Whamcloud - gitweb
34f162704b2bab3dccda38d2c187a574df40b2c5
[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                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY;
226
227 #ifdef HAVE_LRU_RESIZE_SUPPORT
228         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
229                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
230 #endif
231 #ifdef CONFIG_FS_POSIX_ACL
232         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
233                                    OBD_CONNECT_LARGE_ACL;
234 #endif
235
236         data->ocd_cksum_types = obd_cksum_types_supported_client();
237
238         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
239                 /* flag mdc connection as lightweight, only used for test
240                  * purpose, use with care */
241                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
242
243         data->ocd_ibits_known = MDS_INODELOCK_FULL;
244         data->ocd_version = LUSTRE_VERSION_CODE;
245
246         if (sb->s_flags & MS_RDONLY)
247                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
248         if (sbi->ll_flags & LL_SBI_USER_XATTR)
249                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
250
251 #ifdef MS_NOSEC
252         /* Setting this indicates we correctly support S_NOSEC (See kernel
253          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
254          */
255         sb->s_flags |= MS_NOSEC;
256 #endif
257
258         if (sbi->ll_flags & LL_SBI_FLOCK)
259                 sbi->ll_fop = &ll_file_operations_flock;
260         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
261                 sbi->ll_fop = &ll_file_operations;
262         else
263                 sbi->ll_fop = &ll_file_operations_noflock;
264
265         /* always ping even if server suppress_pings */
266         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
267                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
268
269         obd_connect_set_secctx(data);
270
271         data->ocd_brw_size = MD_MAX_BRW_SIZE;
272
273         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
274                           &sbi->ll_sb_uuid, data, NULL);
275         if (err == -EBUSY) {
276                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
277                                    "recovery, of which this client is not a "
278                                    "part. Please wait for recovery to complete,"
279                                    " abort, or time out.\n", md);
280                 GOTO(out, err);
281         } else if (err) {
282                 CERROR("cannot connect to %s: rc = %d\n", md, err);
283                 GOTO(out, err);
284         }
285
286         sbi->ll_md_exp->exp_connect_data = *data;
287
288         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
289                            LUSTRE_SEQ_METADATA);
290         if (err) {
291                 CERROR("%s: Can't init metadata layer FID infrastructure, "
292                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
293                 GOTO(out_md, err);
294         }
295
296         /* For mount, we only need fs info from MDT0, and also in DNE, it
297          * can make sure the client can be mounted as long as MDT0 is
298          * avaible */
299         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
300                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
301                         OBD_STATFS_FOR_MDT0);
302         if (err)
303                 GOTO(out_md_fid, err);
304
305         /* This needs to be after statfs to ensure connect has finished.
306          * Note that "data" does NOT contain the valid connect reply.
307          * If connecting to a 1.8 server there will be no LMV device, so
308          * we can access the MDC export directly and exp_connect_flags will
309          * be non-zero, but if accessing an upgraded 2.1 server it will
310          * have the correct flags filled in.
311          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
312         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
313         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
314             valid != CLIENT_CONNECT_MDT_REQD) {
315                 char *buf;
316
317                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
318                 obd_connect_flags2str(buf, PAGE_SIZE,
319                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
320                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
321                                    "feature(s) needed for correct operation "
322                                    "of this client (%s). Please upgrade "
323                                    "server or downgrade client.\n",
324                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
325                 OBD_FREE(buf, PAGE_SIZE);
326                 GOTO(out_md_fid, err = -EPROTO);
327         }
328
329         size = sizeof(*data);
330         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
331                            KEY_CONN_DATA,  &size, data);
332         if (err) {
333                 CERROR("%s: Get connect data failed: rc = %d\n",
334                        sbi->ll_md_exp->exp_obd->obd_name, err);
335                 GOTO(out_md_fid, err);
336         }
337
338         LASSERT(osfs->os_bsize);
339         sb->s_blocksize = osfs->os_bsize;
340         sb->s_blocksize_bits = log2(osfs->os_bsize);
341         sb->s_magic = LL_SUPER_MAGIC;
342         sb->s_maxbytes = MAX_LFS_FILESIZE;
343         sbi->ll_namelen = osfs->os_namelen;
344         sbi->ll_mnt.mnt = current->fs->root.mnt;
345
346         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
347             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
348                 LCONSOLE_INFO("Disabling user_xattr feature because "
349                               "it is not supported on the server\n");
350                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
351         }
352
353         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
354 #ifdef MS_POSIXACL
355                 sb->s_flags |= MS_POSIXACL;
356 #endif
357                 sbi->ll_flags |= LL_SBI_ACL;
358         } else {
359                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
360 #ifdef MS_POSIXACL
361                 sb->s_flags &= ~MS_POSIXACL;
362 #endif
363                 sbi->ll_flags &= ~LL_SBI_ACL;
364         }
365
366         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
367                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
368
369         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
370                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
371
372         if (obd_connect_has_secctx(data))
373                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
374
375         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
376                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
377                         LCONSOLE_INFO("%s: disabling xattr cache due to "
378                                       "unknown maximum xattr size.\n", dt);
379                 } else if (!sbi->ll_xattr_cache_set) {
380                         /* If xattr_cache is already set (no matter 0 or 1)
381                          * during processing llog, it won't be enabled here. */
382                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
383                         sbi->ll_xattr_cache_enabled = 1;
384                 }
385         }
386
387         sbi->ll_dt_obd = class_name2obd(dt);
388         if (!sbi->ll_dt_obd) {
389                 CERROR("DT %s: not setup or attached\n", dt);
390                 GOTO(out_md_fid, err = -ENODEV);
391         }
392
393         /* pass client page size via ocd_grant_blkbits, the server should report
394          * back its backend blocksize for grant calculation purpose */
395         data->ocd_grant_blkbits = PAGE_SHIFT;
396
397         /* indicate OST features supported by this client */
398         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
399                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
400                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
401                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
402                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
403                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
404                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
405                                   OBD_CONNECT_EINPROGRESS |
406                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
407                                   OBD_CONNECT_LAYOUTLOCK |
408                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
409                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
410                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
411
412 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
413  * can interoperate with an older version of lockahead which was released prior
414  * to landing in master. This support will be dropped when 2.13 development
415  * starts.  At the point, we should not just drop the connect flag (below), we
416  * should also remove the support in the code.
417  *
418  * Removing it means a few things:
419  * 1. Remove this section here
420  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
421  * 3. Remove function exp_connect_lockahead_old
422  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
423  * */
424 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
425         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
426 #endif
427
428         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
429
430         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
431                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
432
433         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
434          * disabled by default, because it can still be enabled on the
435          * fly via /sys. As a consequence, we still need to come to an
436          * agreement on the supported algorithms at connect time
437          */
438         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
439
440         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
441                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
442         else
443                 data->ocd_cksum_types = obd_cksum_types_supported_client();
444
445 #ifdef HAVE_LRU_RESIZE_SUPPORT
446         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
447 #endif
448         /* always ping even if server suppress_pings */
449         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
450                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
451
452         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
453                "ocd_grant: %d\n", data->ocd_connect_flags,
454                data->ocd_version, data->ocd_grant);
455
456         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
457         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
458
459         data->ocd_brw_size = DT_MAX_BRW_SIZE;
460
461         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
462                           &sbi->ll_sb_uuid, data, NULL);
463         if (err == -EBUSY) {
464                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
465                                    "recovery, of which this client is not a "
466                                    "part.  Please wait for recovery to "
467                                    "complete, abort, or time out.\n", dt);
468                 GOTO(out_md, err);
469         } else if (err) {
470                 CERROR("%s: Cannot connect to %s: rc = %d\n",
471                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
472                 GOTO(out_md, err);
473         }
474
475         sbi->ll_dt_exp->exp_connect_data = *data;
476
477         /* Don't change value if it was specified in the config log */
478         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
479                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
480                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
481                               (data->ocd_brw_size >> PAGE_SHIFT));
482
483         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
484                            LUSTRE_SEQ_METADATA);
485         if (err) {
486                 CERROR("%s: Can't init data layer FID infrastructure, "
487                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
488                 GOTO(out_dt, err);
489         }
490
491         mutex_lock(&sbi->ll_lco.lco_lock);
492         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
493         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
494         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
495         mutex_unlock(&sbi->ll_lco.lco_lock);
496
497         fid_zero(&sbi->ll_root_fid);
498         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
499                            &sbi->ll_root_fid);
500         if (err) {
501                 CERROR("cannot mds_connect: rc = %d\n", err);
502                 GOTO(out_lock_cn_cb, err);
503         }
504         if (!fid_is_sane(&sbi->ll_root_fid)) {
505                 CERROR("%s: Invalid root fid "DFID" during mount\n",
506                        sbi->ll_md_exp->exp_obd->obd_name,
507                        PFID(&sbi->ll_root_fid));
508                 GOTO(out_lock_cn_cb, err = -EINVAL);
509         }
510         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
511
512         sb->s_op = &lustre_super_operations;
513 #ifdef HAVE_XATTR_HANDLER_FLAGS
514         sb->s_xattr = ll_xattr_handlers;
515 #endif
516 #if THREAD_SIZE >= 8192 /*b=17630*/
517         sb->s_export_op = &lustre_export_operations;
518 #endif
519
520         /* make root inode
521          * XXX: move this to after cbd setup? */
522         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
523         if (sbi->ll_flags & LL_SBI_ACL)
524                 valid |= OBD_MD_FLACL;
525
526         OBD_ALLOC_PTR(op_data);
527         if (op_data == NULL)
528                 GOTO(out_lock_cn_cb, err = -ENOMEM);
529
530         op_data->op_fid1 = sbi->ll_root_fid;
531         op_data->op_mode = 0;
532         op_data->op_valid = valid;
533
534         err = md_getattr(sbi->ll_md_exp, op_data, &request);
535
536         OBD_FREE_PTR(op_data);
537         if (err) {
538                 CERROR("%s: md_getattr failed for root: rc = %d\n",
539                        sbi->ll_md_exp->exp_obd->obd_name, err);
540                 GOTO(out_lock_cn_cb, err);
541         }
542
543         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
544                                sbi->ll_md_exp, &lmd);
545         if (err) {
546                 CERROR("failed to understand root inode md: rc = %d\n", err);
547                 ptlrpc_req_finished(request);
548                 GOTO(out_lock_cn_cb, err);
549         }
550
551         LASSERT(fid_is_sane(&sbi->ll_root_fid));
552         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
553                                             sbi->ll_flags & LL_SBI_32BIT_API),
554                        &lmd);
555         md_free_lustre_md(sbi->ll_md_exp, &lmd);
556         ptlrpc_req_finished(request);
557
558         if (IS_ERR(root)) {
559 #ifdef CONFIG_FS_POSIX_ACL
560                 if (lmd.posix_acl) {
561                         posix_acl_release(lmd.posix_acl);
562                         lmd.posix_acl = NULL;
563                 }
564 #endif
565                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
566                 root = NULL;
567                 CERROR("lustre_lite: bad iget4 for root\n");
568                 GOTO(out_root, err);
569         }
570
571         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
572         if (sbi->ll_checksum_set) {
573                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
574                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
575                                          sizeof(checksum), &checksum, NULL);
576                 if (err) {
577                         CERROR("%s: Set checksum failed: rc = %d\n",
578                                sbi->ll_dt_exp->exp_obd->obd_name, err);
579                         GOTO(out_root, err);
580                 }
581         }
582         cl_sb_init(sb);
583
584         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
585                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
586                                  sbi->ll_cache, NULL);
587         if (err) {
588                 CERROR("%s: Set cache_set failed: rc = %d\n",
589                        sbi->ll_dt_exp->exp_obd->obd_name, err);
590                 GOTO(out_root, err);
591         }
592
593         sb->s_root = d_make_root(root);
594         if (sb->s_root == NULL) {
595                 CERROR("%s: can't make root dentry\n",
596                         ll_get_fsname(sb, NULL, 0));
597                 GOTO(out_root, err = -ENOMEM);
598         }
599 #ifdef HAVE_DCACHE_LOCK
600         sb->s_root->d_op = &ll_d_ops;
601 #endif
602
603         sbi->ll_sdev_orig = sb->s_dev;
604
605         /* We set sb->s_dev equal on all lustre clients in order to support
606          * NFS export clustering.  NFSD requires that the FSID be the same
607          * on all clients. */
608         /* s_dev is also used in lt_compare() to compare two fs, but that is
609          * only a node-local comparison. */
610         uuid = obd_get_uuid(sbi->ll_md_exp);
611         if (uuid != NULL)
612                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
613
614         if (data != NULL)
615                 OBD_FREE_PTR(data);
616         if (osfs != NULL)
617                 OBD_FREE_PTR(osfs);
618
619         if (sbi->ll_dt_obd) {
620                 err = sysfs_create_link(&sbi->ll_kset.kobj,
621                                         &sbi->ll_dt_obd->obd_kset.kobj,
622                                         sbi->ll_dt_obd->obd_type->typ_name);
623                 if (err < 0) {
624                         CERROR("%s: could not register %s in llite: rc = %d\n",
625                                dt, ll_get_fsname(sb, NULL, 0), err);
626                         err = 0;
627                 }
628         }
629
630         if (sbi->ll_md_obd) {
631                 err = sysfs_create_link(&sbi->ll_kset.kobj,
632                                         &sbi->ll_md_obd->obd_kset.kobj,
633                                         sbi->ll_md_obd->obd_type->typ_name);
634                 if (err < 0) {
635                         CERROR("%s: could not register %s in llite: rc = %d\n",
636                                md, ll_get_fsname(sb, NULL, 0), err);
637                         err = 0;
638                 }
639         }
640
641         RETURN(err);
642 out_root:
643         if (root)
644                 iput(root);
645 out_lock_cn_cb:
646         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
647 out_dt:
648         obd_disconnect(sbi->ll_dt_exp);
649         sbi->ll_dt_exp = NULL;
650         sbi->ll_dt_obd = NULL;
651 out_md_fid:
652         obd_fid_fini(sbi->ll_md_exp->exp_obd);
653 out_md:
654         obd_disconnect(sbi->ll_md_exp);
655         sbi->ll_md_exp = NULL;
656         sbi->ll_md_obd = NULL;
657 out:
658         if (data != NULL)
659                 OBD_FREE_PTR(data);
660         if (osfs != NULL)
661                 OBD_FREE_PTR(osfs);
662         return err;
663 }
664
665 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
666 {
667         int size, rc;
668
669         size = sizeof(*lmmsize);
670         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
671                           KEY_MAX_EASIZE, &size, lmmsize);
672         if (rc != 0) {
673                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
674                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
675                 RETURN(rc);
676         }
677
678         size = sizeof(int);
679         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
680                           KEY_MAX_EASIZE, &size, lmmsize);
681         if (rc)
682                 CERROR("Get max mdsize error rc %d\n", rc);
683
684         RETURN(rc);
685 }
686
687 /**
688  * Get the value of the default_easize parameter.
689  *
690  * \see client_obd::cl_default_mds_easize
691  *
692  * \param[in] sbi       superblock info for this filesystem
693  * \param[out] lmmsize  pointer to storage location for value
694  *
695  * \retval 0            on success
696  * \retval negative     negated errno on failure
697  */
698 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
699 {
700         int size, rc;
701
702         size = sizeof(int);
703         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
704                          KEY_DEFAULT_EASIZE, &size, lmmsize);
705         if (rc)
706                 CERROR("Get default mdsize error rc %d\n", rc);
707
708         RETURN(rc);
709 }
710
711 /**
712  * Set the default_easize parameter to the given value.
713  *
714  * \see client_obd::cl_default_mds_easize
715  *
716  * \param[in] sbi       superblock info for this filesystem
717  * \param[in] lmmsize   the size to set
718  *
719  * \retval 0            on success
720  * \retval negative     negated errno on failure
721  */
722 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
723 {
724         int rc;
725
726         if (lmmsize < sizeof(struct lov_mds_md) ||
727             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
728                 return -EINVAL;
729
730         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
731                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
732                                 sizeof(int), &lmmsize, NULL);
733
734         RETURN(rc);
735 }
736
737 static void client_common_put_super(struct super_block *sb)
738 {
739         struct ll_sb_info *sbi = ll_s2sbi(sb);
740         ENTRY;
741
742         cl_sb_fini(sb);
743
744         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
745         obd_disconnect(sbi->ll_dt_exp);
746         sbi->ll_dt_exp = NULL;
747
748         ll_debugfs_unregister_super(sb);
749
750         obd_fid_fini(sbi->ll_md_exp->exp_obd);
751         obd_disconnect(sbi->ll_md_exp);
752         sbi->ll_md_exp = NULL;
753
754         EXIT;
755 }
756
757 void ll_kill_super(struct super_block *sb)
758 {
759         struct ll_sb_info *sbi;
760         ENTRY;
761
762         /* not init sb ?*/
763         if (!(sb->s_flags & MS_ACTIVE))
764                 return;
765
766         sbi = ll_s2sbi(sb);
767         /* we need restore s_dev from changed for clustred NFS before put_super
768          * because new kernels have cached s_dev and change sb->s_dev in
769          * put_super not affected real removing devices */
770         if (sbi) {
771                 sb->s_dev = sbi->ll_sdev_orig;
772                 sbi->ll_umounting = 1;
773
774                 /* wait running statahead threads to quit */
775                 while (atomic_read(&sbi->ll_sa_running) > 0) {
776                         set_current_state(TASK_UNINTERRUPTIBLE);
777                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
778                 }
779         }
780
781         EXIT;
782 }
783
784 static inline int ll_set_opt(const char *opt, char *data, int fl)
785 {
786         if (strncmp(opt, data, strlen(opt)) != 0)
787                 return 0;
788         else
789                 return fl;
790 }
791
792 /* non-client-specific mount options are parsed in lmd_parse */
793 static int ll_options(char *options, struct ll_sb_info *sbi)
794 {
795         int tmp;
796         char *s1 = options, *s2;
797         int *flags = &sbi->ll_flags;
798         ENTRY;
799
800         if (!options)
801                 RETURN(0);
802
803         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
804
805         while (*s1) {
806                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
807                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
808                 if (tmp) {
809                         *flags |= tmp;
810                         goto next;
811                 }
812                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
813                 if (tmp) {
814                         *flags |= tmp;
815                         goto next;
816                 }
817                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
818                 if (tmp) {
819                         *flags |= tmp;
820                         goto next;
821                 }
822                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
823                 if (tmp) {
824                         *flags &= ~tmp;
825                         goto next;
826                 }
827                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
828                 if (tmp) {
829                         *flags |= tmp;
830                         goto next;
831                 }
832                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
833                 if (tmp) {
834                         *flags &= ~tmp;
835                         goto next;
836                 }
837                 tmp = ll_set_opt("context", s1, 1);
838                 if (tmp)
839                         goto next;
840                 tmp = ll_set_opt("fscontext", s1, 1);
841                 if (tmp)
842                         goto next;
843                 tmp = ll_set_opt("defcontext", s1, 1);
844                 if (tmp)
845                         goto next;
846                 tmp = ll_set_opt("rootcontext", s1, 1);
847                 if (tmp)
848                         goto next;
849                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
850                 if (tmp) {
851                         *flags |= tmp;
852                         goto next;
853                 }
854                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
855                 if (tmp) {
856                         *flags &= ~tmp;
857                         goto next;
858                 }
859
860                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
861                 if (tmp) {
862                         *flags |= tmp;
863                         sbi->ll_checksum_set = 1;
864                         goto next;
865                 }
866                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
867                 if (tmp) {
868                         *flags &= ~tmp;
869                         sbi->ll_checksum_set = 1;
870                         goto next;
871                 }
872                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
873                 if (tmp) {
874                         *flags |= tmp;
875                         goto next;
876                 }
877                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
878                 if (tmp) {
879                         *flags &= ~tmp;
880                         goto next;
881                 }
882                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
883                 if (tmp) {
884                         *flags |= tmp;
885                         goto next;
886                 }
887                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
888                 if (tmp) {
889                         *flags &= ~tmp;
890                         goto next;
891                 }
892                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
893                 if (tmp) {
894                         *flags |= tmp;
895                         goto next;
896                 }
897                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
898                 if (tmp) {
899                         *flags |= tmp;
900                         goto next;
901                 }
902                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
903                 if (tmp) {
904                         *flags &= ~tmp;
905                         goto next;
906                 }
907                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
908                 if (tmp) {
909                         *flags |= tmp;
910                         goto next;
911                 }
912                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
913                                    s1);
914                 RETURN(-EINVAL);
915
916 next:
917                 /* Find next opt */
918                 s2 = strchr(s1, ',');
919                 if (s2 == NULL)
920                         break;
921                 s1 = s2 + 1;
922         }
923         RETURN(0);
924 }
925
926 void ll_lli_init(struct ll_inode_info *lli)
927 {
928         lli->lli_inode_magic = LLI_INODE_MAGIC;
929         lli->lli_flags = 0;
930         spin_lock_init(&lli->lli_lock);
931         lli->lli_posix_acl = NULL;
932         /* Do not set lli_fid, it has been initialized already. */
933         fid_zero(&lli->lli_pfid);
934         lli->lli_mds_read_och = NULL;
935         lli->lli_mds_write_och = NULL;
936         lli->lli_mds_exec_och = NULL;
937         lli->lli_open_fd_read_count = 0;
938         lli->lli_open_fd_write_count = 0;
939         lli->lli_open_fd_exec_count = 0;
940         mutex_init(&lli->lli_och_mutex);
941         spin_lock_init(&lli->lli_agl_lock);
942         spin_lock_init(&lli->lli_layout_lock);
943         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
944         lli->lli_clob = NULL;
945
946         init_rwsem(&lli->lli_xattrs_list_rwsem);
947         mutex_init(&lli->lli_xattrs_enq_lock);
948
949         LASSERT(lli->lli_vfs_inode.i_mode != 0);
950         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
951                 mutex_init(&lli->lli_readdir_mutex);
952                 lli->lli_opendir_key = NULL;
953                 lli->lli_sai = NULL;
954                 spin_lock_init(&lli->lli_sa_lock);
955                 lli->lli_opendir_pid = 0;
956                 lli->lli_sa_enabled = 0;
957                 lli->lli_def_stripe_offset = -1;
958                 init_rwsem(&lli->lli_lsm_sem);
959         } else {
960                 mutex_init(&lli->lli_size_mutex);
961                 lli->lli_symlink_name = NULL;
962                 init_rwsem(&lli->lli_trunc_sem);
963                 range_lock_tree_init(&lli->lli_write_tree);
964                 init_rwsem(&lli->lli_glimpse_sem);
965                 lli->lli_glimpse_time = ktime_set(0, 0);
966                 INIT_LIST_HEAD(&lli->lli_agl_list);
967                 lli->lli_agl_index = 0;
968                 lli->lli_async_rc = 0;
969         }
970         mutex_init(&lli->lli_layout_mutex);
971         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
972 }
973
974 #define MAX_STRING_SIZE 128
975
976 #ifndef HAVE_SUPER_SETUP_BDI_NAME
977
978 #define LSI_BDI_INITIALIZED     0x00400000
979
980 #ifndef HAVE_BDI_CAP_MAP_COPY
981 # define BDI_CAP_MAP_COPY       0
982 #endif
983
984 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
985 {
986         struct  lustre_sb_info *lsi = s2lsi(sb);
987         char buf[MAX_STRING_SIZE];
988         va_list args;
989         int err;
990
991         err = bdi_init(&lsi->lsi_bdi);
992         if (err)
993                 return err;
994
995         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
996         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
997         lsi->lsi_bdi.name = "lustre";
998         va_start(args, fmt);
999         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1000         va_end(args);
1001         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1002         va_end(args);
1003         if (!err)
1004                 sb->s_bdi = &lsi->lsi_bdi;
1005
1006         return err;
1007 }
1008 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1009
1010 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1011 {
1012         struct  lustre_profile *lprof = NULL;
1013         struct  lustre_sb_info *lsi = s2lsi(sb);
1014         struct  ll_sb_info *sbi = NULL;
1015         char    *dt = NULL, *md = NULL;
1016         char    *profilenm = get_profile_name(sb);
1017         struct config_llog_instance *cfg;
1018         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1019         const int instlen = 16 + 2;
1020         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1021         char name[MAX_STRING_SIZE];
1022         int md_len = 0;
1023         int dt_len = 0;
1024         char *ptr;
1025         int len;
1026         int err;
1027
1028         ENTRY;
1029         /* for ASLR, to map between cfg_instance and hashed ptr */
1030         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1031                profilenm, cfg_instance, sb);
1032
1033         try_module_get(THIS_MODULE);
1034
1035         OBD_ALLOC_PTR(cfg);
1036         if (cfg == NULL)
1037                 GOTO(out_free_cfg, err = -ENOMEM);
1038
1039         /* client additional sb info */
1040         lsi->lsi_llsbi = sbi = ll_init_sbi();
1041         if (!sbi)
1042                 GOTO(out_free_cfg, err = -ENOMEM);
1043
1044         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1045         if (err)
1046                 GOTO(out_free_cfg, err);
1047
1048         err = super_setup_bdi_name(sb, "lustre-%016lx", cfg_instance);
1049         if (err)
1050                 GOTO(out_free_cfg, err);
1051
1052 #ifndef HAVE_DCACHE_LOCK
1053         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1054         sb->s_d_op = &ll_d_ops;
1055 #endif
1056         /* Get fsname */
1057         len = strlen(profilenm);
1058         ptr = strrchr(profilenm, '-');
1059         if (ptr && (strcmp(ptr, "-client") == 0))
1060                 len -= 7;
1061
1062         /* Mount info */
1063         snprintf(name, MAX_STRING_SIZE, "%.*s-%016lx", len,
1064                  profilenm, cfg_instance);
1065
1066         /* Call ll_debugfs_register_super() before lustre_process_log()
1067          * so that "llite.*.*" params can be processed correctly.
1068          */
1069         err = ll_debugfs_register_super(sb, name);
1070         if (err < 0) {
1071                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1072                        ll_get_fsname(sb, NULL, 0), err);
1073                 err = 0;
1074         }
1075
1076         /* The cfg_instance is a value unique to this super, in case some
1077          * joker tries to mount the same fs at two mount points.
1078          */
1079         cfg->cfg_instance = cfg_instance;
1080         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1081         cfg->cfg_callback = class_config_llog_handler;
1082         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1083         /* set up client obds */
1084         err = lustre_process_log(sb, profilenm, cfg);
1085         if (err < 0)
1086                 GOTO(out_debugfs, err);
1087
1088         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1089         lprof = class_get_profile(profilenm);
1090         if (lprof == NULL) {
1091                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1092                                    " read from the MGS.  Does that filesystem "
1093                                    "exist?\n", profilenm);
1094                 GOTO(out_debugfs, err = -EINVAL);
1095         }
1096         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1097                lprof->lp_md, lprof->lp_dt);
1098
1099         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1100         OBD_ALLOC(dt, dt_len);
1101         if (!dt)
1102                 GOTO(out_profile, err = -ENOMEM);
1103         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1104
1105         md_len = strlen(lprof->lp_md) + instlen + 2;
1106         OBD_ALLOC(md, md_len);
1107         if (!md)
1108                 GOTO(out_free_dt, err = -ENOMEM);
1109         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1110
1111         /* connections, registrations, sb setup */
1112         err = client_common_fill_super(sb, md, dt, mnt);
1113         if (err < 0)
1114                 GOTO(out_free_md, err);
1115
1116         sbi->ll_client_common_fill_super_succeeded = 1;
1117
1118 out_free_md:
1119         if (md)
1120                 OBD_FREE(md, md_len);
1121 out_free_dt:
1122         if (dt)
1123                 OBD_FREE(dt, dt_len);
1124 out_profile:
1125         if (lprof)
1126                 class_put_profile(lprof);
1127 out_debugfs:
1128         if (err < 0)
1129                 ll_debugfs_unregister_super(sb);
1130 out_free_cfg:
1131         if (cfg)
1132                 OBD_FREE_PTR(cfg);
1133
1134         if (err)
1135                 ll_put_super(sb);
1136         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1137                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1138         RETURN(err);
1139 } /* ll_fill_super */
1140
1141 void ll_put_super(struct super_block *sb)
1142 {
1143         struct config_llog_instance cfg, params_cfg;
1144         struct obd_device *obd;
1145         struct lustre_sb_info *lsi = s2lsi(sb);
1146         struct ll_sb_info *sbi = ll_s2sbi(sb);
1147         char *profilenm = get_profile_name(sb);
1148         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1149         long ccc_count;
1150         int next, force = 1, rc = 0;
1151         ENTRY;
1152
1153         if (!sbi)
1154                 GOTO(out_no_sbi, 0);
1155
1156         /* Should replace instance_id with something better for ASLR */
1157         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1158                profilenm, cfg_instance, sb);
1159
1160         cfg.cfg_instance = cfg_instance;
1161         lustre_end_log(sb, profilenm, &cfg);
1162
1163         params_cfg.cfg_instance = cfg_instance;
1164         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1165
1166         if (sbi->ll_md_exp) {
1167                 obd = class_exp2obd(sbi->ll_md_exp);
1168                 if (obd)
1169                         force = obd->obd_force;
1170         }
1171
1172         /* Wait for unstable pages to be committed to stable storage */
1173         if (force == 0) {
1174                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1175                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1176                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1177                         &lwi);
1178         }
1179
1180         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1181         if (force == 0 && rc != -EINTR)
1182                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1183
1184         /* We need to set force before the lov_disconnect in
1185            lustre_common_put_super, since l_d cleans up osc's as well. */
1186         if (force) {
1187                 next = 0;
1188                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1189                                                      &next)) != NULL) {
1190                         obd->obd_force = force;
1191                 }
1192         }
1193
1194         if (sbi->ll_client_common_fill_super_succeeded) {
1195                 /* Only if client_common_fill_super succeeded */
1196                 client_common_put_super(sb);
1197         }
1198
1199         next = 0;
1200         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1201                 class_manual_cleanup(obd);
1202         }
1203
1204         if (sbi->ll_flags & LL_SBI_VERBOSE)
1205                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1206
1207         if (profilenm)
1208                 class_del_profile(profilenm);
1209
1210 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1211         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1212                 bdi_destroy(&lsi->lsi_bdi);
1213                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1214         }
1215 #endif
1216
1217         ll_free_sbi(sb);
1218         lsi->lsi_llsbi = NULL;
1219 out_no_sbi:
1220         lustre_common_put_super(sb);
1221
1222         cl_env_cache_purge(~0);
1223
1224         module_put(THIS_MODULE);
1225
1226         EXIT;
1227 } /* client_put_super */
1228
1229 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1230 {
1231         struct inode *inode = NULL;
1232
1233         /* NOTE: we depend on atomic igrab() -bzzz */
1234         lock_res_and_lock(lock);
1235         if (lock->l_resource->lr_lvb_inode) {
1236                 struct ll_inode_info * lli;
1237                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1238                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1239                         inode = igrab(lock->l_resource->lr_lvb_inode);
1240                 } else {
1241                         inode = lock->l_resource->lr_lvb_inode;
1242                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1243                                          D_WARNING, lock, "lr_lvb_inode %p is "
1244                                          "bogus: magic %08x",
1245                                          lock->l_resource->lr_lvb_inode,
1246                                          lli->lli_inode_magic);
1247                         inode = NULL;
1248                 }
1249         }
1250         unlock_res_and_lock(lock);
1251         return inode;
1252 }
1253
1254 void ll_dir_clear_lsm_md(struct inode *inode)
1255 {
1256         struct ll_inode_info *lli = ll_i2info(inode);
1257
1258         LASSERT(S_ISDIR(inode->i_mode));
1259
1260         if (lli->lli_lsm_md != NULL) {
1261                 lmv_free_memmd(lli->lli_lsm_md);
1262                 lli->lli_lsm_md = NULL;
1263         }
1264 }
1265
1266 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1267                                       const struct lu_fid *fid,
1268                                       struct lustre_md *md)
1269 {
1270         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1271         struct mdt_body         *body = md->body;
1272         struct inode            *inode;
1273         ino_t                   ino;
1274         ENTRY;
1275
1276         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1277         inode = iget_locked(sb, ino);
1278         if (inode == NULL) {
1279                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1280                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1281                 RETURN(ERR_PTR(-ENOENT));
1282         }
1283
1284         if (inode->i_state & I_NEW) {
1285                 struct ll_inode_info *lli = ll_i2info(inode);
1286                 struct lmv_stripe_md *lsm = md->lmv;
1287
1288                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1289                                 (body->mbo_mode & S_IFMT);
1290                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1291                          PFID(fid));
1292
1293                 LTIME_S(inode->i_mtime) = 0;
1294                 LTIME_S(inode->i_atime) = 0;
1295                 LTIME_S(inode->i_ctime) = 0;
1296                 inode->i_rdev = 0;
1297
1298 #ifdef HAVE_BACKING_DEV_INFO
1299                 /* initializing backing dev info. */
1300                 inode->i_mapping->backing_dev_info =
1301                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1302 #endif
1303                 inode->i_op = &ll_dir_inode_operations;
1304                 inode->i_fop = &ll_dir_operations;
1305                 lli->lli_fid = *fid;
1306                 ll_lli_init(lli);
1307
1308                 LASSERT(lsm != NULL);
1309                 /* master object FID */
1310                 lli->lli_pfid = body->mbo_fid1;
1311                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1312                        lli, PFID(fid), PFID(&lli->lli_pfid));
1313                 unlock_new_inode(inode);
1314         }
1315
1316         RETURN(inode);
1317 }
1318
1319 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1320 {
1321         struct lu_fid *fid;
1322         struct lmv_stripe_md *lsm = md->lmv;
1323         struct ll_inode_info *lli = ll_i2info(inode);
1324         int i;
1325
1326         LASSERT(lsm != NULL);
1327
1328         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1329                 ll_get_fsname(inode->i_sb, NULL, 0),
1330                 PFID(&lli->lli_fid));
1331         lsm_md_dump(D_INODE, lsm);
1332
1333         /* XXX sigh, this lsm_root initialization should be in
1334          * LMV layer, but it needs ll_iget right now, so we
1335          * put this here right now. */
1336         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1337                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1338                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1339                 /* Unfortunately ll_iget will call ll_update_inode,
1340                  * where the initialization of slave inode is slightly
1341                  * different, so it reset lsm_md to NULL to avoid
1342                  * initializing lsm for slave inode. */
1343                 lsm->lsm_md_oinfo[i].lmo_root =
1344                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1345                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1346                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1347
1348                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1349                         while (i-- > 0) {
1350                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1351                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1352                         }
1353                         return rc;
1354                 }
1355         }
1356
1357         lli->lli_lsm_md = lsm;
1358
1359         return 0;
1360 }
1361
1362 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1363 {
1364         struct ll_inode_info *lli = ll_i2info(inode);
1365         struct lmv_stripe_md *lsm = md->lmv;
1366         int rc = 0;
1367
1368         ENTRY;
1369
1370         LASSERT(S_ISDIR(inode->i_mode));
1371         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1372                PFID(ll_inode2fid(inode)));
1373
1374         /*
1375          * no striped information from request, lustre_md from req does not
1376          * include stripeEA, see ll_md_setattr()
1377          */
1378         if (!lsm)
1379                 RETURN(0);
1380
1381         /*
1382          * normally dir layout doesn't change, only take read lock to check
1383          * that to avoid blocking other MD operations.
1384          */
1385         if (lli->lli_lsm_md)
1386                 down_read(&lli->lli_lsm_sem);
1387         else
1388                 down_write(&lli->lli_lsm_sem);
1389
1390         /*
1391          * if dir layout mismatch, check whether version is increased, which
1392          * means layout is changed, this happens in dir migration and lfsck.
1393          */
1394         if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1395                 if (lsm->lsm_md_layout_version <=
1396                     lli->lli_lsm_md->lsm_md_layout_version) {
1397                         CERROR("%s: "DFID" dir layout mismatch:\n",
1398                                 ll_get_fsname(inode->i_sb, NULL, 0),
1399                                 PFID(&lli->lli_fid));
1400                         lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1401                         lsm_md_dump(D_ERROR, lsm);
1402                         GOTO(unlock, rc = -EINVAL);
1403                 }
1404
1405                 /* layout changed, switch to write lock */
1406                 up_read(&lli->lli_lsm_sem);
1407                 down_write(&lli->lli_lsm_sem);
1408                 ll_dir_clear_lsm_md(inode);
1409         }
1410
1411         /* set directory layout */
1412         if (!lli->lli_lsm_md) {
1413                 struct cl_attr  *attr;
1414
1415                 rc = ll_init_lsm_md(inode, md);
1416                 up_write(&lli->lli_lsm_sem);
1417                 if (rc != 0)
1418                         RETURN(rc);
1419
1420                 /* set md->lmv to NULL, so the following free lustre_md
1421                  * will not free this lsm */
1422                 md->lmv = NULL;
1423
1424                 /*
1425                  * md_merge_attr() may take long, since lsm is already set,
1426                  * switch to read lock.
1427                  */
1428                 down_read(&lli->lli_lsm_sem);
1429
1430                 OBD_ALLOC_PTR(attr);
1431                 if (attr == NULL)
1432                         GOTO(unlock, rc = -ENOMEM);
1433
1434                 /* validate the lsm */
1435                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1436                                    ll_md_blocking_ast);
1437                 if (rc != 0) {
1438                         OBD_FREE_PTR(attr);
1439                         GOTO(unlock, rc);
1440                 }
1441
1442                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1443                         md->body->mbo_nlink = attr->cat_nlink;
1444                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1445                         md->body->mbo_size = attr->cat_size;
1446                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1447                         md->body->mbo_atime = attr->cat_atime;
1448                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1449                         md->body->mbo_ctime = attr->cat_ctime;
1450                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1451                         md->body->mbo_mtime = attr->cat_mtime;
1452
1453                 OBD_FREE_PTR(attr);
1454         }
1455 unlock:
1456         up_read(&lli->lli_lsm_sem);
1457
1458         RETURN(rc);
1459 }
1460
1461 void ll_clear_inode(struct inode *inode)
1462 {
1463         struct ll_inode_info *lli = ll_i2info(inode);
1464         struct ll_sb_info *sbi = ll_i2sbi(inode);
1465         ENTRY;
1466
1467         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1468                PFID(ll_inode2fid(inode)), inode);
1469
1470         if (S_ISDIR(inode->i_mode)) {
1471                 /* these should have been cleared in ll_file_release */
1472                 LASSERT(lli->lli_opendir_key == NULL);
1473                 LASSERT(lli->lli_sai == NULL);
1474                 LASSERT(lli->lli_opendir_pid == 0);
1475         }
1476
1477         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1478
1479         LASSERT(!lli->lli_open_fd_write_count);
1480         LASSERT(!lli->lli_open_fd_read_count);
1481         LASSERT(!lli->lli_open_fd_exec_count);
1482
1483         if (lli->lli_mds_write_och)
1484                 ll_md_real_close(inode, FMODE_WRITE);
1485         if (lli->lli_mds_exec_och)
1486                 ll_md_real_close(inode, FMODE_EXEC);
1487         if (lli->lli_mds_read_och)
1488                 ll_md_real_close(inode, FMODE_READ);
1489
1490         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1491                 OBD_FREE(lli->lli_symlink_name,
1492                          strlen(lli->lli_symlink_name) + 1);
1493                 lli->lli_symlink_name = NULL;
1494         }
1495
1496         ll_xattr_cache_destroy(inode);
1497
1498 #ifdef CONFIG_FS_POSIX_ACL
1499         forget_all_cached_acls(inode);
1500         if (lli->lli_posix_acl) {
1501                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1502                 posix_acl_release(lli->lli_posix_acl);
1503                 lli->lli_posix_acl = NULL;
1504         }
1505 #endif
1506         lli->lli_inode_magic = LLI_INODE_DEAD;
1507
1508         if (S_ISDIR(inode->i_mode))
1509                 ll_dir_clear_lsm_md(inode);
1510         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1511                 LASSERT(list_empty(&lli->lli_agl_list));
1512
1513         /*
1514          * XXX This has to be done before lsm is freed below, because
1515          * cl_object still uses inode lsm.
1516          */
1517         cl_inode_fini(inode);
1518
1519         EXIT;
1520 }
1521
1522 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1523 {
1524         struct lustre_md md;
1525         struct inode *inode = dentry->d_inode;
1526         struct ll_sb_info *sbi = ll_i2sbi(inode);
1527         struct ptlrpc_request *request = NULL;
1528         int rc, ia_valid;
1529         ENTRY;
1530
1531         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1532                                      LUSTRE_OPC_ANY, NULL);
1533         if (IS_ERR(op_data))
1534                 RETURN(PTR_ERR(op_data));
1535
1536         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1537         if (rc) {
1538                 ptlrpc_req_finished(request);
1539                 if (rc == -ENOENT) {
1540                         clear_nlink(inode);
1541                         /* Unlinked special device node? Or just a race?
1542                          * Pretend we done everything. */
1543                         if (!S_ISREG(inode->i_mode) &&
1544                             !S_ISDIR(inode->i_mode)) {
1545                                 ia_valid = op_data->op_attr.ia_valid;
1546                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1547                                 rc = simple_setattr(dentry, &op_data->op_attr);
1548                                 op_data->op_attr.ia_valid = ia_valid;
1549                         }
1550                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1551                         CERROR("md_setattr fails: rc = %d\n", rc);
1552                 }
1553                 RETURN(rc);
1554         }
1555
1556         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1557                               sbi->ll_md_exp, &md);
1558         if (rc) {
1559                 ptlrpc_req_finished(request);
1560                 RETURN(rc);
1561         }
1562
1563         ia_valid = op_data->op_attr.ia_valid;
1564         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1565          * cache is not cleared yet. */
1566         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1567         if (S_ISREG(inode->i_mode))
1568                 inode_lock(inode);
1569         rc = simple_setattr(dentry, &op_data->op_attr);
1570         if (S_ISREG(inode->i_mode))
1571                 inode_unlock(inode);
1572         op_data->op_attr.ia_valid = ia_valid;
1573
1574         rc = ll_update_inode(inode, &md);
1575         ptlrpc_req_finished(request);
1576
1577         RETURN(rc);
1578 }
1579
1580 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1581  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1582  * keep these values until such a time that objects are allocated for it.
1583  * We do the MDS operations first, as it is checking permissions for us.
1584  * We don't to the MDS RPC if there is nothing that we want to store there,
1585  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1586  * going to do an RPC anyways.
1587  *
1588  * If we are doing a truncate, we will send the mtime and ctime updates
1589  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1590  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1591  * at the same time.
1592  *
1593  * In case of HSMimport, we only set attr on MDS.
1594  */
1595 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1596                    enum op_xvalid xvalid, bool hsm_import)
1597 {
1598         struct inode *inode = dentry->d_inode;
1599         struct ll_inode_info *lli = ll_i2info(inode);
1600         struct md_op_data *op_data = NULL;
1601         int rc = 0;
1602         ENTRY;
1603
1604         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1605                "valid %x, hsm_import %d\n",
1606                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1607                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1608                hsm_import);
1609
1610         if (attr->ia_valid & ATTR_SIZE) {
1611                 /* Check new size against VFS/VM file size limit and rlimit */
1612                 rc = inode_newsize_ok(inode, attr->ia_size);
1613                 if (rc)
1614                         RETURN(rc);
1615
1616                 /* The maximum Lustre file size is variable, based on the
1617                  * OST maximum object size and number of stripes.  This
1618                  * needs another check in addition to the VFS check above. */
1619                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1620                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1621                                PFID(&lli->lli_fid), attr->ia_size,
1622                                ll_file_maxbytes(inode));
1623                         RETURN(-EFBIG);
1624                 }
1625
1626                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1627         }
1628
1629         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1630         if (attr->ia_valid & TIMES_SET_FLAGS) {
1631                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1632                     !cfs_capable(CFS_CAP_FOWNER))
1633                         RETURN(-EPERM);
1634         }
1635
1636         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1637         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1638              (attr->ia_valid & ATTR_CTIME)) {
1639                 attr->ia_ctime = current_time(inode);
1640                 xvalid |= OP_XVALID_CTIME_SET;
1641         }
1642         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1643             (attr->ia_valid & ATTR_ATIME)) {
1644                 attr->ia_atime = current_time(inode);
1645                 attr->ia_valid |= ATTR_ATIME_SET;
1646         }
1647         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1648             (attr->ia_valid & ATTR_MTIME)) {
1649                 attr->ia_mtime = current_time(inode);
1650                 attr->ia_valid |= ATTR_MTIME_SET;
1651         }
1652
1653         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1654                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1655                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1656                        (s64)ktime_get_real_seconds());
1657
1658         if (S_ISREG(inode->i_mode)) {
1659                 if (attr->ia_valid & ATTR_SIZE)
1660                         inode_dio_write_done(inode);
1661                 inode_unlock(inode);
1662         }
1663
1664         /* We always do an MDS RPC, even if we're only changing the size;
1665          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1666
1667         OBD_ALLOC_PTR(op_data);
1668         if (op_data == NULL)
1669                 GOTO(out, rc = -ENOMEM);
1670
1671         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1672                 /* If we are changing file size, file content is
1673                  * modified, flag it.
1674                  */
1675                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1676                 op_data->op_bias |= MDS_DATA_MODIFIED;
1677                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1678         }
1679
1680         if (attr->ia_valid & ATTR_FILE) {
1681                 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1682
1683                 if (fd->fd_lease_och)
1684                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1685         }
1686
1687         op_data->op_attr = *attr;
1688         op_data->op_xvalid = xvalid;
1689
1690         rc = ll_md_setattr(dentry, op_data);
1691         if (rc)
1692                 GOTO(out, rc);
1693
1694         if (!S_ISREG(inode->i_mode) || hsm_import)
1695                 GOTO(out, rc = 0);
1696
1697         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1698                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1699             xvalid & OP_XVALID_CTIME_SET) {
1700                 /* For truncate and utimes sending attributes to OSTs, setting
1701                  * mtime/atime to the past will be performed under PW [0:EOF]
1702                  * extent lock (new_size:EOF for truncate).  It may seem
1703                  * excessive to send mtime/atime updates to OSTs when not
1704                  * setting times to past, but it is necessary due to possible
1705                  * time de-synchronization between MDT inode and OST objects
1706                  */
1707                 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1708         }
1709
1710         /* If the file was restored, it needs to set dirty flag.
1711          *
1712          * We've already sent MDS_DATA_MODIFIED flag in
1713          * ll_md_setattr() for truncate. However, the MDT refuses to
1714          * set the HS_DIRTY flag on released files, so we have to set
1715          * it again if the file has been restored. Please check how
1716          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1717          *
1718          * Please notice that if the file is not released, the previous
1719          * MDS_DATA_MODIFIED has taken effect and usually
1720          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1721          * This way we can save an RPC for common open + trunc
1722          * operation. */
1723         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1724                 struct hsm_state_set hss = {
1725                         .hss_valid = HSS_SETMASK,
1726                         .hss_setmask = HS_DIRTY,
1727                 };
1728                 int rc2;
1729
1730                 rc2 = ll_hsm_state_set(inode, &hss);
1731                 /* truncate and write can happen at the same time, so that
1732                  * the file can be set modified even though the file is not
1733                  * restored from released state, and ll_hsm_state_set() is
1734                  * not applicable for the file, and rc2 < 0 is normal in this
1735                  * case. */
1736                 if (rc2 < 0)
1737                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1738                                PFID(ll_inode2fid(inode)), rc2);
1739         }
1740
1741         EXIT;
1742 out:
1743         if (op_data != NULL)
1744                 ll_finish_md_op_data(op_data);
1745
1746         if (S_ISREG(inode->i_mode)) {
1747                 inode_lock(inode);
1748                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1749                         inode_dio_wait(inode);
1750                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1751                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1752                  * inode flags, so there is a gap where S_NOSEC is not set.
1753                  * This can cause a writer to take the i_mutex unnecessarily,
1754                  * but this is safe to do and should be rare. */
1755                 inode_has_no_xattr(inode);
1756         }
1757
1758         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1759                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1760
1761         return rc;
1762 }
1763
1764 int ll_setattr(struct dentry *de, struct iattr *attr)
1765 {
1766         int mode = de->d_inode->i_mode;
1767         enum op_xvalid xvalid = 0;
1768
1769         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1770                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1771                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1772
1773         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1774                                (ATTR_SIZE|ATTR_MODE)) &&
1775             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1776              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1777               !(attr->ia_mode & S_ISGID))))
1778                 attr->ia_valid |= ATTR_FORCE;
1779
1780         if ((attr->ia_valid & ATTR_MODE) &&
1781             (mode & S_ISUID) &&
1782             !(attr->ia_mode & S_ISUID) &&
1783             !(attr->ia_valid & ATTR_KILL_SUID))
1784                 attr->ia_valid |= ATTR_KILL_SUID;
1785
1786         if ((attr->ia_valid & ATTR_MODE) &&
1787             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1788             !(attr->ia_mode & S_ISGID) &&
1789             !(attr->ia_valid & ATTR_KILL_SGID))
1790                 attr->ia_valid |= ATTR_KILL_SGID;
1791
1792         return ll_setattr_raw(de, attr, xvalid, false);
1793 }
1794
1795 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1796                        u32 flags)
1797 {
1798         struct obd_statfs obd_osfs;
1799         time64_t max_age;
1800         int rc;
1801
1802         ENTRY;
1803         max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1804
1805         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1806         if (rc) {
1807                 CERROR("md_statfs fails: rc = %d\n", rc);
1808                 RETURN(rc);
1809         }
1810
1811         osfs->os_type = LL_SUPER_MAGIC;
1812
1813         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1814                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1815
1816         if (osfs->os_state & OS_STATE_SUM)
1817                 GOTO(out, rc);
1818
1819         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1820                 flags |= OBD_STATFS_NODELAY;
1821
1822         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1823         if (rc) {
1824                 CERROR("obd_statfs fails: rc = %d\n", rc);
1825                 RETURN(rc);
1826         }
1827
1828         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1829                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1830                obd_osfs.os_files);
1831
1832         osfs->os_bsize = obd_osfs.os_bsize;
1833         osfs->os_blocks = obd_osfs.os_blocks;
1834         osfs->os_bfree = obd_osfs.os_bfree;
1835         osfs->os_bavail = obd_osfs.os_bavail;
1836
1837         /* If we don't have as many objects free on the OST as inodes
1838          * on the MDS, we reduce the total number of inodes to
1839          * compensate, so that the "inodes in use" number is correct.
1840          */
1841         if (obd_osfs.os_ffree < osfs->os_ffree) {
1842                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1843                         obd_osfs.os_ffree;
1844                 osfs->os_ffree = obd_osfs.os_ffree;
1845         }
1846
1847 out:
1848         RETURN(rc);
1849 }
1850 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1851 {
1852         struct super_block *sb = de->d_sb;
1853         struct obd_statfs osfs;
1854         __u64 fsid = huge_encode_dev(sb->s_dev);
1855         int rc;
1856
1857         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1858         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1859
1860         /* Some amount of caching on the client is allowed */
1861         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1862         if (rc)
1863                 return rc;
1864
1865         statfs_unpack(sfs, &osfs);
1866
1867         /* We need to downshift for all 32-bit kernels, because we can't
1868          * tell if the kernel is being called via sys_statfs64() or not.
1869          * Stop before overflowing f_bsize - in which case it is better
1870          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1871         if (sizeof(long) < 8) {
1872                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1873                         sfs->f_bsize <<= 1;
1874
1875                         osfs.os_blocks >>= 1;
1876                         osfs.os_bfree >>= 1;
1877                         osfs.os_bavail >>= 1;
1878                 }
1879         }
1880
1881         sfs->f_blocks = osfs.os_blocks;
1882         sfs->f_bfree = osfs.os_bfree;
1883         sfs->f_bavail = osfs.os_bavail;
1884         sfs->f_fsid.val[0] = (__u32)fsid;
1885         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1886         return 0;
1887 }
1888
1889 void ll_inode_size_lock(struct inode *inode)
1890 {
1891         struct ll_inode_info *lli;
1892
1893         LASSERT(!S_ISDIR(inode->i_mode));
1894
1895         lli = ll_i2info(inode);
1896         mutex_lock(&lli->lli_size_mutex);
1897 }
1898
1899 void ll_inode_size_unlock(struct inode *inode)
1900 {
1901         struct ll_inode_info *lli;
1902
1903         lli = ll_i2info(inode);
1904         mutex_unlock(&lli->lli_size_mutex);
1905 }
1906
1907 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1908 {
1909         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1910         if (ext_flags & LUSTRE_PROJINHERIT_FL)
1911                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1912         else
1913                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1914 }
1915
1916 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1917 {
1918         struct ll_inode_info *lli = ll_i2info(inode);
1919         struct mdt_body *body = md->body;
1920         struct ll_sb_info *sbi = ll_i2sbi(inode);
1921         int rc = 0;
1922
1923         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1924                 rc = cl_file_inode_init(inode, md);
1925                 if (rc)
1926                         return rc;
1927         }
1928
1929         if (S_ISDIR(inode->i_mode)) {
1930                 rc = ll_update_lsm_md(inode, md);
1931                 if (rc != 0)
1932                         return rc;
1933         }
1934
1935 #ifdef CONFIG_FS_POSIX_ACL
1936         if (body->mbo_valid & OBD_MD_FLACL) {
1937                 spin_lock(&lli->lli_lock);
1938                 if (lli->lli_posix_acl)
1939                         posix_acl_release(lli->lli_posix_acl);
1940                 lli->lli_posix_acl = md->posix_acl;
1941                 spin_unlock(&lli->lli_lock);
1942         }
1943 #endif
1944         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1945                                         sbi->ll_flags & LL_SBI_32BIT_API);
1946         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1947
1948         if (body->mbo_valid & OBD_MD_FLATIME) {
1949                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1950                         LTIME_S(inode->i_atime) = body->mbo_atime;
1951                 lli->lli_atime = body->mbo_atime;
1952         }
1953
1954         if (body->mbo_valid & OBD_MD_FLMTIME) {
1955                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1956                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1957                                "to %llu\n", inode->i_ino,
1958                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1959                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1960                 }
1961                 lli->lli_mtime = body->mbo_mtime;
1962         }
1963
1964         if (body->mbo_valid & OBD_MD_FLCTIME) {
1965                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1966                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1967                 lli->lli_ctime = body->mbo_ctime;
1968         }
1969
1970         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1971         if (body->mbo_valid & OBD_MD_FLFLAGS)
1972                 ll_update_inode_flags(inode, body->mbo_flags);
1973         if (body->mbo_valid & OBD_MD_FLMODE)
1974                 inode->i_mode = (inode->i_mode & S_IFMT) |
1975                                 (body->mbo_mode & ~S_IFMT);
1976
1977         if (body->mbo_valid & OBD_MD_FLTYPE)
1978                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1979                                 (body->mbo_mode & S_IFMT);
1980
1981         LASSERT(inode->i_mode != 0);
1982         if (S_ISREG(inode->i_mode))
1983                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1984                                        LL_MAX_BLKSIZE_BITS);
1985         else
1986                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1987
1988         if (body->mbo_valid & OBD_MD_FLUID)
1989                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1990         if (body->mbo_valid & OBD_MD_FLGID)
1991                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1992         if (body->mbo_valid & OBD_MD_FLPROJID)
1993                 lli->lli_projid = body->mbo_projid;
1994         if (body->mbo_valid & OBD_MD_FLNLINK)
1995                 set_nlink(inode, body->mbo_nlink);
1996         if (body->mbo_valid & OBD_MD_FLRDEV)
1997                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1998
1999         if (body->mbo_valid & OBD_MD_FLID) {
2000                 /* FID shouldn't be changed! */
2001                 if (fid_is_sane(&lli->lli_fid)) {
2002                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2003                                  "Trying to change FID "DFID
2004                                  " to the "DFID", inode "DFID"(%p)\n",
2005                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2006                                  PFID(ll_inode2fid(inode)), inode);
2007                 } else {
2008                         lli->lli_fid = body->mbo_fid1;
2009                 }
2010         }
2011
2012         LASSERT(fid_seq(&lli->lli_fid) != 0);
2013
2014         if (body->mbo_valid & OBD_MD_FLSIZE) {
2015                 i_size_write(inode, body->mbo_size);
2016
2017                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2018                        PFID(ll_inode2fid(inode)),
2019                        (unsigned long long)body->mbo_size);
2020
2021                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2022                         inode->i_blocks = body->mbo_blocks;
2023         }
2024
2025         if (body->mbo_valid & OBD_MD_TSTATE) {
2026                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2027                  * clear it when done to ensure to start again
2028                  * glimpsing updated attrs
2029                  */
2030                 if (body->mbo_t_state & MS_RESTORE)
2031                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2032                 else
2033                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2034         }
2035
2036         return 0;
2037 }
2038
2039 int ll_read_inode2(struct inode *inode, void *opaque)
2040 {
2041         struct lustre_md *md = opaque;
2042         struct ll_inode_info *lli = ll_i2info(inode);
2043         int     rc;
2044         ENTRY;
2045
2046         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2047                PFID(&lli->lli_fid), inode);
2048
2049         /* Core attributes from the MDS first.  This is a new inode, and
2050          * the VFS doesn't zero times in the core inode so we have to do
2051          * it ourselves.  They will be overwritten by either MDS or OST
2052          * attributes - we just need to make sure they aren't newer. */
2053         LTIME_S(inode->i_mtime) = 0;
2054         LTIME_S(inode->i_atime) = 0;
2055         LTIME_S(inode->i_ctime) = 0;
2056         inode->i_rdev = 0;
2057         rc = ll_update_inode(inode, md);
2058         if (rc != 0)
2059                 RETURN(rc);
2060
2061         /* OIDEBUG(inode); */
2062
2063 #ifdef HAVE_BACKING_DEV_INFO
2064         /* initializing backing dev info. */
2065         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2066 #endif
2067         if (S_ISREG(inode->i_mode)) {
2068                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2069                 inode->i_op = &ll_file_inode_operations;
2070                 inode->i_fop = sbi->ll_fop;
2071                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2072                 EXIT;
2073         } else if (S_ISDIR(inode->i_mode)) {
2074                 inode->i_op = &ll_dir_inode_operations;
2075                 inode->i_fop = &ll_dir_operations;
2076                 EXIT;
2077         } else if (S_ISLNK(inode->i_mode)) {
2078                 inode->i_op = &ll_fast_symlink_inode_operations;
2079                 EXIT;
2080         } else {
2081                 inode->i_op = &ll_special_inode_operations;
2082
2083                 init_special_inode(inode, inode->i_mode,
2084                                    inode->i_rdev);
2085
2086                 EXIT;
2087         }
2088
2089         return 0;
2090 }
2091
2092 void ll_delete_inode(struct inode *inode)
2093 {
2094         struct ll_inode_info *lli = ll_i2info(inode);
2095         struct address_space *mapping = &inode->i_data;
2096         unsigned long nrpages;
2097         ENTRY;
2098
2099         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2100                 /* It is last chance to write out dirty pages,
2101                  * otherwise we may lose data while umount */
2102                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2103
2104         truncate_inode_pages_final(mapping);
2105
2106         /* Workaround for LU-118: Note nrpages may not be totally updated when
2107          * truncate_inode_pages() returns, as there can be a page in the process
2108          * of deletion (inside __delete_from_page_cache()) in the specified
2109          * range. Thus mapping->nrpages can be non-zero when this function
2110          * returns even after truncation of the whole mapping.  Only do this if
2111          * npages isn't already zero.
2112          */
2113         nrpages = mapping->nrpages;
2114         if (nrpages) {
2115                 spin_lock_irq(&mapping->tree_lock);
2116                 nrpages = mapping->nrpages;
2117                 spin_unlock_irq(&mapping->tree_lock);
2118         } /* Workaround end */
2119
2120         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2121                  "see https://jira.whamcloud.com/browse/LU-118\n",
2122                  ll_get_fsname(inode->i_sb, NULL, 0),
2123                  PFID(ll_inode2fid(inode)), inode, nrpages);
2124
2125 #ifdef HAVE_SBOPS_EVICT_INODE
2126         ll_clear_inode(inode);
2127 #endif
2128         clear_inode(inode);
2129
2130         EXIT;
2131 }
2132
2133 int ll_iocontrol(struct inode *inode, struct file *file,
2134                  unsigned int cmd, unsigned long arg)
2135 {
2136         struct ll_sb_info *sbi = ll_i2sbi(inode);
2137         struct ptlrpc_request *req = NULL;
2138         int rc, flags = 0;
2139         ENTRY;
2140
2141         switch (cmd) {
2142         case FS_IOC_GETFLAGS: {
2143                 struct mdt_body *body;
2144                 struct md_op_data *op_data;
2145
2146                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2147                                              0, 0, LUSTRE_OPC_ANY,
2148                                              NULL);
2149                 if (IS_ERR(op_data))
2150                         RETURN(PTR_ERR(op_data));
2151
2152                 op_data->op_valid = OBD_MD_FLFLAGS;
2153                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2154                 ll_finish_md_op_data(op_data);
2155                 if (rc) {
2156                         CERROR("%s: failure inode "DFID": rc = %d\n",
2157                                sbi->ll_md_exp->exp_obd->obd_name,
2158                                PFID(ll_inode2fid(inode)), rc);
2159                         RETURN(-abs(rc));
2160                 }
2161
2162                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2163
2164                 flags = body->mbo_flags;
2165
2166                 ptlrpc_req_finished(req);
2167
2168                 RETURN(put_user(flags, (int __user *)arg));
2169         }
2170         case FS_IOC_SETFLAGS: {
2171                 struct iattr *attr;
2172                 struct md_op_data *op_data;
2173                 struct cl_object *obj;
2174                 struct fsxattr fa = { 0 };
2175
2176                 if (get_user(flags, (int __user *)arg))
2177                         RETURN(-EFAULT);
2178
2179                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2180                 if (flags & LUSTRE_PROJINHERIT_FL)
2181                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2182
2183                 rc = ll_ioctl_check_project(inode, &fa);
2184                 if (rc)
2185                         RETURN(rc);
2186
2187                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2188                                              LUSTRE_OPC_ANY, NULL);
2189                 if (IS_ERR(op_data))
2190                         RETURN(PTR_ERR(op_data));
2191
2192                 op_data->op_attr_flags = flags;
2193                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2194                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2195                 ll_finish_md_op_data(op_data);
2196                 ptlrpc_req_finished(req);
2197                 if (rc)
2198                         RETURN(rc);
2199
2200                 ll_update_inode_flags(inode, flags);
2201
2202                 obj = ll_i2info(inode)->lli_clob;
2203                 if (obj == NULL)
2204                         RETURN(0);
2205
2206                 OBD_ALLOC_PTR(attr);
2207                 if (attr == NULL)
2208                         RETURN(-ENOMEM);
2209
2210                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2211
2212                 OBD_FREE_PTR(attr);
2213                 RETURN(rc);
2214         }
2215         default:
2216                 RETURN(-ENOSYS);
2217         }
2218
2219         RETURN(0);
2220 }
2221
2222 int ll_flush_ctx(struct inode *inode)
2223 {
2224         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2225
2226         CDEBUG(D_SEC, "flush context for user %d\n",
2227                from_kuid(&init_user_ns, current_uid()));
2228
2229         obd_set_info_async(NULL, sbi->ll_md_exp,
2230                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2231                            0, NULL, NULL);
2232         obd_set_info_async(NULL, sbi->ll_dt_exp,
2233                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2234                            0, NULL, NULL);
2235         return 0;
2236 }
2237
2238 /* umount -f client means force down, don't save state */
2239 void ll_umount_begin(struct super_block *sb)
2240 {
2241         struct ll_sb_info *sbi = ll_s2sbi(sb);
2242         struct obd_device *obd;
2243         struct obd_ioctl_data *ioc_data;
2244         struct l_wait_info lwi;
2245         wait_queue_head_t waitq;
2246         ENTRY;
2247
2248         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2249                sb->s_count, atomic_read(&sb->s_active));
2250
2251         obd = class_exp2obd(sbi->ll_md_exp);
2252         if (obd == NULL) {
2253                 CERROR("Invalid MDC connection handle %#llx\n",
2254                        sbi->ll_md_exp->exp_handle.h_cookie);
2255                 EXIT;
2256                 return;
2257         }
2258         obd->obd_force = 1;
2259
2260         obd = class_exp2obd(sbi->ll_dt_exp);
2261         if (obd == NULL) {
2262                 CERROR("Invalid LOV connection handle %#llx\n",
2263                        sbi->ll_dt_exp->exp_handle.h_cookie);
2264                 EXIT;
2265                 return;
2266         }
2267         obd->obd_force = 1;
2268
2269         OBD_ALLOC_PTR(ioc_data);
2270         if (ioc_data) {
2271                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2272                               sizeof *ioc_data, ioc_data, NULL);
2273
2274                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2275                               sizeof *ioc_data, ioc_data, NULL);
2276
2277                 OBD_FREE_PTR(ioc_data);
2278         }
2279
2280         /* Really, we'd like to wait until there are no requests outstanding,
2281          * and then continue.  For now, we just periodically checking for vfs
2282          * to decrement mnt_cnt and hope to finish it within 10sec.
2283          */
2284         init_waitqueue_head(&waitq);
2285         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2286                                    cfs_time_seconds(1), NULL, NULL);
2287         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2288
2289         EXIT;
2290 }
2291
2292 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2293 {
2294         struct ll_sb_info *sbi = ll_s2sbi(sb);
2295         char *profilenm = get_profile_name(sb);
2296         int err;
2297         __u32 read_only;
2298
2299         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2300                 read_only = *flags & MS_RDONLY;
2301                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2302                                          sizeof(KEY_READ_ONLY),
2303                                          KEY_READ_ONLY, sizeof(read_only),
2304                                          &read_only, NULL);
2305                 if (err) {
2306                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2307                                       profilenm, read_only ?
2308                                       "read-only" : "read-write", err);
2309                         return err;
2310                 }
2311
2312                 if (read_only)
2313                         sb->s_flags |= MS_RDONLY;
2314                 else
2315                         sb->s_flags &= ~MS_RDONLY;
2316
2317                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2318                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2319                                       read_only ?  "read-only" : "read-write");
2320         }
2321         return 0;
2322 }
2323
2324 /**
2325  * Cleanup the open handle that is cached on MDT-side.
2326  *
2327  * For open case, the client side open handling thread may hit error
2328  * after the MDT grant the open. Under such case, the client should
2329  * send close RPC to the MDT as cleanup; otherwise, the open handle
2330  * on the MDT will be leaked there until the client umount or evicted.
2331  *
2332  * In further, if someone unlinked the file, because the open handle
2333  * holds the reference on such file/object, then it will block the
2334  * subsequent threads that want to locate such object via FID.
2335  *
2336  * \param[in] sb        super block for this file-system
2337  * \param[in] open_req  pointer to the original open request
2338  */
2339 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2340 {
2341         struct mdt_body                 *body;
2342         struct md_op_data               *op_data;
2343         struct ptlrpc_request           *close_req = NULL;
2344         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2345         ENTRY;
2346
2347         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2348         OBD_ALLOC_PTR(op_data);
2349         if (op_data == NULL) {
2350                 CWARN("%s: cannot allocate op_data to release open handle for "
2351                       DFID"\n",
2352                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2353
2354                 RETURN_EXIT;
2355         }
2356
2357         op_data->op_fid1 = body->mbo_fid1;
2358         op_data->op_open_handle = body->mbo_open_handle;
2359         op_data->op_mod_time = ktime_get_real_seconds();
2360         md_close(exp, op_data, NULL, &close_req);
2361         ptlrpc_req_finished(close_req);
2362         ll_finish_md_op_data(op_data);
2363
2364         EXIT;
2365 }
2366
2367 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2368                   struct super_block *sb, struct lookup_intent *it)
2369 {
2370         struct ll_sb_info *sbi = NULL;
2371         struct lustre_md md = { NULL };
2372         int rc;
2373         ENTRY;
2374
2375         LASSERT(*inode || sb);
2376         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2377         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2378                               sbi->ll_md_exp, &md);
2379         if (rc != 0)
2380                 GOTO(cleanup, rc);
2381
2382         if (*inode) {
2383                 rc = ll_update_inode(*inode, &md);
2384                 if (rc != 0)
2385                         GOTO(out, rc);
2386         } else {
2387                 LASSERT(sb != NULL);
2388
2389                 /*
2390                  * At this point server returns to client's same fid as client
2391                  * generated for creating. So using ->fid1 is okay here.
2392                  */
2393                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2394                         CERROR("%s: Fid is insane "DFID"\n",
2395                                 ll_get_fsname(sb, NULL, 0),
2396                                 PFID(&md.body->mbo_fid1));
2397                         GOTO(out, rc = -EINVAL);
2398                 }
2399
2400                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2401                                              sbi->ll_flags & LL_SBI_32BIT_API),
2402                                  &md);
2403                 if (IS_ERR(*inode)) {
2404 #ifdef CONFIG_FS_POSIX_ACL
2405                         if (md.posix_acl) {
2406                                 posix_acl_release(md.posix_acl);
2407                                 md.posix_acl = NULL;
2408                         }
2409 #endif
2410                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2411                         *inode = NULL;
2412                         CERROR("new_inode -fatal: rc %d\n", rc);
2413                         GOTO(out, rc);
2414                 }
2415         }
2416
2417         /* Handling piggyback layout lock.
2418          * Layout lock can be piggybacked by getattr and open request.
2419          * The lsm can be applied to inode only if it comes with a layout lock
2420          * otherwise correct layout may be overwritten, for example:
2421          * 1. proc1: mdt returns a lsm but not granting layout
2422          * 2. layout was changed by another client
2423          * 3. proc2: refresh layout and layout lock granted
2424          * 4. proc1: to apply a stale layout */
2425         if (it != NULL && it->it_lock_mode != 0) {
2426                 struct lustre_handle lockh;
2427                 struct ldlm_lock *lock;
2428
2429                 lockh.cookie = it->it_lock_handle;
2430                 lock = ldlm_handle2lock(&lockh);
2431                 LASSERT(lock != NULL);
2432                 if (ldlm_has_layout(lock)) {
2433                         struct cl_object_conf conf;
2434
2435                         memset(&conf, 0, sizeof(conf));
2436                         conf.coc_opc = OBJECT_CONF_SET;
2437                         conf.coc_inode = *inode;
2438                         conf.coc_lock = lock;
2439                         conf.u.coc_layout = md.layout;
2440                         (void)ll_layout_conf(*inode, &conf);
2441                 }
2442                 LDLM_LOCK_PUT(lock);
2443         }
2444
2445         GOTO(out, rc = 0);
2446
2447 out:
2448         md_free_lustre_md(sbi->ll_md_exp, &md);
2449
2450 cleanup:
2451         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2452                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2453
2454         return rc;
2455 }
2456
2457 int ll_obd_statfs(struct inode *inode, void __user *arg)
2458 {
2459         struct ll_sb_info *sbi = NULL;
2460         struct obd_export *exp;
2461         char *buf = NULL;
2462         struct obd_ioctl_data *data = NULL;
2463         __u32 type;
2464         int len = 0, rc;
2465
2466         if (!inode || !(sbi = ll_i2sbi(inode)))
2467                 GOTO(out_statfs, rc = -EINVAL);
2468
2469         rc = obd_ioctl_getdata(&buf, &len, arg);
2470         if (rc)
2471                 GOTO(out_statfs, rc);
2472
2473         data = (void*)buf;
2474         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2475             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2476                 GOTO(out_statfs, rc = -EINVAL);
2477
2478         if (data->ioc_inllen1 != sizeof(__u32) ||
2479             data->ioc_inllen2 != sizeof(__u32) ||
2480             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2481             data->ioc_plen2 != sizeof(struct obd_uuid))
2482                 GOTO(out_statfs, rc = -EINVAL);
2483
2484         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2485         if (type & LL_STATFS_LMV)
2486                 exp = sbi->ll_md_exp;
2487         else if (type & LL_STATFS_LOV)
2488                 exp = sbi->ll_dt_exp;
2489         else
2490                 GOTO(out_statfs, rc = -ENODEV);
2491
2492         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2493         if (rc)
2494                 GOTO(out_statfs, rc);
2495 out_statfs:
2496         OBD_FREE_LARGE(buf, len);
2497         return rc;
2498 }
2499
2500 /*
2501  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2502  * be called early to avoid deadlock.
2503  */
2504 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2505 {
2506         if (op_data->op_mea2_sem) {
2507                 up_read(op_data->op_mea2_sem);
2508                 op_data->op_mea2_sem = NULL;
2509         }
2510
2511         if (op_data->op_mea1_sem) {
2512                 up_read(op_data->op_mea1_sem);
2513                 op_data->op_mea1_sem = NULL;
2514         }
2515 }
2516
2517 /* this function prepares md_op_data hint for passing it down to MD stack. */
2518 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2519                                       struct inode *i1, struct inode *i2,
2520                                       const char *name, size_t namelen,
2521                                       __u32 mode, __u32 opc, void *data)
2522 {
2523         LASSERT(i1 != NULL);
2524
2525         if (name == NULL) {
2526                 /* Do not reuse namelen for something else. */
2527                 if (namelen != 0)
2528                         return ERR_PTR(-EINVAL);
2529         } else {
2530                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2531                         return ERR_PTR(-ENAMETOOLONG);
2532
2533                 if (!lu_name_is_valid_2(name, namelen))
2534                         return ERR_PTR(-EINVAL);
2535         }
2536
2537         if (op_data == NULL)
2538                 OBD_ALLOC_PTR(op_data);
2539
2540         if (op_data == NULL)
2541                 return ERR_PTR(-ENOMEM);
2542
2543         ll_i2gids(op_data->op_suppgids, i1, i2);
2544         op_data->op_fid1 = *ll_inode2fid(i1);
2545         op_data->op_default_stripe_offset = -1;
2546
2547         if (S_ISDIR(i1->i_mode)) {
2548                 down_read(&ll_i2info(i1)->lli_lsm_sem);
2549                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2550                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2551                 if (opc == LUSTRE_OPC_MKDIR)
2552                         op_data->op_default_stripe_offset =
2553                                    ll_i2info(i1)->lli_def_stripe_offset;
2554         }
2555
2556         if (i2) {
2557                 op_data->op_fid2 = *ll_inode2fid(i2);
2558                 if (S_ISDIR(i2->i_mode)) {
2559                         if (i2 != i1) {
2560                                 down_read(&ll_i2info(i2)->lli_lsm_sem);
2561                                 op_data->op_mea2_sem =
2562                                                 &ll_i2info(i2)->lli_lsm_sem;
2563                         }
2564                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2565                 }
2566         } else {
2567                 fid_zero(&op_data->op_fid2);
2568         }
2569
2570         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2571                 op_data->op_cli_flags |= CLI_HASH64;
2572
2573         if (ll_need_32bit_api(ll_i2sbi(i1)))
2574                 op_data->op_cli_flags |= CLI_API32;
2575
2576         op_data->op_name = name;
2577         op_data->op_namelen = namelen;
2578         op_data->op_mode = mode;
2579         op_data->op_mod_time = ktime_get_real_seconds();
2580         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2581         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2582         op_data->op_cap = cfs_curproc_cap_pack();
2583         op_data->op_mds = 0;
2584         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2585              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2586                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2587         }
2588         op_data->op_data = data;
2589
2590         return op_data;
2591 }
2592
2593 void ll_finish_md_op_data(struct md_op_data *op_data)
2594 {
2595         ll_unlock_md_op_lsm(op_data);
2596         security_release_secctx(op_data->op_file_secctx,
2597                                 op_data->op_file_secctx_size);
2598         OBD_FREE_PTR(op_data);
2599 }
2600
2601 #ifdef HAVE_SUPEROPS_USE_DENTRY
2602 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2603 #else
2604 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2605 #endif
2606 {
2607         struct ll_sb_info *sbi;
2608
2609 #ifdef HAVE_SUPEROPS_USE_DENTRY
2610         LASSERT((seq != NULL) && (dentry != NULL));
2611         sbi = ll_s2sbi(dentry->d_sb);
2612 #else
2613         LASSERT((seq != NULL) && (vfs != NULL));
2614         sbi = ll_s2sbi(vfs->mnt_sb);
2615 #endif
2616
2617         if (sbi->ll_flags & LL_SBI_NOLCK)
2618                 seq_puts(seq, ",nolock");
2619
2620         if (sbi->ll_flags & LL_SBI_FLOCK)
2621                 seq_puts(seq, ",flock");
2622
2623         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2624                 seq_puts(seq, ",localflock");
2625
2626         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2627                 seq_puts(seq, ",user_xattr");
2628
2629         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2630                 seq_puts(seq, ",lazystatfs");
2631
2632         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2633                 seq_puts(seq, ",user_fid2path");
2634
2635         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2636                 seq_puts(seq, ",always_ping");
2637
2638         RETURN(0);
2639 }
2640
2641 /**
2642  * Get obd name by cmd, and copy out to user space
2643  */
2644 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2645 {
2646         struct ll_sb_info *sbi = ll_i2sbi(inode);
2647         struct obd_device *obd;
2648         ENTRY;
2649
2650         if (cmd == OBD_IOC_GETDTNAME)
2651                 obd = class_exp2obd(sbi->ll_dt_exp);
2652         else if (cmd == OBD_IOC_GETMDNAME)
2653                 obd = class_exp2obd(sbi->ll_md_exp);
2654         else
2655                 RETURN(-EINVAL);
2656
2657         if (!obd)
2658                 RETURN(-ENOENT);
2659
2660         if (copy_to_user((void __user *)arg, obd->obd_name,
2661                          strlen(obd->obd_name) + 1))
2662                 RETURN(-EFAULT);
2663
2664         RETURN(0);
2665 }
2666
2667 /**
2668  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2669  * fsname will be returned in this buffer; otherwise, a static buffer will be
2670  * used to store the fsname and returned to caller.
2671  */
2672 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2673 {
2674         static char fsname_static[MTI_NAME_MAXLEN];
2675         struct lustre_sb_info *lsi = s2lsi(sb);
2676         char *ptr;
2677         int len;
2678
2679         if (buf == NULL) {
2680                 /* this means the caller wants to use static buffer
2681                  * and it doesn't care about race. Usually this is
2682                  * in error reporting path */
2683                 buf = fsname_static;
2684                 buflen = sizeof(fsname_static);
2685         }
2686
2687         len = strlen(lsi->lsi_lmd->lmd_profile);
2688         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2689         if (ptr && (strcmp(ptr, "-client") == 0))
2690                 len -= 7;
2691
2692         if (unlikely(len >= buflen))
2693                 len = buflen - 1;
2694         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2695         buf[len] = '\0';
2696
2697         return buf;
2698 }
2699
2700 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2701 {
2702         char *path = NULL;
2703
2704         struct path p;
2705
2706         p.dentry = dentry;
2707         p.mnt = current->fs->root.mnt;
2708         path_get(&p);
2709         path = d_path(&p, buf, bufsize);
2710         path_put(&p);
2711         return path;
2712 }
2713
2714 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2715 {
2716         char *buf, *path = NULL;
2717         struct dentry *dentry = NULL;
2718         struct inode *inode = page->mapping->host;
2719
2720         /* this can be called inside spin lock so use GFP_ATOMIC. */
2721         buf = (char *)__get_free_page(GFP_ATOMIC);
2722         if (buf != NULL) {
2723                 dentry = d_find_alias(page->mapping->host);
2724                 if (dentry != NULL)
2725                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2726         }
2727
2728         CDEBUG(D_WARNING,
2729                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2730                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2731                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2732                PFID(ll_inode2fid(inode)),
2733                (path && !IS_ERR(path)) ? path : "", ioret);
2734
2735         if (dentry != NULL)
2736                 dput(dentry);
2737
2738         if (buf != NULL)
2739                 free_page((unsigned long)buf);
2740 }
2741
2742 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2743                         struct lov_user_md **kbuf)
2744 {
2745         struct lov_user_md      lum;
2746         ssize_t                 lum_size;
2747         ENTRY;
2748
2749         if (copy_from_user(&lum, md, sizeof(lum)))
2750                 RETURN(-EFAULT);
2751
2752         lum_size = ll_lov_user_md_size(&lum);
2753         if (lum_size < 0)
2754                 RETURN(lum_size);
2755
2756         OBD_ALLOC(*kbuf, lum_size);
2757         if (*kbuf == NULL)
2758                 RETURN(-ENOMEM);
2759
2760         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2761                 OBD_FREE(*kbuf, lum_size);
2762                 RETURN(-EFAULT);
2763         }
2764
2765         RETURN(lum_size);
2766 }
2767
2768 /*
2769  * Compute llite root squash state after a change of root squash
2770  * configuration setting or add/remove of a lnet nid
2771  */
2772 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2773 {
2774         struct root_squash_info *squash = &sbi->ll_squash;
2775         int i;
2776         bool matched;
2777         struct lnet_process_id id;
2778
2779         /* Update norootsquash flag */
2780         down_write(&squash->rsi_sem);
2781         if (list_empty(&squash->rsi_nosquash_nids))
2782                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2783         else {
2784                 /* Do not apply root squash as soon as one of our NIDs is
2785                  * in the nosquash_nids list */
2786                 matched = false;
2787                 i = 0;
2788                 while (LNetGetId(i++, &id) != -ENOENT) {
2789                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2790                                 continue;
2791                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2792                                 matched = true;
2793                                 break;
2794                         }
2795                 }
2796                 if (matched)
2797                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2798                 else
2799                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2800         }
2801         up_write(&squash->rsi_sem);
2802 }
2803
2804 /**
2805  * Parse linkea content to extract information about a given hardlink
2806  *
2807  * \param[in]   ldata      - Initialized linkea data
2808  * \param[in]   linkno     - Link identifier
2809  * \param[out]  parent_fid - The entry's parent FID
2810  * \param[out]  ln         - Entry name destination buffer
2811  *
2812  * \retval 0 on success
2813  * \retval Appropriate negative error code on failure
2814  */
2815 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2816                             struct lu_fid *parent_fid, struct lu_name *ln)
2817 {
2818         unsigned int    idx;
2819         int             rc;
2820         ENTRY;
2821
2822         rc = linkea_init_with_rec(ldata);
2823         if (rc < 0)
2824                 RETURN(rc);
2825
2826         if (linkno >= ldata->ld_leh->leh_reccount)
2827                 /* beyond last link */
2828                 RETURN(-ENODATA);
2829
2830         linkea_first_entry(ldata);
2831         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2832                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2833                                     parent_fid);
2834                 if (idx == linkno)
2835                         break;
2836
2837                 linkea_next_entry(ldata);
2838         }
2839
2840         if (idx < linkno)
2841                 RETURN(-ENODATA);
2842
2843         RETURN(0);
2844 }
2845
2846 /**
2847  * Get parent FID and name of an identified link. Operation is performed for
2848  * a given link number, letting the caller iterate over linkno to list one or
2849  * all links of an entry.
2850  *
2851  * \param[in]     file - File descriptor against which to perform the operation
2852  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2853  *                       on and the available size. It is eventually filled with
2854  *                       the requested information or left untouched on error
2855  *
2856  * \retval - 0 on success
2857  * \retval - Appropriate negative error code on failure
2858  */
2859 int ll_getparent(struct file *file, struct getparent __user *arg)
2860 {
2861         struct inode            *inode = file_inode(file);
2862         struct linkea_data      *ldata;
2863         struct lu_buf            buf = LU_BUF_NULL;
2864         struct lu_name           ln;
2865         struct lu_fid            parent_fid;
2866         __u32                    linkno;
2867         __u32                    name_size;
2868         int                      rc;
2869
2870         ENTRY;
2871
2872         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2873             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2874                 RETURN(-EPERM);
2875
2876         if (get_user(name_size, &arg->gp_name_size))
2877                 RETURN(-EFAULT);
2878
2879         if (get_user(linkno, &arg->gp_linkno))
2880                 RETURN(-EFAULT);
2881
2882         if (name_size > PATH_MAX)
2883                 RETURN(-EINVAL);
2884
2885         OBD_ALLOC(ldata, sizeof(*ldata));
2886         if (ldata == NULL)
2887                 RETURN(-ENOMEM);
2888
2889         rc = linkea_data_new(ldata, &buf);
2890         if (rc < 0)
2891                 GOTO(ldata_free, rc);
2892
2893 #ifdef HAVE_XATTR_HANDLER_FLAGS
2894         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2895                            buf.lb_len, OBD_MD_FLXATTR);
2896 #else
2897         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2898                          buf.lb_len);
2899 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2900         if (rc < 0)
2901                 GOTO(lb_free, rc);
2902
2903         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2904         if (rc < 0)
2905                 GOTO(lb_free, rc);
2906
2907         if (ln.ln_namelen >= name_size)
2908                 GOTO(lb_free, rc = -EOVERFLOW);
2909
2910         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2911                 GOTO(lb_free, rc = -EFAULT);
2912
2913         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2914                 GOTO(lb_free, rc = -EFAULT);
2915
2916         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2917                 GOTO(lb_free, rc = -EFAULT);
2918
2919 lb_free:
2920         lu_buf_free(&buf);
2921 ldata_free:
2922         OBD_FREE(ldata, sizeof(*ldata));
2923
2924         RETURN(rc);
2925 }