset lsm_md to NULL to avoid
1323                  * initializing lsm for slave inode. */
1324                 lsm->lsm_md_oinfo[i].lmo_root =
1325                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1326                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1327                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1328
1329                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1330                         return rc;
1331                 }
1332         }
1333
1334         return 0;
1335 }
1336
1337 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1338 {
1339         struct ll_inode_info *lli = ll_i2info(inode);
1340         struct lmv_stripe_md *lsm = md->lmv;
1341         int     rc;
1342         ENTRY;
1343
1344         LASSERT(S_ISDIR(inode->i_mode));
1345         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1346                PFID(ll_inode2fid(inode)));
1347
1348         /*
1349          * no striped information from request, lustre_md from req does not
1350          * include stripeEA, see ll_md_setattr()
1351          */
1352         if (!lsm)
1353                 RETURN(0);
1354
1355         /* Compare the old and new stripe information */
1356         if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1357                 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1358                 int idx;
1359                 bool layout_changed = lsm->lsm_md_layout_version >
1360                                       old_lsm->lsm_md_layout_version;
1361
1362                 int mask = layout_changed ? D_INODE : D_ERROR;
1363
1364                 CDEBUG(mask,
1365                         "%s: inode@%p "DFID" lmv layout %s magic %#x/%#x "
1366                         "stripe count %d/%d master_mdt %d/%d "
1367                         "hash_type %#x/%#x version %d/%d migrate offset %d/%d "
1368                         "migrate hash %#x/%#x pool %s/%s\n",
1369                        ll_get_fsname(inode->i_sb, NULL, 0), inode,
1370                        PFID(&lli->lli_fid),
1371                        layout_changed ? "changed" : "mismatch",
1372                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1373                        lsm->lsm_md_stripe_count,
1374                        old_lsm->lsm_md_stripe_count,
1375                        lsm->lsm_md_master_mdt_index,
1376                        old_lsm->lsm_md_master_mdt_index,
1377                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1378                        lsm->lsm_md_layout_version,
1379                        old_lsm->lsm_md_layout_version,
1380                        lsm->lsm_md_migrate_offset,
1381                        old_lsm->lsm_md_migrate_offset,
1382                        lsm->lsm_md_migrate_hash,
1383                        old_lsm->lsm_md_migrate_hash,
1384                        lsm->lsm_md_pool_name,
1385                        old_lsm->lsm_md_pool_name);
1386
1387                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++)
1388                         CDEBUG(mask, "old stripe[%d] "DFID"\n",
1389                                idx, PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1390
1391                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++)
1392                         CDEBUG(mask, "new stripe[%d] "DFID"\n",
1393                                idx, PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1394
1395                 if (!layout_changed)
1396                         RETURN(-EINVAL);
1397
1398                 ll_dir_clear_lsm_md(inode);
1399         }
1400
1401         /* set the directory layout */
1402         if (!lli->lli_lsm_md) {
1403                 struct cl_attr  *attr;
1404
1405                 rc = ll_init_lsm_md(inode, md);
1406                 if (rc != 0)
1407                         RETURN(rc);
1408
1409                 /* set md->lmv to NULL, so the following free lustre_md
1410                  * will not free this lsm */
1411                 md->lmv = NULL;
1412                 lli->lli_lsm_md = lsm;
1413
1414                 OBD_ALLOC_PTR(attr);
1415                 if (attr == NULL)
1416                         RETURN(-ENOMEM);
1417
1418                 /* validate the lsm */
1419                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1420                                    ll_md_blocking_ast);
1421                 if (rc != 0) {
1422                         OBD_FREE_PTR(attr);
1423                         RETURN(rc);
1424                 }
1425
1426                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1427                         md->body->mbo_nlink = attr->cat_nlink;
1428                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1429                         md->body->mbo_size = attr->cat_size;
1430                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1431                         md->body->mbo_atime = attr->cat_atime;
1432                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1433                         md->body->mbo_ctime = attr->cat_ctime;
1434                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1435                         md->body->mbo_mtime = attr->cat_mtime;
1436
1437                 OBD_FREE_PTR(attr);
1438
1439                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1440                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1441                 RETURN(0);
1442         }
1443
1444         /* Compare the old and new stripe information */
1445         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1446                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1447                 int                     idx;
1448
1449                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1450                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1451                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1452                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1453                        inode, lsm, old_lsm,
1454                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1455                        lsm->lsm_md_stripe_count,
1456                        old_lsm->lsm_md_stripe_count,
1457                        lsm->lsm_md_master_mdt_index,
1458                        old_lsm->lsm_md_master_mdt_index,
1459                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1460                        lsm->lsm_md_layout_version,
1461                        old_lsm->lsm_md_layout_version,
1462                        lsm->lsm_md_pool_name,
1463                        old_lsm->lsm_md_pool_name);
1464
1465                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1466                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1467                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1468                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1469                 }
1470
1471                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1472                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1473                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1474                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1475                 }
1476
1477                 RETURN(-EIO);
1478         }
1479
1480         RETURN(0);
1481 }
1482
1483 void ll_clear_inode(struct inode *inode)
1484 {
1485         struct ll_inode_info *lli = ll_i2info(inode);
1486         struct ll_sb_info *sbi = ll_i2sbi(inode);
1487         ENTRY;
1488
1489         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1490                PFID(ll_inode2fid(inode)), inode);
1491
1492         if (S_ISDIR(inode->i_mode)) {
1493                 /* these should have been cleared in ll_file_release */
1494                 LASSERT(lli->lli_opendir_key == NULL);
1495                 LASSERT(lli->lli_sai == NULL);
1496                 LASSERT(lli->lli_opendir_pid == 0);
1497         }
1498
1499         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1500
1501         LASSERT(!lli->lli_open_fd_write_count);
1502         LASSERT(!lli->lli_open_fd_read_count);
1503         LASSERT(!lli->lli_open_fd_exec_count);
1504
1505         if (lli->lli_mds_write_och)
1506                 ll_md_real_close(inode, FMODE_WRITE);
1507         if (lli->lli_mds_exec_och)
1508                 ll_md_real_close(inode, FMODE_EXEC);
1509         if (lli->lli_mds_read_och)
1510                 ll_md_real_close(inode, FMODE_READ);
1511
1512         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1513                 OBD_FREE(lli->lli_symlink_name,
1514                          strlen(lli->lli_symlink_name) + 1);
1515                 lli->lli_symlink_name = NULL;
1516         }
1517
1518         ll_xattr_cache_destroy(inode);
1519
1520 #ifdef CONFIG_FS_POSIX_ACL
1521         forget_all_cached_acls(inode);
1522         if (lli->lli_posix_acl) {
1523                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1524                 posix_acl_release(lli->lli_posix_acl);
1525                 lli->lli_posix_acl = NULL;
1526         }
1527 #endif
1528         lli->lli_inode_magic = LLI_INODE_DEAD;
1529
1530         if (S_ISDIR(inode->i_mode))
1531                 ll_dir_clear_lsm_md(inode);
1532         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1533                 LASSERT(list_empty(&lli->lli_agl_list));
1534
1535         /*
1536          * XXX This has to be done before lsm is freed below, because
1537          * cl_object still uses inode lsm.
1538          */
1539         cl_inode_fini(inode);
1540
1541         EXIT;
1542 }
1543
1544 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1545 {
1546         struct lustre_md md;
1547         struct inode *inode = dentry->d_inode;
1548         struct ll_sb_info *sbi = ll_i2sbi(inode);
1549         struct ptlrpc_request *request = NULL;
1550         int rc, ia_valid;
1551         ENTRY;
1552
1553         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1554                                      LUSTRE_OPC_ANY, NULL);
1555         if (IS_ERR(op_data))
1556                 RETURN(PTR_ERR(op_data));
1557
1558         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1559         if (rc) {
1560                 ptlrpc_req_finished(request);
1561                 if (rc == -ENOENT) {
1562                         clear_nlink(inode);
1563                         /* Unlinked special device node? Or just a race?
1564                          * Pretend we done everything. */
1565                         if (!S_ISREG(inode->i_mode) &&
1566                             !S_ISDIR(inode->i_mode)) {
1567                                 ia_valid = op_data->op_attr.ia_valid;
1568                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1569                                 rc = simple_setattr(dentry, &op_data->op_attr);
1570                                 op_data->op_attr.ia_valid = ia_valid;
1571                         }
1572                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1573                         CERROR("md_setattr fails: rc = %d\n", rc);
1574                 }
1575                 RETURN(rc);
1576         }
1577
1578         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1579                               sbi->ll_md_exp, &md);
1580         if (rc) {
1581                 ptlrpc_req_finished(request);
1582                 RETURN(rc);
1583         }
1584
1585         ia_valid = op_data->op_attr.ia_valid;
1586         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1587          * cache is not cleared yet. */
1588         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1589         if (S_ISREG(inode->i_mode))
1590                 inode_lock(inode);
1591         rc = simple_setattr(dentry, &op_data->op_attr);
1592         if (S_ISREG(inode->i_mode))
1593                 inode_unlock(inode);
1594         op_data->op_attr.ia_valid = ia_valid;
1595
1596         rc = ll_update_inode(inode, &md);
1597         ptlrpc_req_finished(request);
1598
1599         RETURN(rc);
1600 }
1601
1602 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1603  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1604  * keep these values until such a time that objects are allocated for it.
1605  * We do the MDS operations first, as it is checking permissions for us.
1606  * We don't to the MDS RPC if there is nothing that we want to store there,
1607  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1608  * going to do an RPC anyways.
1609  *
1610  * If we are doing a truncate, we will send the mtime and ctime updates
1611  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1612  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1613  * at the same time.
1614  *
1615  * In case of HSMimport, we only set attr on MDS.
1616  */
1617 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1618                    enum op_xvalid xvalid, bool hsm_import)
1619 {
1620         struct inode *inode = dentry->d_inode;
1621         struct ll_inode_info *lli = ll_i2info(inode);
1622         struct md_op_data *op_data = NULL;
1623         int rc = 0;
1624         ENTRY;
1625
1626         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1627                "valid %x, hsm_import %d\n",
1628                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1629                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1630                hsm_import);
1631
1632         if (attr->ia_valid & ATTR_SIZE) {
1633                 /* Check new size against VFS/VM file size limit and rlimit */
1634                 rc = inode_newsize_ok(inode, attr->ia_size);
1635                 if (rc)
1636                         RETURN(rc);
1637
1638                 /* The maximum Lustre file size is variable, based on the
1639                  * OST maximum object size and number of stripes.  This
1640                  * needs another check in addition to the VFS check above. */
1641                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1642                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1643                                PFID(&lli->lli_fid), attr->ia_size,
1644                                ll_file_maxbytes(inode));
1645                         RETURN(-EFBIG);
1646                 }
1647
1648                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1649         }
1650
1651         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1652         if (attr->ia_valid & TIMES_SET_FLAGS) {
1653                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1654                     !cfs_capable(CFS_CAP_FOWNER))
1655                         RETURN(-EPERM);
1656         }
1657
1658         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1659         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1660              (attr->ia_valid & ATTR_CTIME)) {
1661                 attr->ia_ctime = current_time(inode);
1662                 xvalid |= OP_XVALID_CTIME_SET;
1663         }
1664         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1665             (attr->ia_valid & ATTR_ATIME)) {
1666                 attr->ia_atime = current_time(inode);
1667                 attr->ia_valid |= ATTR_ATIME_SET;
1668         }
1669         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1670             (attr->ia_valid & ATTR_MTIME)) {
1671                 attr->ia_mtime = current_time(inode);
1672                 attr->ia_valid |= ATTR_MTIME_SET;
1673         }
1674
1675         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1676                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1677                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1678                        (s64)ktime_get_real_seconds());
1679
1680         if (S_ISREG(inode->i_mode)) {
1681                 if (attr->ia_valid & ATTR_SIZE)
1682                         inode_dio_write_done(inode);
1683                 inode_unlock(inode);
1684         }
1685
1686         /* We always do an MDS RPC, even if we're only changing the size;
1687          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1688
1689         OBD_ALLOC_PTR(op_data);
1690         if (op_data == NULL)
1691                 GOTO(out, rc = -ENOMEM);
1692
1693         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1694                 /* If we are changing file size, file content is
1695                  * modified, flag it.
1696                  */
1697                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1698                 op_data->op_bias |= MDS_DATA_MODIFIED;
1699                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1700         }
1701
1702         if (attr->ia_valid & ATTR_FILE) {
1703                 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1704
1705                 if (fd->fd_lease_och)
1706                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1707         }
1708
1709         op_data->op_attr = *attr;
1710         op_data->op_xvalid = xvalid;
1711
1712         rc = ll_md_setattr(dentry, op_data);
1713         if (rc)
1714                 GOTO(out, rc);
1715
1716         if (!S_ISREG(inode->i_mode) || hsm_import)
1717                 GOTO(out, rc = 0);
1718
1719         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1720                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1721             xvalid & OP_XVALID_CTIME_SET) {
1722                 /* For truncate and utimes sending attributes to OSTs, setting
1723                  * mtime/atime to the past will be performed under PW [0:EOF]
1724                  * extent lock (new_size:EOF for truncate).  It may seem
1725                  * excessive to send mtime/atime updates to OSTs when not
1726                  * setting times to past, but it is necessary due to possible
1727                  * time de-synchronization between MDT inode and OST objects
1728                  */
1729                 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1730         }
1731
1732         /* If the file was restored, it needs to set dirty flag.
1733          *
1734          * We've already sent MDS_DATA_MODIFIED flag in
1735          * ll_md_setattr() for truncate. However, the MDT refuses to
1736          * set the HS_DIRTY flag on released files, so we have to set
1737          * it again if the file has been restored. Please check how
1738          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1739          *
1740          * Please notice that if the file is not released, the previous
1741          * MDS_DATA_MODIFIED has taken effect and usually
1742          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1743          * This way we can save an RPC for common open + trunc
1744          * operation. */
1745         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1746                 struct hsm_state_set hss = {
1747                         .hss_valid = HSS_SETMASK,
1748                         .hss_setmask = HS_DIRTY,
1749                 };
1750                 int rc2;
1751
1752                 rc2 = ll_hsm_state_set(inode, &hss);
1753                 /* truncate and write can happen at the same time, so that
1754                  * the file can be set modified even though the file is not
1755                  * restored from released state, and ll_hsm_state_set() is
1756                  * not applicable for the file, and rc2 < 0 is normal in this
1757                  * case. */
1758                 if (rc2 < 0)
1759                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1760                                PFID(ll_inode2fid(inode)), rc2);
1761         }
1762
1763         EXIT;
1764 out:
1765         if (op_data != NULL)
1766                 ll_finish_md_op_data(op_data);
1767
1768         if (S_ISREG(inode->i_mode)) {
1769                 inode_lock(inode);
1770                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1771                         inode_dio_wait(inode);
1772                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1773                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1774                  * inode flags, so there is a gap where S_NOSEC is not set.
1775                  * This can cause a writer to take the i_mutex unnecessarily,
1776                  * but this is safe to do and should be rare. */
1777                 inode_has_no_xattr(inode);
1778         }
1779
1780         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1781                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1782
1783         return rc;
1784 }
1785
1786 int ll_setattr(struct dentry *de, struct iattr *attr)
1787 {
1788         int mode = de->d_inode->i_mode;
1789         enum op_xvalid xvalid = 0;
1790
1791         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1792                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1793                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1794
1795         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1796                                (ATTR_SIZE|ATTR_MODE)) &&
1797             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1798              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1799               !(attr->ia_mode & S_ISGID))))
1800                 attr->ia_valid |= ATTR_FORCE;
1801
1802         if ((attr->ia_valid & ATTR_MODE) &&
1803             (mode & S_ISUID) &&
1804             !(attr->ia_mode & S_ISUID) &&
1805             !(attr->ia_valid & ATTR_KILL_SUID))
1806                 attr->ia_valid |= ATTR_KILL_SUID;
1807
1808         if ((attr->ia_valid & ATTR_MODE) &&
1809             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1810             !(attr->ia_mode & S_ISGID) &&
1811             !(attr->ia_valid & ATTR_KILL_SGID))
1812                 attr->ia_valid |= ATTR_KILL_SGID;
1813
1814         return ll_setattr_raw(de, attr, xvalid, false);
1815 }
1816
1817 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1818                        u32 flags)
1819 {
1820         struct obd_statfs obd_osfs;
1821         time64_t max_age;
1822         int rc;
1823
1824         ENTRY;
1825         max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1826
1827         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1828         if (rc) {
1829                 CERROR("md_statfs fails: rc = %d\n", rc);
1830                 RETURN(rc);
1831         }
1832
1833         osfs->os_type = LL_SUPER_MAGIC;
1834
1835         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1836                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1837
1838         if (osfs->os_state & OS_STATE_SUM)
1839                 GOTO(out, rc);
1840
1841         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1842                 flags |= OBD_STATFS_NODELAY;
1843
1844         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1845         if (rc) {
1846                 CERROR("obd_statfs fails: rc = %d\n", rc);
1847                 RETURN(rc);
1848         }
1849
1850         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1851                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1852                obd_osfs.os_files);
1853
1854         osfs->os_bsize = obd_osfs.os_bsize;
1855         osfs->os_blocks = obd_osfs.os_blocks;
1856         osfs->os_bfree = obd_osfs.os_bfree;
1857         osfs->os_bavail = obd_osfs.os_bavail;
1858
1859         /* If we don't have as many objects free on the OST as inodes
1860          * on the MDS, we reduce the total number of inodes to
1861          * compensate, so that the "inodes in use" number is correct.
1862          */
1863         if (obd_osfs.os_ffree < osfs->os_ffree) {
1864                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1865                         obd_osfs.os_ffree;
1866                 osfs->os_ffree = obd_osfs.os_ffree;
1867         }
1868
1869 out:
1870         RETURN(rc);
1871 }
1872 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1873 {
1874         struct super_block *sb = de->d_sb;
1875         struct obd_statfs osfs;
1876         __u64 fsid = huge_encode_dev(sb->s_dev);
1877         int rc;
1878
1879         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1880         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1881
1882         /* Some amount of caching on the client is allowed */
1883         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1884         if (rc)
1885                 return rc;
1886
1887         statfs_unpack(sfs, &osfs);
1888
1889         /* We need to downshift for all 32-bit kernels, because we can't
1890          * tell if the kernel is being called via sys_statfs64() or not.
1891          * Stop before overflowing f_bsize - in which case it is better
1892          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1893         if (sizeof(long) < 8) {
1894                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1895                         sfs->f_bsize <<= 1;
1896
1897                         osfs.os_blocks >>= 1;
1898                         osfs.os_bfree >>= 1;
1899                         osfs.os_bavail >>= 1;
1900                 }
1901         }
1902
1903         sfs->f_blocks = osfs.os_blocks;
1904         sfs->f_bfree = osfs.os_bfree;
1905         sfs->f_bavail = osfs.os_bavail;
1906         sfs->f_fsid.val[0] = (__u32)fsid;
1907         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1908         return 0;
1909 }
1910
1911 void ll_inode_size_lock(struct inode *inode)
1912 {
1913         struct ll_inode_info *lli;
1914
1915         LASSERT(!S_ISDIR(inode->i_mode));
1916
1917         lli = ll_i2info(inode);
1918         mutex_lock(&lli->lli_size_mutex);
1919 }
1920
1921 void ll_inode_size_unlock(struct inode *inode)
1922 {
1923         struct ll_inode_info *lli;
1924
1925         lli = ll_i2info(inode);
1926         mutex_unlock(&lli->lli_size_mutex);
1927 }
1928
1929 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1930 {
1931         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1932         if (ext_flags & LUSTRE_PROJINHERIT_FL)
1933                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1934         else
1935                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1936 }
1937
1938 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1939 {
1940         struct ll_inode_info *lli = ll_i2info(inode);
1941         struct mdt_body *body = md->body;
1942         struct ll_sb_info *sbi = ll_i2sbi(inode);
1943         int rc = 0;
1944
1945         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1946                 rc = cl_file_inode_init(inode, md);
1947                 if (rc)
1948                         return rc;
1949         }
1950
1951         if (S_ISDIR(inode->i_mode)) {
1952                 rc = ll_update_lsm_md(inode, md);
1953                 if (rc != 0)
1954                         return rc;
1955         }
1956
1957 #ifdef CONFIG_FS_POSIX_ACL
1958         if (body->mbo_valid & OBD_MD_FLACL) {
1959                 spin_lock(&lli->lli_lock);
1960                 if (lli->lli_posix_acl)
1961                         posix_acl_release(lli->lli_posix_acl);
1962                 lli->lli_posix_acl = md->posix_acl;
1963                 spin_unlock(&lli->lli_lock);
1964         }
1965 #endif
1966         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1967                                         sbi->ll_flags & LL_SBI_32BIT_API);
1968         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1969
1970         if (body->mbo_valid & OBD_MD_FLATIME) {
1971                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1972                         LTIME_S(inode->i_atime) = body->mbo_atime;
1973                 lli->lli_atime = body->mbo_atime;
1974         }
1975
1976         if (body->mbo_valid & OBD_MD_FLMTIME) {
1977                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1978                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1979                                "to %llu\n", inode->i_ino,
1980                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1981                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1982                 }
1983                 lli->lli_mtime = body->mbo_mtime;
1984         }
1985
1986         if (body->mbo_valid & OBD_MD_FLCTIME) {
1987                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1988                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1989                 lli->lli_ctime = body->mbo_ctime;
1990         }
1991
1992         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1993         if (body->mbo_valid & OBD_MD_FLFLAGS)
1994                 ll_update_inode_flags(inode, body->mbo_flags);
1995         if (body->mbo_valid & OBD_MD_FLMODE)
1996                 inode->i_mode = (inode->i_mode & S_IFMT) |
1997                                 (body->mbo_mode & ~S_IFMT);
1998
1999         if (body->mbo_valid & OBD_MD_FLTYPE)
2000                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2001                                 (body->mbo_mode & S_IFMT);
2002
2003         LASSERT(inode->i_mode != 0);
2004         if (S_ISREG(inode->i_mode))
2005                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2006                                        LL_MAX_BLKSIZE_BITS);
2007         else
2008                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2009
2010         if (body->mbo_valid & OBD_MD_FLUID)
2011                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2012         if (body->mbo_valid & OBD_MD_FLGID)
2013                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2014         if (body->mbo_valid & OBD_MD_FLPROJID)
2015                 lli->lli_projid = body->mbo_projid;
2016         if (body->mbo_valid & OBD_MD_FLNLINK)
2017                 set_nlink(inode, body->mbo_nlink);
2018         if (body->mbo_valid & OBD_MD_FLRDEV)
2019                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2020
2021         if (body->mbo_valid & OBD_MD_FLID) {
2022                 /* FID shouldn't be changed! */
2023                 if (fid_is_sane(&lli->lli_fid)) {
2024                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2025                                  "Trying to change FID "DFID
2026                                  " to the "DFID", inode "DFID"(%p)\n",
2027                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2028                                  PFID(ll_inode2fid(inode)), inode);
2029                 } else {
2030                         lli->lli_fid = body->mbo_fid1;
2031                 }
2032         }
2033
2034         LASSERT(fid_seq(&lli->lli_fid) != 0);
2035
2036         if (body->mbo_valid & OBD_MD_FLSIZE) {
2037                 i_size_write(inode, body->mbo_size);
2038
2039                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2040                        PFID(ll_inode2fid(inode)),
2041                        (unsigned long long)body->mbo_size);
2042
2043                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2044                         inode->i_blocks = body->mbo_blocks;
2045         }
2046
2047         if (body->mbo_valid & OBD_MD_TSTATE) {
2048                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2049                  * clear it when done to ensure to start again
2050                  * glimpsing updated attrs
2051                  */
2052                 if (body->mbo_t_state & MS_RESTORE)
2053                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2054                 else
2055                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2056         }
2057
2058         return 0;
2059 }
2060
2061 int ll_read_inode2(struct inode *inode, void *opaque)
2062 {
2063         struct lustre_md *md = opaque;
2064         struct ll_inode_info *lli = ll_i2info(inode);
2065         int     rc;
2066         ENTRY;
2067
2068         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2069                PFID(&lli->lli_fid), inode);
2070
2071         /* Core attributes from the MDS first.  This is a new inode, and
2072          * the VFS doesn't zero times in the core inode so we have to do
2073          * it ourselves.  They will be overwritten by either MDS or OST
2074          * attributes - we just need to make sure they aren't newer. */
2075         LTIME_S(inode->i_mtime) = 0;
2076         LTIME_S(inode->i_atime) = 0;
2077         LTIME_S(inode->i_ctime) = 0;
2078         inode->i_rdev = 0;
2079         rc = ll_update_inode(inode, md);
2080         if (rc != 0)
2081                 RETURN(rc);
2082
2083         /* OIDEBUG(inode); */
2084
2085 #ifdef HAVE_BACKING_DEV_INFO
2086         /* initializing backing dev info. */
2087         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2088 #endif
2089         if (S_ISREG(inode->i_mode)) {
2090                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2091                 inode->i_op = &ll_file_inode_operations;
2092                 inode->i_fop = sbi->ll_fop;
2093                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2094                 EXIT;
2095         } else if (S_ISDIR(inode->i_mode)) {
2096                 inode->i_op = &ll_dir_inode_operations;
2097                 inode->i_fop = &ll_dir_operations;
2098                 EXIT;
2099         } else if (S_ISLNK(inode->i_mode)) {
2100                 inode->i_op = &ll_fast_symlink_inode_operations;
2101                 EXIT;
2102         } else {
2103                 inode->i_op = &ll_special_inode_operations;
2104
2105                 init_special_inode(inode, inode->i_mode,
2106                                    inode->i_rdev);
2107
2108                 EXIT;
2109         }
2110
2111         return 0;
2112 }
2113
2114 void ll_delete_inode(struct inode *inode)
2115 {
2116         struct ll_inode_info *lli = ll_i2info(inode);
2117         ENTRY;
2118
2119         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2120                 /* It is last chance to write out dirty pages,
2121                  * otherwise we may lose data while umount */
2122                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2123
2124         truncate_inode_pages_final(&inode->i_data);
2125
2126         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2127                  "see https://jira.whamcloud.com/browse/LU-118\n",
2128                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2129
2130 #ifdef HAVE_SBOPS_EVICT_INODE
2131         ll_clear_inode(inode);
2132 #endif
2133         clear_inode(inode);
2134
2135         EXIT;
2136 }
2137
2138 int ll_iocontrol(struct inode *inode, struct file *file,
2139                  unsigned int cmd, unsigned long arg)
2140 {
2141         struct ll_sb_info *sbi = ll_i2sbi(inode);
2142         struct ptlrpc_request *req = NULL;
2143         int rc, flags = 0;
2144         ENTRY;
2145
2146         switch (cmd) {
2147         case FS_IOC_GETFLAGS: {
2148                 struct mdt_body *body;
2149                 struct md_op_data *op_data;
2150
2151                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2152                                              0, 0, LUSTRE_OPC_ANY,
2153                                              NULL);
2154                 if (IS_ERR(op_data))
2155                         RETURN(PTR_ERR(op_data));
2156
2157                 op_data->op_valid = OBD_MD_FLFLAGS;
2158                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2159                 ll_finish_md_op_data(op_data);
2160                 if (rc) {
2161                         CERROR("%s: failure inode "DFID": rc = %d\n",
2162                                sbi->ll_md_exp->exp_obd->obd_name,
2163                                PFID(ll_inode2fid(inode)), rc);
2164                         RETURN(-abs(rc));
2165                 }
2166
2167                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2168
2169                 flags = body->mbo_flags;
2170
2171                 ptlrpc_req_finished(req);
2172
2173                 RETURN(put_user(flags, (int __user *)arg));
2174         }
2175         case FS_IOC_SETFLAGS: {
2176                 struct iattr *attr;
2177                 struct md_op_data *op_data;
2178                 struct cl_object *obj;
2179
2180                 if (get_user(flags, (int __user *)arg))
2181                         RETURN(-EFAULT);
2182
2183                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2184                                              LUSTRE_OPC_ANY, NULL);
2185                 if (IS_ERR(op_data))
2186                         RETURN(PTR_ERR(op_data));
2187
2188                 op_data->op_attr_flags = flags;
2189                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2190                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2191                 ll_finish_md_op_data(op_data);
2192                 ptlrpc_req_finished(req);
2193                 if (rc)
2194                         RETURN(rc);
2195
2196                 ll_update_inode_flags(inode, flags);
2197
2198                 obj = ll_i2info(inode)->lli_clob;
2199                 if (obj == NULL)
2200                         RETURN(0);
2201
2202                 OBD_ALLOC_PTR(attr);
2203                 if (attr == NULL)
2204                         RETURN(-ENOMEM);
2205
2206                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2207
2208                 OBD_FREE_PTR(attr);
2209                 RETURN(rc);
2210         }
2211         default:
2212                 RETURN(-ENOSYS);
2213         }
2214
2215         RETURN(0);
2216 }
2217
2218 int ll_flush_ctx(struct inode *inode)
2219 {
2220         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2221
2222         CDEBUG(D_SEC, "flush context for user %d\n",
2223                from_kuid(&init_user_ns, current_uid()));
2224
2225         obd_set_info_async(NULL, sbi->ll_md_exp,
2226                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2227                            0, NULL, NULL);
2228         obd_set_info_async(NULL, sbi->ll_dt_exp,
2229                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2230                            0, NULL, NULL);
2231         return 0;
2232 }
2233
2234 /* umount -f client means force down, don't save state */
2235 void ll_umount_begin(struct super_block *sb)
2236 {
2237         struct ll_sb_info *sbi = ll_s2sbi(sb);
2238         struct obd_device *obd;
2239         struct obd_ioctl_data *ioc_data;
2240         struct l_wait_info lwi;
2241         wait_queue_head_t waitq;
2242         ENTRY;
2243
2244         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2245                sb->s_count, atomic_read(&sb->s_active));
2246
2247         obd = class_exp2obd(sbi->ll_md_exp);
2248         if (obd == NULL) {
2249                 CERROR("Invalid MDC connection handle %#llx\n",
2250                        sbi->ll_md_exp->exp_handle.h_cookie);
2251                 EXIT;
2252                 return;
2253         }
2254         obd->obd_force = 1;
2255
2256         obd = class_exp2obd(sbi->ll_dt_exp);
2257         if (obd == NULL) {
2258                 CERROR("Invalid LOV connection handle %#llx\n",
2259                        sbi->ll_dt_exp->exp_handle.h_cookie);
2260                 EXIT;
2261                 return;
2262         }
2263         obd->obd_force = 1;
2264
2265         OBD_ALLOC_PTR(ioc_data);
2266         if (ioc_data) {
2267                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2268                               sizeof *ioc_data, ioc_data, NULL);
2269
2270                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2271                               sizeof *ioc_data, ioc_data, NULL);
2272
2273                 OBD_FREE_PTR(ioc_data);
2274         }
2275
2276         /* Really, we'd like to wait until there are no requests outstanding,
2277          * and then continue.  For now, we just periodically checking for vfs
2278          * to decrement mnt_cnt and hope to finish it within 10sec.
2279          */
2280         init_waitqueue_head(&waitq);
2281         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2282                                    cfs_time_seconds(1), NULL, NULL);
2283         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2284
2285         EXIT;
2286 }
2287
2288 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2289 {
2290         struct ll_sb_info *sbi = ll_s2sbi(sb);
2291         char *profilenm = get_profile_name(sb);
2292         int err;
2293         __u32 read_only;
2294
2295         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2296                 read_only = *flags & MS_RDONLY;
2297                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2298                                          sizeof(KEY_READ_ONLY),
2299                                          KEY_READ_ONLY, sizeof(read_only),
2300                                          &read_only, NULL);
2301                 if (err) {
2302                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2303                                       profilenm, read_only ?
2304                                       "read-only" : "read-write", err);
2305                         return err;
2306                 }
2307
2308                 if (read_only)
2309                         sb->s_flags |= MS_RDONLY;
2310                 else
2311                         sb->s_flags &= ~MS_RDONLY;
2312
2313                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2314                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2315                                       read_only ?  "read-only" : "read-write");
2316         }
2317         return 0;
2318 }
2319
2320 /**
2321  * Cleanup the open handle that is cached on MDT-side.
2322  *
2323  * For open case, the client side open handling thread may hit error
2324  * after the MDT grant the open. Under such case, the client should
2325  * send close RPC to the MDT as cleanup; otherwise, the open handle
2326  * on the MDT will be leaked there until the client umount or evicted.
2327  *
2328  * In further, if someone unlinked the file, because the open handle
2329  * holds the reference on such file/object, then it will block the
2330  * subsequent threads that want to locate such object via FID.
2331  *
2332  * \param[in] sb        super block for this file-system
2333  * \param[in] open_req  pointer to the original open request
2334  */
2335 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2336 {
2337         struct mdt_body                 *body;
2338         struct md_op_data               *op_data;
2339         struct ptlrpc_request           *close_req = NULL;
2340         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2341         ENTRY;
2342
2343         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2344         OBD_ALLOC_PTR(op_data);
2345         if (op_data == NULL) {
2346                 CWARN("%s: cannot allocate op_data to release open handle for "
2347                       DFID"\n",
2348                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2349
2350                 RETURN_EXIT;
2351         }
2352
2353         op_data->op_fid1 = body->mbo_fid1;
2354         op_data->op_open_handle = body->mbo_open_handle;
2355         op_data->op_mod_time = ktime_get_real_seconds();
2356         md_close(exp, op_data, NULL, &close_req);
2357         ptlrpc_req_finished(close_req);
2358         ll_finish_md_op_data(op_data);
2359
2360         EXIT;
2361 }
2362
2363 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2364                   struct super_block *sb, struct lookup_intent *it)
2365 {
2366         struct ll_sb_info *sbi = NULL;
2367         struct lustre_md md = { NULL };
2368         int rc;
2369         ENTRY;
2370
2371         LASSERT(*inode || sb);
2372         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2373         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2374                               sbi->ll_md_exp, &md);
2375         if (rc != 0)
2376                 GOTO(cleanup, rc);
2377
2378         if (*inode) {
2379                 rc = ll_update_inode(*inode, &md);
2380                 if (rc != 0)
2381                         GOTO(out, rc);
2382         } else {
2383                 LASSERT(sb != NULL);
2384
2385                 /*
2386                  * At this point server returns to client's same fid as client
2387                  * generated for creating. So using ->fid1 is okay here.
2388                  */
2389                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2390                         CERROR("%s: Fid is insane "DFID"\n",
2391                                 ll_get_fsname(sb, NULL, 0),
2392                                 PFID(&md.body->mbo_fid1));
2393                         GOTO(out, rc = -EINVAL);
2394                 }
2395
2396                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2397                                              sbi->ll_flags & LL_SBI_32BIT_API),
2398                                  &md);
2399                 if (IS_ERR(*inode)) {
2400 #ifdef CONFIG_FS_POSIX_ACL
2401                         if (md.posix_acl) {
2402                                 posix_acl_release(md.posix_acl);
2403                                 md.posix_acl = NULL;
2404                         }
2405 #endif
2406                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2407                         *inode = NULL;
2408                         CERROR("new_inode -fatal: rc %d\n", rc);
2409                         GOTO(out, rc);
2410                 }
2411         }
2412
2413         /* Handling piggyback layout lock.
2414          * Layout lock can be piggybacked by getattr and open request.
2415          * The lsm can be applied to inode only if it comes with a layout lock
2416          * otherwise correct layout may be overwritten, for example:
2417          * 1. proc1: mdt returns a lsm but not granting layout
2418          * 2. layout was changed by another client
2419          * 3. proc2: refresh layout and layout lock granted
2420          * 4. proc1: to apply a stale layout */
2421         if (it != NULL && it->it_lock_mode != 0) {
2422                 struct lustre_handle lockh;
2423                 struct ldlm_lock *lock;
2424
2425                 lockh.cookie = it->it_lock_handle;
2426                 lock = ldlm_handle2lock(&lockh);
2427                 LASSERT(lock != NULL);
2428                 if (ldlm_has_layout(lock)) {
2429                         struct cl_object_conf conf;
2430
2431                         memset(&conf, 0, sizeof(conf));
2432                         conf.coc_opc = OBJECT_CONF_SET;
2433                         conf.coc_inode = *inode;
2434                         conf.coc_lock = lock;
2435                         conf.u.coc_layout = md.layout;
2436                         (void)ll_layout_conf(*inode, &conf);
2437                 }
2438                 LDLM_LOCK_PUT(lock);
2439         }
2440
2441         GOTO(out, rc = 0);
2442
2443 out:
2444         md_free_lustre_md(sbi->ll_md_exp, &md);
2445
2446 cleanup:
2447         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2448                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2449
2450         return rc;
2451 }
2452
2453 int ll_obd_statfs(struct inode *inode, void __user *arg)
2454 {
2455         struct ll_sb_info *sbi = NULL;
2456         struct obd_export *exp;
2457         char *buf = NULL;
2458         struct obd_ioctl_data *data = NULL;
2459         __u32 type;
2460         int len = 0, rc;
2461
2462         if (!inode || !(sbi = ll_i2sbi(inode)))
2463                 GOTO(out_statfs, rc = -EINVAL);
2464
2465         rc = obd_ioctl_getdata(&buf, &len, arg);
2466         if (rc)
2467                 GOTO(out_statfs, rc);
2468
2469         data = (void*)buf;
2470         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2471             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2472                 GOTO(out_statfs, rc = -EINVAL);
2473
2474         if (data->ioc_inllen1 != sizeof(__u32) ||
2475             data->ioc_inllen2 != sizeof(__u32) ||
2476             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2477             data->ioc_plen2 != sizeof(struct obd_uuid))
2478                 GOTO(out_statfs, rc = -EINVAL);
2479
2480         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2481         if (type & LL_STATFS_LMV)
2482                 exp = sbi->ll_md_exp;
2483         else if (type & LL_STATFS_LOV)
2484                 exp = sbi->ll_dt_exp;
2485         else
2486                 GOTO(out_statfs, rc = -ENODEV);
2487
2488         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2489         if (rc)
2490                 GOTO(out_statfs, rc);
2491 out_statfs:
2492         OBD_FREE_LARGE(buf, len);
2493         return rc;
2494 }
2495
2496 /* this function prepares md_op_data hint for passing it down to MD stack. */
2497 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2498                                       struct inode *i1, struct inode *i2,
2499                                       const char *name, size_t namelen,
2500                                       __u32 mode, __u32 opc, void *data)
2501 {
2502         LASSERT(i1 != NULL);
2503
2504         if (name == NULL) {
2505                 /* Do not reuse namelen for something else. */
2506                 if (namelen != 0)
2507                         return ERR_PTR(-EINVAL);
2508         } else {
2509                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2510                         return ERR_PTR(-ENAMETOOLONG);
2511
2512                 if (!lu_name_is_valid_2(name, namelen))
2513                         return ERR_PTR(-EINVAL);
2514         }
2515
2516         if (op_data == NULL)
2517                 OBD_ALLOC_PTR(op_data);
2518
2519         if (op_data == NULL)
2520                 return ERR_PTR(-ENOMEM);
2521
2522         ll_i2gids(op_data->op_suppgids, i1, i2);
2523         op_data->op_fid1 = *ll_inode2fid(i1);
2524         op_data->op_default_stripe_offset = -1;
2525         if (S_ISDIR(i1->i_mode)) {
2526                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2527                 if (opc == LUSTRE_OPC_MKDIR)
2528                         op_data->op_default_stripe_offset =
2529                                    ll_i2info(i1)->lli_def_stripe_offset;
2530         }
2531
2532         if (i2) {
2533                 op_data->op_fid2 = *ll_inode2fid(i2);
2534                 if (S_ISDIR(i2->i_mode))
2535                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2536         } else {
2537                 fid_zero(&op_data->op_fid2);
2538         }
2539
2540         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2541                 op_data->op_cli_flags |= CLI_HASH64;
2542
2543         if (ll_need_32bit_api(ll_i2sbi(i1)))
2544                 op_data->op_cli_flags |= CLI_API32;
2545
2546         op_data->op_name = name;
2547         op_data->op_namelen = namelen;
2548         op_data->op_mode = mode;
2549         op_data->op_mod_time = ktime_get_real_seconds();
2550         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2551         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2552         op_data->op_cap = cfs_curproc_cap_pack();
2553         op_data->op_mds = 0;
2554         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2555              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2556                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2557         }
2558         op_data->op_data = data;
2559
2560         return op_data;
2561 }
2562
2563 void ll_finish_md_op_data(struct md_op_data *op_data)
2564 {
2565         security_release_secctx(op_data->op_file_secctx,
2566                                 op_data->op_file_secctx_size);
2567         OBD_FREE_PTR(op_data);
2568 }
2569
2570 #ifdef HAVE_SUPEROPS_USE_DENTRY
2571 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2572 #else
2573 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2574 #endif
2575 {
2576         struct ll_sb_info *sbi;
2577
2578 #ifdef HAVE_SUPEROPS_USE_DENTRY
2579         LASSERT((seq != NULL) && (dentry != NULL));
2580         sbi = ll_s2sbi(dentry->d_sb);
2581 #else
2582         LASSERT((seq != NULL) && (vfs != NULL));
2583         sbi = ll_s2sbi(vfs->mnt_sb);
2584 #endif
2585
2586         if (sbi->ll_flags & LL_SBI_NOLCK)
2587                 seq_puts(seq, ",nolock");
2588
2589         if (sbi->ll_flags & LL_SBI_FLOCK)
2590                 seq_puts(seq, ",flock");
2591
2592         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2593                 seq_puts(seq, ",localflock");
2594
2595         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2596                 seq_puts(seq, ",user_xattr");
2597
2598         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2599                 seq_puts(seq, ",lazystatfs");
2600
2601         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2602                 seq_puts(seq, ",user_fid2path");
2603
2604         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2605                 seq_puts(seq, ",always_ping");
2606
2607         RETURN(0);
2608 }
2609
2610 /**
2611  * Get obd name by cmd, and copy out to user space
2612  */
2613 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2614 {
2615         struct ll_sb_info *sbi = ll_i2sbi(inode);
2616         struct obd_device *obd;
2617         ENTRY;
2618
2619         if (cmd == OBD_IOC_GETDTNAME)
2620                 obd = class_exp2obd(sbi->ll_dt_exp);
2621         else if (cmd == OBD_IOC_GETMDNAME)
2622                 obd = class_exp2obd(sbi->ll_md_exp);
2623         else
2624                 RETURN(-EINVAL);
2625
2626         if (!obd)
2627                 RETURN(-ENOENT);
2628
2629         if (copy_to_user((void __user *)arg, obd->obd_name,
2630                          strlen(obd->obd_name) + 1))
2631                 RETURN(-EFAULT);
2632
2633         RETURN(0);
2634 }
2635
2636 /**
2637  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2638  * fsname will be returned in this buffer; otherwise, a static buffer will be
2639  * used to store the fsname and returned to caller.
2640  */
2641 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2642 {
2643         static char fsname_static[MTI_NAME_MAXLEN];
2644         struct lustre_sb_info *lsi = s2lsi(sb);
2645         char *ptr;
2646         int len;
2647
2648         if (buf == NULL) {
2649                 /* this means the caller wants to use static buffer
2650                  * and it doesn't care about race. Usually this is
2651                  * in error reporting path */
2652                 buf = fsname_static;
2653                 buflen = sizeof(fsname_static);
2654         }
2655
2656         len = strlen(lsi->lsi_lmd->lmd_profile);
2657         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2658         if (ptr && (strcmp(ptr, "-client") == 0))
2659                 len -= 7;
2660
2661         if (unlikely(len >= buflen))
2662                 len = buflen - 1;
2663         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2664         buf[len] = '\0';
2665
2666         return buf;
2667 }
2668
2669 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2670 {
2671         char *path = NULL;
2672
2673         struct path p;
2674
2675         p.dentry = dentry;
2676         p.mnt = current->fs->root.mnt;
2677         path_get(&p);
2678         path = d_path(&p, buf, bufsize);
2679         path_put(&p);
2680         return path;
2681 }
2682
2683 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2684 {
2685         char *buf, *path = NULL;
2686         struct dentry *dentry = NULL;
2687         struct inode *inode = page->mapping->host;
2688
2689         /* this can be called inside spin lock so use GFP_ATOMIC. */
2690         buf = (char *)__get_free_page(GFP_ATOMIC);
2691         if (buf != NULL) {
2692                 dentry = d_find_alias(page->mapping->host);
2693                 if (dentry != NULL)
2694                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2695         }
2696
2697         CDEBUG(D_WARNING,
2698                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2699                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2700                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2701                PFID(ll_inode2fid(inode)),
2702                (path && !IS_ERR(path)) ? path : "", ioret);
2703
2704         if (dentry != NULL)
2705                 dput(dentry);
2706
2707         if (buf != NULL)
2708                 free_page((unsigned long)buf);
2709 }
2710
2711 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2712                         struct lov_user_md **kbuf)
2713 {
2714         struct lov_user_md      lum;
2715         ssize_t                 lum_size;
2716         ENTRY;
2717
2718         if (copy_from_user(&lum, md, sizeof(lum)))
2719                 RETURN(-EFAULT);
2720
2721         lum_size = ll_lov_user_md_size(&lum);
2722         if (lum_size < 0)
2723                 RETURN(lum_size);
2724
2725         OBD_ALLOC(*kbuf, lum_size);
2726         if (*kbuf == NULL)
2727                 RETURN(-ENOMEM);
2728
2729         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2730                 OBD_FREE(*kbuf, lum_size);
2731                 RETURN(-EFAULT);
2732         }
2733
2734         RETURN(lum_size);
2735 }
2736
2737 /*
2738  * Compute llite root squash state after a change of root squash
2739  * configuration setting or add/remove of a lnet nid
2740  */
2741 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2742 {
2743         struct root_squash_info *squash = &sbi->ll_squash;
2744         int i;
2745         bool matched;
2746         struct lnet_process_id id;
2747
2748         /* Update norootsquash flag */
2749         down_write(&squash->rsi_sem);
2750         if (list_empty(&squash->rsi_nosquash_nids))
2751                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2752         else {
2753                 /* Do not apply root squash as soon as one of our NIDs is
2754                  * in the nosquash_nids list */
2755                 matched = false;
2756                 i = 0;
2757                 while (LNetGetId(i++, &id) != -ENOENT) {
2758                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2759                                 continue;
2760                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2761                                 matched = true;
2762                                 break;
2763                         }
2764                 }
2765                 if (matched)
2766                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2767                 else
2768                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2769         }
2770         up_write(&squash->rsi_sem);
2771 }
2772
2773 /**
2774  * Parse linkea content to extract information about a given hardlink
2775  *
2776  * \param[in]   ldata      - Initialized linkea data
2777  * \param[in]   linkno     - Link identifier
2778  * \param[out]  parent_fid - The entry's parent FID
2779  * \param[out]  ln         - Entry name destination buffer
2780  *
2781  * \retval 0 on success
2782  * \retval Appropriate negative error code on failure
2783  */
2784 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2785                             struct lu_fid *parent_fid, struct lu_name *ln)
2786 {
2787         unsigned int    idx;
2788         int             rc;
2789         ENTRY;
2790
2791         rc = linkea_init_with_rec(ldata);
2792         if (rc < 0)
2793                 RETURN(rc);
2794
2795         if (linkno >= ldata->ld_leh->leh_reccount)
2796                 /* beyond last link */
2797                 RETURN(-ENODATA);
2798
2799         linkea_first_entry(ldata);
2800         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2801                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2802                                     parent_fid);
2803                 if (idx == linkno)
2804                         break;
2805
2806                 linkea_next_entry(ldata);
2807         }
2808
2809         if (idx < linkno)
2810                 RETURN(-ENODATA);
2811
2812         RETURN(0);
2813 }
2814
2815 /**
2816  * Get parent FID and name of an identified link. Operation is performed for
2817  * a given link number, letting the caller iterate over linkno to list one or
2818  * all links of an entry.
2819  *
2820  * \param[in]     file - File descriptor against which to perform the operation
2821  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2822  *                       on and the available size. It is eventually filled with
2823  *                       the requested information or left untouched on error
2824  *
2825  * \retval - 0 on success
2826  * \retval - Appropriate negative error code on failure
2827  */
2828 int ll_getparent(struct file *file, struct getparent __user *arg)
2829 {
2830         struct inode            *inode = file_inode(file);
2831         struct linkea_data      *ldata;
2832         struct lu_buf            buf = LU_BUF_NULL;
2833         struct lu_name           ln;
2834         struct lu_fid            parent_fid;
2835         __u32                    linkno;
2836         __u32                    name_size;
2837         int                      rc;
2838
2839         ENTRY;
2840
2841         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2842             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2843                 RETURN(-EPERM);
2844
2845         if (get_user(name_size, &arg->gp_name_size))
2846                 RETURN(-EFAULT);
2847
2848         if (get_user(linkno, &arg->gp_linkno))
2849                 RETURN(-EFAULT);
2850
2851         if (name_size > PATH_MAX)
2852                 RETURN(-EINVAL);
2853
2854         OBD_ALLOC(ldata, sizeof(*ldata));
2855         if (ldata == NULL)
2856                 RETURN(-ENOMEM);
2857
2858         rc = linkea_data_new(ldata, &buf);
2859         if (rc < 0)
2860                 GOTO(ldata_free, rc);
2861
2862 #ifdef HAVE_XATTR_HANDLER_FLAGS
2863         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2864                            buf.lb_len, OBD_MD_FLXATTR);
2865 #else
2866         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2867                          buf.lb_len);
2868 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2869         if (rc < 0)
2870                 GOTO(lb_free, rc);
2871
2872         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2873         if (rc < 0)
2874                 GOTO(lb_free, rc);
2875
2876         if (ln.ln_namelen >= name_size)
2877                 GOTO(lb_free, rc = -EOVERFLOW);
2878
2879         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2880                 GOTO(lb_free, rc = -EFAULT);
2881
2882         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2883                 GOTO(lb_free, rc = -EFAULT);
2884
2885         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2886                 GOTO(lb_free, rc = -EFAULT);
2887
2888 lb_free:
2889         lu_buf_free(&buf);
2890 ldata_free:
2891         OBD_FREE(ldata, sizeof(*ldata));
2892
2893         RETURN(rc);
2894 }