Whamcloud - gitweb
LU-1866 lfsck: enhance otable-based iteration
[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;
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         /* For mount, we only need fs info from MDT0, and also in DNE, it
284          * can make sure the client can be mounted as long as MDT0 is
285          * avaible */
286         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
287                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
288                         OBD_STATFS_FOR_MDT0);
289         if (err)
290                 GOTO(out_md, err);
291
292         /* This needs to be after statfs to ensure connect has finished.
293          * Note that "data" does NOT contain the valid connect reply.
294          * If connecting to a 1.8 server there will be no LMV device, so
295          * we can access the MDC export directly and exp_connect_flags will
296          * be non-zero, but if accessing an upgraded 2.1 server it will
297          * have the correct flags filled in.
298          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
299         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
300         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
301             valid != CLIENT_CONNECT_MDT_REQD) {
302                 char *buf;
303
304                 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
305                 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
306                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
307                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
308                                    "feature(s) needed for correct operation "
309                                    "of this client (%s). Please upgrade "
310                                    "server or downgrade client.\n",
311                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
312                 OBD_FREE(buf, CFS_PAGE_SIZE);
313                 GOTO(out_md, err = -EPROTO);
314         }
315
316         size = sizeof(*data);
317         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
318                            KEY_CONN_DATA,  &size, data, NULL);
319         if (err) {
320                 CERROR("%s: Get connect data failed: rc = %d\n",
321                        sbi->ll_md_exp->exp_obd->obd_name, err);
322                 GOTO(out_md, err);
323         }
324
325         LASSERT(osfs->os_bsize);
326         sb->s_blocksize = osfs->os_bsize;
327         sb->s_blocksize_bits = log2(osfs->os_bsize);
328         sb->s_magic = LL_SUPER_MAGIC;
329         sb->s_maxbytes = MAX_LFS_FILESIZE;
330         sbi->ll_namelen = osfs->os_namelen;
331         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
332
333         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
334             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
335                 LCONSOLE_INFO("Disabling user_xattr feature because "
336                               "it is not supported on the server\n");
337                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
338         }
339
340         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
341 #ifdef MS_POSIXACL
342                 sb->s_flags |= MS_POSIXACL;
343 #endif
344                 sbi->ll_flags |= LL_SBI_ACL;
345         } else {
346                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
347 #ifdef MS_POSIXACL
348                 sb->s_flags &= ~MS_POSIXACL;
349 #endif
350                 sbi->ll_flags &= ~LL_SBI_ACL;
351         }
352
353         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
354                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
355                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
356                         LCONSOLE_INFO("client is set as remote by default.\n");
357                 }
358         } else {
359                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
360                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
361                         LCONSOLE_INFO("client claims to be remote, but server "
362                                       "rejected, forced to be local.\n");
363                 }
364         }
365
366         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
367                 LCONSOLE_INFO("client enabled MDS capability!\n");
368                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
369         }
370
371         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
372                 LCONSOLE_INFO("client enabled OSS capability!\n");
373                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
374         }
375
376         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
377                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
378
379         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
380                 sbi->ll_md_brw_size = data->ocd_brw_size;
381         else
382                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
383
384         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
385                 LCONSOLE_INFO("Layout lock feature supported.\n");
386                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
387         }
388
389         obd = class_name2obd(dt);
390         if (!obd) {
391                 CERROR("DT %s: not setup or attached\n", dt);
392                 GOTO(out_md, err = -ENODEV);
393         }
394
395         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
396                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
397                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
398                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
399                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
400                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
401                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
402                                   OBD_CONNECT_MAXBYTES |
403                                   OBD_CONNECT_EINPROGRESS |
404                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
405                                   OBD_CONNECT_LAYOUTLOCK;
406
407         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
408                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
409
410         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
411                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
412                  * disabled by default, because it can still be enabled on the
413                  * fly via /proc. As a consequence, we still need to come to an
414                  * agreement on the supported algorithms at connect time */
415                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
416
417                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
418                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
419                 else
420                         data->ocd_cksum_types = cksum_types_supported_client();
421         }
422
423 #ifdef HAVE_LRU_RESIZE_SUPPORT
424         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
425 #endif
426         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
427                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
428
429         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
430                "ocd_grant: %d\n", data->ocd_connect_flags,
431                data->ocd_version, data->ocd_grant);
432
433         obd->obd_upcall.onu_owner = &sbi->ll_lco;
434         obd->obd_upcall.onu_upcall = cl_ocd_update;
435
436         data->ocd_brw_size = DT_MAX_BRW_SIZE;
437
438         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
439                           NULL);
440         if (err == -EBUSY) {
441                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
442                                    "recovery, of which this client is not a "
443                                    "part.  Please wait for recovery to "
444                                    "complete, abort, or time out.\n", dt);
445                 GOTO(out_md, err);
446         } else if (err) {
447                 CERROR("%s: Cannot connect to %s: rc = %d\n",
448                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
449                 GOTO(out_md, err);
450         }
451
452         sbi->ll_dt_exp->exp_connect_data = *data;
453
454         mutex_lock(&sbi->ll_lco.lco_lock);
455         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
456         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
457         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
458         mutex_unlock(&sbi->ll_lco.lco_lock);
459
460         fid_zero(&sbi->ll_root_fid);
461         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
462         if (err) {
463                 CERROR("cannot mds_connect: rc = %d\n", err);
464                 GOTO(out_dt, err);
465         }
466         if (!fid_is_sane(&sbi->ll_root_fid)) {
467                 CERROR("%s: Invalid root fid "DFID" during mount\n",
468                        sbi->ll_md_exp->exp_obd->obd_name,
469                        PFID(&sbi->ll_root_fid));
470                 GOTO(out_dt, err = -EINVAL);
471         }
472         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
473
474         sb->s_op = &lustre_super_operations;
475 #if THREAD_SIZE >= 8192 /*b=17630*/
476         sb->s_export_op = &lustre_export_operations;
477 #endif
478
479         /* make root inode
480          * XXX: move this to after cbd setup? */
481         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
482         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
483                 valid |= OBD_MD_FLRMTPERM;
484         else if (sbi->ll_flags & LL_SBI_ACL)
485                 valid |= OBD_MD_FLACL;
486
487         OBD_ALLOC_PTR(op_data);
488         if (op_data == NULL)
489                 GOTO(out_dt, err = -ENOMEM);
490
491         op_data->op_fid1 = sbi->ll_root_fid;
492         op_data->op_mode = 0;
493         op_data->op_capa1 = oc;
494         op_data->op_valid = valid;
495
496         err = md_getattr(sbi->ll_md_exp, op_data, &request);
497         if (oc)
498                 capa_put(oc);
499         OBD_FREE_PTR(op_data);
500         if (err) {
501                 CERROR("%s: md_getattr failed for root: rc = %d\n",
502                        sbi->ll_md_exp->exp_obd->obd_name, err);
503                 GOTO(out_dt, err);
504         }
505
506         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
507                                sbi->ll_md_exp, &lmd);
508         if (err) {
509                 CERROR("failed to understand root inode md: rc = %d\n", err);
510                 ptlrpc_req_finished(request);
511                 GOTO(out_dt, err);
512         }
513
514         LASSERT(fid_is_sane(&sbi->ll_root_fid));
515         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
516         md_free_lustre_md(sbi->ll_md_exp, &lmd);
517         ptlrpc_req_finished(request);
518
519         if (root == NULL || IS_ERR(root)) {
520                 if (lmd.lsm)
521                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
522 #ifdef CONFIG_FS_POSIX_ACL
523                 if (lmd.posix_acl) {
524                         posix_acl_release(lmd.posix_acl);
525                         lmd.posix_acl = NULL;
526                 }
527 #endif
528                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
529                 root = NULL;
530                 CERROR("lustre_lite: bad iget4 for root\n");
531                 GOTO(out_root, err);
532         }
533
534         err = ll_close_thread_start(&sbi->ll_lcq);
535         if (err) {
536                 CERROR("cannot start close thread: rc %d\n", err);
537                 GOTO(out_root, err);
538         }
539
540 #ifdef CONFIG_FS_POSIX_ACL
541         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
542                 rct_init(&sbi->ll_rct);
543                 et_init(&sbi->ll_et);
544         }
545 #endif
546
547         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
548         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
549                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
550                                  NULL);
551         cl_sb_init(sb);
552
553         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
554                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
555                                  &sbi->ll_cache, NULL);
556
557         sb->s_root = d_make_root(root);
558         if (sb->s_root == NULL) {
559                 CERROR("%s: can't make root dentry\n",
560                         ll_get_fsname(sb, NULL, 0));
561                 GOTO(out_root, err = -ENOMEM);
562         }
563
564 #ifdef HAVE_DCACHE_LOCK
565         sb->s_root->d_op = &ll_d_root_ops;
566 #else
567         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
568         d_set_d_op(sb->s_root, &ll_d_root_ops);
569         sb->s_d_op = &ll_d_ops;
570 #endif
571
572         sbi->ll_sdev_orig = sb->s_dev;
573
574         /* We set sb->s_dev equal on all lustre clients in order to support
575          * NFS export clustering.  NFSD requires that the FSID be the same
576          * on all clients. */
577         /* s_dev is also used in lt_compare() to compare two fs, but that is
578          * only a node-local comparison. */
579         uuid = obd_get_uuid(sbi->ll_md_exp);
580         if (uuid != NULL)
581                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
582
583         if (data != NULL)
584                 OBD_FREE_PTR(data);
585         if (osfs != NULL)
586                 OBD_FREE_PTR(osfs);
587
588         RETURN(err);
589 out_root:
590         if (root)
591                 iput(root);
592 out_dt:
593         obd_disconnect(sbi->ll_dt_exp);
594         sbi->ll_dt_exp = NULL;
595         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
596         obd_zombie_barrier();
597 out_md:
598         obd_disconnect(sbi->ll_md_exp);
599         sbi->ll_md_exp = NULL;
600 out:
601         if (data != NULL)
602                 OBD_FREE_PTR(data);
603         if (osfs != NULL)
604                 OBD_FREE_PTR(osfs);
605         lprocfs_unregister_mountpoint(sbi);
606         return err;
607 }
608
609 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
610 {
611         int size, rc;
612
613         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
614         size = sizeof(int);
615         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
616                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
617         if (rc)
618                 CERROR("Get max mdsize error rc %d \n", rc);
619
620         RETURN(rc);
621 }
622
623 void ll_dump_inode(struct inode *inode)
624 {
625         struct ll_d_hlist_node *tmp;
626         int dentry_count = 0;
627
628         LASSERT(inode != NULL);
629
630         ll_d_hlist_for_each(tmp, &inode->i_dentry)
631                 dentry_count++;
632
633         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
634                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
635                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
636 }
637
638 void lustre_dump_dentry(struct dentry *dentry, int recur)
639 {
640         struct list_head *tmp;
641         int subdirs = 0;
642
643         LASSERT(dentry != NULL);
644
645         list_for_each(tmp, &dentry->d_subdirs)
646                 subdirs++;
647
648         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
649                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
650                dentry->d_name.len, dentry->d_name.name,
651                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
652                dentry->d_parent, dentry->d_inode, d_refcount(dentry),
653                dentry->d_flags, dentry->d_fsdata, subdirs);
654         if (dentry->d_inode != NULL)
655                 ll_dump_inode(dentry->d_inode);
656
657         if (recur == 0)
658                 return;
659
660         list_for_each(tmp, &dentry->d_subdirs) {
661                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
662                 lustre_dump_dentry(d, recur - 1);
663         }
664 }
665
666 void client_common_put_super(struct super_block *sb)
667 {
668         struct ll_sb_info *sbi = ll_s2sbi(sb);
669         ENTRY;
670
671 #ifdef CONFIG_FS_POSIX_ACL
672         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
673                 et_fini(&sbi->ll_et);
674                 rct_fini(&sbi->ll_rct);
675         }
676 #endif
677
678         ll_close_thread_shutdown(sbi->ll_lcq);
679
680         cl_sb_fini(sb);
681
682         cfs_list_del(&sbi->ll_conn_chain);
683
684         obd_disconnect(sbi->ll_dt_exp);
685         sbi->ll_dt_exp = NULL;
686         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
687          * see LU-2543. */
688         obd_zombie_barrier();
689
690         lprocfs_unregister_mountpoint(sbi);
691
692         obd_disconnect(sbi->ll_md_exp);
693         sbi->ll_md_exp = NULL;
694
695         EXIT;
696 }
697
698 void ll_kill_super(struct super_block *sb)
699 {
700         struct ll_sb_info *sbi;
701
702         ENTRY;
703
704         /* not init sb ?*/
705         if (!(sb->s_flags & MS_ACTIVE))
706                 return;
707
708         sbi = ll_s2sbi(sb);
709         /* we need restore s_dev from changed for clustred NFS before put_super
710          * because new kernels have cached s_dev and change sb->s_dev in
711          * put_super not affected real removing devices */
712         if (sbi)
713                 sb->s_dev = sbi->ll_sdev_orig;
714         EXIT;
715 }
716
717 char *ll_read_opt(const char *opt, char *data)
718 {
719         char *value;
720         char *retval;
721         ENTRY;
722
723         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
724         if (strncmp(opt, data, strlen(opt)))
725                 RETURN(NULL);
726         if ((value = strchr(data, '=')) == NULL)
727                 RETURN(NULL);
728
729         value++;
730         OBD_ALLOC(retval, strlen(value) + 1);
731         if (!retval) {
732                 CERROR("out of memory!\n");
733                 RETURN(NULL);
734         }
735
736         memcpy(retval, value, strlen(value)+1);
737         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
738         RETURN(retval);
739 }
740
741 static inline int ll_set_opt(const char *opt, char *data, int fl)
742 {
743         if (strncmp(opt, data, strlen(opt)) != 0)
744                 return(0);
745         else
746                 return(fl);
747 }
748
749 /* non-client-specific mount options are parsed in lmd_parse */
750 static int ll_options(char *options, int *flags)
751 {
752         int tmp;
753         char *s1 = options, *s2;
754         ENTRY;
755
756         if (!options)
757                 RETURN(0);
758
759         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
760
761         while (*s1) {
762                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
763                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
764                 if (tmp) {
765                         *flags |= tmp;
766                         goto next;
767                 }
768                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
769                 if (tmp) {
770                         *flags |= tmp;
771                         goto next;
772                 }
773                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
774                 if (tmp) {
775                         *flags |= tmp;
776                         goto next;
777                 }
778                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
779                 if (tmp) {
780                         *flags &= ~tmp;
781                         goto next;
782                 }
783                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
784                 if (tmp) {
785                         *flags |= tmp;
786                         goto next;
787                 }
788                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
789                 if (tmp) {
790                         *flags &= ~tmp;
791                         goto next;
792                 }
793 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
794                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
795                 if (tmp) {
796                         /* Ignore deprecated mount option.  The client will
797                          * always try to mount with ACL support, whether this
798                          * is used depends on whether server supports it. */
799                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
800                                                   "mount option 'acl'.\n");
801                         goto next;
802                 }
803                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
804                 if (tmp) {
805                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
806                                                   "mount option 'noacl'.\n");
807                         goto next;
808                 }
809 #else
810 #warning "{no}acl options have been deprecated since 1.8, please remove them"
811 #endif
812                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
813                 if (tmp) {
814                         *flags |= tmp;
815                         goto next;
816                 }
817                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
818                 if (tmp) {
819                         *flags |= tmp;
820                         goto next;
821                 }
822                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
823                 if (tmp) {
824                         *flags &= ~tmp;
825                         goto next;
826                 }
827
828                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
829                 if (tmp) {
830                         *flags |= tmp;
831                         goto next;
832                 }
833                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
834                 if (tmp) {
835                         *flags &= ~tmp;
836                         goto next;
837                 }
838                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
839                 if (tmp) {
840                         *flags |= tmp;
841                         goto next;
842                 }
843                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
844                 if (tmp) {
845                         *flags &= ~tmp;
846                         goto next;
847                 }
848                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
849                 if (tmp) {
850                         *flags |= tmp;
851                         goto next;
852                 }
853                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
854                 if (tmp) {
855                         *flags &= ~tmp;
856                         goto next;
857                 }
858                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
859                 if (tmp) {
860                         *flags |= tmp;
861                         goto next;
862                 }
863                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
864                 if (tmp) {
865                         *flags |= tmp;
866                         goto next;
867                 }
868                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
869                 if (tmp) {
870                         *flags |= tmp;
871                         goto next;
872                 }
873                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
874                 if (tmp) {
875                         *flags &= ~tmp;
876                         goto next;
877                 }
878                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
879                                    s1);
880                 RETURN(-EINVAL);
881
882 next:
883                 /* Find next opt */
884                 s2 = strchr(s1, ',');
885                 if (s2 == NULL)
886                         break;
887                 s1 = s2 + 1;
888         }
889         RETURN(0);
890 }
891
892 void ll_lli_init(struct ll_inode_info *lli)
893 {
894         lli->lli_inode_magic = LLI_INODE_MAGIC;
895         lli->lli_flags = 0;
896         lli->lli_ioepoch = 0;
897         lli->lli_maxbytes = MAX_LFS_FILESIZE;
898         spin_lock_init(&lli->lli_lock);
899         lli->lli_posix_acl = NULL;
900         lli->lli_remote_perms = NULL;
901         mutex_init(&lli->lli_rmtperm_mutex);
902         /* Do not set lli_fid, it has been initialized already. */
903         fid_zero(&lli->lli_pfid);
904         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
905         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
906         cfs_atomic_set(&lli->lli_open_count, 0);
907         lli->lli_mds_capa = NULL;
908         lli->lli_rmtperm_time = 0;
909         lli->lli_pending_och = NULL;
910         lli->lli_mds_read_och = NULL;
911         lli->lli_mds_write_och = NULL;
912         lli->lli_mds_exec_och = NULL;
913         lli->lli_open_fd_read_count = 0;
914         lli->lli_open_fd_write_count = 0;
915         lli->lli_open_fd_exec_count = 0;
916         mutex_init(&lli->lli_och_mutex);
917         spin_lock_init(&lli->lli_agl_lock);
918         lli->lli_has_smd = false;
919         lli->lli_layout_gen = LL_LAYOUT_GEN_ZERO;
920         lli->lli_clob = NULL;
921
922         LASSERT(lli->lli_vfs_inode.i_mode != 0);
923         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
924                 mutex_init(&lli->lli_readdir_mutex);
925                 lli->lli_opendir_key = NULL;
926                 lli->lli_sai = NULL;
927                 lli->lli_def_acl = NULL;
928                 spin_lock_init(&lli->lli_sa_lock);
929                 lli->lli_opendir_pid = 0;
930         } else {
931                 sema_init(&lli->lli_size_sem, 1);
932                 lli->lli_size_sem_owner = NULL;
933                 lli->lli_symlink_name = NULL;
934                 init_rwsem(&lli->lli_trunc_sem);
935                 mutex_init(&lli->lli_write_mutex);
936                 init_rwsem(&lli->lli_glimpse_sem);
937                 lli->lli_glimpse_time = 0;
938                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
939                 lli->lli_agl_index = 0;
940                 lli->lli_async_rc = 0;
941                 lli->lli_volatile = false;
942         }
943         mutex_init(&lli->lli_layout_mutex);
944 }
945
946 static inline int ll_bdi_register(struct backing_dev_info *bdi)
947 {
948 #ifdef HAVE_BDI_REGISTER
949         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
950
951 #ifdef HAVE_BDI_NAME
952         bdi->name = "lustre";
953 #endif
954         return bdi_register(bdi, NULL, "lustre-%d",
955                             atomic_inc_return(&ll_bdi_num));
956 #else
957         return 0;
958 #endif
959 }
960
961 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
962 {
963         struct lustre_profile *lprof = NULL;
964         struct lustre_sb_info *lsi = s2lsi(sb);
965         struct ll_sb_info *sbi;
966         char  *dt = NULL, *md = NULL;
967         char  *profilenm = get_profile_name(sb);
968         struct config_llog_instance *cfg;
969         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
970         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
971         int    err;
972         ENTRY;
973
974         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
975
976         OBD_ALLOC_PTR(cfg);
977         if (cfg == NULL)
978                 RETURN(-ENOMEM);
979
980         cfs_module_get();
981
982         /* client additional sb info */
983         lsi->lsi_llsbi = sbi = ll_init_sbi();
984         if (!sbi) {
985                 cfs_module_put(THIS_MODULE);
986                 OBD_FREE_PTR(cfg);
987                 RETURN(-ENOMEM);
988         }
989
990         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
991         if (err)
992                 GOTO(out_free, err);
993
994         err = bdi_init(&lsi->lsi_bdi);
995         if (err)
996                 GOTO(out_free, err);
997         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
998         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
999         err = ll_bdi_register(&lsi->lsi_bdi);
1000         if (err)
1001                 GOTO(out_free, err);
1002
1003 #ifdef HAVE_SB_BDI
1004         sb->s_bdi = &lsi->lsi_bdi;
1005 #endif
1006
1007         /* Generate a string unique to this super, in case some joker tries
1008            to mount the same fs at two mount points.
1009            Use the address of the super itself.*/
1010         cfg->cfg_instance = sb;
1011         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1012         cfg->cfg_callback = class_config_llog_handler;
1013         /* set up client obds */
1014         err = lustre_process_log(sb, profilenm, cfg);
1015         if (err < 0) {
1016                 CERROR("Unable to process log: %d\n", err);
1017                 GOTO(out_free, err);
1018         }
1019
1020         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1021         lprof = class_get_profile(profilenm);
1022         if (lprof == NULL) {
1023                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1024                                    " read from the MGS.  Does that filesystem "
1025                                    "exist?\n", profilenm);
1026                 GOTO(out_free, err = -EINVAL);
1027         }
1028         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1029                lprof->lp_md, lprof->lp_dt);
1030
1031         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1032         if (!dt)
1033                 GOTO(out_free, err = -ENOMEM);
1034         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1035
1036         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1037         if (!md)
1038                 GOTO(out_free, err = -ENOMEM);
1039         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1040
1041         /* connections, registrations, sb setup */
1042         err = client_common_fill_super(sb, md, dt, mnt);
1043
1044 out_free:
1045         if (md)
1046                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1047         if (dt)
1048                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1049         if (err)
1050                 ll_put_super(sb);
1051         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1052                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1053
1054         OBD_FREE_PTR(cfg);
1055         RETURN(err);
1056 } /* ll_fill_super */
1057
1058
1059 void lu_context_keys_dump(void);
1060
1061 void ll_put_super(struct super_block *sb)
1062 {
1063         struct config_llog_instance cfg;
1064         struct obd_device *obd;
1065         struct lustre_sb_info *lsi = s2lsi(sb);
1066         struct ll_sb_info *sbi = ll_s2sbi(sb);
1067         char *profilenm = get_profile_name(sb);
1068         int force = 1, next;
1069         ENTRY;
1070
1071         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1072
1073         ll_print_capa_stat(sbi);
1074
1075         cfg.cfg_instance = sb;
1076         lustre_end_log(sb, profilenm, &cfg);
1077
1078         if (sbi->ll_md_exp) {
1079                 obd = class_exp2obd(sbi->ll_md_exp);
1080                 if (obd)
1081                         force = obd->obd_force;
1082         }
1083
1084         /* We need to set force before the lov_disconnect in
1085            lustre_common_put_super, since l_d cleans up osc's as well. */
1086         if (force) {
1087                 next = 0;
1088                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1089                                                      &next)) != NULL) {
1090                         obd->obd_force = force;
1091                 }
1092         }
1093
1094         if (sbi->ll_lcq) {
1095                 /* Only if client_common_fill_super succeeded */
1096                 client_common_put_super(sb);
1097         }
1098
1099         next = 0;
1100         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1101                 class_manual_cleanup(obd);
1102         }
1103
1104         if (sbi->ll_flags & LL_SBI_VERBOSE)
1105                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1106
1107         if (profilenm)
1108                 class_del_profile(profilenm);
1109
1110         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1111                 bdi_destroy(&lsi->lsi_bdi);
1112                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1113         }
1114
1115         ll_free_sbi(sb);
1116         lsi->lsi_llsbi = NULL;
1117
1118         lustre_common_put_super(sb);
1119
1120         cl_env_cache_purge(~0);
1121
1122         cfs_module_put(THIS_MODULE);
1123
1124         EXIT;
1125 } /* client_put_super */
1126
1127 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1128 {
1129         struct inode *inode = NULL;
1130         /* NOTE: we depend on atomic igrab() -bzzz */
1131         lock_res_and_lock(lock);
1132         if (lock->l_ast_data) {
1133                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1134                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1135                         inode = igrab(lock->l_ast_data);
1136                 } else {
1137                         inode = lock->l_ast_data;
1138                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1139                                          D_WARNING, lock, "l_ast_data %p is "
1140                                          "bogus: magic %08x", lock->l_ast_data,
1141                                          lli->lli_inode_magic);
1142                         inode = NULL;
1143                 }
1144         }
1145         unlock_res_and_lock(lock);
1146         return inode;
1147 }
1148
1149 static int null_if_equal(struct ldlm_lock *lock, void *data)
1150 {
1151         if (data == lock->l_ast_data) {
1152                 lock->l_ast_data = NULL;
1153
1154                 if (lock->l_req_mode != lock->l_granted_mode)
1155                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1156         }
1157
1158         return LDLM_ITER_CONTINUE;
1159 }
1160
1161 void ll_clear_inode(struct inode *inode)
1162 {
1163         struct ll_inode_info *lli = ll_i2info(inode);
1164         struct ll_sb_info *sbi = ll_i2sbi(inode);
1165         ENTRY;
1166
1167         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1168                inode->i_generation, inode);
1169
1170         if (S_ISDIR(inode->i_mode)) {
1171                 /* these should have been cleared in ll_file_release */
1172                 LASSERT(lli->lli_opendir_key == NULL);
1173                 LASSERT(lli->lli_sai == NULL);
1174                 LASSERT(lli->lli_opendir_pid == 0);
1175         }
1176
1177         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1178         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1179                          null_if_equal, inode);
1180
1181         LASSERT(!lli->lli_open_fd_write_count);
1182         LASSERT(!lli->lli_open_fd_read_count);
1183         LASSERT(!lli->lli_open_fd_exec_count);
1184
1185         if (lli->lli_mds_write_och)
1186                 ll_md_real_close(inode, FMODE_WRITE);
1187         if (lli->lli_mds_exec_och)
1188                 ll_md_real_close(inode, FMODE_EXEC);
1189         if (lli->lli_mds_read_och)
1190                 ll_md_real_close(inode, FMODE_READ);
1191
1192         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1193                 OBD_FREE(lli->lli_symlink_name,
1194                          strlen(lli->lli_symlink_name) + 1);
1195                 lli->lli_symlink_name = NULL;
1196         }
1197
1198         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1199                 LASSERT(lli->lli_posix_acl == NULL);
1200                 if (lli->lli_remote_perms) {
1201                         free_rmtperm_hash(lli->lli_remote_perms);
1202                         lli->lli_remote_perms = NULL;
1203                 }
1204         }
1205 #ifdef CONFIG_FS_POSIX_ACL
1206         else if (lli->lli_posix_acl) {
1207                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1208                 LASSERT(lli->lli_remote_perms == NULL);
1209                 posix_acl_release(lli->lli_posix_acl);
1210                 lli->lli_posix_acl = NULL;
1211         }
1212 #endif
1213         lli->lli_inode_magic = LLI_INODE_DEAD;
1214
1215         ll_clear_inode_capas(inode);
1216         if (!S_ISDIR(inode->i_mode))
1217                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1218
1219         /*
1220          * XXX This has to be done before lsm is freed below, because
1221          * cl_object still uses inode lsm.
1222          */
1223         cl_inode_fini(inode);
1224         lli->lli_has_smd = false;
1225
1226         EXIT;
1227 }
1228
1229 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1230                   struct md_open_data **mod)
1231 {
1232         struct lustre_md md;
1233         struct inode *inode = dentry->d_inode;
1234         struct ll_sb_info *sbi = ll_i2sbi(inode);
1235         struct ptlrpc_request *request = NULL;
1236         int rc, ia_valid;
1237         ENTRY;
1238
1239         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1240                                      LUSTRE_OPC_ANY, NULL);
1241         if (IS_ERR(op_data))
1242                 RETURN(PTR_ERR(op_data));
1243
1244         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1245                         &request, mod);
1246         if (rc) {
1247                 ptlrpc_req_finished(request);
1248                 if (rc == -ENOENT) {
1249                         clear_nlink(inode);
1250                         /* Unlinked special device node? Or just a race?
1251                          * Pretend we done everything. */
1252                         if (!S_ISREG(inode->i_mode) &&
1253                             !S_ISDIR(inode->i_mode)) {
1254                                 ia_valid = op_data->op_attr.ia_valid;
1255                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1256                                 rc = simple_setattr(dentry, &op_data->op_attr);
1257                                 op_data->op_attr.ia_valid = ia_valid;
1258                         }
1259                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1260                         CERROR("md_setattr fails: rc = %d\n", rc);
1261                 }
1262                 RETURN(rc);
1263         }
1264
1265         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1266                               sbi->ll_md_exp, &md);
1267         if (rc) {
1268                 ptlrpc_req_finished(request);
1269                 RETURN(rc);
1270         }
1271
1272         /* We want to adjust timestamps.
1273          * If there is at least some data in file, we cleared ATTR_SIZE
1274          * to avoid update size, otherwise it is important to do.(SOM case)
1275          * (bug 6196) */
1276         ia_valid = op_data->op_attr.ia_valid;
1277         /* Since we set ATTR_*_SET flags above, and already done permission
1278          * check, So don't let inode_change_ok() check it again.  */
1279         op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1280         rc = simple_setattr(dentry, &op_data->op_attr);
1281         op_data->op_attr.ia_valid = ia_valid;
1282
1283         /* Extract epoch data if obtained. */
1284         op_data->op_handle = md.body->handle;
1285         op_data->op_ioepoch = md.body->ioepoch;
1286
1287         ll_update_inode(inode, &md);
1288         ptlrpc_req_finished(request);
1289
1290         RETURN(rc);
1291 }
1292
1293 /* Close IO epoch and send Size-on-MDS attribute update. */
1294 static int ll_setattr_done_writing(struct inode *inode,
1295                                    struct md_op_data *op_data,
1296                                    struct md_open_data *mod)
1297 {
1298         struct ll_inode_info *lli = ll_i2info(inode);
1299         int rc = 0;
1300         ENTRY;
1301
1302         LASSERT(op_data != NULL);
1303         if (!S_ISREG(inode->i_mode))
1304                 RETURN(0);
1305
1306         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1307                op_data->op_ioepoch, PFID(&lli->lli_fid));
1308
1309         op_data->op_flags = MF_EPOCH_CLOSE;
1310         ll_done_writing_attr(inode, op_data);
1311         ll_pack_inode2opdata(inode, op_data, NULL);
1312
1313         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1314         if (rc == -EAGAIN) {
1315                 /* MDS has instructed us to obtain Size-on-MDS attribute
1316                  * from OSTs and send setattr to back to MDS. */
1317                 rc = ll_som_update(inode, op_data);
1318         } else if (rc) {
1319                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1320                        inode->i_ino, rc);
1321         }
1322         RETURN(rc);
1323 }
1324
1325 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1326 {
1327         struct obd_capa *capa;
1328         int rc;
1329
1330         if (attr->ia_valid & ATTR_SIZE)
1331                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1332         else
1333                 capa = ll_mdscapa_get(inode);
1334
1335         rc = cl_setattr_ost(inode, attr, capa);
1336
1337         if (attr->ia_valid & ATTR_SIZE)
1338                 ll_truncate_free_capa(capa);
1339         else
1340                 capa_put(capa);
1341
1342         return rc;
1343 }
1344
1345 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1346 /**
1347  * inode_newsize_ok - may this inode be truncated to a given size
1348  * @inode:      the inode to be truncated
1349  * @offset:     the new size to assign to the inode
1350  * @Returns:    0 on success, -ve errno on failure
1351  *
1352  * inode_newsize_ok will check filesystem limits and ulimits to check that the
1353  * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1354  * when necessary. Caller must not proceed with inode size change if failure is
1355  * returned. @inode must be a file (not directory), with appropriate
1356  * permissions to allow truncate (inode_newsize_ok does NOT check these
1357  * conditions).
1358  *
1359  * inode_newsize_ok must be called with i_mutex held.
1360  */
1361 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1362 {
1363         if (inode->i_size < offset) {
1364                 unsigned long limit;
1365
1366                 limit = rlimit(RLIMIT_FSIZE);
1367                 if (limit != RLIM_INFINITY && offset > limit)
1368                         goto out_sig;
1369                 if (offset > inode->i_sb->s_maxbytes)
1370                         goto out_big;
1371         } else {
1372                 /*
1373                  * truncation of in-use swapfiles is disallowed - it would
1374                  * cause subsequent swapout to scribble on the now-freed
1375                  * blocks.
1376                  */
1377                 if (IS_SWAPFILE(inode))
1378                         return -ETXTBSY;
1379         }
1380
1381         return 0;
1382 out_sig:
1383         send_sig(SIGXFSZ, current, 0);
1384 out_big:
1385         return -EFBIG;
1386 }
1387 #endif
1388
1389 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1390  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1391  * keep these values until such a time that objects are allocated for it.
1392  * We do the MDS operations first, as it is checking permissions for us.
1393  * We don't to the MDS RPC if there is nothing that we want to store there,
1394  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1395  * going to do an RPC anyways.
1396  *
1397  * If we are doing a truncate, we will send the mtime and ctime updates
1398  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1399  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1400  * at the same time.
1401  */
1402 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1403 {
1404         struct lov_stripe_md *lsm;
1405         struct inode *inode = dentry->d_inode;
1406         struct ll_inode_info *lli = ll_i2info(inode);
1407         struct md_op_data *op_data = NULL;
1408         struct md_open_data *mod = NULL;
1409         int ia_valid = attr->ia_valid;
1410         int rc = 0, rc1 = 0;
1411         ENTRY;
1412
1413         CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1414                 "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
1415                 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1416                 attr->ia_valid);
1417
1418         if (ia_valid & ATTR_SIZE) {
1419                 /* Check new size against VFS/VM file size limit and rlimit */
1420                 rc = inode_newsize_ok(inode, attr->ia_size);
1421                 if (rc)
1422                         RETURN(rc);
1423
1424                 /* The maximum Lustre file size is variable, based on the
1425                  * OST maximum object size and number of stripes.  This
1426                  * needs another check in addition to the VFS check above. */
1427                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1428                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1429                                PFID(&lli->lli_fid), attr->ia_size,
1430                                ll_file_maxbytes(inode));
1431                         RETURN(-EFBIG);
1432                 }
1433
1434                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1435         }
1436
1437         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1438         if (ia_valid & TIMES_SET_FLAGS) {
1439                 if (cfs_curproc_fsuid() != inode->i_uid &&
1440                     !cfs_capable(CFS_CAP_FOWNER))
1441                         RETURN(-EPERM);
1442         }
1443
1444         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1445         if (attr->ia_valid & ATTR_CTIME) {
1446                 attr->ia_ctime = CFS_CURRENT_TIME;
1447                 attr->ia_valid |= ATTR_CTIME_SET;
1448         }
1449         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1450                 attr->ia_atime = CFS_CURRENT_TIME;
1451                 attr->ia_valid |= ATTR_ATIME_SET;
1452         }
1453         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1454                 attr->ia_mtime = CFS_CURRENT_TIME;
1455                 attr->ia_valid |= ATTR_MTIME_SET;
1456         }
1457
1458         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1459                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1460                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1461                        cfs_time_current_sec());
1462
1463         /* If we are changing file size, file content is modified, flag it. */
1464         if (attr->ia_valid & ATTR_SIZE) {
1465                 spin_lock(&lli->lli_lock);
1466                 lli->lli_flags |= LLIF_DATA_MODIFIED;
1467                 spin_unlock(&lli->lli_lock);
1468         }
1469
1470         /* We always do an MDS RPC, even if we're only changing the size;
1471          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1472
1473         OBD_ALLOC_PTR(op_data);
1474         if (op_data == NULL)
1475                 RETURN(-ENOMEM);
1476
1477         if (!S_ISDIR(inode->i_mode)) {
1478                 if (ia_valid & ATTR_SIZE)
1479                         inode_dio_write_done(inode);
1480                 mutex_unlock(&inode->i_mutex);
1481                 down_write(&lli->lli_trunc_sem);
1482         }
1483
1484         /* We need a steady stripe configuration for setattr to avoid
1485          * confusion. */
1486         lsm = ccc_inode_lsm_get(inode);
1487
1488         /* NB: ATTR_SIZE will only be set after this point if the size
1489          * resides on the MDS, ie, this file has no objects. */
1490         if (lsm != NULL)
1491                 attr->ia_valid &= ~ATTR_SIZE;
1492         /* can't call ll_setattr_ost() while holding a refcount of lsm */
1493         ccc_inode_lsm_put(inode, lsm);
1494
1495         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1496
1497         /* Open epoch for truncate. */
1498         if (exp_connect_som(ll_i2mdexp(inode)) &&
1499             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1500                 op_data->op_flags = MF_EPOCH_OPEN;
1501
1502         rc = ll_md_setattr(dentry, op_data, &mod);
1503         if (rc)
1504                 GOTO(out, rc);
1505
1506         /* RPC to MDT is sent, cancel data modification flag */
1507         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1508                 spin_lock(&lli->lli_lock);
1509                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1510                 spin_unlock(&lli->lli_lock);
1511         }
1512
1513         ll_ioepoch_open(lli, op_data->op_ioepoch);
1514         if (!S_ISREG(inode->i_mode))
1515                 GOTO(out, rc = 0);
1516
1517         if (ia_valid & ATTR_SIZE)
1518                 attr->ia_valid |= ATTR_SIZE;
1519         if (ia_valid & (ATTR_SIZE |
1520                         ATTR_ATIME | ATTR_ATIME_SET |
1521                         ATTR_MTIME | ATTR_MTIME_SET))
1522                 /* For truncate and utimes sending attributes to OSTs, setting
1523                  * mtime/atime to the past will be performed under PW [0:EOF]
1524                  * extent lock (new_size:EOF for truncate).  It may seem
1525                  * excessive to send mtime/atime updates to OSTs when not
1526                  * setting times to past, but it is necessary due to possible
1527                  * time de-synchronization between MDT inode and OST objects */
1528                 rc = ll_setattr_ost(inode, attr);
1529         EXIT;
1530 out:
1531         if (op_data) {
1532                 if (op_data->op_ioepoch) {
1533                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1534                         if (!rc)
1535                                 rc = rc1;
1536                 }
1537                 ll_finish_md_op_data(op_data);
1538         }
1539         if (!S_ISDIR(inode->i_mode)) {
1540                 up_write(&lli->lli_trunc_sem);
1541                 mutex_lock(&inode->i_mutex);
1542                 if (ia_valid & ATTR_SIZE)
1543                         inode_dio_wait(inode);
1544         }
1545
1546         ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1547                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1548
1549         return rc;
1550 }
1551
1552 int ll_setattr(struct dentry *de, struct iattr *attr)
1553 {
1554         int mode = de->d_inode->i_mode;
1555
1556         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1557                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1558                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1559
1560         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1561                                (ATTR_SIZE|ATTR_MODE)) &&
1562             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1563              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1564               !(attr->ia_mode & S_ISGID))))
1565                 attr->ia_valid |= ATTR_FORCE;
1566
1567         if ((mode & S_ISUID) &&
1568             !(attr->ia_mode & S_ISUID) &&
1569             !(attr->ia_valid & ATTR_KILL_SUID))
1570                 attr->ia_valid |= ATTR_KILL_SUID;
1571
1572         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1573             !(attr->ia_mode & S_ISGID) &&
1574             !(attr->ia_valid & ATTR_KILL_SGID))
1575                 attr->ia_valid |= ATTR_KILL_SGID;
1576
1577         return ll_setattr_raw(de, attr);
1578 }
1579
1580 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1581                        __u64 max_age, __u32 flags)
1582 {
1583         struct ll_sb_info *sbi = ll_s2sbi(sb);
1584         struct obd_statfs obd_osfs;
1585         int rc;
1586         ENTRY;
1587
1588         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1589         if (rc) {
1590                 CERROR("md_statfs fails: rc = %d\n", rc);
1591                 RETURN(rc);
1592         }
1593
1594         osfs->os_type = sb->s_magic;
1595
1596         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1597                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1598
1599         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1600                 flags |= OBD_STATFS_NODELAY;
1601
1602         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1603         if (rc) {
1604                 CERROR("obd_statfs fails: rc = %d\n", rc);
1605                 RETURN(rc);
1606         }
1607
1608         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1609                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1610                obd_osfs.os_files);
1611
1612         osfs->os_bsize = obd_osfs.os_bsize;
1613         osfs->os_blocks = obd_osfs.os_blocks;
1614         osfs->os_bfree = obd_osfs.os_bfree;
1615         osfs->os_bavail = obd_osfs.os_bavail;
1616
1617         /* If we don't have as many objects free on the OST as inodes
1618          * on the MDS, we reduce the total number of inodes to
1619          * compensate, so that the "inodes in use" number is correct.
1620          */
1621         if (obd_osfs.os_ffree < osfs->os_ffree) {
1622                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1623                         obd_osfs.os_ffree;
1624                 osfs->os_ffree = obd_osfs.os_ffree;
1625         }
1626
1627         RETURN(rc);
1628 }
1629 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1630 {
1631         struct super_block *sb = de->d_sb;
1632         struct obd_statfs osfs;
1633         int rc;
1634
1635         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1636         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1637
1638         /* Some amount of caching on the client is allowed */
1639         rc = ll_statfs_internal(sb, &osfs,
1640                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1641                                 0);
1642         if (rc)
1643                 return rc;
1644
1645         statfs_unpack(sfs, &osfs);
1646
1647         /* We need to downshift for all 32-bit kernels, because we can't
1648          * tell if the kernel is being called via sys_statfs64() or not.
1649          * Stop before overflowing f_bsize - in which case it is better
1650          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1651         if (sizeof(long) < 8) {
1652                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1653                         sfs->f_bsize <<= 1;
1654
1655                         osfs.os_blocks >>= 1;
1656                         osfs.os_bfree >>= 1;
1657                         osfs.os_bavail >>= 1;
1658                 }
1659         }
1660
1661         sfs->f_blocks = osfs.os_blocks;
1662         sfs->f_bfree = osfs.os_bfree;
1663         sfs->f_bavail = osfs.os_bavail;
1664
1665         return 0;
1666 }
1667
1668 void ll_inode_size_lock(struct inode *inode)
1669 {
1670         struct ll_inode_info *lli;
1671
1672         LASSERT(!S_ISDIR(inode->i_mode));
1673
1674         lli = ll_i2info(inode);
1675         LASSERT(lli->lli_size_sem_owner != current);
1676         down(&lli->lli_size_sem);
1677         LASSERT(lli->lli_size_sem_owner == NULL);
1678         lli->lli_size_sem_owner = current;
1679 }
1680
1681 void ll_inode_size_unlock(struct inode *inode)
1682 {
1683         struct ll_inode_info *lli;
1684
1685         lli = ll_i2info(inode);
1686         LASSERT(lli->lli_size_sem_owner == current);
1687         lli->lli_size_sem_owner = NULL;
1688         up(&lli->lli_size_sem);
1689 }
1690
1691 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1692 {
1693         struct ll_inode_info *lli = ll_i2info(inode);
1694         struct mdt_body *body = md->body;
1695         struct lov_stripe_md *lsm = md->lsm;
1696         struct ll_sb_info *sbi = ll_i2sbi(inode);
1697
1698         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1699         if (lsm != NULL) {
1700                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1701                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1702                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1703         }
1704
1705         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1706                 if (body->valid & OBD_MD_FLRMTPERM)
1707                         ll_update_remote_perm(inode, md->remote_perm);
1708         }
1709 #ifdef CONFIG_FS_POSIX_ACL
1710         else if (body->valid & OBD_MD_FLACL) {
1711                 spin_lock(&lli->lli_lock);
1712                 if (lli->lli_posix_acl)
1713                         posix_acl_release(lli->lli_posix_acl);
1714                 lli->lli_posix_acl = md->posix_acl;
1715                 spin_unlock(&lli->lli_lock);
1716         }
1717 #endif
1718         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1719         inode->i_generation = cl_fid_build_gen(&body->fid1);
1720
1721         if (body->valid & OBD_MD_FLATIME) {
1722                 if (body->atime > LTIME_S(inode->i_atime))
1723                         LTIME_S(inode->i_atime) = body->atime;
1724                 lli->lli_lvb.lvb_atime = body->atime;
1725         }
1726         if (body->valid & OBD_MD_FLMTIME) {
1727                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1728                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1729                                "to "LPU64"\n", inode->i_ino,
1730                                LTIME_S(inode->i_mtime), body->mtime);
1731                         LTIME_S(inode->i_mtime) = body->mtime;
1732                 }
1733                 lli->lli_lvb.lvb_mtime = body->mtime;
1734         }
1735         if (body->valid & OBD_MD_FLCTIME) {
1736                 if (body->ctime > LTIME_S(inode->i_ctime))
1737                         LTIME_S(inode->i_ctime) = body->ctime;
1738                 lli->lli_lvb.lvb_ctime = body->ctime;
1739         }
1740         if (body->valid & OBD_MD_FLMODE)
1741                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1742         if (body->valid & OBD_MD_FLTYPE)
1743                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1744         LASSERT(inode->i_mode != 0);
1745         if (S_ISREG(inode->i_mode)) {
1746                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1747         } else {
1748                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1749         }
1750         if (body->valid & OBD_MD_FLUID)
1751                 inode->i_uid = body->uid;
1752         if (body->valid & OBD_MD_FLGID)
1753                 inode->i_gid = body->gid;
1754         if (body->valid & OBD_MD_FLFLAGS)
1755                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1756         if (body->valid & OBD_MD_FLNLINK)
1757                 set_nlink(inode, body->nlink);
1758         if (body->valid & OBD_MD_FLRDEV)
1759                 inode->i_rdev = old_decode_dev(body->rdev);
1760
1761         if (body->valid & OBD_MD_FLID) {
1762                 /* FID shouldn't be changed! */
1763                 if (fid_is_sane(&lli->lli_fid)) {
1764                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1765                                  "Trying to change FID "DFID
1766                                  " to the "DFID", inode %lu/%u(%p)\n",
1767                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1768                                  inode->i_ino, inode->i_generation, inode);
1769                 } else
1770                         lli->lli_fid = body->fid1;
1771         }
1772
1773         LASSERT(fid_seq(&lli->lli_fid) != 0);
1774
1775         if (body->valid & OBD_MD_FLSIZE) {
1776                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1777                     S_ISREG(inode->i_mode) && lli->lli_has_smd) {
1778                         struct lustre_handle lockh;
1779                         ldlm_mode_t mode;
1780
1781                         /* As it is possible a blocking ast has been processed
1782                          * by this time, we need to check there is an UPDATE
1783                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1784                          * it. */
1785                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1786                                                &lockh, LDLM_FL_CBPENDING);
1787                         if (mode) {
1788                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1789                                                       LLIF_EPOCH_PENDING |
1790                                                       LLIF_SOM_DIRTY)) {
1791                                         CERROR("ino %lu flags %u still has "
1792                                                "size authority! do not trust "
1793                                                "the size got from MDS\n",
1794                                                inode->i_ino, lli->lli_flags);
1795                                 } else {
1796                                         /* Use old size assignment to avoid
1797                                          * deadlock bz14138 & bz14326 */
1798                                         i_size_write(inode, body->size);
1799                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1800                                 }
1801                                 ldlm_lock_decref(&lockh, mode);
1802                         }
1803                 } else {
1804                         /* Use old size assignment to avoid
1805                          * deadlock bz14138 & bz14326 */
1806                         i_size_write(inode, body->size);
1807
1808                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1809                                inode->i_ino, (unsigned long long)body->size);
1810                 }
1811
1812                 if (body->valid & OBD_MD_FLBLOCKS)
1813                         inode->i_blocks = body->blocks;
1814         }
1815
1816         if (body->valid & OBD_MD_FLMDSCAPA) {
1817                 LASSERT(md->mds_capa);
1818                 ll_add_capa(inode, md->mds_capa);
1819         }
1820         if (body->valid & OBD_MD_FLOSSCAPA) {
1821                 LASSERT(md->oss_capa);
1822                 ll_add_capa(inode, md->oss_capa);
1823         }
1824 }
1825
1826 void ll_read_inode2(struct inode *inode, void *opaque)
1827 {
1828         struct lustre_md *md = opaque;
1829         struct ll_inode_info *lli = ll_i2info(inode);
1830         ENTRY;
1831
1832         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1833                PFID(&lli->lli_fid), inode);
1834
1835         LASSERT(!lli->lli_has_smd);
1836
1837         /* Core attributes from the MDS first.  This is a new inode, and
1838          * the VFS doesn't zero times in the core inode so we have to do
1839          * it ourselves.  They will be overwritten by either MDS or OST
1840          * attributes - we just need to make sure they aren't newer. */
1841         LTIME_S(inode->i_mtime) = 0;
1842         LTIME_S(inode->i_atime) = 0;
1843         LTIME_S(inode->i_ctime) = 0;
1844         inode->i_rdev = 0;
1845         ll_update_inode(inode, md);
1846
1847         /* OIDEBUG(inode); */
1848
1849         /* initializing backing dev info. */
1850         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1851
1852
1853         if (S_ISREG(inode->i_mode)) {
1854                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1855                 inode->i_op = &ll_file_inode_operations;
1856                 inode->i_fop = sbi->ll_fop;
1857                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1858                 EXIT;
1859         } else if (S_ISDIR(inode->i_mode)) {
1860                 inode->i_op = &ll_dir_inode_operations;
1861                 inode->i_fop = &ll_dir_operations;
1862                 EXIT;
1863         } else if (S_ISLNK(inode->i_mode)) {
1864                 inode->i_op = &ll_fast_symlink_inode_operations;
1865                 EXIT;
1866         } else {
1867                 inode->i_op = &ll_special_inode_operations;
1868
1869                 init_special_inode(inode, inode->i_mode,
1870                                    inode->i_rdev);
1871
1872                 EXIT;
1873         }
1874 }
1875
1876 void ll_delete_inode(struct inode *inode)
1877 {
1878         struct cl_inode_info *lli = cl_i2info(inode);
1879         ENTRY;
1880
1881         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1882                 /* discard all dirty pages before truncating them, required by
1883                  * osc_extent implementation at LU-1030. */
1884                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1885
1886         truncate_inode_pages(&inode->i_data, 0);
1887
1888         /* Workaround for LU-118 */
1889         if (inode->i_data.nrpages) {
1890                 TREE_READ_LOCK_IRQ(&inode->i_data);
1891                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1892                 LASSERTF(inode->i_data.nrpages == 0,
1893                          "inode=%lu/%u(%p) nrpages=%lu, see "
1894                          "http://jira.whamcloud.com/browse/LU-118\n",
1895                          inode->i_ino, inode->i_generation, inode,
1896                          inode->i_data.nrpages);
1897         }
1898         /* Workaround end */
1899
1900 #ifdef HAVE_SBOPS_EVICT_INODE
1901         ll_clear_inode(inode);
1902 #endif
1903         clear_inode(inode);
1904
1905         EXIT;
1906 }
1907
1908 int ll_iocontrol(struct inode *inode, struct file *file,
1909                  unsigned int cmd, unsigned long arg)
1910 {
1911         struct ll_sb_info *sbi = ll_i2sbi(inode);
1912         struct ptlrpc_request *req = NULL;
1913         int rc, flags = 0;
1914         ENTRY;
1915
1916         switch(cmd) {
1917         case FSFILT_IOC_GETFLAGS: {
1918                 struct mdt_body *body;
1919                 struct md_op_data *op_data;
1920
1921                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1922                                              0, 0, LUSTRE_OPC_ANY,
1923                                              NULL);
1924                 if (IS_ERR(op_data))
1925                         RETURN(PTR_ERR(op_data));
1926
1927                 op_data->op_valid = OBD_MD_FLFLAGS;
1928                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1929                 ll_finish_md_op_data(op_data);
1930                 if (rc) {
1931                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1932                         RETURN(-abs(rc));
1933                 }
1934
1935                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1936
1937                 flags = body->flags;
1938
1939                 ptlrpc_req_finished(req);
1940
1941                 RETURN(put_user(flags, (int *)arg));
1942         }
1943         case FSFILT_IOC_SETFLAGS: {
1944                 struct lov_stripe_md *lsm;
1945                 struct obd_info oinfo = { { { 0 } } };
1946                 struct md_op_data *op_data;
1947
1948                 if (get_user(flags, (int *)arg))
1949                         RETURN(-EFAULT);
1950
1951                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1952                                              LUSTRE_OPC_ANY, NULL);
1953                 if (IS_ERR(op_data))
1954                         RETURN(PTR_ERR(op_data));
1955
1956                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1957                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1958                 rc = md_setattr(sbi->ll_md_exp, op_data,
1959                                 NULL, 0, NULL, 0, &req, NULL);
1960                 ll_finish_md_op_data(op_data);
1961                 ptlrpc_req_finished(req);
1962                 if (rc)
1963                         RETURN(rc);
1964
1965                 inode->i_flags = ll_ext_to_inode_flags(flags);
1966
1967                 lsm = ccc_inode_lsm_get(inode);
1968                 if (lsm == NULL)
1969                         RETURN(0);
1970
1971                 OBDO_ALLOC(oinfo.oi_oa);
1972                 if (!oinfo.oi_oa) {
1973                         ccc_inode_lsm_put(inode, lsm);
1974                         RETURN(-ENOMEM);
1975                 }
1976                 oinfo.oi_md = lsm;
1977                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1978                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1979                 oinfo.oi_oa->o_flags = flags;
1980                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1981                                        OBD_MD_FLGROUP;
1982                 oinfo.oi_capa = ll_mdscapa_get(inode);
1983                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1984                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1985                 capa_put(oinfo.oi_capa);
1986                 OBDO_FREE(oinfo.oi_oa);
1987                 ccc_inode_lsm_put(inode, lsm);
1988
1989                 if (rc && rc != -EPERM && rc != -EACCES)
1990                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
1991
1992                 RETURN(rc);
1993         }
1994         default:
1995                 RETURN(-ENOSYS);
1996         }
1997
1998         RETURN(0);
1999 }
2000
2001 int ll_flush_ctx(struct inode *inode)
2002 {
2003         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2004
2005         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2006
2007         obd_set_info_async(NULL, sbi->ll_md_exp,
2008                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2009                            0, NULL, NULL);
2010         obd_set_info_async(NULL, sbi->ll_dt_exp,
2011                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2012                            0, NULL, NULL);
2013         return 0;
2014 }
2015
2016 /* umount -f client means force down, don't save state */
2017 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2018 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2019 {
2020         struct super_block *sb = vfsmnt->mnt_sb;
2021 #else
2022 void ll_umount_begin(struct super_block *sb)
2023 {
2024 #endif
2025         struct ll_sb_info *sbi = ll_s2sbi(sb);
2026         struct obd_device *obd;
2027         struct obd_ioctl_data *ioc_data;
2028         ENTRY;
2029
2030 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2031         if (!(flags & MNT_FORCE)) {
2032                 EXIT;
2033                 return;
2034         }
2035 #endif
2036
2037         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2038                sb->s_count, atomic_read(&sb->s_active));
2039
2040         obd = class_exp2obd(sbi->ll_md_exp);
2041         if (obd == NULL) {
2042                 CERROR("Invalid MDC connection handle "LPX64"\n",
2043                        sbi->ll_md_exp->exp_handle.h_cookie);
2044                 EXIT;
2045                 return;
2046         }
2047         obd->obd_force = 1;
2048
2049         obd = class_exp2obd(sbi->ll_dt_exp);
2050         if (obd == NULL) {
2051                 CERROR("Invalid LOV connection handle "LPX64"\n",
2052                        sbi->ll_dt_exp->exp_handle.h_cookie);
2053                 EXIT;
2054                 return;
2055         }
2056         obd->obd_force = 1;
2057
2058         OBD_ALLOC_PTR(ioc_data);
2059         if (ioc_data) {
2060                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2061                               sizeof ioc_data, ioc_data, NULL);
2062
2063                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2064                               sizeof ioc_data, ioc_data, NULL);
2065
2066                 OBD_FREE_PTR(ioc_data);
2067         }
2068
2069
2070         /* Really, we'd like to wait until there are no requests outstanding,
2071          * and then continue.  For now, we just invalidate the requests,
2072          * schedule() and sleep one second if needed, and hope.
2073          */
2074         cfs_schedule();
2075 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2076         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2077                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2078                                                    cfs_time_seconds(1));
2079                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2080                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2081                                       "may try to umount it a bit later\n",
2082                                       atomic_read(&vfsmnt->mnt_count));
2083         }
2084 #endif
2085
2086         EXIT;
2087 }
2088
2089 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2090 {
2091         struct ll_sb_info *sbi = ll_s2sbi(sb);
2092         char *profilenm = get_profile_name(sb);
2093         int err;
2094         __u32 read_only;
2095
2096         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2097                 read_only = *flags & MS_RDONLY;
2098                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2099                                          sizeof(KEY_READ_ONLY),
2100                                          KEY_READ_ONLY, sizeof(read_only),
2101                                          &read_only, NULL);
2102                 if (err) {
2103                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2104                                       profilenm, read_only ?
2105                                       "read-only" : "read-write", err);
2106                         return err;
2107                 }
2108
2109                 if (read_only)
2110                         sb->s_flags |= MS_RDONLY;
2111                 else
2112                         sb->s_flags &= ~MS_RDONLY;
2113
2114                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2115                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2116                                       read_only ?  "read-only" : "read-write");
2117         }
2118         return 0;
2119 }
2120
2121 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2122                   struct super_block *sb, struct lookup_intent *it)
2123 {
2124         struct ll_sb_info *sbi = NULL;
2125         struct lustre_md md;
2126         int rc;
2127         ENTRY;
2128
2129         LASSERT(*inode || sb);
2130         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2131         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2132                               sbi->ll_md_exp, &md);
2133         if (rc)
2134                 RETURN(rc);
2135
2136         if (*inode) {
2137                 ll_update_inode(*inode, &md);
2138         } else {
2139                 LASSERT(sb != NULL);
2140
2141                 /*
2142                  * At this point server returns to client's same fid as client
2143                  * generated for creating. So using ->fid1 is okay here.
2144                  */
2145                 LASSERT(fid_is_sane(&md.body->fid1));
2146
2147                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2148                 if (*inode == NULL || IS_ERR(*inode)) {
2149 #ifdef CONFIG_FS_POSIX_ACL
2150                         if (md.posix_acl) {
2151                                 posix_acl_release(md.posix_acl);
2152                                 md.posix_acl = NULL;
2153                         }
2154 #endif
2155                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2156                         *inode = NULL;
2157                         CERROR("new_inode -fatal: rc %d\n", rc);
2158                         GOTO(out, rc);
2159                 }
2160         }
2161
2162         /* Handling piggyback layout lock.
2163          * Layout lock can be piggybacked by getattr and open request.
2164          * The lsm can be applied to inode only if it comes with a layout lock
2165          * otherwise correct layout may be overwritten, for example:
2166          * 1. proc1: mdt returns a lsm but not granting layout
2167          * 2. layout was changed by another client
2168          * 3. proc2: refresh layout and layout lock granted
2169          * 4. proc1: to apply a stale layout */
2170         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2171                 struct lustre_handle lockh;
2172                 struct ldlm_lock *lock;
2173
2174                 lockh.cookie = it->d.lustre.it_lock_handle;
2175                 lock = ldlm_handle2lock(&lockh);
2176                 LASSERT(lock != NULL);
2177                 if (ldlm_has_layout(lock)) {
2178                         struct cl_object_conf conf;
2179
2180                         memset(&conf, 0, sizeof(conf));
2181                         conf.coc_opc = OBJECT_CONF_SET;
2182                         conf.coc_inode = *inode;
2183                         conf.coc_lock = lock;
2184                         conf.u.coc_md = &md;
2185                         (void)ll_layout_conf(*inode, &conf);
2186                 }
2187                 LDLM_LOCK_PUT(lock);
2188         }
2189
2190 out:
2191         if (md.lsm != NULL)
2192                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2193         md_free_lustre_md(sbi->ll_md_exp, &md);
2194         RETURN(rc);
2195 }
2196
2197 int ll_obd_statfs(struct inode *inode, void *arg)
2198 {
2199         struct ll_sb_info *sbi = NULL;
2200         struct obd_export *exp;
2201         char *buf = NULL;
2202         struct obd_ioctl_data *data = NULL;
2203         __u32 type;
2204         __u32 flags;
2205         int len = 0, rc;
2206
2207         if (!inode || !(sbi = ll_i2sbi(inode)))
2208                 GOTO(out_statfs, rc = -EINVAL);
2209
2210         rc = obd_ioctl_getdata(&buf, &len, arg);
2211         if (rc)
2212                 GOTO(out_statfs, rc);
2213
2214         data = (void*)buf;
2215         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2216             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2217                 GOTO(out_statfs, rc = -EINVAL);
2218
2219         if (data->ioc_inllen1 != sizeof(__u32) ||
2220             data->ioc_inllen2 != sizeof(__u32) ||
2221             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2222             data->ioc_plen2 != sizeof(struct obd_uuid))
2223                 GOTO(out_statfs, rc = -EINVAL);
2224
2225         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2226         if (type & LL_STATFS_LMV)
2227                 exp = sbi->ll_md_exp;
2228         else if (type & LL_STATFS_LOV)
2229                 exp = sbi->ll_dt_exp;
2230         else
2231                 GOTO(out_statfs, rc = -ENODEV);
2232
2233         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2234         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2235         if (rc)
2236                 GOTO(out_statfs, rc);
2237 out_statfs:
2238         if (buf)
2239                 obd_ioctl_freedata(buf, len);
2240         return rc;
2241 }
2242
2243 int ll_process_config(struct lustre_cfg *lcfg)
2244 {
2245         char *ptr;
2246         void *sb;
2247         struct lprocfs_static_vars lvars;
2248         unsigned long x;
2249         int rc = 0;
2250
2251         lprocfs_llite_init_vars(&lvars);
2252
2253         /* The instance name contains the sb: lustre-client-aacfe000 */
2254         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2255         if (!ptr || !*(++ptr))
2256                 return -EINVAL;
2257         if (sscanf(ptr, "%lx", &x) != 1)
2258                 return -EINVAL;
2259         sb = (void *)x;
2260         /* This better be a real Lustre superblock! */
2261         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2262
2263         /* Note we have not called client_common_fill_super yet, so
2264            proc fns must be able to handle that! */
2265         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2266                                       lcfg, sb);
2267         if (rc > 0)
2268                 rc = 0;
2269         return(rc);
2270 }
2271
2272 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2273 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2274                                        struct inode *i1, struct inode *i2,
2275                                        const char *name, int namelen,
2276                                        int mode, __u32 opc, void *data)
2277 {
2278         LASSERT(i1 != NULL);
2279
2280         if (namelen > ll_i2sbi(i1)->ll_namelen)
2281                 return ERR_PTR(-ENAMETOOLONG);
2282
2283         if (op_data == NULL)
2284                 OBD_ALLOC_PTR(op_data);
2285
2286         if (op_data == NULL)
2287                 return ERR_PTR(-ENOMEM);
2288
2289         ll_i2gids(op_data->op_suppgids, i1, i2);
2290         op_data->op_fid1 = *ll_inode2fid(i1);
2291         op_data->op_capa1 = ll_mdscapa_get(i1);
2292
2293         if (i2) {
2294                 op_data->op_fid2 = *ll_inode2fid(i2);
2295                 op_data->op_capa2 = ll_mdscapa_get(i2);
2296         } else {
2297                 fid_zero(&op_data->op_fid2);
2298                 op_data->op_capa2 = NULL;
2299         }
2300
2301         op_data->op_name = name;
2302         op_data->op_namelen = namelen;
2303         op_data->op_mode = mode;
2304         op_data->op_mod_time = cfs_time_current_sec();
2305         op_data->op_fsuid = cfs_curproc_fsuid();
2306         op_data->op_fsgid = cfs_curproc_fsgid();
2307         op_data->op_cap = cfs_curproc_cap_pack();
2308         op_data->op_bias = 0;
2309         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2310              filename_is_volatile(name, namelen, NULL))
2311                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2312         op_data->op_opc = opc;
2313         op_data->op_mds = 0;
2314         op_data->op_data = data;
2315
2316         /* If the file is being opened after mknod() (normally due to NFS)
2317          * try to use the default stripe data from parent directory for
2318          * allocating OST objects.  Try to pass the parent FID to MDS. */
2319         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2320             !ll_i2info(i2)->lli_has_smd) {
2321                 struct ll_inode_info *lli = ll_i2info(i2);
2322
2323                 spin_lock(&lli->lli_lock);
2324                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2325                         op_data->op_fid1 = lli->lli_pfid;
2326                 spin_unlock(&lli->lli_lock);
2327                 /** We ignore parent's capability temporary. */
2328         }
2329
2330         /* When called by ll_setattr_raw, file is i1. */
2331         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2332                 op_data->op_bias |= MDS_DATA_MODIFIED;
2333
2334         return op_data;
2335 }
2336
2337 void ll_finish_md_op_data(struct md_op_data *op_data)
2338 {
2339         capa_put(op_data->op_capa1);
2340         capa_put(op_data->op_capa2);
2341         OBD_FREE_PTR(op_data);
2342 }
2343
2344 #ifdef HAVE_SUPEROPS_USE_DENTRY
2345 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2346 #else
2347 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2348 #endif
2349 {
2350         struct ll_sb_info *sbi;
2351
2352 #ifdef HAVE_SUPEROPS_USE_DENTRY
2353         LASSERT((seq != NULL) && (dentry != NULL));
2354         sbi = ll_s2sbi(dentry->d_sb);
2355 #else
2356         LASSERT((seq != NULL) && (vfs != NULL));
2357         sbi = ll_s2sbi(vfs->mnt_sb);
2358 #endif
2359
2360         if (sbi->ll_flags & LL_SBI_NOLCK)
2361                 seq_puts(seq, ",nolock");
2362
2363         if (sbi->ll_flags & LL_SBI_FLOCK)
2364                 seq_puts(seq, ",flock");
2365
2366         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2367                 seq_puts(seq, ",localflock");
2368
2369         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2370                 seq_puts(seq, ",user_xattr");
2371
2372         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2373                 seq_puts(seq, ",lazystatfs");
2374
2375         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2376                 seq_puts(seq, ",user_fid2path");
2377
2378         RETURN(0);
2379 }
2380
2381 /**
2382  * Get obd name by cmd, and copy out to user space
2383  */
2384 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2385 {
2386         struct ll_sb_info *sbi = ll_i2sbi(inode);
2387         struct obd_device *obd;
2388         ENTRY;
2389
2390         if (cmd == OBD_IOC_GETDTNAME)
2391                 obd = class_exp2obd(sbi->ll_dt_exp);
2392         else if (cmd == OBD_IOC_GETMDNAME)
2393                 obd = class_exp2obd(sbi->ll_md_exp);
2394         else
2395                 RETURN(-EINVAL);
2396
2397         if (!obd)
2398                 RETURN(-ENOENT);
2399
2400         if (cfs_copy_to_user((void *)arg, obd->obd_name,
2401                              strlen(obd->obd_name) + 1))
2402                 RETURN(-EFAULT);
2403
2404         RETURN(0);
2405 }
2406
2407 /**
2408  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2409  * fsname will be returned in this buffer; otherwise, a static buffer will be
2410  * used to store the fsname and returned to caller.
2411  */
2412 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2413 {
2414         static char fsname_static[MTI_NAME_MAXLEN];
2415         struct lustre_sb_info *lsi = s2lsi(sb);
2416         char *ptr;
2417         int len;
2418
2419         if (buf == NULL) {
2420                 /* this means the caller wants to use static buffer
2421                  * and it doesn't care about race. Usually this is
2422                  * in error reporting path */
2423                 buf = fsname_static;
2424                 buflen = sizeof(fsname_static);
2425         }
2426
2427         len = strlen(lsi->lsi_lmd->lmd_profile);
2428         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2429         if (ptr && (strcmp(ptr, "-client") == 0))
2430                 len -= 7;
2431
2432         if (unlikely(len >= buflen))
2433                 len = buflen - 1;
2434         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2435         buf[len] = '\0';
2436
2437         return buf;
2438 }
2439
2440 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2441 {
2442         char *path = NULL;
2443
2444 #ifdef HAVE_FS_STRUCT_USE_PATH
2445         struct path p;
2446
2447         p.dentry = dentry;
2448         p.mnt = current->fs->root.mnt;
2449         path_get(&p);
2450         path = d_path(&p, buf, bufsize);
2451         path_put(&p);
2452 #else
2453         path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2454 #endif
2455
2456         return path;
2457 }
2458
2459 void ll_dirty_page_discard_warn(cfs_page_t *page, int ioret)
2460 {
2461         char *buf, *path = NULL;
2462         struct dentry *dentry = NULL;
2463         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2464
2465         /* this can be called inside spin lock so use GFP_ATOMIC. */
2466         buf = (char *)__get_free_page(GFP_ATOMIC);
2467         if (buf != NULL) {
2468                 dentry = d_find_alias(page->mapping->host);
2469                 if (dentry != NULL)
2470                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2471         }
2472
2473         CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2474               "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2475               s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2476               PFID(&obj->cob_header.coh_lu.loh_fid),
2477               (path && !IS_ERR(path)) ? path : "", ioret);
2478
2479         if (dentry != NULL)
2480                 dput(dentry);
2481
2482         if (buf != NULL)
2483                 free_page((unsigned long)buf);
2484 }