Whamcloud - gitweb
81916ce845e51b06c28260b83538da7f1cda7f3b
[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.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2012, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/llite/llite_lib.c
37  *
38  * Lustre Light Super operations
39  */
40
41 #define DEBUG_SUBSYSTEM S_LLITE
42
43 #include <linux/module.h>
44 #include <linux/types.h>
45 #include <linux/version.h>
46 #include <linux/mm.h>
47
48 #include <lustre_lite.h>
49 #include <lustre_ha.h>
50 #include <lustre_dlm.h>
51 #include <lprocfs_status.h>
52 #include <lustre_disk.h>
53 #include <lustre_param.h>
54 #include <lustre_log.h>
55 #include <cl_object.h>
56 #include <obd_cksum.h>
57 #include "llite_internal.h"
58
59 cfs_mem_cache_t *ll_file_data_slab;
60
61 CFS_LIST_HEAD(ll_super_blocks);
62 DEFINE_SPINLOCK(ll_sb_lock);
63
64 #ifndef MS_HAS_NEW_AOPS
65 extern struct address_space_operations ll_aops;
66 #else
67 extern struct address_space_operations_ext ll_aops;
68 #endif
69
70 #ifndef log2
71 #define log2(n) ffz(~(n))
72 #endif
73
74 static struct ll_sb_info *ll_init_sbi(void)
75 {
76         struct ll_sb_info *sbi = NULL;
77         unsigned long pages;
78         unsigned long lru_page_max;
79         struct sysinfo si;
80         class_uuid_t uuid;
81         int i;
82         ENTRY;
83
84         OBD_ALLOC(sbi, sizeof(*sbi));
85         if (!sbi)
86                 RETURN(NULL);
87
88         spin_lock_init(&sbi->ll_lock);
89         mutex_init(&sbi->ll_lco.lco_lock);
90         spin_lock_init(&sbi->ll_pp_extent_lock);
91         spin_lock_init(&sbi->ll_process_lock);
92         sbi->ll_rw_stats_on = 0;
93
94         si_meminfo(&si);
95         pages = si.totalram - si.totalhigh;
96         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
97                 lru_page_max = pages / 2;
98         } else {
99                 lru_page_max = (pages / 4) * 3;
100         }
101
102         /* initialize lru data */
103         cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
104         sbi->ll_cache.ccc_lru_max = lru_page_max;
105         cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
106         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
107         CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
108
109         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
110                                            SBI_DEFAULT_READAHEAD_MAX);
111         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
112         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
113                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
114         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
115         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
116
117         ll_generate_random_uuid(uuid);
118         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
119         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
120
121         spin_lock(&ll_sb_lock);
122         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
123         spin_unlock(&ll_sb_lock);
124
125         sbi->ll_flags |= LL_SBI_VERBOSE;
126 #ifdef ENABLE_CHECKSUM
127         sbi->ll_flags |= LL_SBI_CHECKSUM;
128 #endif
129
130 #ifdef HAVE_LRU_RESIZE_SUPPORT
131         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
132 #endif
133
134         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
135                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
136                                pp_r_hist.oh_lock);
137                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
138                                pp_w_hist.oh_lock);
139         }
140
141         /* metadata statahead is enabled by default */
142         sbi->ll_sa_max = LL_SA_RPC_DEF;
143         cfs_atomic_set(&sbi->ll_sa_total, 0);
144         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
145         cfs_atomic_set(&sbi->ll_agl_total, 0);
146         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
147
148         RETURN(sbi);
149 }
150
151 void ll_free_sbi(struct super_block *sb)
152 {
153         struct ll_sb_info *sbi = ll_s2sbi(sb);
154         ENTRY;
155
156         if (sbi != NULL) {
157                 spin_lock(&ll_sb_lock);
158                 cfs_list_del(&sbi->ll_list);
159                 spin_unlock(&ll_sb_lock);
160                 OBD_FREE(sbi, sizeof(*sbi));
161         }
162         EXIT;
163 }
164
165 static struct dentry_operations ll_d_root_ops = {
166         .d_compare = ll_dcompare,
167         .d_revalidate = ll_revalidate_nd,
168 };
169
170 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
171                                     struct vfsmount *mnt)
172 {
173         struct inode *root = 0;
174         struct ll_sb_info *sbi = ll_s2sbi(sb);
175         struct obd_device *obd;
176         struct obd_capa *oc = NULL;
177         struct obd_statfs *osfs = NULL;
178         struct ptlrpc_request *request = NULL;
179         struct obd_connect_data *data = NULL;
180         struct obd_uuid *uuid;
181         struct md_op_data *op_data;
182         struct lustre_md lmd;
183         obd_valid valid;
184         int size, err, checksum;
185         ENTRY;
186
187         obd = class_name2obd(md);
188         if (!obd) {
189                 CERROR("MD %s: not setup or attached\n", md);
190                 RETURN(-EINVAL);
191         }
192
193         OBD_ALLOC_PTR(data);
194         if (data == NULL)
195                 RETURN(-ENOMEM);
196
197         OBD_ALLOC_PTR(osfs);
198         if (osfs == NULL) {
199                 OBD_FREE_PTR(data);
200                 RETURN(-ENOMEM);
201         }
202
203         if (proc_lustre_fs_root) {
204                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
205                                                   dt, md);
206                 if (err < 0)
207                         CERROR("could not register mount in /proc/fs/lustre\n");
208         }
209
210         /* indicate the features supported by this client */
211         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
212                                   OBD_CONNECT_ATTRFID  |
213                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
214                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
215                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
216                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
217                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
218                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
219                                   OBD_CONNECT_EINPROGRESS |
220                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
221                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
222
223         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
224                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
225
226 #ifdef HAVE_LRU_RESIZE_SUPPORT
227         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
228                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
229 #endif
230 #ifdef CONFIG_FS_POSIX_ACL
231         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
232 #endif
233
234         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
235                 /* flag mdc connection as lightweight, only used for test
236                  * purpose, use with care */
237                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
238
239         data->ocd_ibits_known = MDS_INODELOCK_FULL;
240         data->ocd_version = LUSTRE_VERSION_CODE;
241
242         if (sb->s_flags & MS_RDONLY)
243                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
244         if (sbi->ll_flags & LL_SBI_USER_XATTR)
245                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
246
247 #ifdef HAVE_MS_FLOCK_LOCK
248         /* force vfs to use lustre handler for flock() calls - bug 10743 */
249         sb->s_flags |= MS_FLOCK_LOCK;
250 #endif
251 #ifdef MS_HAS_NEW_AOPS
252         sb->s_flags |= MS_HAS_NEW_AOPS;
253 #endif
254
255         if (sbi->ll_flags & LL_SBI_FLOCK)
256                 sbi->ll_fop = &ll_file_operations_flock;
257         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
258                 sbi->ll_fop = &ll_file_operations;
259         else
260                 sbi->ll_fop = &ll_file_operations_noflock;
261
262         /* real client */
263         data->ocd_connect_flags |= OBD_CONNECT_REAL;
264         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
265                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
266
267         data->ocd_brw_size = MD_MAX_BRW_SIZE;
268
269         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
270         if (err == -EBUSY) {
271                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
272                                    "recovery, of which this client is not a "
273                                    "part. Please wait for recovery to complete,"
274                                    " abort, or time out.\n", md);
275                 GOTO(out, err);
276         } else if (err) {
277                 CERROR("cannot connect to %s: rc = %d\n", md, err);
278                 GOTO(out, err);
279         }
280
281         sbi->ll_md_exp->exp_connect_data = *data;
282
283         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
284                            LUSTRE_SEQ_METADATA);
285         if (err) {
286                 CERROR("%s: Can't init metadata layer FID infrastructure, "
287                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
288                 GOTO(out_md, err);
289         }
290
291         /* For mount, we only need fs info from MDT0, and also in DNE, it
292          * can make sure the client can be mounted as long as MDT0 is
293          * avaible */
294         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
295                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
296                         OBD_STATFS_FOR_MDT0);
297         if (err)
298                 GOTO(out_md_fid, err);
299
300         /* This needs to be after statfs to ensure connect has finished.
301          * Note that "data" does NOT contain the valid connect reply.
302          * If connecting to a 1.8 server there will be no LMV device, so
303          * we can access the MDC export directly and exp_connect_flags will
304          * be non-zero, but if accessing an upgraded 2.1 server it will
305          * have the correct flags filled in.
306          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
307         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
308         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
309             valid != CLIENT_CONNECT_MDT_REQD) {
310                 char *buf;
311
312                 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
313                 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
314                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
315                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
316                                    "feature(s) needed for correct operation "
317                                    "of this client (%s). Please upgrade "
318                                    "server or downgrade client.\n",
319                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
320                 OBD_FREE(buf, CFS_PAGE_SIZE);
321                 GOTO(out_md_fid, err = -EPROTO);
322         }
323
324         size = sizeof(*data);
325         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
326                            KEY_CONN_DATA,  &size, data, NULL);
327         if (err) {
328                 CERROR("%s: Get connect data failed: rc = %d\n",
329                        sbi->ll_md_exp->exp_obd->obd_name, err);
330                 GOTO(out_md_fid, err);
331         }
332
333         LASSERT(osfs->os_bsize);
334         sb->s_blocksize = osfs->os_bsize;
335         sb->s_blocksize_bits = log2(osfs->os_bsize);
336         sb->s_magic = LL_SUPER_MAGIC;
337         sb->s_maxbytes = MAX_LFS_FILESIZE;
338         sbi->ll_namelen = osfs->os_namelen;
339         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
340
341         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
342             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
343                 LCONSOLE_INFO("Disabling user_xattr feature because "
344                               "it is not supported on the server\n");
345                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
346         }
347
348         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
349 #ifdef MS_POSIXACL
350                 sb->s_flags |= MS_POSIXACL;
351 #endif
352                 sbi->ll_flags |= LL_SBI_ACL;
353         } else {
354                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
355 #ifdef MS_POSIXACL
356                 sb->s_flags &= ~MS_POSIXACL;
357 #endif
358                 sbi->ll_flags &= ~LL_SBI_ACL;
359         }
360
361         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
362                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
363                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
364                         LCONSOLE_INFO("client is set as remote by default.\n");
365                 }
366         } else {
367                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
368                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
369                         LCONSOLE_INFO("client claims to be remote, but server "
370                                       "rejected, forced to be local.\n");
371                 }
372         }
373
374         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
375                 LCONSOLE_INFO("client enabled MDS capability!\n");
376                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
377         }
378
379         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
380                 LCONSOLE_INFO("client enabled OSS capability!\n");
381                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
382         }
383
384         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
385                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
386
387         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
388                 sbi->ll_md_brw_size = data->ocd_brw_size;
389         else
390                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
391
392         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
393                 LCONSOLE_INFO("Layout lock feature supported.\n");
394                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
395         }
396
397         obd = class_name2obd(dt);
398         if (!obd) {
399                 CERROR("DT %s: not setup or attached\n", dt);
400                 GOTO(out_md_fid, err = -ENODEV);
401         }
402
403         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
404                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
405                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
406                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
407                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
408                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
409                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
410                                   OBD_CONNECT_MAXBYTES |
411                                   OBD_CONNECT_EINPROGRESS |
412                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
413                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
414
415         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
416                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
417
418         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
419                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
420                  * disabled by default, because it can still be enabled on the
421                  * fly via /proc. As a consequence, we still need to come to an
422                  * agreement on the supported algorithms at connect time */
423                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
424
425                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
426                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
427                 else
428                         data->ocd_cksum_types = cksum_types_supported_client();
429         }
430
431 #ifdef HAVE_LRU_RESIZE_SUPPORT
432         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
433 #endif
434         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
435                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
436
437         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
438                "ocd_grant: %d\n", data->ocd_connect_flags,
439                data->ocd_version, data->ocd_grant);
440
441         obd->obd_upcall.onu_owner = &sbi->ll_lco;
442         obd->obd_upcall.onu_upcall = cl_ocd_update;
443
444         data->ocd_brw_size = DT_MAX_BRW_SIZE;
445
446         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
447                           NULL);
448         if (err == -EBUSY) {
449                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
450                                    "recovery, of which this client is not a "
451                                    "part.  Please wait for recovery to "
452                                    "complete, abort, or time out.\n", dt);
453                 GOTO(out_md, err);
454         } else if (err) {
455                 CERROR("%s: Cannot connect to %s: rc = %d\n",
456                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
457                 GOTO(out_md, err);
458         }
459
460         sbi->ll_dt_exp->exp_connect_data = *data;
461
462         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
463                            LUSTRE_SEQ_METADATA);
464         if (err) {
465                 CERROR("%s: Can't init data layer FID infrastructure, "
466                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
467                 GOTO(out_dt, err);
468         }
469
470         mutex_lock(&sbi->ll_lco.lco_lock);
471         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
472         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
473         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
474         mutex_unlock(&sbi->ll_lco.lco_lock);
475
476         fid_zero(&sbi->ll_root_fid);
477         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
478         if (err) {
479                 CERROR("cannot mds_connect: rc = %d\n", err);
480                 GOTO(out_lock_cn_cb, err);
481         }
482         if (!fid_is_sane(&sbi->ll_root_fid)) {
483                 CERROR("%s: Invalid root fid "DFID" during mount\n",
484                        sbi->ll_md_exp->exp_obd->obd_name,
485                        PFID(&sbi->ll_root_fid));
486                 GOTO(out_lock_cn_cb, err = -EINVAL);
487         }
488         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
489
490         sb->s_op = &lustre_super_operations;
491 #if THREAD_SIZE >= 8192 /*b=17630*/
492         sb->s_export_op = &lustre_export_operations;
493 #endif
494
495         /* make root inode
496          * XXX: move this to after cbd setup? */
497         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
498         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
499                 valid |= OBD_MD_FLRMTPERM;
500         else if (sbi->ll_flags & LL_SBI_ACL)
501                 valid |= OBD_MD_FLACL;
502
503         OBD_ALLOC_PTR(op_data);
504         if (op_data == NULL)
505                 GOTO(out_lock_cn_cb, err = -ENOMEM);
506
507         op_data->op_fid1 = sbi->ll_root_fid;
508         op_data->op_mode = 0;
509         op_data->op_capa1 = oc;
510         op_data->op_valid = valid;
511
512         err = md_getattr(sbi->ll_md_exp, op_data, &request);
513         if (oc)
514                 capa_put(oc);
515         OBD_FREE_PTR(op_data);
516         if (err) {
517                 CERROR("%s: md_getattr failed for root: rc = %d\n",
518                        sbi->ll_md_exp->exp_obd->obd_name, err);
519                 GOTO(out_lock_cn_cb, err);
520         }
521
522         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
523                                sbi->ll_md_exp, &lmd);
524         if (err) {
525                 CERROR("failed to understand root inode md: rc = %d\n", err);
526                 ptlrpc_req_finished(request);
527                 GOTO(out_lock_cn_cb, err);
528         }
529
530         LASSERT(fid_is_sane(&sbi->ll_root_fid));
531         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
532                                             ll_need_32bit_api(sbi)),
533                        &lmd);
534         md_free_lustre_md(sbi->ll_md_exp, &lmd);
535         ptlrpc_req_finished(request);
536
537         if (root == NULL || IS_ERR(root)) {
538                 if (lmd.lsm)
539                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
540 #ifdef CONFIG_FS_POSIX_ACL
541                 if (lmd.posix_acl) {
542                         posix_acl_release(lmd.posix_acl);
543                         lmd.posix_acl = NULL;
544                 }
545 #endif
546                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
547                 root = NULL;
548                 CERROR("lustre_lite: bad iget4 for root\n");
549                 GOTO(out_root, err);
550         }
551
552         err = ll_close_thread_start(&sbi->ll_lcq);
553         if (err) {
554                 CERROR("cannot start close thread: rc %d\n", err);
555                 GOTO(out_root, err);
556         }
557
558 #ifdef CONFIG_FS_POSIX_ACL
559         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
560                 rct_init(&sbi->ll_rct);
561                 et_init(&sbi->ll_et);
562         }
563 #endif
564
565         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
566         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
567                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
568                                  NULL);
569         cl_sb_init(sb);
570
571         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
572                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
573                                  &sbi->ll_cache, NULL);
574
575         sb->s_root = d_make_root(root);
576         if (sb->s_root == NULL) {
577                 CERROR("%s: can't make root dentry\n",
578                         ll_get_fsname(sb, NULL, 0));
579                 GOTO(out_root, err = -ENOMEM);
580         }
581
582 #ifdef HAVE_DCACHE_LOCK
583         sb->s_root->d_op = &ll_d_root_ops;
584 #else
585         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
586         d_set_d_op(sb->s_root, &ll_d_root_ops);
587         sb->s_d_op = &ll_d_ops;
588 #endif
589
590         sbi->ll_sdev_orig = sb->s_dev;
591
592         /* We set sb->s_dev equal on all lustre clients in order to support
593          * NFS export clustering.  NFSD requires that the FSID be the same
594          * on all clients. */
595         /* s_dev is also used in lt_compare() to compare two fs, but that is
596          * only a node-local comparison. */
597         uuid = obd_get_uuid(sbi->ll_md_exp);
598         if (uuid != NULL)
599                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
600
601         if (data != NULL)
602                 OBD_FREE_PTR(data);
603         if (osfs != NULL)
604                 OBD_FREE_PTR(osfs);
605
606         RETURN(err);
607 out_root:
608         if (root)
609                 iput(root);
610 out_lock_cn_cb:
611         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
612 out_dt:
613         obd_disconnect(sbi->ll_dt_exp);
614         sbi->ll_dt_exp = NULL;
615         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
616         obd_zombie_barrier();
617 out_md_fid:
618         obd_fid_fini(sbi->ll_md_exp->exp_obd);
619 out_md:
620         obd_disconnect(sbi->ll_md_exp);
621         sbi->ll_md_exp = NULL;
622 out:
623         if (data != NULL)
624                 OBD_FREE_PTR(data);
625         if (osfs != NULL)
626                 OBD_FREE_PTR(osfs);
627         lprocfs_unregister_mountpoint(sbi);
628         return err;
629 }
630
631 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
632 {
633         int size, rc;
634
635         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
636         size = sizeof(int);
637         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
638                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
639         if (rc)
640                 CERROR("Get max mdsize error rc %d \n", rc);
641
642         RETURN(rc);
643 }
644
645 void ll_dump_inode(struct inode *inode)
646 {
647         struct ll_d_hlist_node *tmp;
648         int dentry_count = 0;
649
650         LASSERT(inode != NULL);
651
652         ll_d_hlist_for_each(tmp, &inode->i_dentry)
653                 dentry_count++;
654
655         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
656                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
657                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
658 }
659
660 void lustre_dump_dentry(struct dentry *dentry, int recur)
661 {
662         struct list_head *tmp;
663         int subdirs = 0;
664
665         LASSERT(dentry != NULL);
666
667         list_for_each(tmp, &dentry->d_subdirs)
668                 subdirs++;
669
670         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
671                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
672                dentry->d_name.len, dentry->d_name.name,
673                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
674                dentry->d_parent, dentry->d_inode, d_refcount(dentry),
675                dentry->d_flags, dentry->d_fsdata, subdirs);
676         if (dentry->d_inode != NULL)
677                 ll_dump_inode(dentry->d_inode);
678
679         if (recur == 0)
680                 return;
681
682         list_for_each(tmp, &dentry->d_subdirs) {
683                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
684                 lustre_dump_dentry(d, recur - 1);
685         }
686 }
687
688 void client_common_put_super(struct super_block *sb)
689 {
690         struct ll_sb_info *sbi = ll_s2sbi(sb);
691         ENTRY;
692
693 #ifdef CONFIG_FS_POSIX_ACL
694         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
695                 et_fini(&sbi->ll_et);
696                 rct_fini(&sbi->ll_rct);
697         }
698 #endif
699
700         ll_close_thread_shutdown(sbi->ll_lcq);
701
702         cl_sb_fini(sb);
703
704         cfs_list_del(&sbi->ll_conn_chain);
705
706         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
707         obd_disconnect(sbi->ll_dt_exp);
708         sbi->ll_dt_exp = NULL;
709         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
710          * see LU-2543. */
711         obd_zombie_barrier();
712
713         lprocfs_unregister_mountpoint(sbi);
714
715         obd_fid_fini(sbi->ll_md_exp->exp_obd);
716         obd_disconnect(sbi->ll_md_exp);
717         sbi->ll_md_exp = NULL;
718
719         EXIT;
720 }
721
722 void ll_kill_super(struct super_block *sb)
723 {
724         struct ll_sb_info *sbi;
725
726         ENTRY;
727
728         /* not init sb ?*/
729         if (!(sb->s_flags & MS_ACTIVE))
730                 return;
731
732         sbi = ll_s2sbi(sb);
733         /* we need restore s_dev from changed for clustred NFS before put_super
734          * because new kernels have cached s_dev and change sb->s_dev in
735          * put_super not affected real removing devices */
736         if (sbi)
737                 sb->s_dev = sbi->ll_sdev_orig;
738         EXIT;
739 }
740
741 char *ll_read_opt(const char *opt, char *data)
742 {
743         char *value;
744         char *retval;
745         ENTRY;
746
747         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
748         if (strncmp(opt, data, strlen(opt)))
749                 RETURN(NULL);
750         if ((value = strchr(data, '=')) == NULL)
751                 RETURN(NULL);
752
753         value++;
754         OBD_ALLOC(retval, strlen(value) + 1);
755         if (!retval) {
756                 CERROR("out of memory!\n");
757                 RETURN(NULL);
758         }
759
760         memcpy(retval, value, strlen(value)+1);
761         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
762         RETURN(retval);
763 }
764
765 static inline int ll_set_opt(const char *opt, char *data, int fl)
766 {
767         if (strncmp(opt, data, strlen(opt)) != 0)
768                 return(0);
769         else
770                 return(fl);
771 }
772
773 /* non-client-specific mount options are parsed in lmd_parse */
774 static int ll_options(char *options, int *flags)
775 {
776         int tmp;
777         char *s1 = options, *s2;
778         ENTRY;
779
780         if (!options)
781                 RETURN(0);
782
783         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
784
785         while (*s1) {
786                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
787                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
788                 if (tmp) {
789                         *flags |= tmp;
790                         goto next;
791                 }
792                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
793                 if (tmp) {
794                         *flags |= tmp;
795                         goto next;
796                 }
797                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
798                 if (tmp) {
799                         *flags |= tmp;
800                         goto next;
801                 }
802                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
803                 if (tmp) {
804                         *flags &= ~tmp;
805                         goto next;
806                 }
807                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
808                 if (tmp) {
809                         *flags |= tmp;
810                         goto next;
811                 }
812                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
813                 if (tmp) {
814                         *flags &= ~tmp;
815                         goto next;
816                 }
817 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
818                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
819                 if (tmp) {
820                         /* Ignore deprecated mount option.  The client will
821                          * always try to mount with ACL support, whether this
822                          * is used depends on whether server supports it. */
823                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
824                                                   "mount option 'acl'.\n");
825                         goto next;
826                 }
827                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
828                 if (tmp) {
829                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
830                                                   "mount option 'noacl'.\n");
831                         goto next;
832                 }
833 #else
834 #warning "{no}acl options have been deprecated since 1.8, please remove them"
835 #endif
836                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
837                 if (tmp) {
838                         *flags |= tmp;
839                         goto next;
840                 }
841                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
842                 if (tmp) {
843                         *flags |= tmp;
844                         goto next;
845                 }
846                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
847                 if (tmp) {
848                         *flags &= ~tmp;
849                         goto next;
850                 }
851
852                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
853                 if (tmp) {
854                         *flags |= tmp;
855                         goto next;
856                 }
857                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
858                 if (tmp) {
859                         *flags &= ~tmp;
860                         goto next;
861                 }
862                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
863                 if (tmp) {
864                         *flags |= tmp;
865                         goto next;
866                 }
867                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
868                 if (tmp) {
869                         *flags &= ~tmp;
870                         goto next;
871                 }
872                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
873                 if (tmp) {
874                         *flags |= tmp;
875                         goto next;
876                 }
877                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
878                 if (tmp) {
879                         *flags &= ~tmp;
880                         goto next;
881                 }
882                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
883                 if (tmp) {
884                         *flags |= tmp;
885                         goto next;
886                 }
887                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
888                 if (tmp) {
889                         *flags |= tmp;
890                         goto next;
891                 }
892                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
893                 if (tmp) {
894                         *flags |= tmp;
895                         goto next;
896                 }
897                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
898                 if (tmp) {
899                         *flags &= ~tmp;
900                         goto next;
901                 }
902                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
903                                    s1);
904                 RETURN(-EINVAL);
905
906 next:
907                 /* Find next opt */
908                 s2 = strchr(s1, ',');
909                 if (s2 == NULL)
910                         break;
911                 s1 = s2 + 1;
912         }
913         RETURN(0);
914 }
915
916 void ll_lli_init(struct ll_inode_info *lli)
917 {
918         lli->lli_inode_magic = LLI_INODE_MAGIC;
919         lli->lli_flags = 0;
920         lli->lli_ioepoch = 0;
921         lli->lli_maxbytes = MAX_LFS_FILESIZE;
922         spin_lock_init(&lli->lli_lock);
923         lli->lli_posix_acl = NULL;
924         lli->lli_remote_perms = NULL;
925         mutex_init(&lli->lli_rmtperm_mutex);
926         /* Do not set lli_fid, it has been initialized already. */
927         fid_zero(&lli->lli_pfid);
928         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
929         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
930         cfs_atomic_set(&lli->lli_open_count, 0);
931         lli->lli_mds_capa = NULL;
932         lli->lli_rmtperm_time = 0;
933         lli->lli_pending_och = NULL;
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         lli->lli_has_smd = false;
943         lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
944         lli->lli_clob = NULL;
945
946         LASSERT(lli->lli_vfs_inode.i_mode != 0);
947         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
948                 mutex_init(&lli->lli_readdir_mutex);
949                 lli->lli_opendir_key = NULL;
950                 lli->lli_sai = NULL;
951                 lli->lli_def_acl = NULL;
952                 spin_lock_init(&lli->lli_sa_lock);
953                 lli->lli_opendir_pid = 0;
954         } else {
955                 sema_init(&lli->lli_size_sem, 1);
956                 lli->lli_size_sem_owner = NULL;
957                 lli->lli_symlink_name = NULL;
958                 init_rwsem(&lli->lli_trunc_sem);
959                 mutex_init(&lli->lli_write_mutex);
960                 init_rwsem(&lli->lli_glimpse_sem);
961                 lli->lli_glimpse_time = 0;
962                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
963                 lli->lli_agl_index = 0;
964                 lli->lli_async_rc = 0;
965                 lli->lli_volatile = false;
966         }
967         mutex_init(&lli->lli_layout_mutex);
968 }
969
970 static inline int ll_bdi_register(struct backing_dev_info *bdi)
971 {
972 #ifdef HAVE_BDI_REGISTER
973         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
974
975 #ifdef HAVE_BDI_NAME
976         bdi->name = "lustre";
977 #endif
978         return bdi_register(bdi, NULL, "lustre-%d",
979                             atomic_inc_return(&ll_bdi_num));
980 #else
981         return 0;
982 #endif
983 }
984
985 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
986 {
987         struct lustre_profile *lprof = NULL;
988         struct lustre_sb_info *lsi = s2lsi(sb);
989         struct ll_sb_info *sbi;
990         char  *dt = NULL, *md = NULL;
991         char  *profilenm = get_profile_name(sb);
992         struct config_llog_instance *cfg;
993         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
994         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
995         int    err;
996         ENTRY;
997
998         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
999
1000         OBD_ALLOC_PTR(cfg);
1001         if (cfg == NULL)
1002                 RETURN(-ENOMEM);
1003
1004         cfs_module_get();
1005
1006         /* client additional sb info */
1007         lsi->lsi_llsbi = sbi = ll_init_sbi();
1008         if (!sbi) {
1009                 cfs_module_put(THIS_MODULE);
1010                 OBD_FREE_PTR(cfg);
1011                 RETURN(-ENOMEM);
1012         }
1013
1014         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1015         if (err)
1016                 GOTO(out_free, err);
1017
1018         err = bdi_init(&lsi->lsi_bdi);
1019         if (err)
1020                 GOTO(out_free, err);
1021         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1022         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1023         err = ll_bdi_register(&lsi->lsi_bdi);
1024         if (err)
1025                 GOTO(out_free, err);
1026
1027 #ifdef HAVE_SB_BDI
1028         sb->s_bdi = &lsi->lsi_bdi;
1029 #endif
1030
1031         /* Generate a string unique to this super, in case some joker tries
1032            to mount the same fs at two mount points.
1033            Use the address of the super itself.*/
1034         cfg->cfg_instance = sb;
1035         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1036         cfg->cfg_callback = class_config_llog_handler;
1037         /* set up client obds */
1038         err = lustre_process_log(sb, profilenm, cfg);
1039         if (err < 0) {
1040                 CERROR("Unable to process log: %d\n", err);
1041                 GOTO(out_free, err);
1042         }
1043
1044         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1045         lprof = class_get_profile(profilenm);
1046         if (lprof == NULL) {
1047                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1048                                    " read from the MGS.  Does that filesystem "
1049                                    "exist?\n", profilenm);
1050                 GOTO(out_free, err = -EINVAL);
1051         }
1052         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1053                lprof->lp_md, lprof->lp_dt);
1054
1055         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1056         if (!dt)
1057                 GOTO(out_free, err = -ENOMEM);
1058         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1059
1060         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1061         if (!md)
1062                 GOTO(out_free, err = -ENOMEM);
1063         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1064
1065         /* connections, registrations, sb setup */
1066         err = client_common_fill_super(sb, md, dt, mnt);
1067
1068 out_free:
1069         if (md)
1070                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1071         if (dt)
1072                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1073         if (err)
1074                 ll_put_super(sb);
1075         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1076                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1077
1078         OBD_FREE_PTR(cfg);
1079         RETURN(err);
1080 } /* ll_fill_super */
1081
1082
1083 void lu_context_keys_dump(void);
1084
1085 void ll_put_super(struct super_block *sb)
1086 {
1087         struct config_llog_instance cfg;
1088         struct obd_device *obd;
1089         struct lustre_sb_info *lsi = s2lsi(sb);
1090         struct ll_sb_info *sbi = ll_s2sbi(sb);
1091         char *profilenm = get_profile_name(sb);
1092         int force = 1, next;
1093         ENTRY;
1094
1095         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1096
1097         ll_print_capa_stat(sbi);
1098
1099         cfg.cfg_instance = sb;
1100         lustre_end_log(sb, profilenm, &cfg);
1101
1102         if (sbi->ll_md_exp) {
1103                 obd = class_exp2obd(sbi->ll_md_exp);
1104                 if (obd)
1105                         force = obd->obd_force;
1106         }
1107
1108         /* We need to set force before the lov_disconnect in
1109            lustre_common_put_super, since l_d cleans up osc's as well. */
1110         if (force) {
1111                 next = 0;
1112                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1113                                                      &next)) != NULL) {
1114                         obd->obd_force = force;
1115                 }
1116         }
1117
1118         if (sbi->ll_lcq) {
1119                 /* Only if client_common_fill_super succeeded */
1120                 client_common_put_super(sb);
1121         }
1122
1123         next = 0;
1124         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1125                 class_manual_cleanup(obd);
1126         }
1127
1128         if (sbi->ll_flags & LL_SBI_VERBOSE)
1129                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1130
1131         if (profilenm)
1132                 class_del_profile(profilenm);
1133
1134         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1135                 bdi_destroy(&lsi->lsi_bdi);
1136                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1137         }
1138
1139         ll_free_sbi(sb);
1140         lsi->lsi_llsbi = NULL;
1141
1142         lustre_common_put_super(sb);
1143
1144         cfs_module_put(THIS_MODULE);
1145
1146         EXIT;
1147 } /* client_put_super */
1148
1149 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1150 {
1151         struct inode *inode = NULL;
1152
1153         /* NOTE: we depend on atomic igrab() -bzzz */
1154         lock_res_and_lock(lock);
1155         if (lock->l_resource->lr_lvb_inode) {
1156                 struct ll_inode_info * lli;
1157                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1158                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1159                         inode = igrab(lock->l_resource->lr_lvb_inode);
1160                 } else {
1161                         inode = lock->l_resource->lr_lvb_inode;
1162                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1163                                          D_WARNING, lock, "lr_lvb_inode %p is "
1164                                          "bogus: magic %08x",
1165                                          lock->l_resource->lr_lvb_inode,
1166                                          lli->lli_inode_magic);
1167                         inode = NULL;
1168                 }
1169         }
1170         unlock_res_and_lock(lock);
1171         return inode;
1172 }
1173
1174 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1175 {
1176         struct inode *inode = NULL;
1177         /* NOTE: we depend on atomic igrab() -bzzz */
1178         lock_res_and_lock(lock);
1179         if (lock->l_ast_data) {
1180                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1181                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1182                         inode = igrab(lock->l_ast_data);
1183                 } else {
1184                         inode = lock->l_ast_data;
1185                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1186                                          D_WARNING, lock, "l_ast_data %p is "
1187                                          "bogus: magic %08x", lock->l_ast_data,
1188                                          lli->lli_inode_magic);
1189                         inode = NULL;
1190                 }
1191         }
1192         unlock_res_and_lock(lock);
1193         return inode;
1194 }
1195
1196 void ll_clear_inode(struct inode *inode)
1197 {
1198         struct ll_inode_info *lli = ll_i2info(inode);
1199         struct ll_sb_info *sbi = ll_i2sbi(inode);
1200         ENTRY;
1201
1202         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1203                inode->i_generation, inode);
1204
1205         if (S_ISDIR(inode->i_mode)) {
1206                 /* these should have been cleared in ll_file_release */
1207                 LASSERT(lli->lli_opendir_key == NULL);
1208                 LASSERT(lli->lli_sai == NULL);
1209                 LASSERT(lli->lli_opendir_pid == 0);
1210         }
1211
1212         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1213         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1214
1215         LASSERT(!lli->lli_open_fd_write_count);
1216         LASSERT(!lli->lli_open_fd_read_count);
1217         LASSERT(!lli->lli_open_fd_exec_count);
1218
1219         if (lli->lli_mds_write_och)
1220                 ll_md_real_close(inode, FMODE_WRITE);
1221         if (lli->lli_mds_exec_och)
1222                 ll_md_real_close(inode, FMODE_EXEC);
1223         if (lli->lli_mds_read_och)
1224                 ll_md_real_close(inode, FMODE_READ);
1225
1226         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1227                 OBD_FREE(lli->lli_symlink_name,
1228                          strlen(lli->lli_symlink_name) + 1);
1229                 lli->lli_symlink_name = NULL;
1230         }
1231
1232         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1233                 LASSERT(lli->lli_posix_acl == NULL);
1234                 if (lli->lli_remote_perms) {
1235                         free_rmtperm_hash(lli->lli_remote_perms);
1236                         lli->lli_remote_perms = NULL;
1237                 }
1238         }
1239 #ifdef CONFIG_FS_POSIX_ACL
1240         else if (lli->lli_posix_acl) {
1241                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1242                 LASSERT(lli->lli_remote_perms == NULL);
1243                 posix_acl_release(lli->lli_posix_acl);
1244                 lli->lli_posix_acl = NULL;
1245         }
1246 #endif
1247         lli->lli_inode_magic = LLI_INODE_DEAD;
1248
1249         ll_clear_inode_capas(inode);
1250         if (!S_ISDIR(inode->i_mode))
1251                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1252
1253         /*
1254          * XXX This has to be done before lsm is freed below, because
1255          * cl_object still uses inode lsm.
1256          */
1257         cl_inode_fini(inode);
1258         lli->lli_has_smd = false;
1259
1260         EXIT;
1261 }
1262
1263 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1264                   struct md_open_data **mod)
1265 {
1266         struct lustre_md md;
1267         struct inode *inode = dentry->d_inode;
1268         struct ll_sb_info *sbi = ll_i2sbi(inode);
1269         struct ptlrpc_request *request = NULL;
1270         int rc, ia_valid;
1271         ENTRY;
1272
1273         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1274                                      LUSTRE_OPC_ANY, NULL);
1275         if (IS_ERR(op_data))
1276                 RETURN(PTR_ERR(op_data));
1277
1278         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1279                         &request, mod);
1280         if (rc) {
1281                 ptlrpc_req_finished(request);
1282                 if (rc == -ENOENT) {
1283                         clear_nlink(inode);
1284                         /* Unlinked special device node? Or just a race?
1285                          * Pretend we done everything. */
1286                         if (!S_ISREG(inode->i_mode) &&
1287                             !S_ISDIR(inode->i_mode)) {
1288                                 ia_valid = op_data->op_attr.ia_valid;
1289                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1290                                 rc = simple_setattr(dentry, &op_data->op_attr);
1291                                 op_data->op_attr.ia_valid = ia_valid;
1292                         }
1293                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1294                         CERROR("md_setattr fails: rc = %d\n", rc);
1295                 }
1296                 RETURN(rc);
1297         }
1298
1299         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1300                               sbi->ll_md_exp, &md);
1301         if (rc) {
1302                 ptlrpc_req_finished(request);
1303                 RETURN(rc);
1304         }
1305
1306         ia_valid = op_data->op_attr.ia_valid;
1307         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1308          * cache is not cleared yet. */
1309         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1310         rc = simple_setattr(dentry, &op_data->op_attr);
1311         op_data->op_attr.ia_valid = ia_valid;
1312
1313         /* Extract epoch data if obtained. */
1314         op_data->op_handle = md.body->handle;
1315         op_data->op_ioepoch = md.body->ioepoch;
1316
1317         ll_update_inode(inode, &md);
1318         ptlrpc_req_finished(request);
1319
1320         RETURN(rc);
1321 }
1322
1323 /* Close IO epoch and send Size-on-MDS attribute update. */
1324 static int ll_setattr_done_writing(struct inode *inode,
1325                                    struct md_op_data *op_data,
1326                                    struct md_open_data *mod)
1327 {
1328         struct ll_inode_info *lli = ll_i2info(inode);
1329         int rc = 0;
1330         ENTRY;
1331
1332         LASSERT(op_data != NULL);
1333         if (!S_ISREG(inode->i_mode))
1334                 RETURN(0);
1335
1336         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1337                op_data->op_ioepoch, PFID(&lli->lli_fid));
1338
1339         op_data->op_flags = MF_EPOCH_CLOSE;
1340         ll_done_writing_attr(inode, op_data);
1341         ll_pack_inode2opdata(inode, op_data, NULL);
1342
1343         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1344         if (rc == -EAGAIN) {
1345                 /* MDS has instructed us to obtain Size-on-MDS attribute
1346                  * from OSTs and send setattr to back to MDS. */
1347                 rc = ll_som_update(inode, op_data);
1348         } else if (rc) {
1349                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1350                        inode->i_ino, rc);
1351         }
1352         RETURN(rc);
1353 }
1354
1355 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1356 {
1357         struct obd_capa *capa;
1358         int rc;
1359
1360         if (attr->ia_valid & ATTR_SIZE)
1361                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1362         else
1363                 capa = ll_mdscapa_get(inode);
1364
1365         rc = cl_setattr_ost(inode, attr, capa);
1366
1367         if (attr->ia_valid & ATTR_SIZE)
1368                 ll_truncate_free_capa(capa);
1369         else
1370                 capa_put(capa);
1371
1372         return rc;
1373 }
1374
1375 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1376 /**
1377  * inode_newsize_ok - may this inode be truncated to a given size
1378  * @inode:      the inode to be truncated
1379  * @offset:     the new size to assign to the inode
1380  * @Returns:    0 on success, -ve errno on failure
1381  *
1382  * inode_newsize_ok will check filesystem limits and ulimits to check that the
1383  * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1384  * when necessary. Caller must not proceed with inode size change if failure is
1385  * returned. @inode must be a file (not directory), with appropriate
1386  * permissions to allow truncate (inode_newsize_ok does NOT check these
1387  * conditions).
1388  *
1389  * inode_newsize_ok must be called with i_mutex held.
1390  */
1391 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1392 {
1393         if (inode->i_size < offset) {
1394                 unsigned long limit;
1395
1396                 limit = rlimit(RLIMIT_FSIZE);
1397                 if (limit != RLIM_INFINITY && offset > limit)
1398                         goto out_sig;
1399                 if (offset > inode->i_sb->s_maxbytes)
1400                         goto out_big;
1401         } else {
1402                 /*
1403                  * truncation of in-use swapfiles is disallowed - it would
1404                  * cause subsequent swapout to scribble on the now-freed
1405                  * blocks.
1406                  */
1407                 if (IS_SWAPFILE(inode))
1408                         return -ETXTBSY;
1409         }
1410
1411         return 0;
1412 out_sig:
1413         send_sig(SIGXFSZ, current, 0);
1414 out_big:
1415         return -EFBIG;
1416 }
1417 #endif
1418
1419 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1420  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1421  * keep these values until such a time that objects are allocated for it.
1422  * We do the MDS operations first, as it is checking permissions for us.
1423  * We don't to the MDS RPC if there is nothing that we want to store there,
1424  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1425  * going to do an RPC anyways.
1426  *
1427  * If we are doing a truncate, we will send the mtime and ctime updates
1428  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1429  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1430  * at the same time.
1431  */
1432 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1433 {
1434         struct inode *inode = dentry->d_inode;
1435         struct ll_inode_info *lli = ll_i2info(inode);
1436         struct md_op_data *op_data = NULL;
1437         struct md_open_data *mod = NULL;
1438         int rc = 0, rc1 = 0;
1439         ENTRY;
1440
1441         CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1442                 "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
1443                 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1444                 attr->ia_valid);
1445
1446         if (attr->ia_valid & ATTR_SIZE) {
1447                 /* Check new size against VFS/VM file size limit and rlimit */
1448                 rc = inode_newsize_ok(inode, attr->ia_size);
1449                 if (rc)
1450                         RETURN(rc);
1451
1452                 /* The maximum Lustre file size is variable, based on the
1453                  * OST maximum object size and number of stripes.  This
1454                  * needs another check in addition to the VFS check above. */
1455                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1456                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1457                                PFID(&lli->lli_fid), attr->ia_size,
1458                                ll_file_maxbytes(inode));
1459                         RETURN(-EFBIG);
1460                 }
1461
1462                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1463         }
1464
1465         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1466         if (attr->ia_valid & TIMES_SET_FLAGS) {
1467                 if (cfs_curproc_fsuid() != inode->i_uid &&
1468                     !cfs_capable(CFS_CAP_FOWNER))
1469                         RETURN(-EPERM);
1470         }
1471
1472         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1473         if (attr->ia_valid & ATTR_CTIME) {
1474                 attr->ia_ctime = CFS_CURRENT_TIME;
1475                 attr->ia_valid |= ATTR_CTIME_SET;
1476         }
1477         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1478             (attr->ia_valid & ATTR_ATIME)) {
1479                 attr->ia_atime = CFS_CURRENT_TIME;
1480                 attr->ia_valid |= ATTR_ATIME_SET;
1481         }
1482         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1483             (attr->ia_valid & ATTR_MTIME)) {
1484                 attr->ia_mtime = CFS_CURRENT_TIME;
1485                 attr->ia_valid |= ATTR_MTIME_SET;
1486         }
1487
1488         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1489                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1490                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1491                        cfs_time_current_sec());
1492
1493         /* If we are changing file size, file content is modified, flag it. */
1494         if (attr->ia_valid & ATTR_SIZE) {
1495                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1496                 spin_lock(&lli->lli_lock);
1497                 lli->lli_flags |= LLIF_DATA_MODIFIED;
1498                 spin_unlock(&lli->lli_lock);
1499         }
1500
1501         /* We always do an MDS RPC, even if we're only changing the size;
1502          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1503
1504         OBD_ALLOC_PTR(op_data);
1505         if (op_data == NULL)
1506                 RETURN(-ENOMEM);
1507
1508         if (!S_ISDIR(inode->i_mode)) {
1509                 if (attr->ia_valid & ATTR_SIZE)
1510                         inode_dio_write_done(inode);
1511                 mutex_unlock(&inode->i_mutex);
1512                 down_write(&lli->lli_trunc_sem);
1513         }
1514
1515         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1516
1517         /* Open epoch for truncate. */
1518         if (exp_connect_som(ll_i2mdexp(inode)) &&
1519             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1520                 op_data->op_flags = MF_EPOCH_OPEN;
1521
1522         rc = ll_md_setattr(dentry, op_data, &mod);
1523         if (rc)
1524                 GOTO(out, rc);
1525
1526         /* RPC to MDT is sent, cancel data modification flag */
1527         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1528                 spin_lock(&lli->lli_lock);
1529                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1530                 spin_unlock(&lli->lli_lock);
1531         }
1532
1533         ll_ioepoch_open(lli, op_data->op_ioepoch);
1534         if (!S_ISREG(inode->i_mode))
1535                 GOTO(out, rc = 0);
1536
1537         if (attr->ia_valid & (ATTR_SIZE |
1538                               ATTR_ATIME | ATTR_ATIME_SET |
1539                               ATTR_MTIME | ATTR_MTIME_SET))
1540                 /* For truncate and utimes sending attributes to OSTs, setting
1541                  * mtime/atime to the past will be performed under PW [0:EOF]
1542                  * extent lock (new_size:EOF for truncate).  It may seem
1543                  * excessive to send mtime/atime updates to OSTs when not
1544                  * setting times to past, but it is necessary due to possible
1545                  * time de-synchronization between MDT inode and OST objects */
1546                 rc = ll_setattr_ost(inode, attr);
1547         EXIT;
1548 out:
1549         if (op_data) {
1550                 if (op_data->op_ioepoch) {
1551                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1552                         if (!rc)
1553                                 rc = rc1;
1554                 }
1555                 ll_finish_md_op_data(op_data);
1556         }
1557         if (!S_ISDIR(inode->i_mode)) {
1558                 up_write(&lli->lli_trunc_sem);
1559                 mutex_lock(&inode->i_mutex);
1560                 if (attr->ia_valid & ATTR_SIZE)
1561                         inode_dio_wait(inode);
1562         }
1563
1564         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1565                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1566
1567         return rc;
1568 }
1569
1570 int ll_setattr(struct dentry *de, struct iattr *attr)
1571 {
1572         int mode = de->d_inode->i_mode;
1573
1574         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1575                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1576                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1577
1578         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1579                                (ATTR_SIZE|ATTR_MODE)) &&
1580             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1581              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1582               !(attr->ia_mode & S_ISGID))))
1583                 attr->ia_valid |= ATTR_FORCE;
1584
1585         if ((mode & S_ISUID) &&
1586             !(attr->ia_mode & S_ISUID) &&
1587             !(attr->ia_valid & ATTR_KILL_SUID))
1588                 attr->ia_valid |= ATTR_KILL_SUID;
1589
1590         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1591             !(attr->ia_mode & S_ISGID) &&
1592             !(attr->ia_valid & ATTR_KILL_SGID))
1593                 attr->ia_valid |= ATTR_KILL_SGID;
1594
1595         return ll_setattr_raw(de, attr);
1596 }
1597
1598 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1599                        __u64 max_age, __u32 flags)
1600 {
1601         struct ll_sb_info *sbi = ll_s2sbi(sb);
1602         struct obd_statfs obd_osfs;
1603         int rc;
1604         ENTRY;
1605
1606         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1607         if (rc) {
1608                 CERROR("md_statfs fails: rc = %d\n", rc);
1609                 RETURN(rc);
1610         }
1611
1612         osfs->os_type = sb->s_magic;
1613
1614         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1615                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1616
1617         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1618                 flags |= OBD_STATFS_NODELAY;
1619
1620         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1621         if (rc) {
1622                 CERROR("obd_statfs fails: rc = %d\n", rc);
1623                 RETURN(rc);
1624         }
1625
1626         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1627                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1628                obd_osfs.os_files);
1629
1630         osfs->os_bsize = obd_osfs.os_bsize;
1631         osfs->os_blocks = obd_osfs.os_blocks;
1632         osfs->os_bfree = obd_osfs.os_bfree;
1633         osfs->os_bavail = obd_osfs.os_bavail;
1634
1635         /* If we don't have as many objects free on the OST as inodes
1636          * on the MDS, we reduce the total number of inodes to
1637          * compensate, so that the "inodes in use" number is correct.
1638          */
1639         if (obd_osfs.os_ffree < osfs->os_ffree) {
1640                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1641                         obd_osfs.os_ffree;
1642                 osfs->os_ffree = obd_osfs.os_ffree;
1643         }
1644
1645         RETURN(rc);
1646 }
1647 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1648 {
1649         struct super_block *sb = de->d_sb;
1650         struct obd_statfs osfs;
1651         int rc;
1652
1653         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1654         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1655
1656         /* Some amount of caching on the client is allowed */
1657         rc = ll_statfs_internal(sb, &osfs,
1658                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1659                                 0);
1660         if (rc)
1661                 return rc;
1662
1663         statfs_unpack(sfs, &osfs);
1664
1665         /* We need to downshift for all 32-bit kernels, because we can't
1666          * tell if the kernel is being called via sys_statfs64() or not.
1667          * Stop before overflowing f_bsize - in which case it is better
1668          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1669         if (sizeof(long) < 8) {
1670                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1671                         sfs->f_bsize <<= 1;
1672
1673                         osfs.os_blocks >>= 1;
1674                         osfs.os_bfree >>= 1;
1675                         osfs.os_bavail >>= 1;
1676                 }
1677         }
1678
1679         sfs->f_blocks = osfs.os_blocks;
1680         sfs->f_bfree = osfs.os_bfree;
1681         sfs->f_bavail = osfs.os_bavail;
1682
1683         return 0;
1684 }
1685
1686 void ll_inode_size_lock(struct inode *inode)
1687 {
1688         struct ll_inode_info *lli;
1689
1690         LASSERT(!S_ISDIR(inode->i_mode));
1691
1692         lli = ll_i2info(inode);
1693         LASSERT(lli->lli_size_sem_owner != current);
1694         down(&lli->lli_size_sem);
1695         LASSERT(lli->lli_size_sem_owner == NULL);
1696         lli->lli_size_sem_owner = current;
1697 }
1698
1699 void ll_inode_size_unlock(struct inode *inode)
1700 {
1701         struct ll_inode_info *lli;
1702
1703         lli = ll_i2info(inode);
1704         LASSERT(lli->lli_size_sem_owner == current);
1705         lli->lli_size_sem_owner = NULL;
1706         up(&lli->lli_size_sem);
1707 }
1708
1709 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1710 {
1711         struct ll_inode_info *lli = ll_i2info(inode);
1712         struct mdt_body *body = md->body;
1713         struct lov_stripe_md *lsm = md->lsm;
1714         struct ll_sb_info *sbi = ll_i2sbi(inode);
1715
1716         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1717         if (lsm != NULL) {
1718                 if (!lli->lli_has_smd &&
1719                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1720                         cl_file_inode_init(inode, md);
1721
1722                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1723                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1724                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1725         }
1726
1727         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1728                 if (body->valid & OBD_MD_FLRMTPERM)
1729                         ll_update_remote_perm(inode, md->remote_perm);
1730         }
1731 #ifdef CONFIG_FS_POSIX_ACL
1732         else if (body->valid & OBD_MD_FLACL) {
1733                 spin_lock(&lli->lli_lock);
1734                 if (lli->lli_posix_acl)
1735                         posix_acl_release(lli->lli_posix_acl);
1736                 lli->lli_posix_acl = md->posix_acl;
1737                 spin_unlock(&lli->lli_lock);
1738         }
1739 #endif
1740         inode->i_ino = cl_fid_build_ino(&body->fid1, ll_need_32bit_api(sbi));
1741         inode->i_generation = cl_fid_build_gen(&body->fid1);
1742
1743         if (body->valid & OBD_MD_FLATIME) {
1744                 if (body->atime > LTIME_S(inode->i_atime))
1745                         LTIME_S(inode->i_atime) = body->atime;
1746                 lli->lli_lvb.lvb_atime = body->atime;
1747         }
1748         if (body->valid & OBD_MD_FLMTIME) {
1749                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1750                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1751                                "to "LPU64"\n", inode->i_ino,
1752                                LTIME_S(inode->i_mtime), body->mtime);
1753                         LTIME_S(inode->i_mtime) = body->mtime;
1754                 }
1755                 lli->lli_lvb.lvb_mtime = body->mtime;
1756         }
1757         if (body->valid & OBD_MD_FLCTIME) {
1758                 if (body->ctime > LTIME_S(inode->i_ctime))
1759                         LTIME_S(inode->i_ctime) = body->ctime;
1760                 lli->lli_lvb.lvb_ctime = body->ctime;
1761         }
1762         if (body->valid & OBD_MD_FLMODE)
1763                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1764         if (body->valid & OBD_MD_FLTYPE)
1765                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1766         LASSERT(inode->i_mode != 0);
1767         if (S_ISREG(inode->i_mode)) {
1768                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1769         } else {
1770                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1771         }
1772         if (body->valid & OBD_MD_FLUID)
1773                 inode->i_uid = body->uid;
1774         if (body->valid & OBD_MD_FLGID)
1775                 inode->i_gid = body->gid;
1776         if (body->valid & OBD_MD_FLFLAGS)
1777                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1778         if (body->valid & OBD_MD_FLNLINK)
1779                 set_nlink(inode, body->nlink);
1780         if (body->valid & OBD_MD_FLRDEV)
1781                 inode->i_rdev = old_decode_dev(body->rdev);
1782
1783         if (body->valid & OBD_MD_FLID) {
1784                 /* FID shouldn't be changed! */
1785                 if (fid_is_sane(&lli->lli_fid)) {
1786                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1787                                  "Trying to change FID "DFID
1788                                  " to the "DFID", inode %lu/%u(%p)\n",
1789                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1790                                  inode->i_ino, inode->i_generation, inode);
1791                 } else
1792                         lli->lli_fid = body->fid1;
1793         }
1794
1795         LASSERT(fid_seq(&lli->lli_fid) != 0);
1796
1797         if (body->valid & OBD_MD_FLSIZE) {
1798                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1799                     S_ISREG(inode->i_mode)) {
1800                         struct lustre_handle lockh;
1801                         ldlm_mode_t mode;
1802
1803                         /* As it is possible a blocking ast has been processed
1804                          * by this time, we need to check there is an UPDATE
1805                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1806                          * it. */
1807                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1808                                                &lockh, LDLM_FL_CBPENDING);
1809                         if (mode) {
1810                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1811                                                       LLIF_EPOCH_PENDING |
1812                                                       LLIF_SOM_DIRTY)) {
1813                                         CERROR("ino %lu flags %u still has "
1814                                                "size authority! do not trust "
1815                                                "the size got from MDS\n",
1816                                                inode->i_ino, lli->lli_flags);
1817                                 } else {
1818                                         /* Use old size assignment to avoid
1819                                          * deadlock bz14138 & bz14326 */
1820                                         i_size_write(inode, body->size);
1821                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1822                                 }
1823                                 ldlm_lock_decref(&lockh, mode);
1824                         }
1825                 } else {
1826                         /* Use old size assignment to avoid
1827                          * deadlock bz14138 & bz14326 */
1828                         i_size_write(inode, body->size);
1829
1830                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1831                                inode->i_ino, (unsigned long long)body->size);
1832                 }
1833
1834                 if (body->valid & OBD_MD_FLBLOCKS)
1835                         inode->i_blocks = body->blocks;
1836         }
1837
1838         if (body->valid & OBD_MD_FLMDSCAPA) {
1839                 LASSERT(md->mds_capa);
1840                 ll_add_capa(inode, md->mds_capa);
1841         }
1842         if (body->valid & OBD_MD_FLOSSCAPA) {
1843                 LASSERT(md->oss_capa);
1844                 ll_add_capa(inode, md->oss_capa);
1845         }
1846 }
1847
1848 void ll_read_inode2(struct inode *inode, void *opaque)
1849 {
1850         struct lustre_md *md = opaque;
1851         struct ll_inode_info *lli = ll_i2info(inode);
1852         ENTRY;
1853
1854         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1855                PFID(&lli->lli_fid), inode);
1856
1857         LASSERT(!lli->lli_has_smd);
1858
1859         /* Core attributes from the MDS first.  This is a new inode, and
1860          * the VFS doesn't zero times in the core inode so we have to do
1861          * it ourselves.  They will be overwritten by either MDS or OST
1862          * attributes - we just need to make sure they aren't newer. */
1863         LTIME_S(inode->i_mtime) = 0;
1864         LTIME_S(inode->i_atime) = 0;
1865         LTIME_S(inode->i_ctime) = 0;
1866         inode->i_rdev = 0;
1867         ll_update_inode(inode, md);
1868
1869         /* OIDEBUG(inode); */
1870
1871         /* initializing backing dev info. */
1872         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1873
1874
1875         if (S_ISREG(inode->i_mode)) {
1876                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1877                 inode->i_op = &ll_file_inode_operations;
1878                 inode->i_fop = sbi->ll_fop;
1879                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1880                 EXIT;
1881         } else if (S_ISDIR(inode->i_mode)) {
1882                 inode->i_op = &ll_dir_inode_operations;
1883                 inode->i_fop = &ll_dir_operations;
1884                 EXIT;
1885         } else if (S_ISLNK(inode->i_mode)) {
1886                 inode->i_op = &ll_fast_symlink_inode_operations;
1887                 EXIT;
1888         } else {
1889                 inode->i_op = &ll_special_inode_operations;
1890
1891                 init_special_inode(inode, inode->i_mode,
1892                                    inode->i_rdev);
1893
1894                 EXIT;
1895         }
1896 }
1897
1898 void ll_delete_inode(struct inode *inode)
1899 {
1900         struct cl_inode_info *lli = cl_i2info(inode);
1901         ENTRY;
1902
1903         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1904                 /* discard all dirty pages before truncating them, required by
1905                  * osc_extent implementation at LU-1030. */
1906                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1907
1908         truncate_inode_pages(&inode->i_data, 0);
1909
1910         /* Workaround for LU-118 */
1911         if (inode->i_data.nrpages) {
1912                 TREE_READ_LOCK_IRQ(&inode->i_data);
1913                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1914                 LASSERTF(inode->i_data.nrpages == 0,
1915                          "inode=%lu/%u(%p) nrpages=%lu, see "
1916                          "http://jira.whamcloud.com/browse/LU-118\n",
1917                          inode->i_ino, inode->i_generation, inode,
1918                          inode->i_data.nrpages);
1919         }
1920         /* Workaround end */
1921
1922 #ifdef HAVE_SBOPS_EVICT_INODE
1923         ll_clear_inode(inode);
1924 #endif
1925         clear_inode(inode);
1926
1927         EXIT;
1928 }
1929
1930 int ll_iocontrol(struct inode *inode, struct file *file,
1931                  unsigned int cmd, unsigned long arg)
1932 {
1933         struct ll_sb_info *sbi = ll_i2sbi(inode);
1934         struct ptlrpc_request *req = NULL;
1935         int rc, flags = 0;
1936         ENTRY;
1937
1938         switch(cmd) {
1939         case FSFILT_IOC_GETFLAGS: {
1940                 struct mdt_body *body;
1941                 struct md_op_data *op_data;
1942
1943                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1944                                              0, 0, LUSTRE_OPC_ANY,
1945                                              NULL);
1946                 if (IS_ERR(op_data))
1947                         RETURN(PTR_ERR(op_data));
1948
1949                 op_data->op_valid = OBD_MD_FLFLAGS;
1950                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1951                 ll_finish_md_op_data(op_data);
1952                 if (rc) {
1953                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1954                         RETURN(-abs(rc));
1955                 }
1956
1957                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1958
1959                 flags = body->flags;
1960
1961                 ptlrpc_req_finished(req);
1962
1963                 RETURN(put_user(flags, (int *)arg));
1964         }
1965         case FSFILT_IOC_SETFLAGS: {
1966                 struct lov_stripe_md *lsm;
1967                 struct obd_info oinfo = { { { 0 } } };
1968                 struct md_op_data *op_data;
1969
1970                 if (get_user(flags, (int *)arg))
1971                         RETURN(-EFAULT);
1972
1973                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1974                                              LUSTRE_OPC_ANY, NULL);
1975                 if (IS_ERR(op_data))
1976                         RETURN(PTR_ERR(op_data));
1977
1978                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1979                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1980                 rc = md_setattr(sbi->ll_md_exp, op_data,
1981                                 NULL, 0, NULL, 0, &req, NULL);
1982                 ll_finish_md_op_data(op_data);
1983                 ptlrpc_req_finished(req);
1984                 if (rc)
1985                         RETURN(rc);
1986
1987                 inode->i_flags = ll_ext_to_inode_flags(flags);
1988
1989                 lsm = ccc_inode_lsm_get(inode);
1990                 if (lsm == NULL)
1991                         RETURN(0);
1992
1993                 OBDO_ALLOC(oinfo.oi_oa);
1994                 if (!oinfo.oi_oa) {
1995                         ccc_inode_lsm_put(inode, lsm);
1996                         RETURN(-ENOMEM);
1997                 }
1998                 oinfo.oi_md = lsm;
1999                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2000                 oinfo.oi_oa->o_flags = flags;
2001                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2002                                        OBD_MD_FLGROUP;
2003                 oinfo.oi_capa = ll_mdscapa_get(inode);
2004                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2005                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2006                 capa_put(oinfo.oi_capa);
2007                 OBDO_FREE(oinfo.oi_oa);
2008                 ccc_inode_lsm_put(inode, lsm);
2009
2010                 if (rc && rc != -EPERM && rc != -EACCES)
2011                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2012
2013                 RETURN(rc);
2014         }
2015         default:
2016                 RETURN(-ENOSYS);
2017         }
2018
2019         RETURN(0);
2020 }
2021
2022 int ll_flush_ctx(struct inode *inode)
2023 {
2024         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2025
2026         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2027
2028         obd_set_info_async(NULL, sbi->ll_md_exp,
2029                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2030                            0, NULL, NULL);
2031         obd_set_info_async(NULL, sbi->ll_dt_exp,
2032                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2033                            0, NULL, NULL);
2034         return 0;
2035 }
2036
2037 /* umount -f client means force down, don't save state */
2038 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2039 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2040 {
2041         struct super_block *sb = vfsmnt->mnt_sb;
2042 #else
2043 void ll_umount_begin(struct super_block *sb)
2044 {
2045 #endif
2046         struct ll_sb_info *sbi = ll_s2sbi(sb);
2047         struct obd_device *obd;
2048         struct obd_ioctl_data *ioc_data;
2049         ENTRY;
2050
2051 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2052         if (!(flags & MNT_FORCE)) {
2053                 EXIT;
2054                 return;
2055         }
2056 #endif
2057
2058         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2059                sb->s_count, atomic_read(&sb->s_active));
2060
2061         obd = class_exp2obd(sbi->ll_md_exp);
2062         if (obd == NULL) {
2063                 CERROR("Invalid MDC connection handle "LPX64"\n",
2064                        sbi->ll_md_exp->exp_handle.h_cookie);
2065                 EXIT;
2066                 return;
2067         }
2068         obd->obd_force = 1;
2069
2070         obd = class_exp2obd(sbi->ll_dt_exp);
2071         if (obd == NULL) {
2072                 CERROR("Invalid LOV connection handle "LPX64"\n",
2073                        sbi->ll_dt_exp->exp_handle.h_cookie);
2074                 EXIT;
2075                 return;
2076         }
2077         obd->obd_force = 1;
2078
2079         OBD_ALLOC_PTR(ioc_data);
2080         if (ioc_data) {
2081                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2082                               sizeof *ioc_data, ioc_data, NULL);
2083
2084                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2085                               sizeof *ioc_data, ioc_data, NULL);
2086
2087                 OBD_FREE_PTR(ioc_data);
2088         }
2089
2090
2091         /* Really, we'd like to wait until there are no requests outstanding,
2092          * and then continue.  For now, we just invalidate the requests,
2093          * schedule() and sleep one second if needed, and hope.
2094          */
2095         cfs_schedule();
2096 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2097         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2098                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2099                                                    cfs_time_seconds(1));
2100                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2101                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2102                                       "may try to umount it a bit later\n",
2103                                       atomic_read(&vfsmnt->mnt_count));
2104         }
2105 #endif
2106
2107         EXIT;
2108 }
2109
2110 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2111 {
2112         struct ll_sb_info *sbi = ll_s2sbi(sb);
2113         char *profilenm = get_profile_name(sb);
2114         int err;
2115         __u32 read_only;
2116
2117         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2118                 read_only = *flags & MS_RDONLY;
2119                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2120                                          sizeof(KEY_READ_ONLY),
2121                                          KEY_READ_ONLY, sizeof(read_only),
2122                                          &read_only, NULL);
2123                 if (err) {
2124                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2125                                       profilenm, read_only ?
2126                                       "read-only" : "read-write", err);
2127                         return err;
2128                 }
2129
2130                 if (read_only)
2131                         sb->s_flags |= MS_RDONLY;
2132                 else
2133                         sb->s_flags &= ~MS_RDONLY;
2134
2135                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2136                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2137                                       read_only ?  "read-only" : "read-write");
2138         }
2139         return 0;
2140 }
2141
2142 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2143                   struct super_block *sb, struct lookup_intent *it)
2144 {
2145         struct ll_sb_info *sbi = NULL;
2146         struct lustre_md md;
2147         int rc;
2148         ENTRY;
2149
2150         LASSERT(*inode || sb);
2151         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2152         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2153                               sbi->ll_md_exp, &md);
2154         if (rc)
2155                 RETURN(rc);
2156
2157         if (*inode) {
2158                 ll_update_inode(*inode, &md);
2159         } else {
2160                 LASSERT(sb != NULL);
2161
2162                 /*
2163                  * At this point server returns to client's same fid as client
2164                  * generated for creating. So using ->fid1 is okay here.
2165                  */
2166                 LASSERT(fid_is_sane(&md.body->fid1));
2167
2168                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2169                                                       ll_need_32bit_api(sbi)),
2170                                  &md);
2171                 if (*inode == NULL || IS_ERR(*inode)) {
2172 #ifdef CONFIG_FS_POSIX_ACL
2173                         if (md.posix_acl) {
2174                                 posix_acl_release(md.posix_acl);
2175                                 md.posix_acl = NULL;
2176                         }
2177 #endif
2178                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2179                         *inode = NULL;
2180                         CERROR("new_inode -fatal: rc %d\n", rc);
2181                         GOTO(out, rc);
2182                 }
2183         }
2184
2185         /* Handling piggyback layout lock.
2186          * Layout lock can be piggybacked by getattr and open request.
2187          * The lsm can be applied to inode only if it comes with a layout lock
2188          * otherwise correct layout may be overwritten, for example:
2189          * 1. proc1: mdt returns a lsm but not granting layout
2190          * 2. layout was changed by another client
2191          * 3. proc2: refresh layout and layout lock granted
2192          * 4. proc1: to apply a stale layout */
2193         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2194                 struct lustre_handle lockh;
2195                 struct ldlm_lock *lock;
2196
2197                 lockh.cookie = it->d.lustre.it_lock_handle;
2198                 lock = ldlm_handle2lock(&lockh);
2199                 LASSERT(lock != NULL);
2200                 if (ldlm_has_layout(lock)) {
2201                         struct cl_object_conf conf;
2202
2203                         memset(&conf, 0, sizeof(conf));
2204                         conf.coc_opc = OBJECT_CONF_SET;
2205                         conf.coc_inode = *inode;
2206                         conf.coc_lock = lock;
2207                         conf.u.coc_md = &md;
2208                         (void)ll_layout_conf(*inode, &conf);
2209                 }
2210                 LDLM_LOCK_PUT(lock);
2211         }
2212
2213 out:
2214         if (md.lsm != NULL)
2215                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2216         md_free_lustre_md(sbi->ll_md_exp, &md);
2217         RETURN(rc);
2218 }
2219
2220 int ll_obd_statfs(struct inode *inode, void *arg)
2221 {
2222         struct ll_sb_info *sbi = NULL;
2223         struct obd_export *exp;
2224         char *buf = NULL;
2225         struct obd_ioctl_data *data = NULL;
2226         __u32 type;
2227         __u32 flags;
2228         int len = 0, rc;
2229
2230         if (!inode || !(sbi = ll_i2sbi(inode)))
2231                 GOTO(out_statfs, rc = -EINVAL);
2232
2233         rc = obd_ioctl_getdata(&buf, &len, arg);
2234         if (rc)
2235                 GOTO(out_statfs, rc);
2236
2237         data = (void*)buf;
2238         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2239             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2240                 GOTO(out_statfs, rc = -EINVAL);
2241
2242         if (data->ioc_inllen1 != sizeof(__u32) ||
2243             data->ioc_inllen2 != sizeof(__u32) ||
2244             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2245             data->ioc_plen2 != sizeof(struct obd_uuid))
2246                 GOTO(out_statfs, rc = -EINVAL);
2247
2248         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2249         if (type & LL_STATFS_LMV)
2250                 exp = sbi->ll_md_exp;
2251         else if (type & LL_STATFS_LOV)
2252                 exp = sbi->ll_dt_exp;
2253         else
2254                 GOTO(out_statfs, rc = -ENODEV);
2255
2256         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2257         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2258         if (rc)
2259                 GOTO(out_statfs, rc);
2260 out_statfs:
2261         if (buf)
2262                 obd_ioctl_freedata(buf, len);
2263         return rc;
2264 }
2265
2266 int ll_process_config(struct lustre_cfg *lcfg)
2267 {
2268         char *ptr;
2269         void *sb;
2270         struct lprocfs_static_vars lvars;
2271         unsigned long x;
2272         int rc = 0;
2273
2274         lprocfs_llite_init_vars(&lvars);
2275
2276         /* The instance name contains the sb: lustre-client-aacfe000 */
2277         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2278         if (!ptr || !*(++ptr))
2279                 return -EINVAL;
2280         if (sscanf(ptr, "%lx", &x) != 1)
2281                 return -EINVAL;
2282         sb = (void *)x;
2283         /* This better be a real Lustre superblock! */
2284         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2285
2286         /* Note we have not called client_common_fill_super yet, so
2287            proc fns must be able to handle that! */
2288         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2289                                       lcfg, sb);
2290         if (rc > 0)
2291                 rc = 0;
2292         return(rc);
2293 }
2294
2295 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2296 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2297                                        struct inode *i1, struct inode *i2,
2298                                        const char *name, int namelen,
2299                                        int mode, __u32 opc, void *data)
2300 {
2301         LASSERT(i1 != NULL);
2302
2303         if (namelen > ll_i2sbi(i1)->ll_namelen)
2304                 return ERR_PTR(-ENAMETOOLONG);
2305
2306         if (op_data == NULL)
2307                 OBD_ALLOC_PTR(op_data);
2308
2309         if (op_data == NULL)
2310                 return ERR_PTR(-ENOMEM);
2311
2312         ll_i2gids(op_data->op_suppgids, i1, i2);
2313         op_data->op_fid1 = *ll_inode2fid(i1);
2314         op_data->op_capa1 = ll_mdscapa_get(i1);
2315
2316         if (i2) {
2317                 op_data->op_fid2 = *ll_inode2fid(i2);
2318                 op_data->op_capa2 = ll_mdscapa_get(i2);
2319         } else {
2320                 fid_zero(&op_data->op_fid2);
2321                 op_data->op_capa2 = NULL;
2322         }
2323
2324         op_data->op_name = name;
2325         op_data->op_namelen = namelen;
2326         op_data->op_mode = mode;
2327         op_data->op_mod_time = cfs_time_current_sec();
2328         op_data->op_fsuid = cfs_curproc_fsuid();
2329         op_data->op_fsgid = cfs_curproc_fsgid();
2330         op_data->op_cap = cfs_curproc_cap_pack();
2331         op_data->op_bias = 0;
2332         op_data->op_cli_flags = 0;
2333         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2334              filename_is_volatile(name, namelen, NULL))
2335                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2336         op_data->op_opc = opc;
2337         op_data->op_mds = 0;
2338         op_data->op_data = data;
2339
2340         /* If the file is being opened after mknod() (normally due to NFS)
2341          * try to use the default stripe data from parent directory for
2342          * allocating OST objects.  Try to pass the parent FID to MDS. */
2343         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2344             !ll_i2info(i2)->lli_has_smd) {
2345                 struct ll_inode_info *lli = ll_i2info(i2);
2346
2347                 spin_lock(&lli->lli_lock);
2348                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2349                         op_data->op_fid1 = lli->lli_pfid;
2350                 spin_unlock(&lli->lli_lock);
2351                 /** We ignore parent's capability temporary. */
2352         }
2353
2354         /* When called by ll_setattr_raw, file is i1. */
2355         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2356                 op_data->op_bias |= MDS_DATA_MODIFIED;
2357
2358         return op_data;
2359 }
2360
2361 void ll_finish_md_op_data(struct md_op_data *op_data)
2362 {
2363         capa_put(op_data->op_capa1);
2364         capa_put(op_data->op_capa2);
2365         OBD_FREE_PTR(op_data);
2366 }
2367
2368 #ifdef HAVE_SUPEROPS_USE_DENTRY
2369 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2370 #else
2371 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2372 #endif
2373 {
2374         struct ll_sb_info *sbi;
2375
2376 #ifdef HAVE_SUPEROPS_USE_DENTRY
2377         LASSERT((seq != NULL) && (dentry != NULL));
2378         sbi = ll_s2sbi(dentry->d_sb);
2379 #else
2380         LASSERT((seq != NULL) && (vfs != NULL));
2381         sbi = ll_s2sbi(vfs->mnt_sb);
2382 #endif
2383
2384         if (sbi->ll_flags & LL_SBI_NOLCK)
2385                 seq_puts(seq, ",nolock");
2386
2387         if (sbi->ll_flags & LL_SBI_FLOCK)
2388                 seq_puts(seq, ",flock");
2389
2390         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2391                 seq_puts(seq, ",localflock");
2392
2393         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2394                 seq_puts(seq, ",user_xattr");
2395
2396         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2397                 seq_puts(seq, ",lazystatfs");
2398
2399         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2400                 seq_puts(seq, ",user_fid2path");
2401
2402         RETURN(0);
2403 }
2404
2405 /**
2406  * Get obd name by cmd, and copy out to user space
2407  */
2408 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2409 {
2410         struct ll_sb_info *sbi = ll_i2sbi(inode);
2411         struct obd_device *obd;
2412         ENTRY;
2413
2414         if (cmd == OBD_IOC_GETDTNAME)
2415                 obd = class_exp2obd(sbi->ll_dt_exp);
2416         else if (cmd == OBD_IOC_GETMDNAME)
2417                 obd = class_exp2obd(sbi->ll_md_exp);
2418         else
2419                 RETURN(-EINVAL);
2420
2421         if (!obd)
2422                 RETURN(-ENOENT);
2423
2424         if (cfs_copy_to_user((void *)arg, obd->obd_name,
2425                              strlen(obd->obd_name) + 1))
2426                 RETURN(-EFAULT);
2427
2428         RETURN(0);
2429 }
2430
2431 /**
2432  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2433  * fsname will be returned in this buffer; otherwise, a static buffer will be
2434  * used to store the fsname and returned to caller.
2435  */
2436 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2437 {
2438         static char fsname_static[MTI_NAME_MAXLEN];
2439         struct lustre_sb_info *lsi = s2lsi(sb);
2440         char *ptr;
2441         int len;
2442
2443         if (buf == NULL) {
2444                 /* this means the caller wants to use static buffer
2445                  * and it doesn't care about race. Usually this is
2446                  * in error reporting path */
2447                 buf = fsname_static;
2448                 buflen = sizeof(fsname_static);
2449         }
2450
2451         len = strlen(lsi->lsi_lmd->lmd_profile);
2452         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2453         if (ptr && (strcmp(ptr, "-client") == 0))
2454                 len -= 7;
2455
2456         if (unlikely(len >= buflen))
2457                 len = buflen - 1;
2458         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2459         buf[len] = '\0';
2460
2461         return buf;
2462 }
2463
2464 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2465 {
2466         char *path = NULL;
2467
2468 #ifdef HAVE_FS_STRUCT_USE_PATH
2469         struct path p;
2470
2471         p.dentry = dentry;
2472         p.mnt = current->fs->root.mnt;
2473         path_get(&p);
2474         path = d_path(&p, buf, bufsize);
2475         path_put(&p);
2476 #else
2477         path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2478 #endif
2479
2480         return path;
2481 }
2482
2483 void ll_dirty_page_discard_warn(cfs_page_t *page, int ioret)
2484 {
2485         char *buf, *path = NULL;
2486         struct dentry *dentry = NULL;
2487         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2488
2489         /* this can be called inside spin lock so use GFP_ATOMIC. */
2490         buf = (char *)__get_free_page(GFP_ATOMIC);
2491         if (buf != NULL) {
2492                 dentry = d_find_alias(page->mapping->host);
2493                 if (dentry != NULL)
2494                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2495         }
2496
2497         CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2498               "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2499               s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2500               PFID(&obj->cob_header.coh_lu.loh_fid),
2501               (path && !IS_ERR(path)) ? path : "", ioret);
2502
2503         if (dentry != NULL)
2504                 dput(dentry);
2505
2506         if (buf != NULL)
2507                 free_page((unsigned long)buf);
2508 }